linux/mm/page-writeback.c

3238 lines
96 KiB
C
Raw Permalink Normal View History

// SPDX-License-Identifier: GPL-2.0-only
/*
* mm/page-writeback.c
*
* Copyright (C) 2002, Linus Torvalds.
* Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
*
* Contains functions related to writing back dirty pages at the
* address_space level.
*
* 10Apr2002 Andrew Morton
* Initial version
*/
#include <linux/kernel.h>
#include <linux/math64.h>
#include <linux/export.h>
#include <linux/spinlock.h>
#include <linux/fs.h>
#include <linux/mm.h>
#include <linux/swap.h>
#include <linux/slab.h>
#include <linux/pagemap.h>
#include <linux/writeback.h>
#include <linux/init.h>
#include <linux/backing-dev.h>
#include <linux/task_io_accounting_ops.h>
#include <linux/blkdev.h>
#include <linux/mpage.h>
[PATCH] mm: tracking shared dirty pages Tracking of dirty pages in shared writeable mmap()s. The idea is simple: write protect clean shared writeable pages, catch the write-fault, make writeable and set dirty. On page write-back clean all the PTE dirty bits and write protect them once again. The implementation is a tad harder, mainly because the default backing_dev_info capabilities were too loosely maintained. Hence it is not enough to test the backing_dev_info for cap_account_dirty. The current heuristic is as follows, a VMA is eligible when: - its shared writeable (vm_flags & (VM_WRITE|VM_SHARED)) == (VM_WRITE|VM_SHARED) - it is not a 'special' mapping (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) == 0 - the backing_dev_info is cap_account_dirty mapping_cap_account_dirty(vma->vm_file->f_mapping) - f_op->mmap() didn't change the default page protection Page from remap_pfn_range() are explicitly excluded because their COW semantics are already horrid enough (see vm_normal_page() in do_wp_page()) and because they don't have a backing store anyway. mprotect() is taught about the new behaviour as well. However it overrides the last condition. Cleaning the pages on write-back is done with page_mkclean() a new rmap call. It can be called on any page, but is currently only implemented for mapped pages, if the page is found the be of a VMA that accounts dirty pages it will also wrprotect the PTE. Finally, in fs/buffers.c:try_to_free_buffers(); remove clear_page_dirty() from under ->private_lock. This seems to be safe, since ->private_lock is used to serialize access to the buffers, not the page itself. This is needed because clear_page_dirty() will call into page_mkclean() and would thereby violate locking order. [dhowells@redhat.com: Provide a page_mkclean() implementation for NOMMU] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-26 06:30:57 +00:00
#include <linux/rmap.h>
#include <linux/percpu.h>
#include <linux/smp.h>
#include <linux/sysctl.h>
#include <linux/cpu.h>
#include <linux/syscalls.h>
#include <linux/pagevec.h>
#include <linux/timer.h>
#include <linux/sched/rt.h>
#include <linux/sched/signal.h>
mm: vmscan: fix do_try_to_free_pages() livelock This patch is based on KOSAKI's work and I add a little more description, please refer https://lkml.org/lkml/2012/6/14/74. Currently, I found system can enter a state that there are lots of free pages in a zone but only order-0 and order-1 pages which means the zone is heavily fragmented, then high order allocation could make direct reclaim path's long stall(ex, 60 seconds) especially in no swap and no compaciton enviroment. This problem happened on v3.4, but it seems issue still lives in current tree, the reason is do_try_to_free_pages enter live lock: kswapd will go to sleep if the zones have been fully scanned and are still not balanced. As kswapd thinks there's little point trying all over again to avoid infinite loop. Instead it changes order from high-order to 0-order because kswapd think order-0 is the most important. Look at 73ce02e9 in detail. If watermarks are ok, kswapd will go back to sleep and may leave zone->all_unreclaimable =3D 0. It assume high-order users can still perform direct reclaim if they wish. Direct reclaim continue to reclaim for a high order which is not a COSTLY_ORDER without oom-killer until kswapd turn on zone->all_unreclaimble= . This is because to avoid too early oom-kill. So it means direct_reclaim depends on kswapd to break this loop. In worst case, direct-reclaim may continue to page reclaim forever when kswapd sleeps forever until someone like watchdog detect and finally kill the process. As described in: http://thread.gmane.org/gmane.linux.kernel.mm/103737 We can't turn on zone->all_unreclaimable from direct reclaim path because direct reclaim path don't take any lock and this way is racy. Thus this patch removes zone->all_unreclaimable field completely and recalculates zone reclaimable state every time. Note: we can't take the idea that direct-reclaim see zone->pages_scanned directly and kswapd continue to use zone->all_unreclaimable. Because, it is racy. commit 929bea7c71 (vmscan: all_unreclaimable() use zone->all_unreclaimable as a name) describes the detail. [akpm@linux-foundation.org: uninline zone_reclaimable_pages() and zone_reclaimable()] Cc: Aaditya Kumar <aaditya.kumar.30@gmail.com> Cc: Ying Han <yinghan@google.com> Cc: Nick Piggin <npiggin@gmail.com> Acked-by: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Lameter <cl@linux.com> Cc: Bob Liu <lliubbo@gmail.com> Cc: Neil Zhang <zhangwm@marvell.com> Cc: Russell King - ARM Linux <linux@arm.linux.org.uk> Reviewed-by: Michal Hocko <mhocko@suse.cz> Acked-by: Minchan Kim <minchan@kernel.org> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Lisa Du <cldu@marvell.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:36 +00:00
#include <linux/mm_inline.h>
#include <trace/events/writeback.h>
mm: vmscan: fix do_try_to_free_pages() livelock This patch is based on KOSAKI's work and I add a little more description, please refer https://lkml.org/lkml/2012/6/14/74. Currently, I found system can enter a state that there are lots of free pages in a zone but only order-0 and order-1 pages which means the zone is heavily fragmented, then high order allocation could make direct reclaim path's long stall(ex, 60 seconds) especially in no swap and no compaciton enviroment. This problem happened on v3.4, but it seems issue still lives in current tree, the reason is do_try_to_free_pages enter live lock: kswapd will go to sleep if the zones have been fully scanned and are still not balanced. As kswapd thinks there's little point trying all over again to avoid infinite loop. Instead it changes order from high-order to 0-order because kswapd think order-0 is the most important. Look at 73ce02e9 in detail. If watermarks are ok, kswapd will go back to sleep and may leave zone->all_unreclaimable =3D 0. It assume high-order users can still perform direct reclaim if they wish. Direct reclaim continue to reclaim for a high order which is not a COSTLY_ORDER without oom-killer until kswapd turn on zone->all_unreclaimble= . This is because to avoid too early oom-kill. So it means direct_reclaim depends on kswapd to break this loop. In worst case, direct-reclaim may continue to page reclaim forever when kswapd sleeps forever until someone like watchdog detect and finally kill the process. As described in: http://thread.gmane.org/gmane.linux.kernel.mm/103737 We can't turn on zone->all_unreclaimable from direct reclaim path because direct reclaim path don't take any lock and this way is racy. Thus this patch removes zone->all_unreclaimable field completely and recalculates zone reclaimable state every time. Note: we can't take the idea that direct-reclaim see zone->pages_scanned directly and kswapd continue to use zone->all_unreclaimable. Because, it is racy. commit 929bea7c71 (vmscan: all_unreclaimable() use zone->all_unreclaimable as a name) describes the detail. [akpm@linux-foundation.org: uninline zone_reclaimable_pages() and zone_reclaimable()] Cc: Aaditya Kumar <aaditya.kumar.30@gmail.com> Cc: Ying Han <yinghan@google.com> Cc: Nick Piggin <npiggin@gmail.com> Acked-by: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Lameter <cl@linux.com> Cc: Bob Liu <lliubbo@gmail.com> Cc: Neil Zhang <zhangwm@marvell.com> Cc: Russell King - ARM Linux <linux@arm.linux.org.uk> Reviewed-by: Michal Hocko <mhocko@suse.cz> Acked-by: Minchan Kim <minchan@kernel.org> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Lisa Du <cldu@marvell.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:36 +00:00
#include "internal.h"
writeback: introduce max-pause and pass-good dirty limits The max-pause limit helps to keep the sleep time inside balance_dirty_pages() within MAX_PAUSE=200ms. The 200ms max sleep means per task rate limit of 8pages/200ms=160KB/s when dirty exceeded, which normally is enough to stop dirtiers from continue pushing the dirty pages high, unless there are a sufficient large number of slow dirtiers (eg. 500 tasks doing 160KB/s will still sum up to 80MB/s, exceeding the write bandwidth of a slow disk and hence accumulating more and more dirty pages). The pass-good limit helps to let go of the good bdi's in the presence of a blocked bdi (ie. NFS server not responding) or slow USB disk which for some reason build up a large number of initial dirty pages that refuse to go away anytime soon. For example, given two bdi's A and B and the initial state bdi_thresh_A = dirty_thresh / 2 bdi_thresh_B = dirty_thresh / 2 bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 Then A get blocked, after a dozen seconds bdi_thresh_A = 0 bdi_thresh_B = dirty_thresh bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 The (bdi_dirty_B < bdi_thresh_B) test is now useless and the dirty pages will be effectively throttled by condition (nr_dirty < dirty_thresh). This has two problems: (1) we lose the protections for light dirtiers (2) balance_dirty_pages() effectively becomes IO-less because the (bdi_nr_reclaimable > bdi_thresh) test won't be true. This is good for IO, but balance_dirty_pages() loses an important way to break out of the loop which leads to more spread out throttle delays. DIRTY_PASSGOOD_AREA can eliminate the above issues. The only problem is, DIRTY_PASSGOOD_AREA needs to be defined as 2 to fully cover the above example while this patch uses the more conservative value 8 so as not to surprise people with too many dirty pages than expected. The max-pause limit won't noticeably impact the speed dirty pages are knocked down when there is a sudden drop of global/bdi dirty thresholds. Because the heavy dirties will be throttled below 160KB/s which is slow enough. It does help to avoid long dirty throttle delays and especially will make light dirtiers more responsive. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-20 04:18:42 +00:00
/*
* Sleep at most 200ms at a time in balance_dirty_pages().
*/
#define MAX_PAUSE max(HZ/5, 1)
/*
* Try to keep balance_dirty_pages() call intervals higher than this many pages
* by raising pause time to max_pause when falls below it.
*/
#define DIRTY_POLL_THRESH (128 >> (PAGE_SHIFT - 10))
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
/*
* Estimate write bandwidth at 200ms intervals.
*/
#define BANDWIDTH_INTERVAL max(HZ/5, 1)
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
#define RATELIMIT_CALC_SHIFT 10
/*
* After a CPU has dirtied this many pages, balance_dirty_pages_ratelimited
* will look to see if it needs to force writeback or throttling.
*/
static long ratelimit_pages = 32;
/* The following parameters are exported via /proc/sys/vm */
/*
* Start background writeback (via writeback threads) at this percentage
*/
static int dirty_background_ratio = 10;
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
/*
* dirty_background_bytes starts at 0 (disabled) so that it is a function of
* dirty_background_ratio * the amount of dirtyable memory
*/
static unsigned long dirty_background_bytes;
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
/*
* free highmem will not be subtracted from the total free memory
* for calculating free ratios if vm_highmem_is_dirtyable is true
*/
static int vm_highmem_is_dirtyable;
/*
* The generator of dirty data starts writeback at this percentage
*/
static int vm_dirty_ratio = 20;
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
/*
* vm_dirty_bytes starts at 0 (disabled) so that it is a function of
* vm_dirty_ratio * the amount of dirtyable memory
*/
static unsigned long vm_dirty_bytes;
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
/*
* The interval between `kupdate'-style writebacks
*/
unsigned int dirty_writeback_interval = 5 * 100; /* centiseconds */
EXPORT_SYMBOL_GPL(dirty_writeback_interval);
/*
* The longest time for which data is allowed to remain dirty
*/
unsigned int dirty_expire_interval = 30 * 100; /* centiseconds */
/*
* Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies:
* a full sync is triggered after this time elapses without any disk activity.
*/
int laptop_mode;
EXPORT_SYMBOL(laptop_mode);
/* End of sysctl-exported parameters */
struct wb_domain global_wb_domain;
/* consolidated parameters for balance_dirty_pages() and its subroutines */
struct dirty_throttle_control {
#ifdef CONFIG_CGROUP_WRITEBACK
struct wb_domain *dom;
struct dirty_throttle_control *gdtc; /* only set in memcg dtc's */
#endif
struct bdi_writeback *wb;
struct fprop_local_percpu *wb_completions;
unsigned long avail; /* dirtyable */
unsigned long dirty; /* file_dirty + write + nfs */
unsigned long thresh; /* dirty threshold */
unsigned long bg_thresh; /* dirty background threshold */
unsigned long wb_dirty; /* per-wb counterparts */
unsigned long wb_thresh;
unsigned long wb_bg_thresh;
unsigned long pos_ratio;
bool freerun;
bool dirty_exceeded;
};
/*
* Length of period for aging writeout fractions of bdis. This is an
* arbitrarily chosen number. The longer the period, the slower fractions will
* reflect changes in current writeout rate.
*/
#define VM_COMPLETIONS_PERIOD_LEN (3*HZ)
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
#ifdef CONFIG_CGROUP_WRITEBACK
#define GDTC_INIT(__wb) .wb = (__wb), \
.dom = &global_wb_domain, \
.wb_completions = &(__wb)->completions
#define GDTC_INIT_NO_WB .dom = &global_wb_domain
#define MDTC_INIT(__wb, __gdtc) .wb = (__wb), \
.dom = mem_cgroup_wb_domain(__wb), \
.wb_completions = &(__wb)->memcg_completions, \
.gdtc = __gdtc
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
static bool mdtc_valid(struct dirty_throttle_control *dtc)
{
return dtc->dom;
}
static struct wb_domain *dtc_dom(struct dirty_throttle_control *dtc)
{
return dtc->dom;
}
static struct dirty_throttle_control *mdtc_gdtc(struct dirty_throttle_control *mdtc)
{
return mdtc->gdtc;
}
static struct fprop_local_percpu *wb_memcg_completions(struct bdi_writeback *wb)
{
return &wb->memcg_completions;
}
static void wb_min_max_ratio(struct bdi_writeback *wb,
unsigned long *minp, unsigned long *maxp)
{
unsigned long this_bw = READ_ONCE(wb->avg_write_bandwidth);
unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth);
unsigned long long min = wb->bdi->min_ratio;
unsigned long long max = wb->bdi->max_ratio;
/*
* @wb may already be clean by the time control reaches here and
* the total may not include its bw.
*/
if (this_bw < tot_bw) {
if (min) {
min *= this_bw;
mm/page-writeback.c: avoid potential division by zero in wb_min_max_ratio() Patch series "use div64_ul() instead of div_u64() if the divisor is unsigned long". We were first inspired by commit b0ab99e7736a ("sched: Fix possible divide by zero in avg_atom () calculation"), then refer to the recently analyzed mm code, we found this suspicious place. 201 if (min) { 202 min *= this_bw; 203 do_div(min, tot_bw); 204 } And we also disassembled and confirmed it: /usr/src/debug/kernel-4.9.168-016.ali3000/linux-4.9.168-016.ali3000.alios7.x86_64/mm/page-writeback.c: 201 0xffffffff811c37da <__wb_calc_thresh+234>: xor %r10d,%r10d 0xffffffff811c37dd <__wb_calc_thresh+237>: test %rax,%rax 0xffffffff811c37e0 <__wb_calc_thresh+240>: je 0xffffffff811c3800 <__wb_calc_thresh+272> /usr/src/debug/kernel-4.9.168-016.ali3000/linux-4.9.168-016.ali3000.alios7.x86_64/mm/page-writeback.c: 202 0xffffffff811c37e2 <__wb_calc_thresh+242>: imul %r8,%rax /usr/src/debug/kernel-4.9.168-016.ali3000/linux-4.9.168-016.ali3000.alios7.x86_64/mm/page-writeback.c: 203 0xffffffff811c37e6 <__wb_calc_thresh+246>: mov %r9d,%r10d ---> truncates it to 32 bits here 0xffffffff811c37e9 <__wb_calc_thresh+249>: xor %edx,%edx 0xffffffff811c37eb <__wb_calc_thresh+251>: div %r10 0xffffffff811c37ee <__wb_calc_thresh+254>: imul %rbx,%rax 0xffffffff811c37f2 <__wb_calc_thresh+258>: shr $0x2,%rax 0xffffffff811c37f6 <__wb_calc_thresh+262>: mul %rcx 0xffffffff811c37f9 <__wb_calc_thresh+265>: shr $0x2,%rdx 0xffffffff811c37fd <__wb_calc_thresh+269>: mov %rdx,%r10 This series uses div64_ul() instead of div_u64() if the divisor is unsigned long, to avoid truncation to 32-bit on 64-bit platforms. This patch (of 3): The variables 'min' and 'max' are unsigned long and do_div truncates them to 32 bits, which means it can test non-zero and be truncated to zero for division. Fix this issue by using div64_ul() instead. Link: http://lkml.kernel.org/r/20200102081442.8273-2-wenyang@linux.alibaba.com Fixes: 693108a8a667 ("writeback: make bdi->min/max_ratio handling cgroup writeback aware") Signed-off-by: Wen Yang <wenyang@linux.alibaba.com> Reviewed-by: Andrew Morton <akpm@linux-foundation.org> Cc: Qian Cai <cai@lca.pw> Cc: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-14 00:29:23 +00:00
min = div64_ul(min, tot_bw);
}
if (max < 100 * BDI_RATIO_SCALE) {
max *= this_bw;
mm/page-writeback.c: avoid potential division by zero in wb_min_max_ratio() Patch series "use div64_ul() instead of div_u64() if the divisor is unsigned long". We were first inspired by commit b0ab99e7736a ("sched: Fix possible divide by zero in avg_atom () calculation"), then refer to the recently analyzed mm code, we found this suspicious place. 201 if (min) { 202 min *= this_bw; 203 do_div(min, tot_bw); 204 } And we also disassembled and confirmed it: /usr/src/debug/kernel-4.9.168-016.ali3000/linux-4.9.168-016.ali3000.alios7.x86_64/mm/page-writeback.c: 201 0xffffffff811c37da <__wb_calc_thresh+234>: xor %r10d,%r10d 0xffffffff811c37dd <__wb_calc_thresh+237>: test %rax,%rax 0xffffffff811c37e0 <__wb_calc_thresh+240>: je 0xffffffff811c3800 <__wb_calc_thresh+272> /usr/src/debug/kernel-4.9.168-016.ali3000/linux-4.9.168-016.ali3000.alios7.x86_64/mm/page-writeback.c: 202 0xffffffff811c37e2 <__wb_calc_thresh+242>: imul %r8,%rax /usr/src/debug/kernel-4.9.168-016.ali3000/linux-4.9.168-016.ali3000.alios7.x86_64/mm/page-writeback.c: 203 0xffffffff811c37e6 <__wb_calc_thresh+246>: mov %r9d,%r10d ---> truncates it to 32 bits here 0xffffffff811c37e9 <__wb_calc_thresh+249>: xor %edx,%edx 0xffffffff811c37eb <__wb_calc_thresh+251>: div %r10 0xffffffff811c37ee <__wb_calc_thresh+254>: imul %rbx,%rax 0xffffffff811c37f2 <__wb_calc_thresh+258>: shr $0x2,%rax 0xffffffff811c37f6 <__wb_calc_thresh+262>: mul %rcx 0xffffffff811c37f9 <__wb_calc_thresh+265>: shr $0x2,%rdx 0xffffffff811c37fd <__wb_calc_thresh+269>: mov %rdx,%r10 This series uses div64_ul() instead of div_u64() if the divisor is unsigned long, to avoid truncation to 32-bit on 64-bit platforms. This patch (of 3): The variables 'min' and 'max' are unsigned long and do_div truncates them to 32 bits, which means it can test non-zero and be truncated to zero for division. Fix this issue by using div64_ul() instead. Link: http://lkml.kernel.org/r/20200102081442.8273-2-wenyang@linux.alibaba.com Fixes: 693108a8a667 ("writeback: make bdi->min/max_ratio handling cgroup writeback aware") Signed-off-by: Wen Yang <wenyang@linux.alibaba.com> Reviewed-by: Andrew Morton <akpm@linux-foundation.org> Cc: Qian Cai <cai@lca.pw> Cc: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-14 00:29:23 +00:00
max = div64_ul(max, tot_bw);
}
}
*minp = min;
*maxp = max;
}
#else /* CONFIG_CGROUP_WRITEBACK */
#define GDTC_INIT(__wb) .wb = (__wb), \
.wb_completions = &(__wb)->completions
#define GDTC_INIT_NO_WB
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
#define MDTC_INIT(__wb, __gdtc)
static bool mdtc_valid(struct dirty_throttle_control *dtc)
{
return false;
}
static struct wb_domain *dtc_dom(struct dirty_throttle_control *dtc)
{
return &global_wb_domain;
}
static struct dirty_throttle_control *mdtc_gdtc(struct dirty_throttle_control *mdtc)
{
return NULL;
}
static struct fprop_local_percpu *wb_memcg_completions(struct bdi_writeback *wb)
{
return NULL;
}
static void wb_min_max_ratio(struct bdi_writeback *wb,
unsigned long *minp, unsigned long *maxp)
{
*minp = wb->bdi->min_ratio;
*maxp = wb->bdi->max_ratio;
}
#endif /* CONFIG_CGROUP_WRITEBACK */
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
/*
* In a memory zone, there is a certain amount of pages we consider
* available for the page cache, which is essentially the number of
* free and reclaimable pages, minus some zone reserves to protect
* lowmem and the ability to uphold the zone's watermarks without
* requiring writeback.
*
* This number of dirtyable pages is the base value of which the
* user-configurable dirty ratio is the effective number of pages that
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
* are allowed to be actually dirtied. Per individual zone, or
* globally by using the sum of dirtyable pages over all zones.
*
* Because the user is allowed to specify the dirty limit globally as
* absolute number of bytes, calculating the per-zone dirty limit can
* require translating the configured limit into a percentage of
* global dirtyable memory first.
*/
mm/page-writeback.c: fix dirty_balance_reserve subtraction from dirtyable memory Tejun reported stuttering and latency spikes on a system where random tasks would enter direct reclaim and get stuck on dirty pages. Around 50% of memory was occupied by tmpfs backed by an SSD, and another disk (rotating) was reading and writing at max speed to shrink a partition. : The problem was pretty ridiculous. It's a 8gig machine w/ one ssd and 10k : rpm harddrive and I could reliably reproduce constant stuttering every : several seconds for as long as buffered IO was going on on the hard drive : either with tmpfs occupying somewhere above 4gig or a test program which : allocates about the same amount of anon memory. Although swap usage was : zero, turning off swap also made the problem go away too. : : The trigger conditions seem quite plausible - high anon memory usage w/ : heavy buffered IO and swap configured - and it's highly likely that this : is happening in the wild too. (this can happen with copying large files : to usb sticks too, right?) This patch (of 2): The dirty_balance_reserve is an approximation of the fraction of free pages that the page allocator does not make available for page cache allocations. As a result, it has to be taken into account when calculating the amount of "dirtyable memory", the baseline to which dirty_background_ratio and dirty_ratio are applied. However, currently the reserve is subtracted from the sum of free and reclaimable pages, which is non-sensical and leads to erroneous results when the system is dominated by unreclaimable pages and the dirty_balance_reserve is bigger than free+reclaimable. In that case, at least the already allocated cache should be considered dirtyable. Fix the calculation by subtracting the reserve from the amount of free pages, then adding the reclaimable pages on top. [akpm@linux-foundation.org: fix CONFIG_HIGHMEM build] Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Reported-by: Tejun Heo <tj@kernel.org> Tested-by: Tejun Heo <tj@kernel.org> Reviewed-by: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Wu Fengguang <fengguang.wu@intel.com> Reviewed-by: Michal Hocko <mhocko@suse.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-29 22:05:39 +00:00
/**
* node_dirtyable_memory - number of dirtyable pages in a node
* @pgdat: the node
mm/page-writeback.c: fix dirty_balance_reserve subtraction from dirtyable memory Tejun reported stuttering and latency spikes on a system where random tasks would enter direct reclaim and get stuck on dirty pages. Around 50% of memory was occupied by tmpfs backed by an SSD, and another disk (rotating) was reading and writing at max speed to shrink a partition. : The problem was pretty ridiculous. It's a 8gig machine w/ one ssd and 10k : rpm harddrive and I could reliably reproduce constant stuttering every : several seconds for as long as buffered IO was going on on the hard drive : either with tmpfs occupying somewhere above 4gig or a test program which : allocates about the same amount of anon memory. Although swap usage was : zero, turning off swap also made the problem go away too. : : The trigger conditions seem quite plausible - high anon memory usage w/ : heavy buffered IO and swap configured - and it's highly likely that this : is happening in the wild too. (this can happen with copying large files : to usb sticks too, right?) This patch (of 2): The dirty_balance_reserve is an approximation of the fraction of free pages that the page allocator does not make available for page cache allocations. As a result, it has to be taken into account when calculating the amount of "dirtyable memory", the baseline to which dirty_background_ratio and dirty_ratio are applied. However, currently the reserve is subtracted from the sum of free and reclaimable pages, which is non-sensical and leads to erroneous results when the system is dominated by unreclaimable pages and the dirty_balance_reserve is bigger than free+reclaimable. In that case, at least the already allocated cache should be considered dirtyable. Fix the calculation by subtracting the reserve from the amount of free pages, then adding the reclaimable pages on top. [akpm@linux-foundation.org: fix CONFIG_HIGHMEM build] Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Reported-by: Tejun Heo <tj@kernel.org> Tested-by: Tejun Heo <tj@kernel.org> Reviewed-by: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Wu Fengguang <fengguang.wu@intel.com> Reviewed-by: Michal Hocko <mhocko@suse.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-29 22:05:39 +00:00
*
* Return: the node's number of pages potentially available for dirty
* page cache. This is the base value for the per-node dirty limits.
mm/page-writeback.c: fix dirty_balance_reserve subtraction from dirtyable memory Tejun reported stuttering and latency spikes on a system where random tasks would enter direct reclaim and get stuck on dirty pages. Around 50% of memory was occupied by tmpfs backed by an SSD, and another disk (rotating) was reading and writing at max speed to shrink a partition. : The problem was pretty ridiculous. It's a 8gig machine w/ one ssd and 10k : rpm harddrive and I could reliably reproduce constant stuttering every : several seconds for as long as buffered IO was going on on the hard drive : either with tmpfs occupying somewhere above 4gig or a test program which : allocates about the same amount of anon memory. Although swap usage was : zero, turning off swap also made the problem go away too. : : The trigger conditions seem quite plausible - high anon memory usage w/ : heavy buffered IO and swap configured - and it's highly likely that this : is happening in the wild too. (this can happen with copying large files : to usb sticks too, right?) This patch (of 2): The dirty_balance_reserve is an approximation of the fraction of free pages that the page allocator does not make available for page cache allocations. As a result, it has to be taken into account when calculating the amount of "dirtyable memory", the baseline to which dirty_background_ratio and dirty_ratio are applied. However, currently the reserve is subtracted from the sum of free and reclaimable pages, which is non-sensical and leads to erroneous results when the system is dominated by unreclaimable pages and the dirty_balance_reserve is bigger than free+reclaimable. In that case, at least the already allocated cache should be considered dirtyable. Fix the calculation by subtracting the reserve from the amount of free pages, then adding the reclaimable pages on top. [akpm@linux-foundation.org: fix CONFIG_HIGHMEM build] Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Reported-by: Tejun Heo <tj@kernel.org> Tested-by: Tejun Heo <tj@kernel.org> Reviewed-by: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Wu Fengguang <fengguang.wu@intel.com> Reviewed-by: Michal Hocko <mhocko@suse.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-29 22:05:39 +00:00
*/
static unsigned long node_dirtyable_memory(struct pglist_data *pgdat)
mm/page-writeback.c: fix dirty_balance_reserve subtraction from dirtyable memory Tejun reported stuttering and latency spikes on a system where random tasks would enter direct reclaim and get stuck on dirty pages. Around 50% of memory was occupied by tmpfs backed by an SSD, and another disk (rotating) was reading and writing at max speed to shrink a partition. : The problem was pretty ridiculous. It's a 8gig machine w/ one ssd and 10k : rpm harddrive and I could reliably reproduce constant stuttering every : several seconds for as long as buffered IO was going on on the hard drive : either with tmpfs occupying somewhere above 4gig or a test program which : allocates about the same amount of anon memory. Although swap usage was : zero, turning off swap also made the problem go away too. : : The trigger conditions seem quite plausible - high anon memory usage w/ : heavy buffered IO and swap configured - and it's highly likely that this : is happening in the wild too. (this can happen with copying large files : to usb sticks too, right?) This patch (of 2): The dirty_balance_reserve is an approximation of the fraction of free pages that the page allocator does not make available for page cache allocations. As a result, it has to be taken into account when calculating the amount of "dirtyable memory", the baseline to which dirty_background_ratio and dirty_ratio are applied. However, currently the reserve is subtracted from the sum of free and reclaimable pages, which is non-sensical and leads to erroneous results when the system is dominated by unreclaimable pages and the dirty_balance_reserve is bigger than free+reclaimable. In that case, at least the already allocated cache should be considered dirtyable. Fix the calculation by subtracting the reserve from the amount of free pages, then adding the reclaimable pages on top. [akpm@linux-foundation.org: fix CONFIG_HIGHMEM build] Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Reported-by: Tejun Heo <tj@kernel.org> Tested-by: Tejun Heo <tj@kernel.org> Reviewed-by: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Wu Fengguang <fengguang.wu@intel.com> Reviewed-by: Michal Hocko <mhocko@suse.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-29 22:05:39 +00:00
{
unsigned long nr_pages = 0;
int z;
for (z = 0; z < MAX_NR_ZONES; z++) {
struct zone *zone = pgdat->node_zones + z;
if (!populated_zone(zone))
continue;
nr_pages += zone_page_state(zone, NR_FREE_PAGES);
}
mm/page-writeback.c: fix dirty_balance_reserve subtraction from dirtyable memory Tejun reported stuttering and latency spikes on a system where random tasks would enter direct reclaim and get stuck on dirty pages. Around 50% of memory was occupied by tmpfs backed by an SSD, and another disk (rotating) was reading and writing at max speed to shrink a partition. : The problem was pretty ridiculous. It's a 8gig machine w/ one ssd and 10k : rpm harddrive and I could reliably reproduce constant stuttering every : several seconds for as long as buffered IO was going on on the hard drive : either with tmpfs occupying somewhere above 4gig or a test program which : allocates about the same amount of anon memory. Although swap usage was : zero, turning off swap also made the problem go away too. : : The trigger conditions seem quite plausible - high anon memory usage w/ : heavy buffered IO and swap configured - and it's highly likely that this : is happening in the wild too. (this can happen with copying large files : to usb sticks too, right?) This patch (of 2): The dirty_balance_reserve is an approximation of the fraction of free pages that the page allocator does not make available for page cache allocations. As a result, it has to be taken into account when calculating the amount of "dirtyable memory", the baseline to which dirty_background_ratio and dirty_ratio are applied. However, currently the reserve is subtracted from the sum of free and reclaimable pages, which is non-sensical and leads to erroneous results when the system is dominated by unreclaimable pages and the dirty_balance_reserve is bigger than free+reclaimable. In that case, at least the already allocated cache should be considered dirtyable. Fix the calculation by subtracting the reserve from the amount of free pages, then adding the reclaimable pages on top. [akpm@linux-foundation.org: fix CONFIG_HIGHMEM build] Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Reported-by: Tejun Heo <tj@kernel.org> Tested-by: Tejun Heo <tj@kernel.org> Reviewed-by: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Wu Fengguang <fengguang.wu@intel.com> Reviewed-by: Michal Hocko <mhocko@suse.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-29 22:05:39 +00:00
/*
* Pages reserved for the kernel should not be considered
* dirtyable, to prevent a situation where reclaim has to
* clean pages in order to balance the zones.
*/
nr_pages -= min(nr_pages, pgdat->totalreserve_pages);
mm/page-writeback.c: fix dirty_balance_reserve subtraction from dirtyable memory Tejun reported stuttering and latency spikes on a system where random tasks would enter direct reclaim and get stuck on dirty pages. Around 50% of memory was occupied by tmpfs backed by an SSD, and another disk (rotating) was reading and writing at max speed to shrink a partition. : The problem was pretty ridiculous. It's a 8gig machine w/ one ssd and 10k : rpm harddrive and I could reliably reproduce constant stuttering every : several seconds for as long as buffered IO was going on on the hard drive : either with tmpfs occupying somewhere above 4gig or a test program which : allocates about the same amount of anon memory. Although swap usage was : zero, turning off swap also made the problem go away too. : : The trigger conditions seem quite plausible - high anon memory usage w/ : heavy buffered IO and swap configured - and it's highly likely that this : is happening in the wild too. (this can happen with copying large files : to usb sticks too, right?) This patch (of 2): The dirty_balance_reserve is an approximation of the fraction of free pages that the page allocator does not make available for page cache allocations. As a result, it has to be taken into account when calculating the amount of "dirtyable memory", the baseline to which dirty_background_ratio and dirty_ratio are applied. However, currently the reserve is subtracted from the sum of free and reclaimable pages, which is non-sensical and leads to erroneous results when the system is dominated by unreclaimable pages and the dirty_balance_reserve is bigger than free+reclaimable. In that case, at least the already allocated cache should be considered dirtyable. Fix the calculation by subtracting the reserve from the amount of free pages, then adding the reclaimable pages on top. [akpm@linux-foundation.org: fix CONFIG_HIGHMEM build] Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Reported-by: Tejun Heo <tj@kernel.org> Tested-by: Tejun Heo <tj@kernel.org> Reviewed-by: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Wu Fengguang <fengguang.wu@intel.com> Reviewed-by: Michal Hocko <mhocko@suse.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-29 22:05:39 +00:00
nr_pages += node_page_state(pgdat, NR_INACTIVE_FILE);
nr_pages += node_page_state(pgdat, NR_ACTIVE_FILE);
mm/page-writeback.c: fix dirty_balance_reserve subtraction from dirtyable memory Tejun reported stuttering and latency spikes on a system where random tasks would enter direct reclaim and get stuck on dirty pages. Around 50% of memory was occupied by tmpfs backed by an SSD, and another disk (rotating) was reading and writing at max speed to shrink a partition. : The problem was pretty ridiculous. It's a 8gig machine w/ one ssd and 10k : rpm harddrive and I could reliably reproduce constant stuttering every : several seconds for as long as buffered IO was going on on the hard drive : either with tmpfs occupying somewhere above 4gig or a test program which : allocates about the same amount of anon memory. Although swap usage was : zero, turning off swap also made the problem go away too. : : The trigger conditions seem quite plausible - high anon memory usage w/ : heavy buffered IO and swap configured - and it's highly likely that this : is happening in the wild too. (this can happen with copying large files : to usb sticks too, right?) This patch (of 2): The dirty_balance_reserve is an approximation of the fraction of free pages that the page allocator does not make available for page cache allocations. As a result, it has to be taken into account when calculating the amount of "dirtyable memory", the baseline to which dirty_background_ratio and dirty_ratio are applied. However, currently the reserve is subtracted from the sum of free and reclaimable pages, which is non-sensical and leads to erroneous results when the system is dominated by unreclaimable pages and the dirty_balance_reserve is bigger than free+reclaimable. In that case, at least the already allocated cache should be considered dirtyable. Fix the calculation by subtracting the reserve from the amount of free pages, then adding the reclaimable pages on top. [akpm@linux-foundation.org: fix CONFIG_HIGHMEM build] Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Reported-by: Tejun Heo <tj@kernel.org> Tested-by: Tejun Heo <tj@kernel.org> Reviewed-by: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Wu Fengguang <fengguang.wu@intel.com> Reviewed-by: Michal Hocko <mhocko@suse.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-29 22:05:39 +00:00
return nr_pages;
}
static unsigned long highmem_dirtyable_memory(unsigned long total)
{
#ifdef CONFIG_HIGHMEM
int node;
unsigned long x = 0;
int i;
for_each_node_state(node, N_HIGH_MEMORY) {
for (i = ZONE_NORMAL + 1; i < MAX_NR_ZONES; i++) {
struct zone *z;
unsigned long nr_pages;
if (!is_highmem_idx(i))
continue;
z = &NODE_DATA(node)->node_zones[i];
if (!populated_zone(z))
continue;
nr_pages = zone_page_state(z, NR_FREE_PAGES);
/* watch for underflows */
nr_pages -= min(nr_pages, high_wmark_pages(z));
nr_pages += zone_page_state(z, NR_ZONE_INACTIVE_FILE);
nr_pages += zone_page_state(z, NR_ZONE_ACTIVE_FILE);
x += nr_pages;
}
}
/*
* Make sure that the number of highmem pages is never larger
* than the number of the total dirtyable memory. This can only
* occur in very strange VM situations but we want to make sure
* that this does not occur.
*/
return min(x, total);
#else
return 0;
#endif
}
/**
* global_dirtyable_memory - number of globally dirtyable pages
*
* Return: the global number of pages potentially available for dirty
* page cache. This is the base value for the global dirty limits.
*/
static unsigned long global_dirtyable_memory(void)
{
unsigned long x;
x = global_zone_page_state(NR_FREE_PAGES);
/*
* Pages reserved for the kernel should not be considered
* dirtyable, to prevent a situation where reclaim has to
* clean pages in order to balance the zones.
*/
x -= min(x, totalreserve_pages);
mm, vmscan: move LRU lists to node This moves the LRU lists from the zone to the node and related data such as counters, tracing, congestion tracking and writeback tracking. Unfortunately, due to reclaim and compaction retry logic, it is necessary to account for the number of LRU pages on both zone and node logic. Most reclaim logic is based on the node counters but the retry logic uses the zone counters which do not distinguish inactive and active sizes. It would be possible to leave the LRU counters on a per-zone basis but it's a heavier calculation across multiple cache lines that is much more frequent than the retry checks. Other than the LRU counters, this is mostly a mechanical patch but note that it introduces a number of anomalies. For example, the scans are per-zone but using per-node counters. We also mark a node as congested when a zone is congested. This causes weird problems that are fixed later but is easier to review. In the event that there is excessive overhead on 32-bit systems due to the nodes being on LRU then there are two potential solutions 1. Long-term isolation of highmem pages when reclaim is lowmem When pages are skipped, they are immediately added back onto the LRU list. If lowmem reclaim persisted for long periods of time, the same highmem pages get continually scanned. The idea would be that lowmem keeps those pages on a separate list until a reclaim for highmem pages arrives that splices the highmem pages back onto the LRU. It potentially could be implemented similar to the UNEVICTABLE list. That would reduce the skip rate with the potential corner case is that highmem pages have to be scanned and reclaimed to free lowmem slab pages. 2. Linear scan lowmem pages if the initial LRU shrink fails This will break LRU ordering but may be preferable and faster during memory pressure than skipping LRU pages. Link: http://lkml.kernel.org/r/1467970510-21195-4-git-send-email-mgorman@techsingularity.net Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Hillf Danton <hillf.zj@alibaba-inc.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@surriel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-28 22:45:31 +00:00
x += global_node_page_state(NR_INACTIVE_FILE);
x += global_node_page_state(NR_ACTIVE_FILE);
mm/page-writeback.c: fix dirty_balance_reserve subtraction from dirtyable memory Tejun reported stuttering and latency spikes on a system where random tasks would enter direct reclaim and get stuck on dirty pages. Around 50% of memory was occupied by tmpfs backed by an SSD, and another disk (rotating) was reading and writing at max speed to shrink a partition. : The problem was pretty ridiculous. It's a 8gig machine w/ one ssd and 10k : rpm harddrive and I could reliably reproduce constant stuttering every : several seconds for as long as buffered IO was going on on the hard drive : either with tmpfs occupying somewhere above 4gig or a test program which : allocates about the same amount of anon memory. Although swap usage was : zero, turning off swap also made the problem go away too. : : The trigger conditions seem quite plausible - high anon memory usage w/ : heavy buffered IO and swap configured - and it's highly likely that this : is happening in the wild too. (this can happen with copying large files : to usb sticks too, right?) This patch (of 2): The dirty_balance_reserve is an approximation of the fraction of free pages that the page allocator does not make available for page cache allocations. As a result, it has to be taken into account when calculating the amount of "dirtyable memory", the baseline to which dirty_background_ratio and dirty_ratio are applied. However, currently the reserve is subtracted from the sum of free and reclaimable pages, which is non-sensical and leads to erroneous results when the system is dominated by unreclaimable pages and the dirty_balance_reserve is bigger than free+reclaimable. In that case, at least the already allocated cache should be considered dirtyable. Fix the calculation by subtracting the reserve from the amount of free pages, then adding the reclaimable pages on top. [akpm@linux-foundation.org: fix CONFIG_HIGHMEM build] Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Reported-by: Tejun Heo <tj@kernel.org> Tested-by: Tejun Heo <tj@kernel.org> Reviewed-by: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Wu Fengguang <fengguang.wu@intel.com> Reviewed-by: Michal Hocko <mhocko@suse.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-29 22:05:39 +00:00
if (!vm_highmem_is_dirtyable)
x -= highmem_dirtyable_memory(x);
return x + 1; /* Ensure that we never return 0 */
}
/**
* domain_dirty_limits - calculate thresh and bg_thresh for a wb_domain
* @dtc: dirty_throttle_control of interest
*
* Calculate @dtc->thresh and ->bg_thresh considering
* vm_dirty_{bytes|ratio} and dirty_background_{bytes|ratio}. The caller
* must ensure that @dtc->avail is set before calling this function. The
mm/writeback: replace PF_LESS_THROTTLE with PF_LOCAL_THROTTLE PF_LESS_THROTTLE exists for loop-back nfsd (and a similar need in the loop block driver and callers of prctl(PR_SET_IO_FLUSHER)), where a daemon needs to write to one bdi (the final bdi) in order to free up writes queued to another bdi (the client bdi). The daemon sets PF_LESS_THROTTLE and gets a larger allowance of dirty pages, so that it can still dirty pages after other processses have been throttled. The purpose of this is to avoid deadlock that happen when the PF_LESS_THROTTLE process must write for any dirty pages to be freed, but it is being thottled and cannot write. This approach was designed when all threads were blocked equally, independently on which device they were writing to, or how fast it was. Since that time the writeback algorithm has changed substantially with different threads getting different allowances based on non-trivial heuristics. This means the simple "add 25%" heuristic is no longer reliable. The important issue is not that the daemon needs a *larger* dirty page allowance, but that it needs a *private* dirty page allowance, so that dirty pages for the "client" bdi that it is helping to clear (the bdi for an NFS filesystem or loop block device etc) do not affect the throttling of the daemon writing to the "final" bdi. This patch changes the heuristic so that the task is not throttled when the bdi it is writing to has a dirty page count below below (or equal to) the free-run threshold for that bdi. This ensures it will always be able to have some pages in flight, and so will not deadlock. In a steady-state, it is expected that PF_LOCAL_THROTTLE tasks might still be throttled by global threshold, but that is acceptable as it is only the deadlock state that is interesting for this flag. This approach of "only throttle when target bdi is busy" is consistent with the other use of PF_LESS_THROTTLE in current_may_throttle(), were it causes attention to be focussed only on the target bdi. So this patch - renames PF_LESS_THROTTLE to PF_LOCAL_THROTTLE, - removes the 25% bonus that that flag gives, and - If PF_LOCAL_THROTTLE is set, don't delay at all unless the global and the local free-run thresholds are exceeded. Note that previously realtime threads were treated the same as PF_LESS_THROTTLE threads. This patch does *not* change the behvaiour for real-time threads, so it is now different from the behaviour of nfsd and loop tasks. I don't know what is wanted for realtime. [akpm@linux-foundation.org: coding style fixes] Signed-off-by: NeilBrown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Jan Kara <jack@suse.cz> Acked-by: Chuck Lever <chuck.lever@oracle.com> [nfsd] Cc: Christoph Hellwig <hch@lst.de> Cc: Michal Hocko <mhocko@suse.com> Cc: Trond Myklebust <trond.myklebust@hammerspace.com> Link: http://lkml.kernel.org/r/87ftbf7gs3.fsf@notabene.neil.brown.name Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-02 04:48:18 +00:00
* dirty limits will be lifted by 1/4 for real-time tasks.
*/
static void domain_dirty_limits(struct dirty_throttle_control *dtc)
{
const unsigned long available_memory = dtc->avail;
struct dirty_throttle_control *gdtc = mdtc_gdtc(dtc);
unsigned long bytes = vm_dirty_bytes;
unsigned long bg_bytes = dirty_background_bytes;
/* convert ratios to per-PAGE_SIZE for higher precision */
unsigned long ratio = (vm_dirty_ratio * PAGE_SIZE) / 100;
unsigned long bg_ratio = (dirty_background_ratio * PAGE_SIZE) / 100;
unsigned long thresh;
unsigned long bg_thresh;
struct task_struct *tsk;
/* gdtc is !NULL iff @dtc is for memcg domain */
if (gdtc) {
unsigned long global_avail = gdtc->avail;
/*
* The byte settings can't be applied directly to memcg
* domains. Convert them to ratios by scaling against
* globally available memory. As the ratios are in
* per-PAGE_SIZE, they can be obtained by dividing bytes by
* number of pages.
*/
if (bytes)
ratio = min(DIV_ROUND_UP(bytes, global_avail),
PAGE_SIZE);
if (bg_bytes)
bg_ratio = min(DIV_ROUND_UP(bg_bytes, global_avail),
PAGE_SIZE);
bytes = bg_bytes = 0;
}
if (bytes)
thresh = DIV_ROUND_UP(bytes, PAGE_SIZE);
else
thresh = (ratio * available_memory) / PAGE_SIZE;
if (bg_bytes)
bg_thresh = DIV_ROUND_UP(bg_bytes, PAGE_SIZE);
else
bg_thresh = (bg_ratio * available_memory) / PAGE_SIZE;
tsk = current;
if (rt_or_dl_task(tsk)) {
MM: increase safety margin provided by PF_LESS_THROTTLE When nfsd is exporting a filesystem over NFS which is then NFS-mounted on the local machine there is a risk of deadlock. This happens when there are lots of dirty pages in the NFS filesystem and they cause NFSD to be throttled, either in throttle_vm_writeout() or in balance_dirty_pages(). To avoid this problem the PF_LESS_THROTTLE flag is set for NFSD threads and it provides a 25% increase to the limits that affect NFSD. Any process writing to an NFS filesystem will be throttled well before the number of dirty NFS pages reaches the limit imposed on NFSD, so NFSD will not deadlock on pages that it needs to write out. At least it shouldn't. All processes are allowed a small excess margin to avoid performing too many calculations: ratelimit_pages. ratelimit_pages is set so that if a thread on every CPU uses the entire margin, the total will only go 3% over the limit, and this is much less than the 25% bonus that PF_LESS_THROTTLE provides, so this margin shouldn't be a problem. But it is. The "total memory" that these 3% and 25% are calculated against are not really total memory but are "global_dirtyable_memory()" which doesn't include anonymous memory, just free memory and page-cache memory. The "ratelimit_pages" number is based on whatever the global_dirtyable_memory was on the last CPU hot-plug, which might not be what you expect, but is probably close to the total freeable memory. The throttle threshold uses the global_dirtable_memory at the moment when the throttling happens, which could be much less than at the last CPU hotplug. So if lots of anonymous memory has been allocated, thus pushing out lots of page-cache pages, then NFSD might end up being throttled due to dirty NFS pages because the "25%" bonus it gets is calculated against a rather small amount of dirtyable memory, while the "3%" margin that other processes are allowed to dirty without penalty is calculated against a much larger number. To remove this possibility of deadlock we need to make sure that the margin granted to PF_LESS_THROTTLE exceeds that rate-limit margin. Simply adding ratelimit_pages isn't enough as that should be multiplied by the number of cpus. So add "global_wb_domain.dirty_limit / 32" as that more accurately reflects the current total over-shoot margin. This ensures that the number of dirty NFS pages never gets so high that nfsd will be throttled waiting for them to be written. Link: http://lkml.kernel.org/r/87futgowwv.fsf@notabene.neil.brown.name Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 23:58:53 +00:00
bg_thresh += bg_thresh / 4 + global_wb_domain.dirty_limit / 32;
thresh += thresh / 4 + global_wb_domain.dirty_limit / 32;
}
/*
* Dirty throttling logic assumes the limits in page units fit into
* 32-bits. This gives 16TB dirty limits max which is hopefully enough.
*/
if (thresh > UINT_MAX)
thresh = UINT_MAX;
/* This makes sure bg_thresh is within 32-bits as well */
if (bg_thresh >= thresh)
bg_thresh = thresh / 2;
dtc->thresh = thresh;
dtc->bg_thresh = bg_thresh;
/* we should eventually report the domain in the TP */
if (!gdtc)
trace_global_dirty_state(bg_thresh, thresh);
}
/**
* global_dirty_limits - background-writeback and dirty-throttling thresholds
* @pbackground: out parameter for bg_thresh
* @pdirty: out parameter for thresh
*
* Calculate bg_thresh and thresh for global_wb_domain. See
* domain_dirty_limits() for details.
*/
void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty)
{
struct dirty_throttle_control gdtc = { GDTC_INIT_NO_WB };
gdtc.avail = global_dirtyable_memory();
domain_dirty_limits(&gdtc);
*pbackground = gdtc.bg_thresh;
*pdirty = gdtc.thresh;
}
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
/**
* node_dirty_limit - maximum number of dirty pages allowed in a node
* @pgdat: the node
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
*
* Return: the maximum number of dirty pages allowed in a node, based
* on the node's dirtyable memory.
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
*/
static unsigned long node_dirty_limit(struct pglist_data *pgdat)
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
{
unsigned long node_memory = node_dirtyable_memory(pgdat);
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
struct task_struct *tsk = current;
unsigned long dirty;
if (vm_dirty_bytes)
dirty = DIV_ROUND_UP(vm_dirty_bytes, PAGE_SIZE) *
node_memory / global_dirtyable_memory();
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
else
dirty = vm_dirty_ratio * node_memory / 100;
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
if (rt_or_dl_task(tsk))
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
dirty += dirty / 4;
/*
* Dirty throttling logic assumes the limits in page units fit into
* 32-bits. This gives 16TB dirty limits max which is hopefully enough.
*/
return min_t(unsigned long, dirty, UINT_MAX);
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
}
/**
* node_dirty_ok - tells whether a node is within its dirty limits
* @pgdat: the node to check
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
*
* Return: %true when the dirty pages in @pgdat are within the node's
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
* dirty limit, %false if the limit is exceeded.
*/
bool node_dirty_ok(struct pglist_data *pgdat)
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
{
unsigned long limit = node_dirty_limit(pgdat);
unsigned long nr_pages = 0;
nr_pages += node_page_state(pgdat, NR_FILE_DIRTY);
nr_pages += node_page_state(pgdat, NR_WRITEBACK);
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
return nr_pages <= limit;
mm: try to distribute dirty pages fairly across zones The maximum number of dirty pages that exist in the system at any time is determined by a number of pages considered dirtyable and a user-configured percentage of those, or an absolute number in bytes. This number of dirtyable pages is the sum of memory provided by all the zones in the system minus their lowmem reserves and high watermarks, so that the system can retain a healthy number of free pages without having to reclaim dirty pages. But there is a flaw in that we have a zoned page allocator which does not care about the global state but rather the state of individual memory zones. And right now there is nothing that prevents one zone from filling up with dirty pages while other zones are spared, which frequently leads to situations where kswapd, in order to restore the watermark of free pages, does indeed have to write pages from that zone's LRU list. This can interfere so badly with IO from the flusher threads that major filesystems (btrfs, xfs, ext4) mostly ignore write requests from reclaim already, taking away the VM's only possibility to keep such a zone balanced, aside from hoping the flushers will soon clean pages from that zone. Enter per-zone dirty limits. They are to a zone's dirtyable memory what the global limit is to the global amount of dirtyable memory, and try to make sure that no single zone receives more than its fair share of the globally allowed dirty pages in the first place. As the number of pages considered dirtyable excludes the zones' lowmem reserves and high watermarks, the maximum number of dirty pages in a zone is such that the zone can always be balanced without requiring page cleaning. As this is a placement decision in the page allocator and pages are dirtied only after the allocation, this patch allows allocators to pass __GFP_WRITE when they know in advance that the page will be written to and become dirty soon. The page allocator will then attempt to allocate from the first zone of the zonelist - which on NUMA is determined by the task's NUMA memory policy - that has not exceeded its dirty limit. At first glance, it would appear that the diversion to lower zones can increase pressure on them, but this is not the case. With a full high zone, allocations will be diverted to lower zones eventually, so it is more of a shift in timing of the lower zone allocations. Workloads that previously could fit their dirty pages completely in the higher zone may be forced to allocate from lower zones, but the amount of pages that "spill over" are limited themselves by the lower zones' dirty constraints, and thus unlikely to become a problem. For now, the problem of unfair dirty page distribution remains for NUMA configurations where the zones allowed for allocation are in sum not big enough to trigger the global dirty limits, wake up the flusher threads and remedy the situation. Because of this, an allocation that could not succeed on any of the considered zones is allowed to ignore the dirty limits before going into direct reclaim or even failing the allocation, until a future patch changes the global dirty throttling and flusher thread activation so that they take individual zone states into account. Test results 15M DMA + 3246M DMA32 + 504 Normal = 3765M memory 40% dirty ratio 16G USB thumb drive 10 runs of dd if=/dev/zero of=disk/zeroes bs=32k count=$((10 << 15)) seconds nr_vmscan_write (stddev) min| median| max xfs vanilla: 549.747( 3.492) 0.000| 0.000| 0.000 patched: 550.996( 3.802) 0.000| 0.000| 0.000 fuse-ntfs vanilla: 1183.094(53.178) 54349.000| 59341.000| 65163.000 patched: 558.049(17.914) 0.000| 0.000| 43.000 btrfs vanilla: 573.679(14.015) 156657.000| 460178.000| 606926.000 patched: 563.365(11.368) 0.000| 0.000| 1362.000 ext4 vanilla: 561.197(15.782) 0.000|2725438.000|4143837.000 patched: 568.806(17.496) 0.000| 0.000| 0.000 Signed-off-by: Johannes Weiner <jweiner@redhat.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Acked-by: Mel Gorman <mgorman@suse.de> Reviewed-by: Michal Hocko <mhocko@suse.cz> Tested-by: Wu Fengguang <fengguang.wu@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Jan Kara <jack@suse.cz> Cc: Shaohua Li <shaohua.li@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Chris Mason <chris.mason@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-10 23:07:49 +00:00
}
#ifdef CONFIG_SYSCTL
sysctl: treewide: constify the ctl_table argument of proc_handlers const qualify the struct ctl_table argument in the proc_handler function signatures. This is a prerequisite to moving the static ctl_table structs into .rodata data which will ensure that proc_handler function pointers cannot be modified. This patch has been generated by the following coccinelle script: ``` virtual patch @r1@ identifier ctl, write, buffer, lenp, ppos; identifier func !~ "appldata_(timer|interval)_handler|sched_(rt|rr)_handler|rds_tcp_skbuf_handler|proc_sctp_do_(hmac_alg|rto_min|rto_max|udp_port|alpha_beta|auth|probe_interval)"; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int write, void *buffer, size_t *lenp, loff_t *ppos); @r2@ identifier func, ctl, write, buffer, lenp, ppos; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int write, void *buffer, size_t *lenp, loff_t *ppos) { ... } @r3@ identifier func; @@ int func( - struct ctl_table * + const struct ctl_table * ,int , void *, size_t *, loff_t *); @r4@ identifier func, ctl; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int , void *, size_t *, loff_t *); @r5@ identifier func, write, buffer, lenp, ppos; @@ int func( - struct ctl_table * + const struct ctl_table * ,int write, void *buffer, size_t *lenp, loff_t *ppos); ``` * Code formatting was adjusted in xfs_sysctl.c to comply with code conventions. The xfs_stats_clear_proc_handler, xfs_panic_mask_proc_handler and xfs_deprecated_dointvec_minmax where adjusted. * The ctl_table argument in proc_watchdog_common was const qualified. This is called from a proc_handler itself and is calling back into another proc_handler, making it necessary to change it as part of the proc_handler migration. Co-developed-by: Thomas Weißschuh <linux@weissschuh.net> Signed-off-by: Thomas Weißschuh <linux@weissschuh.net> Co-developed-by: Joel Granados <j.granados@samsung.com> Signed-off-by: Joel Granados <j.granados@samsung.com>
2024-07-24 18:59:29 +00:00
static int dirty_background_ratio_handler(const struct ctl_table *table, int write,
void *buffer, size_t *lenp, loff_t *ppos)
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
{
int ret;
ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
if (ret == 0 && write)
dirty_background_bytes = 0;
return ret;
}
sysctl: treewide: constify the ctl_table argument of proc_handlers const qualify the struct ctl_table argument in the proc_handler function signatures. This is a prerequisite to moving the static ctl_table structs into .rodata data which will ensure that proc_handler function pointers cannot be modified. This patch has been generated by the following coccinelle script: ``` virtual patch @r1@ identifier ctl, write, buffer, lenp, ppos; identifier func !~ "appldata_(timer|interval)_handler|sched_(rt|rr)_handler|rds_tcp_skbuf_handler|proc_sctp_do_(hmac_alg|rto_min|rto_max|udp_port|alpha_beta|auth|probe_interval)"; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int write, void *buffer, size_t *lenp, loff_t *ppos); @r2@ identifier func, ctl, write, buffer, lenp, ppos; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int write, void *buffer, size_t *lenp, loff_t *ppos) { ... } @r3@ identifier func; @@ int func( - struct ctl_table * + const struct ctl_table * ,int , void *, size_t *, loff_t *); @r4@ identifier func, ctl; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int , void *, size_t *, loff_t *); @r5@ identifier func, write, buffer, lenp, ppos; @@ int func( - struct ctl_table * + const struct ctl_table * ,int write, void *buffer, size_t *lenp, loff_t *ppos); ``` * Code formatting was adjusted in xfs_sysctl.c to comply with code conventions. The xfs_stats_clear_proc_handler, xfs_panic_mask_proc_handler and xfs_deprecated_dointvec_minmax where adjusted. * The ctl_table argument in proc_watchdog_common was const qualified. This is called from a proc_handler itself and is calling back into another proc_handler, making it necessary to change it as part of the proc_handler migration. Co-developed-by: Thomas Weißschuh <linux@weissschuh.net> Signed-off-by: Thomas Weißschuh <linux@weissschuh.net> Co-developed-by: Joel Granados <j.granados@samsung.com> Signed-off-by: Joel Granados <j.granados@samsung.com>
2024-07-24 18:59:29 +00:00
static int dirty_background_bytes_handler(const struct ctl_table *table, int write,
void *buffer, size_t *lenp, loff_t *ppos)
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
{
int ret;
unsigned long old_bytes = dirty_background_bytes;
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
if (ret == 0 && write) {
if (DIV_ROUND_UP(dirty_background_bytes, PAGE_SIZE) >
UINT_MAX) {
dirty_background_bytes = old_bytes;
return -ERANGE;
}
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
dirty_background_ratio = 0;
}
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
return ret;
}
sysctl: treewide: constify the ctl_table argument of proc_handlers const qualify the struct ctl_table argument in the proc_handler function signatures. This is a prerequisite to moving the static ctl_table structs into .rodata data which will ensure that proc_handler function pointers cannot be modified. This patch has been generated by the following coccinelle script: ``` virtual patch @r1@ identifier ctl, write, buffer, lenp, ppos; identifier func !~ "appldata_(timer|interval)_handler|sched_(rt|rr)_handler|rds_tcp_skbuf_handler|proc_sctp_do_(hmac_alg|rto_min|rto_max|udp_port|alpha_beta|auth|probe_interval)"; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int write, void *buffer, size_t *lenp, loff_t *ppos); @r2@ identifier func, ctl, write, buffer, lenp, ppos; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int write, void *buffer, size_t *lenp, loff_t *ppos) { ... } @r3@ identifier func; @@ int func( - struct ctl_table * + const struct ctl_table * ,int , void *, size_t *, loff_t *); @r4@ identifier func, ctl; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int , void *, size_t *, loff_t *); @r5@ identifier func, write, buffer, lenp, ppos; @@ int func( - struct ctl_table * + const struct ctl_table * ,int write, void *buffer, size_t *lenp, loff_t *ppos); ``` * Code formatting was adjusted in xfs_sysctl.c to comply with code conventions. The xfs_stats_clear_proc_handler, xfs_panic_mask_proc_handler and xfs_deprecated_dointvec_minmax where adjusted. * The ctl_table argument in proc_watchdog_common was const qualified. This is called from a proc_handler itself and is calling back into another proc_handler, making it necessary to change it as part of the proc_handler migration. Co-developed-by: Thomas Weißschuh <linux@weissschuh.net> Signed-off-by: Thomas Weißschuh <linux@weissschuh.net> Co-developed-by: Joel Granados <j.granados@samsung.com> Signed-off-by: Joel Granados <j.granados@samsung.com>
2024-07-24 18:59:29 +00:00
static int dirty_ratio_handler(const struct ctl_table *table, int write, void *buffer,
size_t *lenp, loff_t *ppos)
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
{
int old_ratio = vm_dirty_ratio;
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
int ret;
ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
if (ret == 0 && write && vm_dirty_ratio != old_ratio) {
writeback_set_ratelimit();
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
vm_dirty_bytes = 0;
}
return ret;
}
sysctl: treewide: constify the ctl_table argument of proc_handlers const qualify the struct ctl_table argument in the proc_handler function signatures. This is a prerequisite to moving the static ctl_table structs into .rodata data which will ensure that proc_handler function pointers cannot be modified. This patch has been generated by the following coccinelle script: ``` virtual patch @r1@ identifier ctl, write, buffer, lenp, ppos; identifier func !~ "appldata_(timer|interval)_handler|sched_(rt|rr)_handler|rds_tcp_skbuf_handler|proc_sctp_do_(hmac_alg|rto_min|rto_max|udp_port|alpha_beta|auth|probe_interval)"; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int write, void *buffer, size_t *lenp, loff_t *ppos); @r2@ identifier func, ctl, write, buffer, lenp, ppos; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int write, void *buffer, size_t *lenp, loff_t *ppos) { ... } @r3@ identifier func; @@ int func( - struct ctl_table * + const struct ctl_table * ,int , void *, size_t *, loff_t *); @r4@ identifier func, ctl; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int , void *, size_t *, loff_t *); @r5@ identifier func, write, buffer, lenp, ppos; @@ int func( - struct ctl_table * + const struct ctl_table * ,int write, void *buffer, size_t *lenp, loff_t *ppos); ``` * Code formatting was adjusted in xfs_sysctl.c to comply with code conventions. The xfs_stats_clear_proc_handler, xfs_panic_mask_proc_handler and xfs_deprecated_dointvec_minmax where adjusted. * The ctl_table argument in proc_watchdog_common was const qualified. This is called from a proc_handler itself and is calling back into another proc_handler, making it necessary to change it as part of the proc_handler migration. Co-developed-by: Thomas Weißschuh <linux@weissschuh.net> Signed-off-by: Thomas Weißschuh <linux@weissschuh.net> Co-developed-by: Joel Granados <j.granados@samsung.com> Signed-off-by: Joel Granados <j.granados@samsung.com>
2024-07-24 18:59:29 +00:00
static int dirty_bytes_handler(const struct ctl_table *table, int write,
void *buffer, size_t *lenp, loff_t *ppos)
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
{
unsigned long old_bytes = vm_dirty_bytes;
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
int ret;
ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
if (ret == 0 && write && vm_dirty_bytes != old_bytes) {
if (DIV_ROUND_UP(vm_dirty_bytes, PAGE_SIZE) > UINT_MAX) {
vm_dirty_bytes = old_bytes;
return -ERANGE;
}
writeback_set_ratelimit();
mm: add dirty_background_bytes and dirty_bytes sysctls This change introduces two new sysctls to /proc/sys/vm: dirty_background_bytes and dirty_bytes. dirty_background_bytes is the counterpart to dirty_background_ratio and dirty_bytes is the counterpart to dirty_ratio. With growing memory capacities of individual machines, it's no longer sufficient to specify dirty thresholds as a percentage of the amount of dirtyable memory over the entire system. dirty_background_bytes and dirty_bytes specify quantities of memory, in bytes, that represent the dirty limits for the entire system. If either of these values is set, its value represents the amount of dirty memory that is needed to commence either background or direct writeback. When a `bytes' or `ratio' file is written, its counterpart becomes a function of the written value. For example, if dirty_bytes is written to be 8096, 8K of memory is required to commence direct writeback. dirty_ratio is then functionally equivalent to 8K / the amount of dirtyable memory: dirtyable_memory = free pages + mapped pages + file cache dirty_background_bytes = dirty_background_ratio * dirtyable_memory -or- dirty_background_ratio = dirty_background_bytes / dirtyable_memory AND dirty_bytes = dirty_ratio * dirtyable_memory -or- dirty_ratio = dirty_bytes / dirtyable_memory Only one of dirty_background_bytes and dirty_background_ratio may be specified at a time, and only one of dirty_bytes and dirty_ratio may be specified. When one sysctl is written, the other appears as 0 when read. The `bytes' files operate on a page size granularity since dirty limits are compared with ZVC values, which are in page units. Prior to this change, the minimum dirty_ratio was 5 as implemented by get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user written value between 0 and 100. This restriction is maintained, but dirty_bytes has a lower limit of only one page. Also prior to this change, the dirty_background_ratio could not equal or exceed dirty_ratio. This restriction is maintained in addition to restricting dirty_background_bytes. If either background threshold equals or exceeds that of the dirty threshold, it is implicitly set to half the dirty threshold. Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: David Rientjes <rientjes@google.com> Cc: Andrea Righi <righi.andrea@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:31 +00:00
vm_dirty_ratio = 0;
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
}
return ret;
}
#endif
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
static unsigned long wp_next_time(unsigned long cur_time)
{
cur_time += VM_COMPLETIONS_PERIOD_LEN;
/* 0 has a special meaning... */
if (!cur_time)
return 1;
return cur_time;
}
static void wb_domain_writeout_add(struct wb_domain *dom,
struct fprop_local_percpu *completions,
unsigned int max_prop_frac, long nr)
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
{
__fprop_add_percpu_max(&dom->completions, completions,
max_prop_frac, nr);
/* First event after period switching was turned off? */
if (unlikely(!dom->period_time)) {
/*
* We can race with other __bdi_writeout_inc calls here but
* it does not cause any harm since the resulting time when
* timer will fire and what is in writeout_period_time will be
* roughly the same.
*/
dom->period_time = wp_next_time(jiffies);
mod_timer(&dom->period_timer, dom->period_time);
}
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
}
/*
* Increment @wb's writeout completion count and the global writeout
* completion count. Called from __folio_end_writeback().
*/
static inline void __wb_writeout_add(struct bdi_writeback *wb, long nr)
{
struct wb_domain *cgdom;
wb_stat_mod(wb, WB_WRITTEN, nr);
wb_domain_writeout_add(&global_wb_domain, &wb->completions,
wb->bdi->max_prop_frac, nr);
cgdom = mem_cgroup_wb_domain(wb);
if (cgdom)
wb_domain_writeout_add(cgdom, wb_memcg_completions(wb),
wb->bdi->max_prop_frac, nr);
}
void wb_writeout_inc(struct bdi_writeback *wb)
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
{
unsigned long flags;
local_irq_save(flags);
__wb_writeout_add(wb, 1);
local_irq_restore(flags);
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
}
EXPORT_SYMBOL_GPL(wb_writeout_inc);
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
/*
* On idle system, we can be called long after we scheduled because we use
* deferred timers so count with missed periods.
*/
static void writeout_period(struct timer_list *t)
{
struct wb_domain *dom = from_timer(dom, t, period_timer);
int miss_periods = (jiffies - dom->period_time) /
VM_COMPLETIONS_PERIOD_LEN;
if (fprop_new_period(&dom->completions, miss_periods + 1)) {
dom->period_time = wp_next_time(dom->period_time +
miss_periods * VM_COMPLETIONS_PERIOD_LEN);
mod_timer(&dom->period_timer, dom->period_time);
} else {
/*
* Aging has zeroed all fractions. Stop wasting CPU on period
* updates.
*/
dom->period_time = 0;
}
}
int wb_domain_init(struct wb_domain *dom, gfp_t gfp)
{
memset(dom, 0, sizeof(*dom));
spin_lock_init(&dom->lock);
timer_setup(&dom->period_timer, writeout_period, TIMER_DEFERRABLE);
dom->dirty_limit_tstamp = jiffies;
return fprop_global_init(&dom->completions, gfp);
}
#ifdef CONFIG_CGROUP_WRITEBACK
void wb_domain_exit(struct wb_domain *dom)
{
del_timer_sync(&dom->period_timer);
fprop_global_destroy(&dom->completions);
}
#endif
/*
* bdi_min_ratio keeps the sum of the minimum dirty shares of all
* registered backing devices, which, for obvious reasons, can not
* exceed 100%.
*/
static unsigned int bdi_min_ratio;
static int bdi_check_pages_limit(unsigned long pages)
{
unsigned long max_dirty_pages = global_dirtyable_memory();
if (pages > max_dirty_pages)
return -EINVAL;
return 0;
}
static unsigned long bdi_ratio_from_pages(unsigned long pages)
{
unsigned long background_thresh;
unsigned long dirty_thresh;
unsigned long ratio;
global_dirty_limits(&background_thresh, &dirty_thresh);
ratio = div64_u64(pages * 100ULL * BDI_RATIO_SCALE, dirty_thresh);
return ratio;
}
static u64 bdi_get_bytes(unsigned int ratio)
{
unsigned long background_thresh;
unsigned long dirty_thresh;
u64 bytes;
global_dirty_limits(&background_thresh, &dirty_thresh);
bytes = (dirty_thresh * PAGE_SIZE * ratio) / BDI_RATIO_SCALE / 100;
return bytes;
}
static int __bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio)
{
unsigned int delta;
int ret = 0;
if (min_ratio > 100 * BDI_RATIO_SCALE)
return -EINVAL;
spin_lock_bh(&bdi_lock);
if (min_ratio > bdi->max_ratio) {
ret = -EINVAL;
} else {
if (min_ratio < bdi->min_ratio) {
delta = bdi->min_ratio - min_ratio;
bdi_min_ratio -= delta;
bdi->min_ratio = min_ratio;
} else {
delta = min_ratio - bdi->min_ratio;
if (bdi_min_ratio + delta < 100 * BDI_RATIO_SCALE) {
bdi_min_ratio += delta;
bdi->min_ratio = min_ratio;
} else {
ret = -EINVAL;
}
}
}
spin_unlock_bh(&bdi_lock);
return ret;
}
static int __bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio)
{
int ret = 0;
if (max_ratio > 100 * BDI_RATIO_SCALE)
return -EINVAL;
spin_lock_bh(&bdi_lock);
if (bdi->min_ratio > max_ratio) {
ret = -EINVAL;
} else {
bdi->max_ratio = max_ratio;
bdi->max_prop_frac = (FPROP_FRAC_BASE * max_ratio) /
(100 * BDI_RATIO_SCALE);
}
spin_unlock_bh(&bdi_lock);
return ret;
}
int bdi_set_min_ratio_no_scale(struct backing_dev_info *bdi, unsigned int min_ratio)
{
return __bdi_set_min_ratio(bdi, min_ratio);
}
int bdi_set_max_ratio_no_scale(struct backing_dev_info *bdi, unsigned int max_ratio)
{
return __bdi_set_max_ratio(bdi, max_ratio);
}
int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio)
{
return __bdi_set_min_ratio(bdi, min_ratio * BDI_RATIO_SCALE);
}
int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio)
{
return __bdi_set_max_ratio(bdi, max_ratio * BDI_RATIO_SCALE);
}
EXPORT_SYMBOL(bdi_set_max_ratio);
u64 bdi_get_min_bytes(struct backing_dev_info *bdi)
{
return bdi_get_bytes(bdi->min_ratio);
}
int bdi_set_min_bytes(struct backing_dev_info *bdi, u64 min_bytes)
{
int ret;
unsigned long pages = min_bytes >> PAGE_SHIFT;
unsigned long min_ratio;
ret = bdi_check_pages_limit(pages);
if (ret)
return ret;
min_ratio = bdi_ratio_from_pages(pages);
return __bdi_set_min_ratio(bdi, min_ratio);
}
u64 bdi_get_max_bytes(struct backing_dev_info *bdi)
{
return bdi_get_bytes(bdi->max_ratio);
}
int bdi_set_max_bytes(struct backing_dev_info *bdi, u64 max_bytes)
{
int ret;
unsigned long pages = max_bytes >> PAGE_SHIFT;
unsigned long max_ratio;
ret = bdi_check_pages_limit(pages);
if (ret)
return ret;
max_ratio = bdi_ratio_from_pages(pages);
return __bdi_set_max_ratio(bdi, max_ratio);
}
mm: add bdi_set_strict_limit() function Patch series "mm/block: add bdi sysfs knobs", v4. At meta network block devices (nbd) are used to implement remote block storage. In testing and during production it has been observed that these network block devices can consume a huge portion of the dirty writeback cache and writeback can take a considerable time. To be able to give stricter limits, I'm proposing the following changes: 1) introduce strictlimit knob Currently the max_ratio knob exists to limit the dirty_memory. However this knob only applies once (dirty_ratio + dirty_background_ratio) / 2 has been reached. With the BDI_CAP_STRICTLIMIT flag, the max_ratio can be applied without reaching that limit. This change exposes that knob. This knob can also be useful for NFS, fuse filesystems and USB devices. 2) Use part of 1000000 internal calculation The max_ratio is based on percentage. With the current machine sizes percentage values can be very high (1% of a 256GB main memory is already 2.5GB). This change uses part of 1000000 instead of percentages for the internal calculations. 3) Introduce two new sysfs knobs: min_bytes and max_bytes. Currently all calculations are based on ratio, but for a user it often more convenient to specify a limit in bytes. The new knobs will not store bytes values, instead they will translate the byte value to a corresponding ratio. As the internal values are now part of 1000, the ratio is closer to the specified value. However the value should be more seen as an approximation as it can fluctuate over time. 3) Introduce two new sysfs knobs: min_ratio_fine and max_ratio_fine. The granularity for the existing sysfs bdi knobs min_ratio and max_ratio is based on percentage values. The new sysfs bdi knobs min_ratio_fine and max_ratio_fine allow to specify the ratio as part of 1 million. This patch (of 20): This adds the bdi_set_strict_limit function to be able to set/unset the BDI_CAP_STRICTLIMIT flag. Link: https://lkml.kernel.org/r/20221119005215.3052436-1-shr@devkernel.io Link: https://lkml.kernel.org/r/20221119005215.3052436-2-shr@devkernel.io Signed-off-by: Stefan Roesch <shr@devkernel.io> Cc: Jens Axboe <axboe@kernel.dk> Cc: Chris Mason <clm@meta.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-11-19 00:51:56 +00:00
int bdi_set_strict_limit(struct backing_dev_info *bdi, unsigned int strict_limit)
{
if (strict_limit > 1)
return -EINVAL;
spin_lock_bh(&bdi_lock);
if (strict_limit)
bdi->capabilities |= BDI_CAP_STRICTLIMIT;
else
bdi->capabilities &= ~BDI_CAP_STRICTLIMIT;
spin_unlock_bh(&bdi_lock);
return 0;
}
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
static unsigned long dirty_freerun_ceiling(unsigned long thresh,
unsigned long bg_thresh)
{
return (thresh + bg_thresh) / 2;
}
static unsigned long hard_dirty_limit(struct wb_domain *dom,
unsigned long thresh)
writeback: introduce max-pause and pass-good dirty limits The max-pause limit helps to keep the sleep time inside balance_dirty_pages() within MAX_PAUSE=200ms. The 200ms max sleep means per task rate limit of 8pages/200ms=160KB/s when dirty exceeded, which normally is enough to stop dirtiers from continue pushing the dirty pages high, unless there are a sufficient large number of slow dirtiers (eg. 500 tasks doing 160KB/s will still sum up to 80MB/s, exceeding the write bandwidth of a slow disk and hence accumulating more and more dirty pages). The pass-good limit helps to let go of the good bdi's in the presence of a blocked bdi (ie. NFS server not responding) or slow USB disk which for some reason build up a large number of initial dirty pages that refuse to go away anytime soon. For example, given two bdi's A and B and the initial state bdi_thresh_A = dirty_thresh / 2 bdi_thresh_B = dirty_thresh / 2 bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 Then A get blocked, after a dozen seconds bdi_thresh_A = 0 bdi_thresh_B = dirty_thresh bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 The (bdi_dirty_B < bdi_thresh_B) test is now useless and the dirty pages will be effectively throttled by condition (nr_dirty < dirty_thresh). This has two problems: (1) we lose the protections for light dirtiers (2) balance_dirty_pages() effectively becomes IO-less because the (bdi_nr_reclaimable > bdi_thresh) test won't be true. This is good for IO, but balance_dirty_pages() loses an important way to break out of the loop which leads to more spread out throttle delays. DIRTY_PASSGOOD_AREA can eliminate the above issues. The only problem is, DIRTY_PASSGOOD_AREA needs to be defined as 2 to fully cover the above example while this patch uses the more conservative value 8 so as not to surprise people with too many dirty pages than expected. The max-pause limit won't noticeably impact the speed dirty pages are knocked down when there is a sudden drop of global/bdi dirty thresholds. Because the heavy dirties will be throttled below 160KB/s which is slow enough. It does help to avoid long dirty throttle delays and especially will make light dirtiers more responsive. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-20 04:18:42 +00:00
{
return max(thresh, dom->dirty_limit);
writeback: introduce max-pause and pass-good dirty limits The max-pause limit helps to keep the sleep time inside balance_dirty_pages() within MAX_PAUSE=200ms. The 200ms max sleep means per task rate limit of 8pages/200ms=160KB/s when dirty exceeded, which normally is enough to stop dirtiers from continue pushing the dirty pages high, unless there are a sufficient large number of slow dirtiers (eg. 500 tasks doing 160KB/s will still sum up to 80MB/s, exceeding the write bandwidth of a slow disk and hence accumulating more and more dirty pages). The pass-good limit helps to let go of the good bdi's in the presence of a blocked bdi (ie. NFS server not responding) or slow USB disk which for some reason build up a large number of initial dirty pages that refuse to go away anytime soon. For example, given two bdi's A and B and the initial state bdi_thresh_A = dirty_thresh / 2 bdi_thresh_B = dirty_thresh / 2 bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 Then A get blocked, after a dozen seconds bdi_thresh_A = 0 bdi_thresh_B = dirty_thresh bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 The (bdi_dirty_B < bdi_thresh_B) test is now useless and the dirty pages will be effectively throttled by condition (nr_dirty < dirty_thresh). This has two problems: (1) we lose the protections for light dirtiers (2) balance_dirty_pages() effectively becomes IO-less because the (bdi_nr_reclaimable > bdi_thresh) test won't be true. This is good for IO, but balance_dirty_pages() loses an important way to break out of the loop which leads to more spread out throttle delays. DIRTY_PASSGOOD_AREA can eliminate the above issues. The only problem is, DIRTY_PASSGOOD_AREA needs to be defined as 2 to fully cover the above example while this patch uses the more conservative value 8 so as not to surprise people with too many dirty pages than expected. The max-pause limit won't noticeably impact the speed dirty pages are knocked down when there is a sudden drop of global/bdi dirty thresholds. Because the heavy dirties will be throttled below 160KB/s which is slow enough. It does help to avoid long dirty throttle delays and especially will make light dirtiers more responsive. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-20 04:18:42 +00:00
}
2015-09-29 17:04:26 +00:00
/*
* Memory which can be further allocated to a memcg domain is capped by
* system-wide clean memory excluding the amount being used in the domain.
*/
static void mdtc_calc_avail(struct dirty_throttle_control *mdtc,
unsigned long filepages, unsigned long headroom)
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
{
struct dirty_throttle_control *gdtc = mdtc_gdtc(mdtc);
2015-09-29 17:04:26 +00:00
unsigned long clean = filepages - min(filepages, mdtc->dirty);
unsigned long global_clean = gdtc->avail - min(gdtc->avail, gdtc->dirty);
unsigned long other_clean = global_clean - min(global_clean, clean);
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
2015-09-29 17:04:26 +00:00
mdtc->avail = filepages + min(headroom, other_clean);
writeback: introduce max-pause and pass-good dirty limits The max-pause limit helps to keep the sleep time inside balance_dirty_pages() within MAX_PAUSE=200ms. The 200ms max sleep means per task rate limit of 8pages/200ms=160KB/s when dirty exceeded, which normally is enough to stop dirtiers from continue pushing the dirty pages high, unless there are a sufficient large number of slow dirtiers (eg. 500 tasks doing 160KB/s will still sum up to 80MB/s, exceeding the write bandwidth of a slow disk and hence accumulating more and more dirty pages). The pass-good limit helps to let go of the good bdi's in the presence of a blocked bdi (ie. NFS server not responding) or slow USB disk which for some reason build up a large number of initial dirty pages that refuse to go away anytime soon. For example, given two bdi's A and B and the initial state bdi_thresh_A = dirty_thresh / 2 bdi_thresh_B = dirty_thresh / 2 bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 Then A get blocked, after a dozen seconds bdi_thresh_A = 0 bdi_thresh_B = dirty_thresh bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 The (bdi_dirty_B < bdi_thresh_B) test is now useless and the dirty pages will be effectively throttled by condition (nr_dirty < dirty_thresh). This has two problems: (1) we lose the protections for light dirtiers (2) balance_dirty_pages() effectively becomes IO-less because the (bdi_nr_reclaimable > bdi_thresh) test won't be true. This is good for IO, but balance_dirty_pages() loses an important way to break out of the loop which leads to more spread out throttle delays. DIRTY_PASSGOOD_AREA can eliminate the above issues. The only problem is, DIRTY_PASSGOOD_AREA needs to be defined as 2 to fully cover the above example while this patch uses the more conservative value 8 so as not to surprise people with too many dirty pages than expected. The max-pause limit won't noticeably impact the speed dirty pages are knocked down when there is a sudden drop of global/bdi dirty thresholds. Because the heavy dirties will be throttled below 160KB/s which is slow enough. It does help to avoid long dirty throttle delays and especially will make light dirtiers more responsive. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-20 04:18:42 +00:00
}
static inline bool dtc_is_global(struct dirty_throttle_control *dtc)
{
return mdtc_gdtc(dtc) == NULL;
}
/*
* Dirty background will ignore pages being written as we're trying to
* decide whether to put more under writeback.
*/
static void domain_dirty_avail(struct dirty_throttle_control *dtc,
bool include_writeback)
{
if (dtc_is_global(dtc)) {
dtc->avail = global_dirtyable_memory();
dtc->dirty = global_node_page_state(NR_FILE_DIRTY);
if (include_writeback)
dtc->dirty += global_node_page_state(NR_WRITEBACK);
} else {
unsigned long filepages = 0, headroom = 0, writeback = 0;
mem_cgroup_wb_stats(dtc->wb, &filepages, &headroom, &dtc->dirty,
&writeback);
if (include_writeback)
dtc->dirty += writeback;
mdtc_calc_avail(dtc, filepages, headroom);
}
}
/**
* __wb_calc_thresh - @wb's share of dirty threshold
* @dtc: dirty_throttle_context of interest
* @thresh: dirty throttling or dirty background threshold of wb_domain in @dtc
*
* Note that balance_dirty_pages() will only seriously take dirty throttling
* threshold as a hard limit when sleeping max_pause per page is not enough
* to keep the dirty pages under control. For example, when the device is
* completely stalled due to some error conditions, or when there are 1000
* dd tasks writing to a slow 10MB/s USB key.
* In the other normal situations, it acts more gently by throttling the tasks
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
* more (rather than completely block them) when the wb dirty pages go high.
*
* It allocates high/low dirty limits to fast/slow devices, in order to prevent
* - starving fast devices
* - piling up dirty pages (that will take long time to sync) on slow devices
*
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
* The wb's share of dirty limit will be adapting to its throughput and
* bounded by the bdi->min_ratio and/or bdi->max_ratio parameters, if set.
*
* Return: @wb's dirty limit in pages. For dirty throttling limit, the term
* "dirty" in the context of dirty balancing includes all PG_dirty and
* PG_writeback pages.
*/
static unsigned long __wb_calc_thresh(struct dirty_throttle_control *dtc,
unsigned long thresh)
{
struct wb_domain *dom = dtc_dom(dtc);
u64 wb_thresh;
unsigned long numerator, denominator;
unsigned long wb_min_ratio, wb_max_ratio;
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
/*
* Calculate this wb's share of the thresh ratio.
*/
fprop_fraction_percpu(&dom->completions, dtc->wb_completions,
&numerator, &denominator);
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
wb_thresh = (thresh * (100 * BDI_RATIO_SCALE - bdi_min_ratio)) / (100 * BDI_RATIO_SCALE);
wb_thresh *= numerator;
wb_thresh = div64_ul(wb_thresh, denominator);
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
wb_min_max_ratio(dtc->wb, &wb_min_ratio, &wb_max_ratio);
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
wb_thresh += (thresh * wb_min_ratio) / (100 * BDI_RATIO_SCALE);
if (wb_thresh > (thresh * wb_max_ratio) / (100 * BDI_RATIO_SCALE))
wb_thresh = thresh * wb_max_ratio / (100 * BDI_RATIO_SCALE);
return wb_thresh;
}
unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh)
{
struct dirty_throttle_control gdtc = { GDTC_INIT(wb) };
return __wb_calc_thresh(&gdtc, thresh);
}
writeback: support retrieving per group debug writeback stats of bdi Add /sys/kernel/debug/bdi/xxx/wb_stats to show per group writeback stats of bdi. Following domain hierarchy is tested: global domain (320G) / \ cgroup domain1(10G) cgroup domain2(10G) | | bdi wb1 wb2 /* per wb writeback info of bdi is collected */ cat wb_stats WbCgIno: 1 WbWriteback: 0 kB WbReclaimable: 0 kB WbDirtyThresh: 0 kB WbDirtied: 0 kB WbWritten: 0 kB WbWriteBandwidth: 102400 kBps b_dirty: 0 b_io: 0 b_more_io: 0 b_dirty_time: 0 state: 1 WbCgIno: 4091 WbWriteback: 1792 kB WbReclaimable: 820512 kB WbDirtyThresh: 6004692 kB WbDirtied: 1820448 kB WbWritten: 999488 kB WbWriteBandwidth: 169020 kBps b_dirty: 0 b_io: 0 b_more_io: 1 b_dirty_time: 0 state: 5 WbCgIno: 4131 WbWriteback: 1120 kB WbReclaimable: 820064 kB WbDirtyThresh: 6004728 kB WbDirtied: 1822688 kB WbWritten: 1002400 kB WbWriteBandwidth: 153520 kBps b_dirty: 0 b_io: 0 b_more_io: 1 b_dirty_time: 0 state: 5 [shikemeng@huaweicloud.com: fix build problems] Link: https://lkml.kernel.org/r/20240423034643.141219-4-shikemeng@huaweicloud.com Link: https://lkml.kernel.org/r/20240423034643.141219-3-shikemeng@huaweicloud.com Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com> Cc: Brian Foster <bfoster@redhat.com> Cc: David Howells <dhowells@redhat.com> Cc: David Sterba <dsterba@suse.com> Cc: Jan Kara <jack@suse.cz> Cc: Mateusz Guzik <mjguzik@gmail.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: SeongJae Park <sj@kernel.org> Cc: Stephen Rothwell <sfr@canb.auug.org.au> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-04-23 03:46:40 +00:00
unsigned long cgwb_calc_thresh(struct bdi_writeback *wb)
{
struct dirty_throttle_control gdtc = { GDTC_INIT_NO_WB };
struct dirty_throttle_control mdtc = { MDTC_INIT(wb, &gdtc) };
domain_dirty_avail(&gdtc, true);
domain_dirty_avail(&mdtc, true);
writeback: support retrieving per group debug writeback stats of bdi Add /sys/kernel/debug/bdi/xxx/wb_stats to show per group writeback stats of bdi. Following domain hierarchy is tested: global domain (320G) / \ cgroup domain1(10G) cgroup domain2(10G) | | bdi wb1 wb2 /* per wb writeback info of bdi is collected */ cat wb_stats WbCgIno: 1 WbWriteback: 0 kB WbReclaimable: 0 kB WbDirtyThresh: 0 kB WbDirtied: 0 kB WbWritten: 0 kB WbWriteBandwidth: 102400 kBps b_dirty: 0 b_io: 0 b_more_io: 0 b_dirty_time: 0 state: 1 WbCgIno: 4091 WbWriteback: 1792 kB WbReclaimable: 820512 kB WbDirtyThresh: 6004692 kB WbDirtied: 1820448 kB WbWritten: 999488 kB WbWriteBandwidth: 169020 kBps b_dirty: 0 b_io: 0 b_more_io: 1 b_dirty_time: 0 state: 5 WbCgIno: 4131 WbWriteback: 1120 kB WbReclaimable: 820064 kB WbDirtyThresh: 6004728 kB WbDirtied: 1822688 kB WbWritten: 1002400 kB WbWriteBandwidth: 153520 kBps b_dirty: 0 b_io: 0 b_more_io: 1 b_dirty_time: 0 state: 5 [shikemeng@huaweicloud.com: fix build problems] Link: https://lkml.kernel.org/r/20240423034643.141219-4-shikemeng@huaweicloud.com Link: https://lkml.kernel.org/r/20240423034643.141219-3-shikemeng@huaweicloud.com Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com> Cc: Brian Foster <bfoster@redhat.com> Cc: David Howells <dhowells@redhat.com> Cc: David Sterba <dsterba@suse.com> Cc: Jan Kara <jack@suse.cz> Cc: Mateusz Guzik <mjguzik@gmail.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: SeongJae Park <sj@kernel.org> Cc: Stephen Rothwell <sfr@canb.auug.org.au> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-04-23 03:46:40 +00:00
domain_dirty_limits(&mdtc);
return __wb_calc_thresh(&mdtc, mdtc.thresh);
}
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
/*
* setpoint - dirty 3
* f(dirty) := 1.0 + (----------------)
* limit - setpoint
*
* it's a 3rd order polynomial that subjects to
*
* (1) f(freerun) = 2.0 => rampup dirty_ratelimit reasonably fast
* (2) f(setpoint) = 1.0 => the balance point
* (3) f(limit) = 0 => the hard limit
* (4) df/dx <= 0 => negative feedback control
* (5) the closer to setpoint, the smaller |df/dx| (and the reverse)
* => fast response on large errors; small oscillation near setpoint
*/
static long long pos_ratio_polynom(unsigned long setpoint,
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
unsigned long dirty,
unsigned long limit)
{
long long pos_ratio;
long x;
x = div64_s64(((s64)setpoint - (s64)dirty) << RATELIMIT_CALC_SHIFT,
(limit - setpoint) | 1);
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
pos_ratio = x;
pos_ratio = pos_ratio * x >> RATELIMIT_CALC_SHIFT;
pos_ratio = pos_ratio * x >> RATELIMIT_CALC_SHIFT;
pos_ratio += 1 << RATELIMIT_CALC_SHIFT;
return clamp(pos_ratio, 0LL, 2LL << RATELIMIT_CALC_SHIFT);
}
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
/*
* Dirty position control.
*
* (o) global/bdi setpoints
*
* We want the dirty pages be balanced around the global/wb setpoints.
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
* When the number of dirty pages is higher/lower than the setpoint, the
* dirty position control ratio (and hence task dirty ratelimit) will be
* decreased/increased to bring the dirty pages back to the setpoint.
*
* pos_ratio = 1 << RATELIMIT_CALC_SHIFT
*
* if (dirty < setpoint) scale up pos_ratio
* if (dirty > setpoint) scale down pos_ratio
*
* if (wb_dirty < wb_setpoint) scale up pos_ratio
* if (wb_dirty > wb_setpoint) scale down pos_ratio
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*
* task_ratelimit = dirty_ratelimit * pos_ratio >> RATELIMIT_CALC_SHIFT
*
* (o) global control line
*
* ^ pos_ratio
* |
* | |<===== global dirty control scope ======>|
* 2.0 * * * * * * *
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
* | .*
* | . *
* | . *
* | . *
* | . *
* | . *
* 1.0 ................................*
* | . . *
* | . . *
* | . . *
* | . . *
* | . . *
* 0 +------------.------------------.----------------------*------------->
* freerun^ setpoint^ limit^ dirty pages
*
* (o) wb control line
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*
* ^ pos_ratio
* |
* | *
* | *
* | *
* | *
* | * |<=========== span ============>|
* 1.0 .......................*
* | . *
* | . *
* | . *
* | . *
* | . *
* | . *
* | . *
* | . *
* | . *
* | . *
* | . *
* 1/4 ...............................................* * * * * * * * * * * *
* | . .
* | . .
* | . .
* 0 +----------------------.-------------------------------.------------->
* wb_setpoint^ x_intercept^
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*
* The wb control line won't drop below pos_ratio=1/4, so that wb_dirty can
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
* be smoothly throttled down to normal if it starts high in situations like
* - start writing to a slow SD card and a fast disk at the same time. The SD
* card's wb_dirty may rush to many times higher than wb_setpoint.
* - the wb dirty thresh drops quickly due to change of JBOD workload
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*/
static void wb_position_ratio(struct dirty_throttle_control *dtc)
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
{
struct bdi_writeback *wb = dtc->wb;
unsigned long write_bw = READ_ONCE(wb->avg_write_bandwidth);
unsigned long freerun = dirty_freerun_ceiling(dtc->thresh, dtc->bg_thresh);
unsigned long limit = hard_dirty_limit(dtc_dom(dtc), dtc->thresh);
unsigned long wb_thresh = dtc->wb_thresh;
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
unsigned long x_intercept;
unsigned long setpoint; /* dirty pages' target balance point */
unsigned long wb_setpoint;
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
unsigned long span;
long long pos_ratio; /* for scaling up/down the rate limit */
long x;
dtc->pos_ratio = 0;
if (unlikely(dtc->dirty >= limit))
return;
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
/*
* global setpoint
*
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* See comment for pos_ratio_polynom().
*/
setpoint = (freerun + limit) / 2;
pos_ratio = pos_ratio_polynom(setpoint, dtc->dirty, limit);
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
/*
* The strictlimit feature is a tool preventing mistrusted filesystems
* from growing a large number of dirty pages before throttling. For
* such filesystems balance_dirty_pages always checks wb counters
* against wb limits. Even if global "nr_dirty" is under "freerun".
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* This is especially important for fuse which sets bdi->max_ratio to
* 1% by default. Without strictlimit feature, fuse writeback may
* consume arbitrary amount of RAM because it is accounted in
* NR_WRITEBACK_TEMP which is not involved in calculating "nr_dirty".
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
* Here, in wb_position_ratio(), we calculate pos_ratio based on
* two values: wb_dirty and wb_thresh. Let's consider an example:
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* total amount of RAM is 16GB, bdi->max_ratio is equal to 1%, global
* limits are set by default to 10% and 20% (background and throttle).
* Then wb_thresh is 1% of 20% of 16GB. This amounts to ~8K pages.
* wb_calc_thresh(wb, bg_thresh) is about ~4K pages. wb_setpoint is
* about ~6K pages (as the average of background and throttle wb
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* limits). The 3rd order polynomial will provide positive feedback if
* wb_dirty is under wb_setpoint and vice versa.
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* Note, that we cannot use global counters in these calculations
* because we want to throttle process writing to a strictlimit wb
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* much earlier than global "freerun" is reached (~23MB vs. ~2.3GB
* in the example above).
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*/
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
if (unlikely(wb->bdi->capabilities & BDI_CAP_STRICTLIMIT)) {
long long wb_pos_ratio;
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
if (dtc->wb_dirty < 8) {
dtc->pos_ratio = min_t(long long, pos_ratio * 2,
2 << RATELIMIT_CALC_SHIFT);
return;
}
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
if (dtc->wb_dirty >= wb_thresh)
return;
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
wb_setpoint = dirty_freerun_ceiling(wb_thresh,
dtc->wb_bg_thresh);
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
if (wb_setpoint == 0 || wb_setpoint == wb_thresh)
return;
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
wb_pos_ratio = pos_ratio_polynom(wb_setpoint, dtc->wb_dirty,
wb_thresh);
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
/*
* Typically, for strictlimit case, wb_setpoint << setpoint
* and pos_ratio >> wb_pos_ratio. In the other words global
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* state ("dirty") is not limiting factor and we have to
* make decision based on wb counters. But there is an
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* important case when global pos_ratio should get precedence:
* global limits are exceeded (e.g. due to activities on other
* wb's) while given strictlimit wb is below limit.
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
*
* "pos_ratio * wb_pos_ratio" would work for the case above,
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* but it would look too non-natural for the case of all
* activity in the system coming from a single strictlimit wb
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* with bdi->max_ratio == 100%.
*
* Note that min() below somewhat changes the dynamics of the
* control system. Normally, pos_ratio value can be well over 3
* (when globally we are at freerun and wb is well below wb
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* setpoint). Now the maximum pos_ratio in the same situation
* is 2. We might want to tweak this if we observe the control
* system is too slow to adapt.
*/
dtc->pos_ratio = min(pos_ratio, wb_pos_ratio);
return;
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
}
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
/*
* We have computed basic pos_ratio above based on global situation. If
* the wb is over/under its share of dirty pages, we want to scale
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
* pos_ratio further down/up. That is done by the following mechanism.
*/
/*
* wb setpoint
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*
* f(wb_dirty) := 1.0 + k * (wb_dirty - wb_setpoint)
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*
* x_intercept - wb_dirty
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
* := --------------------------
* x_intercept - wb_setpoint
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*
* The main wb control line is a linear function that subjects to
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*
* (1) f(wb_setpoint) = 1.0
* (2) k = - 1 / (8 * write_bw) (in single wb case)
* or equally: x_intercept = wb_setpoint + 8 * write_bw
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*
* For single wb case, the dirty pages are observed to fluctuate
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
* regularly within range
* [wb_setpoint - write_bw/2, wb_setpoint + write_bw/2]
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
* for various filesystems, where (2) can yield in a reasonable 12.5%
* fluctuation range for pos_ratio.
*
* For JBOD case, wb_thresh (not wb_dirty!) could fluctuate up to its
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
* own size, so move the slope over accordingly and choose a slope that
* yields 100% pos_ratio fluctuation on suddenly doubled wb_thresh.
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*/
if (unlikely(wb_thresh > dtc->thresh))
wb_thresh = dtc->thresh;
/*
* It's very possible that wb_thresh is close to 0 not because the
* device is slow, but that it has remained inactive for long time.
* Honour such devices a reasonable good (hopefully IO efficient)
* threshold, so that the occasional writes won't be blocked and active
* writes can rampup the threshold quickly.
*/
wb_thresh = max(wb_thresh, (limit - dtc->dirty) / 8);
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
/*
* scale global setpoint to wb's:
* wb_setpoint = setpoint * wb_thresh / thresh
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*/
Merge branch 'for-4.2/writeback' of git://git.kernel.dk/linux-block Pull cgroup writeback support from Jens Axboe: "This is the big pull request for adding cgroup writeback support. This code has been in development for a long time, and it has been simmering in for-next for a good chunk of this cycle too. This is one of those problems that has been talked about for at least half a decade, finally there's a solution and code to go with it. Also see last weeks writeup on LWN: http://lwn.net/Articles/648292/" * 'for-4.2/writeback' of git://git.kernel.dk/linux-block: (85 commits) writeback, blkio: add documentation for cgroup writeback support vfs, writeback: replace FS_CGROUP_WRITEBACK with SB_I_CGROUPWB writeback: do foreign inode detection iff cgroup writeback is enabled v9fs: fix error handling in v9fs_session_init() bdi: fix wrong error return value in cgwb_create() buffer: remove unusued 'ret' variable writeback: disassociate inodes from dying bdi_writebacks writeback: implement foreign cgroup inode bdi_writeback switching writeback: add lockdep annotation to inode_to_wb() writeback: use unlocked_inode_to_wb transaction in inode_congested() writeback: implement unlocked_inode_to_wb transaction and use it for stat updates writeback: implement [locked_]inode_to_wb_and_lock_list() writeback: implement foreign cgroup inode detection writeback: make writeback_control track the inode being written back writeback: relocate wb[_try]_get(), wb_put(), inode_{attach|detach}_wb() mm: vmscan: disable memcg direct reclaim stalling if cgroup writeback support is in use writeback: implement memcg writeback domain based throttling writeback: reset wb_domain->dirty_limit[_tstmp] when memcg domain size changes writeback: implement memcg wb_domain writeback: update wb_over_bg_thresh() to use wb_domain aware operations ...
2015-06-25 23:00:17 +00:00
x = div_u64((u64)wb_thresh << 16, dtc->thresh | 1);
wb_setpoint = setpoint * (u64)x >> 16;
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
/*
* Use span=(8*write_bw) in single wb case as indicated by
* (thresh - wb_thresh ~= 0) and transit to wb_thresh in JBOD case.
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*
* wb_thresh thresh - wb_thresh
* span = --------- * (8 * write_bw) + ------------------ * wb_thresh
* thresh thresh
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
*/
span = (dtc->thresh - wb_thresh + 8 * write_bw) * (u64)x >> 16;
x_intercept = wb_setpoint + span;
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
if (dtc->wb_dirty < x_intercept - span / 4) {
pos_ratio = div64_u64(pos_ratio * (x_intercept - dtc->wb_dirty),
Merge branch 'for-4.2/writeback' of git://git.kernel.dk/linux-block Pull cgroup writeback support from Jens Axboe: "This is the big pull request for adding cgroup writeback support. This code has been in development for a long time, and it has been simmering in for-next for a good chunk of this cycle too. This is one of those problems that has been talked about for at least half a decade, finally there's a solution and code to go with it. Also see last weeks writeup on LWN: http://lwn.net/Articles/648292/" * 'for-4.2/writeback' of git://git.kernel.dk/linux-block: (85 commits) writeback, blkio: add documentation for cgroup writeback support vfs, writeback: replace FS_CGROUP_WRITEBACK with SB_I_CGROUPWB writeback: do foreign inode detection iff cgroup writeback is enabled v9fs: fix error handling in v9fs_session_init() bdi: fix wrong error return value in cgwb_create() buffer: remove unusued 'ret' variable writeback: disassociate inodes from dying bdi_writebacks writeback: implement foreign cgroup inode bdi_writeback switching writeback: add lockdep annotation to inode_to_wb() writeback: use unlocked_inode_to_wb transaction in inode_congested() writeback: implement unlocked_inode_to_wb transaction and use it for stat updates writeback: implement [locked_]inode_to_wb_and_lock_list() writeback: implement foreign cgroup inode detection writeback: make writeback_control track the inode being written back writeback: relocate wb[_try]_get(), wb_put(), inode_{attach|detach}_wb() mm: vmscan: disable memcg direct reclaim stalling if cgroup writeback support is in use writeback: implement memcg writeback domain based throttling writeback: reset wb_domain->dirty_limit[_tstmp] when memcg domain size changes writeback: implement memcg wb_domain writeback: update wb_over_bg_thresh() to use wb_domain aware operations ...
2015-06-25 23:00:17 +00:00
(x_intercept - wb_setpoint) | 1);
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
} else
pos_ratio /= 4;
/*
* wb reserve area, safeguard against dirty pool underrun and disk idle
* It may push the desired control point of global dirty pages higher
* than setpoint.
*/
x_intercept = wb_thresh / 2;
if (dtc->wb_dirty < x_intercept) {
if (dtc->wb_dirty > x_intercept / 8)
pos_ratio = div_u64(pos_ratio * x_intercept,
dtc->wb_dirty);
else
pos_ratio *= 8;
}
dtc->pos_ratio = pos_ratio;
writeback: dirty position control bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so that the resulted task rate limit can drive the dirty pages back to the global/bdi setpoints. Old scheme is, | free run area | throttle area ----------------------------------------+----------------------------> thresh^ dirty pages New scheme is, ^ task rate limit | | * | * | * |[free run] * [smooth throttled] | * | * | * ..bdi->dirty_ratelimit..........* | . * | . * | . * | . * | . * +-------------------------------.-----------------------*------------> setpoint^ limit^ dirty pages The slope of the bdi control line should be 1) large enough to pull the dirty pages to setpoint reasonably fast 2) small enough to avoid big fluctuations in the resulted pos_ratio and hence task ratelimit Since the fluctuation range of the bdi dirty pages is typically observed to be within 1-second worth of data, the bdi control line's slope is selected to be a linear function of bdi write bandwidth, so that it can adapt to slow/fast storage devices well. Assume the bdi control line pos_ratio = 1.0 + k * (dirty - bdi_setpoint) where k is the negative slope. If targeting for 12.5% fluctuation range in pos_ratio when dirty pages are fluctuating in range [bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2], we get slope k = - 1 / (8 * write_bw) Let pos_ratio(x_intercept) = 0, we get the parameter used in code: x_intercept = bdi_setpoint + 8 * write_bw The global/bdi slopes are nicely complementing each other when the system has only one major bdi (indicated by bdi_thresh ~= thresh): 1) slope of global control line => scaling to the control scope size 2) slope of main bdi control line => scaling to the writeout bandwidth so that - in memory tight systems, (1) becomes strong enough to squeeze dirty pages inside the control scope - in large memory systems where the "gravity" of (1) for pulling the dirty pages to setpoint is too weak, (2) can back (1) up and drive dirty pages to bdi_setpoint ~= setpoint reasonably fast. Unfortunately in JBOD setups, the fluctuation range of bdi threshold is related to memory size due to the interferences between disks. In this case, the bdi slope will be weighted sum of write_bw and bdi_thresh. Given equations span = x_intercept - bdi_setpoint k = df/dx = - 1 / span and the extremum values span = bdi_thresh dx = bdi_thresh we get df = - dx / span = - 1.0 That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence task ratelimit will fluctuate by -100%. peter: use 3rd order polynomial for the global control line CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 22:04:18 +00:00
}
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
static void wb_update_write_bandwidth(struct bdi_writeback *wb,
unsigned long elapsed,
unsigned long written)
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
{
const unsigned long period = roundup_pow_of_two(3 * HZ);
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
unsigned long avg = wb->avg_write_bandwidth;
unsigned long old = wb->write_bandwidth;
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
u64 bw;
/*
* bw = written * HZ / elapsed
*
* bw * elapsed + write_bandwidth * (period - elapsed)
* write_bandwidth = ---------------------------------------------------
* period
*
* @written may have decreased due to folio_redirty_for_writepage().
* Avoid underflowing @bw calculation.
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
*/
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
bw = written - min(written, wb->written_stamp);
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
bw *= HZ;
if (unlikely(elapsed > period)) {
bw = div64_ul(bw, elapsed);
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
avg = bw;
goto out;
}
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
bw += (u64)wb->write_bandwidth * (period - elapsed);
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
bw >>= ilog2(period);
/*
* one more level of smoothing, for filtering out sudden spikes
*/
if (avg > old && old >= (unsigned long)bw)
avg -= (avg - old) >> 3;
if (avg < old && old <= (unsigned long)bw)
avg += (old - avg) >> 3;
out:
/* keep avg > 0 to guarantee that tot > 0 if there are dirty wbs */
avg = max(avg, 1LU);
if (wb_has_dirty_io(wb)) {
long delta = avg - wb->avg_write_bandwidth;
WARN_ON_ONCE(atomic_long_add_return(delta,
&wb->bdi->tot_write_bandwidth) <= 0);
}
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
wb->write_bandwidth = bw;
WRITE_ONCE(wb->avg_write_bandwidth, avg);
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
}
static void update_dirty_limit(struct dirty_throttle_control *dtc)
writeback: introduce smoothed global dirty limit The start of a heavy weight application (ie. KVM) may instantly knock down determine_dirtyable_memory() if the swap is not enabled or full. global_dirty_limits() and bdi_dirty_limit() will in turn get global/bdi dirty thresholds that are _much_ lower than the global/bdi dirty pages. balance_dirty_pages() will then heavily throttle all dirtiers including the light ones, until the dirty pages drop below the new dirty thresholds. During this _deep_ dirty-exceeded state, the system may appear rather unresponsive to the users. About "deep" dirty-exceeded: task_dirty_limit() assigns 1/8 lower dirty threshold to heavy dirtiers than light ones, and the dirty pages will be throttled around the heavy dirtiers' dirty threshold and reasonably below the light dirtiers' dirty threshold. In this state, only the heavy dirtiers will be throttled and the dirty pages are carefully controlled to not exceed the light dirtiers' dirty threshold. However if the threshold itself suddenly drops below the number of dirty pages, the light dirtiers will get heavily throttled. So introduce global_dirty_limit for tracking the global dirty threshold with policies - follow downwards slowly - follow up in one shot global_dirty_limit can effectively mask out the impact of sudden drop of dirtyable memory. It will be used in the next patch for two new type of dirty limits. Note that the new dirty limits are not going to avoid throttling the light dirtiers, but could limit their sleep time to 200ms. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 21:54:09 +00:00
{
struct wb_domain *dom = dtc_dom(dtc);
unsigned long thresh = dtc->thresh;
unsigned long limit = dom->dirty_limit;
writeback: introduce smoothed global dirty limit The start of a heavy weight application (ie. KVM) may instantly knock down determine_dirtyable_memory() if the swap is not enabled or full. global_dirty_limits() and bdi_dirty_limit() will in turn get global/bdi dirty thresholds that are _much_ lower than the global/bdi dirty pages. balance_dirty_pages() will then heavily throttle all dirtiers including the light ones, until the dirty pages drop below the new dirty thresholds. During this _deep_ dirty-exceeded state, the system may appear rather unresponsive to the users. About "deep" dirty-exceeded: task_dirty_limit() assigns 1/8 lower dirty threshold to heavy dirtiers than light ones, and the dirty pages will be throttled around the heavy dirtiers' dirty threshold and reasonably below the light dirtiers' dirty threshold. In this state, only the heavy dirtiers will be throttled and the dirty pages are carefully controlled to not exceed the light dirtiers' dirty threshold. However if the threshold itself suddenly drops below the number of dirty pages, the light dirtiers will get heavily throttled. So introduce global_dirty_limit for tracking the global dirty threshold with policies - follow downwards slowly - follow up in one shot global_dirty_limit can effectively mask out the impact of sudden drop of dirtyable memory. It will be used in the next patch for two new type of dirty limits. Note that the new dirty limits are not going to avoid throttling the light dirtiers, but could limit their sleep time to 200ms. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 21:54:09 +00:00
/*
* Follow up in one step.
*/
if (limit < thresh) {
limit = thresh;
goto update;
}
/*
* Follow down slowly. Use the higher one as the target, because thresh
* may drop below dirty. This is exactly the reason to introduce
* dom->dirty_limit which is guaranteed to lie above the dirty pages.
writeback: introduce smoothed global dirty limit The start of a heavy weight application (ie. KVM) may instantly knock down determine_dirtyable_memory() if the swap is not enabled or full. global_dirty_limits() and bdi_dirty_limit() will in turn get global/bdi dirty thresholds that are _much_ lower than the global/bdi dirty pages. balance_dirty_pages() will then heavily throttle all dirtiers including the light ones, until the dirty pages drop below the new dirty thresholds. During this _deep_ dirty-exceeded state, the system may appear rather unresponsive to the users. About "deep" dirty-exceeded: task_dirty_limit() assigns 1/8 lower dirty threshold to heavy dirtiers than light ones, and the dirty pages will be throttled around the heavy dirtiers' dirty threshold and reasonably below the light dirtiers' dirty threshold. In this state, only the heavy dirtiers will be throttled and the dirty pages are carefully controlled to not exceed the light dirtiers' dirty threshold. However if the threshold itself suddenly drops below the number of dirty pages, the light dirtiers will get heavily throttled. So introduce global_dirty_limit for tracking the global dirty threshold with policies - follow downwards slowly - follow up in one shot global_dirty_limit can effectively mask out the impact of sudden drop of dirtyable memory. It will be used in the next patch for two new type of dirty limits. Note that the new dirty limits are not going to avoid throttling the light dirtiers, but could limit their sleep time to 200ms. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 21:54:09 +00:00
*/
thresh = max(thresh, dtc->dirty);
writeback: introduce smoothed global dirty limit The start of a heavy weight application (ie. KVM) may instantly knock down determine_dirtyable_memory() if the swap is not enabled or full. global_dirty_limits() and bdi_dirty_limit() will in turn get global/bdi dirty thresholds that are _much_ lower than the global/bdi dirty pages. balance_dirty_pages() will then heavily throttle all dirtiers including the light ones, until the dirty pages drop below the new dirty thresholds. During this _deep_ dirty-exceeded state, the system may appear rather unresponsive to the users. About "deep" dirty-exceeded: task_dirty_limit() assigns 1/8 lower dirty threshold to heavy dirtiers than light ones, and the dirty pages will be throttled around the heavy dirtiers' dirty threshold and reasonably below the light dirtiers' dirty threshold. In this state, only the heavy dirtiers will be throttled and the dirty pages are carefully controlled to not exceed the light dirtiers' dirty threshold. However if the threshold itself suddenly drops below the number of dirty pages, the light dirtiers will get heavily throttled. So introduce global_dirty_limit for tracking the global dirty threshold with policies - follow downwards slowly - follow up in one shot global_dirty_limit can effectively mask out the impact of sudden drop of dirtyable memory. It will be used in the next patch for two new type of dirty limits. Note that the new dirty limits are not going to avoid throttling the light dirtiers, but could limit their sleep time to 200ms. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 21:54:09 +00:00
if (limit > thresh) {
limit -= (limit - thresh) >> 5;
goto update;
}
return;
update:
dom->dirty_limit = limit;
writeback: introduce smoothed global dirty limit The start of a heavy weight application (ie. KVM) may instantly knock down determine_dirtyable_memory() if the swap is not enabled or full. global_dirty_limits() and bdi_dirty_limit() will in turn get global/bdi dirty thresholds that are _much_ lower than the global/bdi dirty pages. balance_dirty_pages() will then heavily throttle all dirtiers including the light ones, until the dirty pages drop below the new dirty thresholds. During this _deep_ dirty-exceeded state, the system may appear rather unresponsive to the users. About "deep" dirty-exceeded: task_dirty_limit() assigns 1/8 lower dirty threshold to heavy dirtiers than light ones, and the dirty pages will be throttled around the heavy dirtiers' dirty threshold and reasonably below the light dirtiers' dirty threshold. In this state, only the heavy dirtiers will be throttled and the dirty pages are carefully controlled to not exceed the light dirtiers' dirty threshold. However if the threshold itself suddenly drops below the number of dirty pages, the light dirtiers will get heavily throttled. So introduce global_dirty_limit for tracking the global dirty threshold with policies - follow downwards slowly - follow up in one shot global_dirty_limit can effectively mask out the impact of sudden drop of dirtyable memory. It will be used in the next patch for two new type of dirty limits. Note that the new dirty limits are not going to avoid throttling the light dirtiers, but could limit their sleep time to 200ms. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 21:54:09 +00:00
}
static void domain_update_dirty_limit(struct dirty_throttle_control *dtc,
unsigned long now)
writeback: introduce smoothed global dirty limit The start of a heavy weight application (ie. KVM) may instantly knock down determine_dirtyable_memory() if the swap is not enabled or full. global_dirty_limits() and bdi_dirty_limit() will in turn get global/bdi dirty thresholds that are _much_ lower than the global/bdi dirty pages. balance_dirty_pages() will then heavily throttle all dirtiers including the light ones, until the dirty pages drop below the new dirty thresholds. During this _deep_ dirty-exceeded state, the system may appear rather unresponsive to the users. About "deep" dirty-exceeded: task_dirty_limit() assigns 1/8 lower dirty threshold to heavy dirtiers than light ones, and the dirty pages will be throttled around the heavy dirtiers' dirty threshold and reasonably below the light dirtiers' dirty threshold. In this state, only the heavy dirtiers will be throttled and the dirty pages are carefully controlled to not exceed the light dirtiers' dirty threshold. However if the threshold itself suddenly drops below the number of dirty pages, the light dirtiers will get heavily throttled. So introduce global_dirty_limit for tracking the global dirty threshold with policies - follow downwards slowly - follow up in one shot global_dirty_limit can effectively mask out the impact of sudden drop of dirtyable memory. It will be used in the next patch for two new type of dirty limits. Note that the new dirty limits are not going to avoid throttling the light dirtiers, but could limit their sleep time to 200ms. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 21:54:09 +00:00
{
struct wb_domain *dom = dtc_dom(dtc);
writeback: introduce smoothed global dirty limit The start of a heavy weight application (ie. KVM) may instantly knock down determine_dirtyable_memory() if the swap is not enabled or full. global_dirty_limits() and bdi_dirty_limit() will in turn get global/bdi dirty thresholds that are _much_ lower than the global/bdi dirty pages. balance_dirty_pages() will then heavily throttle all dirtiers including the light ones, until the dirty pages drop below the new dirty thresholds. During this _deep_ dirty-exceeded state, the system may appear rather unresponsive to the users. About "deep" dirty-exceeded: task_dirty_limit() assigns 1/8 lower dirty threshold to heavy dirtiers than light ones, and the dirty pages will be throttled around the heavy dirtiers' dirty threshold and reasonably below the light dirtiers' dirty threshold. In this state, only the heavy dirtiers will be throttled and the dirty pages are carefully controlled to not exceed the light dirtiers' dirty threshold. However if the threshold itself suddenly drops below the number of dirty pages, the light dirtiers will get heavily throttled. So introduce global_dirty_limit for tracking the global dirty threshold with policies - follow downwards slowly - follow up in one shot global_dirty_limit can effectively mask out the impact of sudden drop of dirtyable memory. It will be used in the next patch for two new type of dirty limits. Note that the new dirty limits are not going to avoid throttling the light dirtiers, but could limit their sleep time to 200ms. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 21:54:09 +00:00
/*
* check locklessly first to optimize away locking for the most time
*/
if (time_before(now, dom->dirty_limit_tstamp + BANDWIDTH_INTERVAL))
writeback: introduce smoothed global dirty limit The start of a heavy weight application (ie. KVM) may instantly knock down determine_dirtyable_memory() if the swap is not enabled or full. global_dirty_limits() and bdi_dirty_limit() will in turn get global/bdi dirty thresholds that are _much_ lower than the global/bdi dirty pages. balance_dirty_pages() will then heavily throttle all dirtiers including the light ones, until the dirty pages drop below the new dirty thresholds. During this _deep_ dirty-exceeded state, the system may appear rather unresponsive to the users. About "deep" dirty-exceeded: task_dirty_limit() assigns 1/8 lower dirty threshold to heavy dirtiers than light ones, and the dirty pages will be throttled around the heavy dirtiers' dirty threshold and reasonably below the light dirtiers' dirty threshold. In this state, only the heavy dirtiers will be throttled and the dirty pages are carefully controlled to not exceed the light dirtiers' dirty threshold. However if the threshold itself suddenly drops below the number of dirty pages, the light dirtiers will get heavily throttled. So introduce global_dirty_limit for tracking the global dirty threshold with policies - follow downwards slowly - follow up in one shot global_dirty_limit can effectively mask out the impact of sudden drop of dirtyable memory. It will be used in the next patch for two new type of dirty limits. Note that the new dirty limits are not going to avoid throttling the light dirtiers, but could limit their sleep time to 200ms. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 21:54:09 +00:00
return;
spin_lock(&dom->lock);
if (time_after_eq(now, dom->dirty_limit_tstamp + BANDWIDTH_INTERVAL)) {
update_dirty_limit(dtc);
dom->dirty_limit_tstamp = now;
writeback: introduce smoothed global dirty limit The start of a heavy weight application (ie. KVM) may instantly knock down determine_dirtyable_memory() if the swap is not enabled or full. global_dirty_limits() and bdi_dirty_limit() will in turn get global/bdi dirty thresholds that are _much_ lower than the global/bdi dirty pages. balance_dirty_pages() will then heavily throttle all dirtiers including the light ones, until the dirty pages drop below the new dirty thresholds. During this _deep_ dirty-exceeded state, the system may appear rather unresponsive to the users. About "deep" dirty-exceeded: task_dirty_limit() assigns 1/8 lower dirty threshold to heavy dirtiers than light ones, and the dirty pages will be throttled around the heavy dirtiers' dirty threshold and reasonably below the light dirtiers' dirty threshold. In this state, only the heavy dirtiers will be throttled and the dirty pages are carefully controlled to not exceed the light dirtiers' dirty threshold. However if the threshold itself suddenly drops below the number of dirty pages, the light dirtiers will get heavily throttled. So introduce global_dirty_limit for tracking the global dirty threshold with policies - follow downwards slowly - follow up in one shot global_dirty_limit can effectively mask out the impact of sudden drop of dirtyable memory. It will be used in the next patch for two new type of dirty limits. Note that the new dirty limits are not going to avoid throttling the light dirtiers, but could limit their sleep time to 200ms. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 21:54:09 +00:00
}
spin_unlock(&dom->lock);
writeback: introduce smoothed global dirty limit The start of a heavy weight application (ie. KVM) may instantly knock down determine_dirtyable_memory() if the swap is not enabled or full. global_dirty_limits() and bdi_dirty_limit() will in turn get global/bdi dirty thresholds that are _much_ lower than the global/bdi dirty pages. balance_dirty_pages() will then heavily throttle all dirtiers including the light ones, until the dirty pages drop below the new dirty thresholds. During this _deep_ dirty-exceeded state, the system may appear rather unresponsive to the users. About "deep" dirty-exceeded: task_dirty_limit() assigns 1/8 lower dirty threshold to heavy dirtiers than light ones, and the dirty pages will be throttled around the heavy dirtiers' dirty threshold and reasonably below the light dirtiers' dirty threshold. In this state, only the heavy dirtiers will be throttled and the dirty pages are carefully controlled to not exceed the light dirtiers' dirty threshold. However if the threshold itself suddenly drops below the number of dirty pages, the light dirtiers will get heavily throttled. So introduce global_dirty_limit for tracking the global dirty threshold with policies - follow downwards slowly - follow up in one shot global_dirty_limit can effectively mask out the impact of sudden drop of dirtyable memory. It will be used in the next patch for two new type of dirty limits. Note that the new dirty limits are not going to avoid throttling the light dirtiers, but could limit their sleep time to 200ms. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-03-02 21:54:09 +00:00
}
writeback: dirty rate control It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N) when there are N dd tasks. On write() syscall, use bdi->dirty_ratelimit ============================================ balance_dirty_pages(pages_dirtied) { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); pause = pages_dirtied / task_ratelimit; sleep(pause); } On every 200ms, update bdi->dirty_ratelimit =========================================== bdi_update_dirty_ratelimit() { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate; bdi->dirty_ratelimit = balanced_dirty_ratelimit } Estimation of balanced bdi->dirty_ratelimit =========================================== balanced task_ratelimit ----------------------- balance_dirty_pages() needs to throttle tasks dirtying pages such that the total amount of dirty pages stays below the specified dirty limit in order to avoid memory deadlocks. Furthermore we desire fairness in that tasks get throttled proportionally to the amount of pages they dirty. IOW we want to throttle tasks such that we match the dirty rate to the writeout bandwidth, this yields a stable amount of dirty pages: dirty_rate == write_bw (1) The fairness requirement gives us: task_ratelimit = balanced_dirty_ratelimit == write_bw / N (2) where N is the number of dd tasks. We don't know N beforehand, but still can estimate balanced_dirty_ratelimit within 200ms. Start by throttling each dd task at rate task_ratelimit = task_ratelimit_0 (3) (any non-zero initial value is OK) After 200ms, we measured dirty_rate = # of pages dirtied by all dd's / 200ms write_bw = # of pages written to the disk / 200ms For the aggressive dd dirtiers, the equality holds dirty_rate == N * task_rate == N * task_ratelimit_0 (4) Or task_ratelimit_0 == dirty_rate / N (5) Now we conclude that the balanced task ratelimit can be estimated by write_bw balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6) dirty_rate Because with (4) and (5) we can get the desired equality (1): write_bw balanced_dirty_ratelimit == (dirty_rate / N) * ---------- dirty_rate == write_bw / N Then using the balanced task ratelimit we can compute task pause times like: task_pause = task->nr_dirtied / task_ratelimit task_ratelimit with position control ------------------------------------ However, while the above gives us means of matching the dirty rate to the writeout bandwidth, it at best provides us with a stable dirty page count (assuming a static system). In order to control the dirty page count such that it is high enough to provide performance, but does not exceed the specified limit we need another control. The dirty position control works by extending (2) to task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7) where pos_ratio is a negative feedback function that subjects to 1) f(setpoint) = 1.0 2) df/dx < 0 That is, if the dirty pages are ABOVE the setpoint, we throttle each task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty pages are created less fast than they are cleaned, thus DROP to the setpoints (and the reverse). Based on (7) and the assumption that both dirty_ratelimit and pos_ratio remains CONSTANT for the past 200ms, we get task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8) Putting (8) into (6), we get the formula used in bdi_update_dirty_ratelimit(): write_bw balanced_dirty_ratelimit *= pos_ratio * ---------- (9) dirty_rate Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 16:51:31 +00:00
/*
* Maintain wb->dirty_ratelimit, the base dirty throttle rate.
writeback: dirty rate control It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N) when there are N dd tasks. On write() syscall, use bdi->dirty_ratelimit ============================================ balance_dirty_pages(pages_dirtied) { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); pause = pages_dirtied / task_ratelimit; sleep(pause); } On every 200ms, update bdi->dirty_ratelimit =========================================== bdi_update_dirty_ratelimit() { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate; bdi->dirty_ratelimit = balanced_dirty_ratelimit } Estimation of balanced bdi->dirty_ratelimit =========================================== balanced task_ratelimit ----------------------- balance_dirty_pages() needs to throttle tasks dirtying pages such that the total amount of dirty pages stays below the specified dirty limit in order to avoid memory deadlocks. Furthermore we desire fairness in that tasks get throttled proportionally to the amount of pages they dirty. IOW we want to throttle tasks such that we match the dirty rate to the writeout bandwidth, this yields a stable amount of dirty pages: dirty_rate == write_bw (1) The fairness requirement gives us: task_ratelimit = balanced_dirty_ratelimit == write_bw / N (2) where N is the number of dd tasks. We don't know N beforehand, but still can estimate balanced_dirty_ratelimit within 200ms. Start by throttling each dd task at rate task_ratelimit = task_ratelimit_0 (3) (any non-zero initial value is OK) After 200ms, we measured dirty_rate = # of pages dirtied by all dd's / 200ms write_bw = # of pages written to the disk / 200ms For the aggressive dd dirtiers, the equality holds dirty_rate == N * task_rate == N * task_ratelimit_0 (4) Or task_ratelimit_0 == dirty_rate / N (5) Now we conclude that the balanced task ratelimit can be estimated by write_bw balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6) dirty_rate Because with (4) and (5) we can get the desired equality (1): write_bw balanced_dirty_ratelimit == (dirty_rate / N) * ---------- dirty_rate == write_bw / N Then using the balanced task ratelimit we can compute task pause times like: task_pause = task->nr_dirtied / task_ratelimit task_ratelimit with position control ------------------------------------ However, while the above gives us means of matching the dirty rate to the writeout bandwidth, it at best provides us with a stable dirty page count (assuming a static system). In order to control the dirty page count such that it is high enough to provide performance, but does not exceed the specified limit we need another control. The dirty position control works by extending (2) to task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7) where pos_ratio is a negative feedback function that subjects to 1) f(setpoint) = 1.0 2) df/dx < 0 That is, if the dirty pages are ABOVE the setpoint, we throttle each task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty pages are created less fast than they are cleaned, thus DROP to the setpoints (and the reverse). Based on (7) and the assumption that both dirty_ratelimit and pos_ratio remains CONSTANT for the past 200ms, we get task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8) Putting (8) into (6), we get the formula used in bdi_update_dirty_ratelimit(): write_bw balanced_dirty_ratelimit *= pos_ratio * ---------- (9) dirty_rate Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 16:51:31 +00:00
*
* Normal wb tasks will be curbed at or below it in long term.
writeback: dirty rate control It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N) when there are N dd tasks. On write() syscall, use bdi->dirty_ratelimit ============================================ balance_dirty_pages(pages_dirtied) { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); pause = pages_dirtied / task_ratelimit; sleep(pause); } On every 200ms, update bdi->dirty_ratelimit =========================================== bdi_update_dirty_ratelimit() { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate; bdi->dirty_ratelimit = balanced_dirty_ratelimit } Estimation of balanced bdi->dirty_ratelimit =========================================== balanced task_ratelimit ----------------------- balance_dirty_pages() needs to throttle tasks dirtying pages such that the total amount of dirty pages stays below the specified dirty limit in order to avoid memory deadlocks. Furthermore we desire fairness in that tasks get throttled proportionally to the amount of pages they dirty. IOW we want to throttle tasks such that we match the dirty rate to the writeout bandwidth, this yields a stable amount of dirty pages: dirty_rate == write_bw (1) The fairness requirement gives us: task_ratelimit = balanced_dirty_ratelimit == write_bw / N (2) where N is the number of dd tasks. We don't know N beforehand, but still can estimate balanced_dirty_ratelimit within 200ms. Start by throttling each dd task at rate task_ratelimit = task_ratelimit_0 (3) (any non-zero initial value is OK) After 200ms, we measured dirty_rate = # of pages dirtied by all dd's / 200ms write_bw = # of pages written to the disk / 200ms For the aggressive dd dirtiers, the equality holds dirty_rate == N * task_rate == N * task_ratelimit_0 (4) Or task_ratelimit_0 == dirty_rate / N (5) Now we conclude that the balanced task ratelimit can be estimated by write_bw balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6) dirty_rate Because with (4) and (5) we can get the desired equality (1): write_bw balanced_dirty_ratelimit == (dirty_rate / N) * ---------- dirty_rate == write_bw / N Then using the balanced task ratelimit we can compute task pause times like: task_pause = task->nr_dirtied / task_ratelimit task_ratelimit with position control ------------------------------------ However, while the above gives us means of matching the dirty rate to the writeout bandwidth, it at best provides us with a stable dirty page count (assuming a static system). In order to control the dirty page count such that it is high enough to provide performance, but does not exceed the specified limit we need another control. The dirty position control works by extending (2) to task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7) where pos_ratio is a negative feedback function that subjects to 1) f(setpoint) = 1.0 2) df/dx < 0 That is, if the dirty pages are ABOVE the setpoint, we throttle each task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty pages are created less fast than they are cleaned, thus DROP to the setpoints (and the reverse). Based on (7) and the assumption that both dirty_ratelimit and pos_ratio remains CONSTANT for the past 200ms, we get task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8) Putting (8) into (6), we get the formula used in bdi_update_dirty_ratelimit(): write_bw balanced_dirty_ratelimit *= pos_ratio * ---------- (9) dirty_rate Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 16:51:31 +00:00
* Obviously it should be around (write_bw / N) when there are N dd tasks.
*/
static void wb_update_dirty_ratelimit(struct dirty_throttle_control *dtc,
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
unsigned long dirtied,
unsigned long elapsed)
writeback: dirty rate control It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N) when there are N dd tasks. On write() syscall, use bdi->dirty_ratelimit ============================================ balance_dirty_pages(pages_dirtied) { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); pause = pages_dirtied / task_ratelimit; sleep(pause); } On every 200ms, update bdi->dirty_ratelimit =========================================== bdi_update_dirty_ratelimit() { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate; bdi->dirty_ratelimit = balanced_dirty_ratelimit } Estimation of balanced bdi->dirty_ratelimit =========================================== balanced task_ratelimit ----------------------- balance_dirty_pages() needs to throttle tasks dirtying pages such that the total amount of dirty pages stays below the specified dirty limit in order to avoid memory deadlocks. Furthermore we desire fairness in that tasks get throttled proportionally to the amount of pages they dirty. IOW we want to throttle tasks such that we match the dirty rate to the writeout bandwidth, this yields a stable amount of dirty pages: dirty_rate == write_bw (1) The fairness requirement gives us: task_ratelimit = balanced_dirty_ratelimit == write_bw / N (2) where N is the number of dd tasks. We don't know N beforehand, but still can estimate balanced_dirty_ratelimit within 200ms. Start by throttling each dd task at rate task_ratelimit = task_ratelimit_0 (3) (any non-zero initial value is OK) After 200ms, we measured dirty_rate = # of pages dirtied by all dd's / 200ms write_bw = # of pages written to the disk / 200ms For the aggressive dd dirtiers, the equality holds dirty_rate == N * task_rate == N * task_ratelimit_0 (4) Or task_ratelimit_0 == dirty_rate / N (5) Now we conclude that the balanced task ratelimit can be estimated by write_bw balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6) dirty_rate Because with (4) and (5) we can get the desired equality (1): write_bw balanced_dirty_ratelimit == (dirty_rate / N) * ---------- dirty_rate == write_bw / N Then using the balanced task ratelimit we can compute task pause times like: task_pause = task->nr_dirtied / task_ratelimit task_ratelimit with position control ------------------------------------ However, while the above gives us means of matching the dirty rate to the writeout bandwidth, it at best provides us with a stable dirty page count (assuming a static system). In order to control the dirty page count such that it is high enough to provide performance, but does not exceed the specified limit we need another control. The dirty position control works by extending (2) to task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7) where pos_ratio is a negative feedback function that subjects to 1) f(setpoint) = 1.0 2) df/dx < 0 That is, if the dirty pages are ABOVE the setpoint, we throttle each task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty pages are created less fast than they are cleaned, thus DROP to the setpoints (and the reverse). Based on (7) and the assumption that both dirty_ratelimit and pos_ratio remains CONSTANT for the past 200ms, we get task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8) Putting (8) into (6), we get the formula used in bdi_update_dirty_ratelimit(): write_bw balanced_dirty_ratelimit *= pos_ratio * ---------- (9) dirty_rate Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 16:51:31 +00:00
{
struct bdi_writeback *wb = dtc->wb;
unsigned long dirty = dtc->dirty;
unsigned long freerun = dirty_freerun_ceiling(dtc->thresh, dtc->bg_thresh);
unsigned long limit = hard_dirty_limit(dtc_dom(dtc), dtc->thresh);
writeback: stabilize bdi->dirty_ratelimit There are some imperfections in balanced_dirty_ratelimit. 1) large fluctuations The dirty_rate used for computing balanced_dirty_ratelimit is merely averaged in the past 200ms (very small comparing to the 3s estimation period for write_bw), which makes rather dispersed distribution of balanced_dirty_ratelimit. It's pretty hard to average out the singular points by increasing the estimation period. Considering that the averaging technique will introduce very undesirable time lags, I give it up totally. (btw, the 3s write_bw averaging time lag is much more acceptable because its impact is one-way and therefore won't lead to oscillations.) The more practical way is filtering -- most singular balanced_dirty_ratelimit points can be filtered out by remembering some prev_balanced_rate and prev_prev_balanced_rate. However the more reliable way is to guard balanced_dirty_ratelimit with task_ratelimit. 2) due to truncates and fs redirties, the (write_bw <=> dirty_rate) match could become unbalanced, which may lead to large systematical errors in balanced_dirty_ratelimit. The truncates, due to its possibly bumpy nature, can hardly be compensated smoothly. So let's face it. When some over-estimated balanced_dirty_ratelimit brings dirty_ratelimit high, dirty pages will go higher than the setpoint. task_ratelimit will in turn become lower than dirty_ratelimit. So if we consider both balanced_dirty_ratelimit and task_ratelimit and update dirty_ratelimit only when they are on the same side of dirty_ratelimit, the systematical errors in balanced_dirty_ratelimit won't be able to bring dirty_ratelimit far away. The balanced_dirty_ratelimit estimation may also be inaccurate near @limit or @freerun, however is less an issue. 3) since we ultimately want to - keep the fluctuations of task ratelimit as small as possible - keep the dirty pages around the setpoint as long time as possible the update policy used for (2) also serves the above goals nicely: if for some reason the dirty pages are high (task_ratelimit < dirty_ratelimit), and dirty_ratelimit is low (dirty_ratelimit < balanced_dirty_ratelimit), there is no point to bring up dirty_ratelimit in a hurry only to hurt both the above two goals. So, we make use of task_ratelimit to limit the update of dirty_ratelimit in two ways: 1) avoid changing dirty rate when it's against the position control target (the adjusted rate will slow down the progress of dirty pages going back to setpoint). 2) limit the step size. task_ratelimit is changing values step by step, leaving a consistent trace comparing to the randomly jumping balanced_dirty_ratelimit. task_ratelimit also has the nice smaller errors in stable state and typically larger errors when there are big errors in rate. So it's a pretty good limiting factor for the step size of dirty_ratelimit. Note that bdi->dirty_ratelimit is always tracking balanced_dirty_ratelimit. task_ratelimit is merely used as a limiting factor. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-08-26 21:53:24 +00:00
unsigned long setpoint = (freerun + limit) / 2;
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
unsigned long write_bw = wb->avg_write_bandwidth;
unsigned long dirty_ratelimit = wb->dirty_ratelimit;
writeback: dirty rate control It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N) when there are N dd tasks. On write() syscall, use bdi->dirty_ratelimit ============================================ balance_dirty_pages(pages_dirtied) { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); pause = pages_dirtied / task_ratelimit; sleep(pause); } On every 200ms, update bdi->dirty_ratelimit =========================================== bdi_update_dirty_ratelimit() { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate; bdi->dirty_ratelimit = balanced_dirty_ratelimit } Estimation of balanced bdi->dirty_ratelimit =========================================== balanced task_ratelimit ----------------------- balance_dirty_pages() needs to throttle tasks dirtying pages such that the total amount of dirty pages stays below the specified dirty limit in order to avoid memory deadlocks. Furthermore we desire fairness in that tasks get throttled proportionally to the amount of pages they dirty. IOW we want to throttle tasks such that we match the dirty rate to the writeout bandwidth, this yields a stable amount of dirty pages: dirty_rate == write_bw (1) The fairness requirement gives us: task_ratelimit = balanced_dirty_ratelimit == write_bw / N (2) where N is the number of dd tasks. We don't know N beforehand, but still can estimate balanced_dirty_ratelimit within 200ms. Start by throttling each dd task at rate task_ratelimit = task_ratelimit_0 (3) (any non-zero initial value is OK) After 200ms, we measured dirty_rate = # of pages dirtied by all dd's / 200ms write_bw = # of pages written to the disk / 200ms For the aggressive dd dirtiers, the equality holds dirty_rate == N * task_rate == N * task_ratelimit_0 (4) Or task_ratelimit_0 == dirty_rate / N (5) Now we conclude that the balanced task ratelimit can be estimated by write_bw balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6) dirty_rate Because with (4) and (5) we can get the desired equality (1): write_bw balanced_dirty_ratelimit == (dirty_rate / N) * ---------- dirty_rate == write_bw / N Then using the balanced task ratelimit we can compute task pause times like: task_pause = task->nr_dirtied / task_ratelimit task_ratelimit with position control ------------------------------------ However, while the above gives us means of matching the dirty rate to the writeout bandwidth, it at best provides us with a stable dirty page count (assuming a static system). In order to control the dirty page count such that it is high enough to provide performance, but does not exceed the specified limit we need another control. The dirty position control works by extending (2) to task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7) where pos_ratio is a negative feedback function that subjects to 1) f(setpoint) = 1.0 2) df/dx < 0 That is, if the dirty pages are ABOVE the setpoint, we throttle each task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty pages are created less fast than they are cleaned, thus DROP to the setpoints (and the reverse). Based on (7) and the assumption that both dirty_ratelimit and pos_ratio remains CONSTANT for the past 200ms, we get task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8) Putting (8) into (6), we get the formula used in bdi_update_dirty_ratelimit(): write_bw balanced_dirty_ratelimit *= pos_ratio * ---------- (9) dirty_rate Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 16:51:31 +00:00
unsigned long dirty_rate;
unsigned long task_ratelimit;
unsigned long balanced_dirty_ratelimit;
writeback: stabilize bdi->dirty_ratelimit There are some imperfections in balanced_dirty_ratelimit. 1) large fluctuations The dirty_rate used for computing balanced_dirty_ratelimit is merely averaged in the past 200ms (very small comparing to the 3s estimation period for write_bw), which makes rather dispersed distribution of balanced_dirty_ratelimit. It's pretty hard to average out the singular points by increasing the estimation period. Considering that the averaging technique will introduce very undesirable time lags, I give it up totally. (btw, the 3s write_bw averaging time lag is much more acceptable because its impact is one-way and therefore won't lead to oscillations.) The more practical way is filtering -- most singular balanced_dirty_ratelimit points can be filtered out by remembering some prev_balanced_rate and prev_prev_balanced_rate. However the more reliable way is to guard balanced_dirty_ratelimit with task_ratelimit. 2) due to truncates and fs redirties, the (write_bw <=> dirty_rate) match could become unbalanced, which may lead to large systematical errors in balanced_dirty_ratelimit. The truncates, due to its possibly bumpy nature, can hardly be compensated smoothly. So let's face it. When some over-estimated balanced_dirty_ratelimit brings dirty_ratelimit high, dirty pages will go higher than the setpoint. task_ratelimit will in turn become lower than dirty_ratelimit. So if we consider both balanced_dirty_ratelimit and task_ratelimit and update dirty_ratelimit only when they are on the same side of dirty_ratelimit, the systematical errors in balanced_dirty_ratelimit won't be able to bring dirty_ratelimit far away. The balanced_dirty_ratelimit estimation may also be inaccurate near @limit or @freerun, however is less an issue. 3) since we ultimately want to - keep the fluctuations of task ratelimit as small as possible - keep the dirty pages around the setpoint as long time as possible the update policy used for (2) also serves the above goals nicely: if for some reason the dirty pages are high (task_ratelimit < dirty_ratelimit), and dirty_ratelimit is low (dirty_ratelimit < balanced_dirty_ratelimit), there is no point to bring up dirty_ratelimit in a hurry only to hurt both the above two goals. So, we make use of task_ratelimit to limit the update of dirty_ratelimit in two ways: 1) avoid changing dirty rate when it's against the position control target (the adjusted rate will slow down the progress of dirty pages going back to setpoint). 2) limit the step size. task_ratelimit is changing values step by step, leaving a consistent trace comparing to the randomly jumping balanced_dirty_ratelimit. task_ratelimit also has the nice smaller errors in stable state and typically larger errors when there are big errors in rate. So it's a pretty good limiting factor for the step size of dirty_ratelimit. Note that bdi->dirty_ratelimit is always tracking balanced_dirty_ratelimit. task_ratelimit is merely used as a limiting factor. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-08-26 21:53:24 +00:00
unsigned long step;
unsigned long x;
unsigned long shift;
writeback: dirty rate control It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N) when there are N dd tasks. On write() syscall, use bdi->dirty_ratelimit ============================================ balance_dirty_pages(pages_dirtied) { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); pause = pages_dirtied / task_ratelimit; sleep(pause); } On every 200ms, update bdi->dirty_ratelimit =========================================== bdi_update_dirty_ratelimit() { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate; bdi->dirty_ratelimit = balanced_dirty_ratelimit } Estimation of balanced bdi->dirty_ratelimit =========================================== balanced task_ratelimit ----------------------- balance_dirty_pages() needs to throttle tasks dirtying pages such that the total amount of dirty pages stays below the specified dirty limit in order to avoid memory deadlocks. Furthermore we desire fairness in that tasks get throttled proportionally to the amount of pages they dirty. IOW we want to throttle tasks such that we match the dirty rate to the writeout bandwidth, this yields a stable amount of dirty pages: dirty_rate == write_bw (1) The fairness requirement gives us: task_ratelimit = balanced_dirty_ratelimit == write_bw / N (2) where N is the number of dd tasks. We don't know N beforehand, but still can estimate balanced_dirty_ratelimit within 200ms. Start by throttling each dd task at rate task_ratelimit = task_ratelimit_0 (3) (any non-zero initial value is OK) After 200ms, we measured dirty_rate = # of pages dirtied by all dd's / 200ms write_bw = # of pages written to the disk / 200ms For the aggressive dd dirtiers, the equality holds dirty_rate == N * task_rate == N * task_ratelimit_0 (4) Or task_ratelimit_0 == dirty_rate / N (5) Now we conclude that the balanced task ratelimit can be estimated by write_bw balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6) dirty_rate Because with (4) and (5) we can get the desired equality (1): write_bw balanced_dirty_ratelimit == (dirty_rate / N) * ---------- dirty_rate == write_bw / N Then using the balanced task ratelimit we can compute task pause times like: task_pause = task->nr_dirtied / task_ratelimit task_ratelimit with position control ------------------------------------ However, while the above gives us means of matching the dirty rate to the writeout bandwidth, it at best provides us with a stable dirty page count (assuming a static system). In order to control the dirty page count such that it is high enough to provide performance, but does not exceed the specified limit we need another control. The dirty position control works by extending (2) to task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7) where pos_ratio is a negative feedback function that subjects to 1) f(setpoint) = 1.0 2) df/dx < 0 That is, if the dirty pages are ABOVE the setpoint, we throttle each task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty pages are created less fast than they are cleaned, thus DROP to the setpoints (and the reverse). Based on (7) and the assumption that both dirty_ratelimit and pos_ratio remains CONSTANT for the past 200ms, we get task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8) Putting (8) into (6), we get the formula used in bdi_update_dirty_ratelimit(): write_bw balanced_dirty_ratelimit *= pos_ratio * ---------- (9) dirty_rate Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 16:51:31 +00:00
/*
* The dirty rate will match the writeout rate in long term, except
* when dirty pages are truncated by userspace or re-dirtied by FS.
*/
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
dirty_rate = (dirtied - wb->dirtied_stamp) * HZ / elapsed;
writeback: dirty rate control It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N) when there are N dd tasks. On write() syscall, use bdi->dirty_ratelimit ============================================ balance_dirty_pages(pages_dirtied) { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); pause = pages_dirtied / task_ratelimit; sleep(pause); } On every 200ms, update bdi->dirty_ratelimit =========================================== bdi_update_dirty_ratelimit() { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate; bdi->dirty_ratelimit = balanced_dirty_ratelimit } Estimation of balanced bdi->dirty_ratelimit =========================================== balanced task_ratelimit ----------------------- balance_dirty_pages() needs to throttle tasks dirtying pages such that the total amount of dirty pages stays below the specified dirty limit in order to avoid memory deadlocks. Furthermore we desire fairness in that tasks get throttled proportionally to the amount of pages they dirty. IOW we want to throttle tasks such that we match the dirty rate to the writeout bandwidth, this yields a stable amount of dirty pages: dirty_rate == write_bw (1) The fairness requirement gives us: task_ratelimit = balanced_dirty_ratelimit == write_bw / N (2) where N is the number of dd tasks. We don't know N beforehand, but still can estimate balanced_dirty_ratelimit within 200ms. Start by throttling each dd task at rate task_ratelimit = task_ratelimit_0 (3) (any non-zero initial value is OK) After 200ms, we measured dirty_rate = # of pages dirtied by all dd's / 200ms write_bw = # of pages written to the disk / 200ms For the aggressive dd dirtiers, the equality holds dirty_rate == N * task_rate == N * task_ratelimit_0 (4) Or task_ratelimit_0 == dirty_rate / N (5) Now we conclude that the balanced task ratelimit can be estimated by write_bw balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6) dirty_rate Because with (4) and (5) we can get the desired equality (1): write_bw balanced_dirty_ratelimit == (dirty_rate / N) * ---------- dirty_rate == write_bw / N Then using the balanced task ratelimit we can compute task pause times like: task_pause = task->nr_dirtied / task_ratelimit task_ratelimit with position control ------------------------------------ However, while the above gives us means of matching the dirty rate to the writeout bandwidth, it at best provides us with a stable dirty page count (assuming a static system). In order to control the dirty page count such that it is high enough to provide performance, but does not exceed the specified limit we need another control. The dirty position control works by extending (2) to task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7) where pos_ratio is a negative feedback function that subjects to 1) f(setpoint) = 1.0 2) df/dx < 0 That is, if the dirty pages are ABOVE the setpoint, we throttle each task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty pages are created less fast than they are cleaned, thus DROP to the setpoints (and the reverse). Based on (7) and the assumption that both dirty_ratelimit and pos_ratio remains CONSTANT for the past 200ms, we get task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8) Putting (8) into (6), we get the formula used in bdi_update_dirty_ratelimit(): write_bw balanced_dirty_ratelimit *= pos_ratio * ---------- (9) dirty_rate Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 16:51:31 +00:00
/*
* task_ratelimit reflects each dd's dirty rate for the past 200ms.
*/
task_ratelimit = (u64)dirty_ratelimit *
dtc->pos_ratio >> RATELIMIT_CALC_SHIFT;
writeback: dirty rate control It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N) when there are N dd tasks. On write() syscall, use bdi->dirty_ratelimit ============================================ balance_dirty_pages(pages_dirtied) { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); pause = pages_dirtied / task_ratelimit; sleep(pause); } On every 200ms, update bdi->dirty_ratelimit =========================================== bdi_update_dirty_ratelimit() { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate; bdi->dirty_ratelimit = balanced_dirty_ratelimit } Estimation of balanced bdi->dirty_ratelimit =========================================== balanced task_ratelimit ----------------------- balance_dirty_pages() needs to throttle tasks dirtying pages such that the total amount of dirty pages stays below the specified dirty limit in order to avoid memory deadlocks. Furthermore we desire fairness in that tasks get throttled proportionally to the amount of pages they dirty. IOW we want to throttle tasks such that we match the dirty rate to the writeout bandwidth, this yields a stable amount of dirty pages: dirty_rate == write_bw (1) The fairness requirement gives us: task_ratelimit = balanced_dirty_ratelimit == write_bw / N (2) where N is the number of dd tasks. We don't know N beforehand, but still can estimate balanced_dirty_ratelimit within 200ms. Start by throttling each dd task at rate task_ratelimit = task_ratelimit_0 (3) (any non-zero initial value is OK) After 200ms, we measured dirty_rate = # of pages dirtied by all dd's / 200ms write_bw = # of pages written to the disk / 200ms For the aggressive dd dirtiers, the equality holds dirty_rate == N * task_rate == N * task_ratelimit_0 (4) Or task_ratelimit_0 == dirty_rate / N (5) Now we conclude that the balanced task ratelimit can be estimated by write_bw balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6) dirty_rate Because with (4) and (5) we can get the desired equality (1): write_bw balanced_dirty_ratelimit == (dirty_rate / N) * ---------- dirty_rate == write_bw / N Then using the balanced task ratelimit we can compute task pause times like: task_pause = task->nr_dirtied / task_ratelimit task_ratelimit with position control ------------------------------------ However, while the above gives us means of matching the dirty rate to the writeout bandwidth, it at best provides us with a stable dirty page count (assuming a static system). In order to control the dirty page count such that it is high enough to provide performance, but does not exceed the specified limit we need another control. The dirty position control works by extending (2) to task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7) where pos_ratio is a negative feedback function that subjects to 1) f(setpoint) = 1.0 2) df/dx < 0 That is, if the dirty pages are ABOVE the setpoint, we throttle each task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty pages are created less fast than they are cleaned, thus DROP to the setpoints (and the reverse). Based on (7) and the assumption that both dirty_ratelimit and pos_ratio remains CONSTANT for the past 200ms, we get task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8) Putting (8) into (6), we get the formula used in bdi_update_dirty_ratelimit(): write_bw balanced_dirty_ratelimit *= pos_ratio * ---------- (9) dirty_rate Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 16:51:31 +00:00
task_ratelimit++; /* it helps rampup dirty_ratelimit from tiny values */
/*
* A linear estimation of the "balanced" throttle rate. The theory is,
* if there are N dd tasks, each throttled at task_ratelimit, the wb's
writeback: dirty rate control It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N) when there are N dd tasks. On write() syscall, use bdi->dirty_ratelimit ============================================ balance_dirty_pages(pages_dirtied) { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); pause = pages_dirtied / task_ratelimit; sleep(pause); } On every 200ms, update bdi->dirty_ratelimit =========================================== bdi_update_dirty_ratelimit() { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate; bdi->dirty_ratelimit = balanced_dirty_ratelimit } Estimation of balanced bdi->dirty_ratelimit =========================================== balanced task_ratelimit ----------------------- balance_dirty_pages() needs to throttle tasks dirtying pages such that the total amount of dirty pages stays below the specified dirty limit in order to avoid memory deadlocks. Furthermore we desire fairness in that tasks get throttled proportionally to the amount of pages they dirty. IOW we want to throttle tasks such that we match the dirty rate to the writeout bandwidth, this yields a stable amount of dirty pages: dirty_rate == write_bw (1) The fairness requirement gives us: task_ratelimit = balanced_dirty_ratelimit == write_bw / N (2) where N is the number of dd tasks. We don't know N beforehand, but still can estimate balanced_dirty_ratelimit within 200ms. Start by throttling each dd task at rate task_ratelimit = task_ratelimit_0 (3) (any non-zero initial value is OK) After 200ms, we measured dirty_rate = # of pages dirtied by all dd's / 200ms write_bw = # of pages written to the disk / 200ms For the aggressive dd dirtiers, the equality holds dirty_rate == N * task_rate == N * task_ratelimit_0 (4) Or task_ratelimit_0 == dirty_rate / N (5) Now we conclude that the balanced task ratelimit can be estimated by write_bw balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6) dirty_rate Because with (4) and (5) we can get the desired equality (1): write_bw balanced_dirty_ratelimit == (dirty_rate / N) * ---------- dirty_rate == write_bw / N Then using the balanced task ratelimit we can compute task pause times like: task_pause = task->nr_dirtied / task_ratelimit task_ratelimit with position control ------------------------------------ However, while the above gives us means of matching the dirty rate to the writeout bandwidth, it at best provides us with a stable dirty page count (assuming a static system). In order to control the dirty page count such that it is high enough to provide performance, but does not exceed the specified limit we need another control. The dirty position control works by extending (2) to task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7) where pos_ratio is a negative feedback function that subjects to 1) f(setpoint) = 1.0 2) df/dx < 0 That is, if the dirty pages are ABOVE the setpoint, we throttle each task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty pages are created less fast than they are cleaned, thus DROP to the setpoints (and the reverse). Based on (7) and the assumption that both dirty_ratelimit and pos_ratio remains CONSTANT for the past 200ms, we get task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8) Putting (8) into (6), we get the formula used in bdi_update_dirty_ratelimit(): write_bw balanced_dirty_ratelimit *= pos_ratio * ---------- (9) dirty_rate Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 16:51:31 +00:00
* dirty_rate will be measured to be (N * task_ratelimit). So the below
* formula will yield the balanced rate limit (write_bw / N).
*
* Note that the expanded form is not a pure rate feedback:
* rate_(i+1) = rate_(i) * (write_bw / dirty_rate) (1)
* but also takes pos_ratio into account:
* rate_(i+1) = rate_(i) * (write_bw / dirty_rate) * pos_ratio (2)
*
* (1) is not realistic because pos_ratio also takes part in balancing
* the dirty rate. Consider the state
* pos_ratio = 0.5 (3)
* rate = 2 * (write_bw / N) (4)
* If (1) is used, it will stuck in that state! Because each dd will
* be throttled at
* task_ratelimit = pos_ratio * rate = (write_bw / N) (5)
* yielding
* dirty_rate = N * task_ratelimit = write_bw (6)
* put (6) into (1) we get
* rate_(i+1) = rate_(i) (7)
*
* So we end up using (2) to always keep
* rate_(i+1) ~= (write_bw / N) (8)
* regardless of the value of pos_ratio. As long as (8) is satisfied,
* pos_ratio is able to drive itself to 1.0, which is not only where
* the dirty count meet the setpoint, but also where the slope of
* pos_ratio is most flat and hence task_ratelimit is least fluctuated.
*/
balanced_dirty_ratelimit = div_u64((u64)task_ratelimit * write_bw,
dirty_rate | 1);
writeback: balanced_rate cannot exceed write bandwidth Add an upper limit to balanced_rate according to the below inequality. This filters out some rare but huge singular points, which at least enables more readable gnuplot figures. When there are N dd dirtiers, balanced_dirty_ratelimit = write_bw / N So it holds that balanced_dirty_ratelimit <= write_bw The singular points originate from dirty_rate in the below formular: balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate where dirty_rate = (number of page dirties in the past 200ms) / 200ms In the extreme case, if all dd tasks suddenly get blocked on something else and hence no pages are dirtied at all, dirty_rate will be 0 and balanced_dirty_ratelimit will be inf. This could happen in reality. Note that these huge singular points are not a real threat, since they are _guaranteed_ to be filtered out by the min(balanced_dirty_ratelimit, task_ratelimit) line in bdi_update_dirty_ratelimit(). task_ratelimit is based on the number of dirty pages, which will never _suddenly_ fly away like balanced_dirty_ratelimit. So any weirdly large balanced_dirty_ratelimit will be cut down to the level of task_ratelimit. There won't be tiny singular points though, as long as the dirty pages lie inside the dirty throttling region (above the freerun region). Because there the dd tasks will be throttled by balanced_dirty_pages() and won't be able to suddenly dirty much more pages than average. Acked-by: Jan Kara <jack@suse.cz> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-08-03 20:30:36 +00:00
/*
* balanced_dirty_ratelimit ~= (write_bw / N) <= write_bw
*/
if (unlikely(balanced_dirty_ratelimit > write_bw))
balanced_dirty_ratelimit = write_bw;
writeback: dirty rate control It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N) when there are N dd tasks. On write() syscall, use bdi->dirty_ratelimit ============================================ balance_dirty_pages(pages_dirtied) { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); pause = pages_dirtied / task_ratelimit; sleep(pause); } On every 200ms, update bdi->dirty_ratelimit =========================================== bdi_update_dirty_ratelimit() { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate; bdi->dirty_ratelimit = balanced_dirty_ratelimit } Estimation of balanced bdi->dirty_ratelimit =========================================== balanced task_ratelimit ----------------------- balance_dirty_pages() needs to throttle tasks dirtying pages such that the total amount of dirty pages stays below the specified dirty limit in order to avoid memory deadlocks. Furthermore we desire fairness in that tasks get throttled proportionally to the amount of pages they dirty. IOW we want to throttle tasks such that we match the dirty rate to the writeout bandwidth, this yields a stable amount of dirty pages: dirty_rate == write_bw (1) The fairness requirement gives us: task_ratelimit = balanced_dirty_ratelimit == write_bw / N (2) where N is the number of dd tasks. We don't know N beforehand, but still can estimate balanced_dirty_ratelimit within 200ms. Start by throttling each dd task at rate task_ratelimit = task_ratelimit_0 (3) (any non-zero initial value is OK) After 200ms, we measured dirty_rate = # of pages dirtied by all dd's / 200ms write_bw = # of pages written to the disk / 200ms For the aggressive dd dirtiers, the equality holds dirty_rate == N * task_rate == N * task_ratelimit_0 (4) Or task_ratelimit_0 == dirty_rate / N (5) Now we conclude that the balanced task ratelimit can be estimated by write_bw balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6) dirty_rate Because with (4) and (5) we can get the desired equality (1): write_bw balanced_dirty_ratelimit == (dirty_rate / N) * ---------- dirty_rate == write_bw / N Then using the balanced task ratelimit we can compute task pause times like: task_pause = task->nr_dirtied / task_ratelimit task_ratelimit with position control ------------------------------------ However, while the above gives us means of matching the dirty rate to the writeout bandwidth, it at best provides us with a stable dirty page count (assuming a static system). In order to control the dirty page count such that it is high enough to provide performance, but does not exceed the specified limit we need another control. The dirty position control works by extending (2) to task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7) where pos_ratio is a negative feedback function that subjects to 1) f(setpoint) = 1.0 2) df/dx < 0 That is, if the dirty pages are ABOVE the setpoint, we throttle each task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty pages are created less fast than they are cleaned, thus DROP to the setpoints (and the reverse). Based on (7) and the assumption that both dirty_ratelimit and pos_ratio remains CONSTANT for the past 200ms, we get task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8) Putting (8) into (6), we get the formula used in bdi_update_dirty_ratelimit(): write_bw balanced_dirty_ratelimit *= pos_ratio * ---------- (9) dirty_rate Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 16:51:31 +00:00
writeback: stabilize bdi->dirty_ratelimit There are some imperfections in balanced_dirty_ratelimit. 1) large fluctuations The dirty_rate used for computing balanced_dirty_ratelimit is merely averaged in the past 200ms (very small comparing to the 3s estimation period for write_bw), which makes rather dispersed distribution of balanced_dirty_ratelimit. It's pretty hard to average out the singular points by increasing the estimation period. Considering that the averaging technique will introduce very undesirable time lags, I give it up totally. (btw, the 3s write_bw averaging time lag is much more acceptable because its impact is one-way and therefore won't lead to oscillations.) The more practical way is filtering -- most singular balanced_dirty_ratelimit points can be filtered out by remembering some prev_balanced_rate and prev_prev_balanced_rate. However the more reliable way is to guard balanced_dirty_ratelimit with task_ratelimit. 2) due to truncates and fs redirties, the (write_bw <=> dirty_rate) match could become unbalanced, which may lead to large systematical errors in balanced_dirty_ratelimit. The truncates, due to its possibly bumpy nature, can hardly be compensated smoothly. So let's face it. When some over-estimated balanced_dirty_ratelimit brings dirty_ratelimit high, dirty pages will go higher than the setpoint. task_ratelimit will in turn become lower than dirty_ratelimit. So if we consider both balanced_dirty_ratelimit and task_ratelimit and update dirty_ratelimit only when they are on the same side of dirty_ratelimit, the systematical errors in balanced_dirty_ratelimit won't be able to bring dirty_ratelimit far away. The balanced_dirty_ratelimit estimation may also be inaccurate near @limit or @freerun, however is less an issue. 3) since we ultimately want to - keep the fluctuations of task ratelimit as small as possible - keep the dirty pages around the setpoint as long time as possible the update policy used for (2) also serves the above goals nicely: if for some reason the dirty pages are high (task_ratelimit < dirty_ratelimit), and dirty_ratelimit is low (dirty_ratelimit < balanced_dirty_ratelimit), there is no point to bring up dirty_ratelimit in a hurry only to hurt both the above two goals. So, we make use of task_ratelimit to limit the update of dirty_ratelimit in two ways: 1) avoid changing dirty rate when it's against the position control target (the adjusted rate will slow down the progress of dirty pages going back to setpoint). 2) limit the step size. task_ratelimit is changing values step by step, leaving a consistent trace comparing to the randomly jumping balanced_dirty_ratelimit. task_ratelimit also has the nice smaller errors in stable state and typically larger errors when there are big errors in rate. So it's a pretty good limiting factor for the step size of dirty_ratelimit. Note that bdi->dirty_ratelimit is always tracking balanced_dirty_ratelimit. task_ratelimit is merely used as a limiting factor. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-08-26 21:53:24 +00:00
/*
* We could safely do this and return immediately:
*
* wb->dirty_ratelimit = balanced_dirty_ratelimit;
writeback: stabilize bdi->dirty_ratelimit There are some imperfections in balanced_dirty_ratelimit. 1) large fluctuations The dirty_rate used for computing balanced_dirty_ratelimit is merely averaged in the past 200ms (very small comparing to the 3s estimation period for write_bw), which makes rather dispersed distribution of balanced_dirty_ratelimit. It's pretty hard to average out the singular points by increasing the estimation period. Considering that the averaging technique will introduce very undesirable time lags, I give it up totally. (btw, the 3s write_bw averaging time lag is much more acceptable because its impact is one-way and therefore won't lead to oscillations.) The more practical way is filtering -- most singular balanced_dirty_ratelimit points can be filtered out by remembering some prev_balanced_rate and prev_prev_balanced_rate. However the more reliable way is to guard balanced_dirty_ratelimit with task_ratelimit. 2) due to truncates and fs redirties, the (write_bw <=> dirty_rate) match could become unbalanced, which may lead to large systematical errors in balanced_dirty_ratelimit. The truncates, due to its possibly bumpy nature, can hardly be compensated smoothly. So let's face it. When some over-estimated balanced_dirty_ratelimit brings dirty_ratelimit high, dirty pages will go higher than the setpoint. task_ratelimit will in turn become lower than dirty_ratelimit. So if we consider both balanced_dirty_ratelimit and task_ratelimit and update dirty_ratelimit only when they are on the same side of dirty_ratelimit, the systematical errors in balanced_dirty_ratelimit won't be able to bring dirty_ratelimit far away. The balanced_dirty_ratelimit estimation may also be inaccurate near @limit or @freerun, however is less an issue. 3) since we ultimately want to - keep the fluctuations of task ratelimit as small as possible - keep the dirty pages around the setpoint as long time as possible the update policy used for (2) also serves the above goals nicely: if for some reason the dirty pages are high (task_ratelimit < dirty_ratelimit), and dirty_ratelimit is low (dirty_ratelimit < balanced_dirty_ratelimit), there is no point to bring up dirty_ratelimit in a hurry only to hurt both the above two goals. So, we make use of task_ratelimit to limit the update of dirty_ratelimit in two ways: 1) avoid changing dirty rate when it's against the position control target (the adjusted rate will slow down the progress of dirty pages going back to setpoint). 2) limit the step size. task_ratelimit is changing values step by step, leaving a consistent trace comparing to the randomly jumping balanced_dirty_ratelimit. task_ratelimit also has the nice smaller errors in stable state and typically larger errors when there are big errors in rate. So it's a pretty good limiting factor for the step size of dirty_ratelimit. Note that bdi->dirty_ratelimit is always tracking balanced_dirty_ratelimit. task_ratelimit is merely used as a limiting factor. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-08-26 21:53:24 +00:00
*
* However to get a more stable dirty_ratelimit, the below elaborated
* code makes use of task_ratelimit to filter out singular points and
writeback: stabilize bdi->dirty_ratelimit There are some imperfections in balanced_dirty_ratelimit. 1) large fluctuations The dirty_rate used for computing balanced_dirty_ratelimit is merely averaged in the past 200ms (very small comparing to the 3s estimation period for write_bw), which makes rather dispersed distribution of balanced_dirty_ratelimit. It's pretty hard to average out the singular points by increasing the estimation period. Considering that the averaging technique will introduce very undesirable time lags, I give it up totally. (btw, the 3s write_bw averaging time lag is much more acceptable because its impact is one-way and therefore won't lead to oscillations.) The more practical way is filtering -- most singular balanced_dirty_ratelimit points can be filtered out by remembering some prev_balanced_rate and prev_prev_balanced_rate. However the more reliable way is to guard balanced_dirty_ratelimit with task_ratelimit. 2) due to truncates and fs redirties, the (write_bw <=> dirty_rate) match could become unbalanced, which may lead to large systematical errors in balanced_dirty_ratelimit. The truncates, due to its possibly bumpy nature, can hardly be compensated smoothly. So let's face it. When some over-estimated balanced_dirty_ratelimit brings dirty_ratelimit high, dirty pages will go higher than the setpoint. task_ratelimit will in turn become lower than dirty_ratelimit. So if we consider both balanced_dirty_ratelimit and task_ratelimit and update dirty_ratelimit only when they are on the same side of dirty_ratelimit, the systematical errors in balanced_dirty_ratelimit won't be able to bring dirty_ratelimit far away. The balanced_dirty_ratelimit estimation may also be inaccurate near @limit or @freerun, however is less an issue. 3) since we ultimately want to - keep the fluctuations of task ratelimit as small as possible - keep the dirty pages around the setpoint as long time as possible the update policy used for (2) also serves the above goals nicely: if for some reason the dirty pages are high (task_ratelimit < dirty_ratelimit), and dirty_ratelimit is low (dirty_ratelimit < balanced_dirty_ratelimit), there is no point to bring up dirty_ratelimit in a hurry only to hurt both the above two goals. So, we make use of task_ratelimit to limit the update of dirty_ratelimit in two ways: 1) avoid changing dirty rate when it's against the position control target (the adjusted rate will slow down the progress of dirty pages going back to setpoint). 2) limit the step size. task_ratelimit is changing values step by step, leaving a consistent trace comparing to the randomly jumping balanced_dirty_ratelimit. task_ratelimit also has the nice smaller errors in stable state and typically larger errors when there are big errors in rate. So it's a pretty good limiting factor for the step size of dirty_ratelimit. Note that bdi->dirty_ratelimit is always tracking balanced_dirty_ratelimit. task_ratelimit is merely used as a limiting factor. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-08-26 21:53:24 +00:00
* limit the step size.
*
* The below code essentially only uses the relative value of
*
* task_ratelimit - dirty_ratelimit
* = (pos_ratio - 1) * dirty_ratelimit
*
* which reflects the direction and size of dirty position error.
*/
/*
* dirty_ratelimit will follow balanced_dirty_ratelimit iff
* task_ratelimit is on the same side of dirty_ratelimit, too.
* For example, when
* - dirty_ratelimit > balanced_dirty_ratelimit
* - dirty_ratelimit > task_ratelimit (dirty pages are above setpoint)
* lowering dirty_ratelimit will help meet both the position and rate
* control targets. Otherwise, don't update dirty_ratelimit if it will
* only help meet the rate target. After all, what the users ultimately
* feel and care are stable dirty rate and small position error.
*
* |task_ratelimit - dirty_ratelimit| is used to limit the step size
* and filter out the singular points of balanced_dirty_ratelimit. Which
writeback: stabilize bdi->dirty_ratelimit There are some imperfections in balanced_dirty_ratelimit. 1) large fluctuations The dirty_rate used for computing balanced_dirty_ratelimit is merely averaged in the past 200ms (very small comparing to the 3s estimation period for write_bw), which makes rather dispersed distribution of balanced_dirty_ratelimit. It's pretty hard to average out the singular points by increasing the estimation period. Considering that the averaging technique will introduce very undesirable time lags, I give it up totally. (btw, the 3s write_bw averaging time lag is much more acceptable because its impact is one-way and therefore won't lead to oscillations.) The more practical way is filtering -- most singular balanced_dirty_ratelimit points can be filtered out by remembering some prev_balanced_rate and prev_prev_balanced_rate. However the more reliable way is to guard balanced_dirty_ratelimit with task_ratelimit. 2) due to truncates and fs redirties, the (write_bw <=> dirty_rate) match could become unbalanced, which may lead to large systematical errors in balanced_dirty_ratelimit. The truncates, due to its possibly bumpy nature, can hardly be compensated smoothly. So let's face it. When some over-estimated balanced_dirty_ratelimit brings dirty_ratelimit high, dirty pages will go higher than the setpoint. task_ratelimit will in turn become lower than dirty_ratelimit. So if we consider both balanced_dirty_ratelimit and task_ratelimit and update dirty_ratelimit only when they are on the same side of dirty_ratelimit, the systematical errors in balanced_dirty_ratelimit won't be able to bring dirty_ratelimit far away. The balanced_dirty_ratelimit estimation may also be inaccurate near @limit or @freerun, however is less an issue. 3) since we ultimately want to - keep the fluctuations of task ratelimit as small as possible - keep the dirty pages around the setpoint as long time as possible the update policy used for (2) also serves the above goals nicely: if for some reason the dirty pages are high (task_ratelimit < dirty_ratelimit), and dirty_ratelimit is low (dirty_ratelimit < balanced_dirty_ratelimit), there is no point to bring up dirty_ratelimit in a hurry only to hurt both the above two goals. So, we make use of task_ratelimit to limit the update of dirty_ratelimit in two ways: 1) avoid changing dirty rate when it's against the position control target (the adjusted rate will slow down the progress of dirty pages going back to setpoint). 2) limit the step size. task_ratelimit is changing values step by step, leaving a consistent trace comparing to the randomly jumping balanced_dirty_ratelimit. task_ratelimit also has the nice smaller errors in stable state and typically larger errors when there are big errors in rate. So it's a pretty good limiting factor for the step size of dirty_ratelimit. Note that bdi->dirty_ratelimit is always tracking balanced_dirty_ratelimit. task_ratelimit is merely used as a limiting factor. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-08-26 21:53:24 +00:00
* keeps jumping around randomly and can even leap far away at times
* due to the small 200ms estimation period of dirty_rate (we want to
* keep that period small to reduce time lags).
*/
step = 0;
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
/*
* For strictlimit case, calculations above were based on wb counters
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
* and limits (starting from pos_ratio = wb_position_ratio() and up to
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate).
* Hence, to calculate "step" properly, we have to use wb_dirty as
* "dirty" and wb_setpoint as "setpoint".
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
*
* We rampup dirty_ratelimit forcibly if wb_dirty is low because
* it's possible that wb_thresh is close to zero due to inactivity
* of backing device.
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
*/
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
if (unlikely(wb->bdi->capabilities & BDI_CAP_STRICTLIMIT)) {
dirty = dtc->wb_dirty;
if (dtc->wb_dirty < 8)
setpoint = dtc->wb_dirty + 1;
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
else
setpoint = (dtc->wb_thresh + dtc->wb_bg_thresh) / 2;
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
}
writeback: stabilize bdi->dirty_ratelimit There are some imperfections in balanced_dirty_ratelimit. 1) large fluctuations The dirty_rate used for computing balanced_dirty_ratelimit is merely averaged in the past 200ms (very small comparing to the 3s estimation period for write_bw), which makes rather dispersed distribution of balanced_dirty_ratelimit. It's pretty hard to average out the singular points by increasing the estimation period. Considering that the averaging technique will introduce very undesirable time lags, I give it up totally. (btw, the 3s write_bw averaging time lag is much more acceptable because its impact is one-way and therefore won't lead to oscillations.) The more practical way is filtering -- most singular balanced_dirty_ratelimit points can be filtered out by remembering some prev_balanced_rate and prev_prev_balanced_rate. However the more reliable way is to guard balanced_dirty_ratelimit with task_ratelimit. 2) due to truncates and fs redirties, the (write_bw <=> dirty_rate) match could become unbalanced, which may lead to large systematical errors in balanced_dirty_ratelimit. The truncates, due to its possibly bumpy nature, can hardly be compensated smoothly. So let's face it. When some over-estimated balanced_dirty_ratelimit brings dirty_ratelimit high, dirty pages will go higher than the setpoint. task_ratelimit will in turn become lower than dirty_ratelimit. So if we consider both balanced_dirty_ratelimit and task_ratelimit and update dirty_ratelimit only when they are on the same side of dirty_ratelimit, the systematical errors in balanced_dirty_ratelimit won't be able to bring dirty_ratelimit far away. The balanced_dirty_ratelimit estimation may also be inaccurate near @limit or @freerun, however is less an issue. 3) since we ultimately want to - keep the fluctuations of task ratelimit as small as possible - keep the dirty pages around the setpoint as long time as possible the update policy used for (2) also serves the above goals nicely: if for some reason the dirty pages are high (task_ratelimit < dirty_ratelimit), and dirty_ratelimit is low (dirty_ratelimit < balanced_dirty_ratelimit), there is no point to bring up dirty_ratelimit in a hurry only to hurt both the above two goals. So, we make use of task_ratelimit to limit the update of dirty_ratelimit in two ways: 1) avoid changing dirty rate when it's against the position control target (the adjusted rate will slow down the progress of dirty pages going back to setpoint). 2) limit the step size. task_ratelimit is changing values step by step, leaving a consistent trace comparing to the randomly jumping balanced_dirty_ratelimit. task_ratelimit also has the nice smaller errors in stable state and typically larger errors when there are big errors in rate. So it's a pretty good limiting factor for the step size of dirty_ratelimit. Note that bdi->dirty_ratelimit is always tracking balanced_dirty_ratelimit. task_ratelimit is merely used as a limiting factor. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-08-26 21:53:24 +00:00
if (dirty < setpoint) {
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
x = min3(wb->balanced_dirty_ratelimit,
balanced_dirty_ratelimit, task_ratelimit);
writeback: stabilize bdi->dirty_ratelimit There are some imperfections in balanced_dirty_ratelimit. 1) large fluctuations The dirty_rate used for computing balanced_dirty_ratelimit is merely averaged in the past 200ms (very small comparing to the 3s estimation period for write_bw), which makes rather dispersed distribution of balanced_dirty_ratelimit. It's pretty hard to average out the singular points by increasing the estimation period. Considering that the averaging technique will introduce very undesirable time lags, I give it up totally. (btw, the 3s write_bw averaging time lag is much more acceptable because its impact is one-way and therefore won't lead to oscillations.) The more practical way is filtering -- most singular balanced_dirty_ratelimit points can be filtered out by remembering some prev_balanced_rate and prev_prev_balanced_rate. However the more reliable way is to guard balanced_dirty_ratelimit with task_ratelimit. 2) due to truncates and fs redirties, the (write_bw <=> dirty_rate) match could become unbalanced, which may lead to large systematical errors in balanced_dirty_ratelimit. The truncates, due to its possibly bumpy nature, can hardly be compensated smoothly. So let's face it. When some over-estimated balanced_dirty_ratelimit brings dirty_ratelimit high, dirty pages will go higher than the setpoint. task_ratelimit will in turn become lower than dirty_ratelimit. So if we consider both balanced_dirty_ratelimit and task_ratelimit and update dirty_ratelimit only when they are on the same side of dirty_ratelimit, the systematical errors in balanced_dirty_ratelimit won't be able to bring dirty_ratelimit far away. The balanced_dirty_ratelimit estimation may also be inaccurate near @limit or @freerun, however is less an issue. 3) since we ultimately want to - keep the fluctuations of task ratelimit as small as possible - keep the dirty pages around the setpoint as long time as possible the update policy used for (2) also serves the above goals nicely: if for some reason the dirty pages are high (task_ratelimit < dirty_ratelimit), and dirty_ratelimit is low (dirty_ratelimit < balanced_dirty_ratelimit), there is no point to bring up dirty_ratelimit in a hurry only to hurt both the above two goals. So, we make use of task_ratelimit to limit the update of dirty_ratelimit in two ways: 1) avoid changing dirty rate when it's against the position control target (the adjusted rate will slow down the progress of dirty pages going back to setpoint). 2) limit the step size. task_ratelimit is changing values step by step, leaving a consistent trace comparing to the randomly jumping balanced_dirty_ratelimit. task_ratelimit also has the nice smaller errors in stable state and typically larger errors when there are big errors in rate. So it's a pretty good limiting factor for the step size of dirty_ratelimit. Note that bdi->dirty_ratelimit is always tracking balanced_dirty_ratelimit. task_ratelimit is merely used as a limiting factor. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-08-26 21:53:24 +00:00
if (dirty_ratelimit < x)
step = x - dirty_ratelimit;
} else {
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
x = max3(wb->balanced_dirty_ratelimit,
balanced_dirty_ratelimit, task_ratelimit);
writeback: stabilize bdi->dirty_ratelimit There are some imperfections in balanced_dirty_ratelimit. 1) large fluctuations The dirty_rate used for computing balanced_dirty_ratelimit is merely averaged in the past 200ms (very small comparing to the 3s estimation period for write_bw), which makes rather dispersed distribution of balanced_dirty_ratelimit. It's pretty hard to average out the singular points by increasing the estimation period. Considering that the averaging technique will introduce very undesirable time lags, I give it up totally. (btw, the 3s write_bw averaging time lag is much more acceptable because its impact is one-way and therefore won't lead to oscillations.) The more practical way is filtering -- most singular balanced_dirty_ratelimit points can be filtered out by remembering some prev_balanced_rate and prev_prev_balanced_rate. However the more reliable way is to guard balanced_dirty_ratelimit with task_ratelimit. 2) due to truncates and fs redirties, the (write_bw <=> dirty_rate) match could become unbalanced, which may lead to large systematical errors in balanced_dirty_ratelimit. The truncates, due to its possibly bumpy nature, can hardly be compensated smoothly. So let's face it. When some over-estimated balanced_dirty_ratelimit brings dirty_ratelimit high, dirty pages will go higher than the setpoint. task_ratelimit will in turn become lower than dirty_ratelimit. So if we consider both balanced_dirty_ratelimit and task_ratelimit and update dirty_ratelimit only when they are on the same side of dirty_ratelimit, the systematical errors in balanced_dirty_ratelimit won't be able to bring dirty_ratelimit far away. The balanced_dirty_ratelimit estimation may also be inaccurate near @limit or @freerun, however is less an issue. 3) since we ultimately want to - keep the fluctuations of task ratelimit as small as possible - keep the dirty pages around the setpoint as long time as possible the update policy used for (2) also serves the above goals nicely: if for some reason the dirty pages are high (task_ratelimit < dirty_ratelimit), and dirty_ratelimit is low (dirty_ratelimit < balanced_dirty_ratelimit), there is no point to bring up dirty_ratelimit in a hurry only to hurt both the above two goals. So, we make use of task_ratelimit to limit the update of dirty_ratelimit in two ways: 1) avoid changing dirty rate when it's against the position control target (the adjusted rate will slow down the progress of dirty pages going back to setpoint). 2) limit the step size. task_ratelimit is changing values step by step, leaving a consistent trace comparing to the randomly jumping balanced_dirty_ratelimit. task_ratelimit also has the nice smaller errors in stable state and typically larger errors when there are big errors in rate. So it's a pretty good limiting factor for the step size of dirty_ratelimit. Note that bdi->dirty_ratelimit is always tracking balanced_dirty_ratelimit. task_ratelimit is merely used as a limiting factor. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-08-26 21:53:24 +00:00
if (dirty_ratelimit > x)
step = dirty_ratelimit - x;
}
/*
* Don't pursue 100% rate matching. It's impossible since the balanced
* rate itself is constantly fluctuating. So decrease the track speed
* when it gets close to the target. Helps eliminate pointless tremors.
*/
shift = dirty_ratelimit / (2 * step + 1);
if (shift < BITS_PER_LONG)
step = DIV_ROUND_UP(step >> shift, 8);
else
step = 0;
writeback: stabilize bdi->dirty_ratelimit There are some imperfections in balanced_dirty_ratelimit. 1) large fluctuations The dirty_rate used for computing balanced_dirty_ratelimit is merely averaged in the past 200ms (very small comparing to the 3s estimation period for write_bw), which makes rather dispersed distribution of balanced_dirty_ratelimit. It's pretty hard to average out the singular points by increasing the estimation period. Considering that the averaging technique will introduce very undesirable time lags, I give it up totally. (btw, the 3s write_bw averaging time lag is much more acceptable because its impact is one-way and therefore won't lead to oscillations.) The more practical way is filtering -- most singular balanced_dirty_ratelimit points can be filtered out by remembering some prev_balanced_rate and prev_prev_balanced_rate. However the more reliable way is to guard balanced_dirty_ratelimit with task_ratelimit. 2) due to truncates and fs redirties, the (write_bw <=> dirty_rate) match could become unbalanced, which may lead to large systematical errors in balanced_dirty_ratelimit. The truncates, due to its possibly bumpy nature, can hardly be compensated smoothly. So let's face it. When some over-estimated balanced_dirty_ratelimit brings dirty_ratelimit high, dirty pages will go higher than the setpoint. task_ratelimit will in turn become lower than dirty_ratelimit. So if we consider both balanced_dirty_ratelimit and task_ratelimit and update dirty_ratelimit only when they are on the same side of dirty_ratelimit, the systematical errors in balanced_dirty_ratelimit won't be able to bring dirty_ratelimit far away. The balanced_dirty_ratelimit estimation may also be inaccurate near @limit or @freerun, however is less an issue. 3) since we ultimately want to - keep the fluctuations of task ratelimit as small as possible - keep the dirty pages around the setpoint as long time as possible the update policy used for (2) also serves the above goals nicely: if for some reason the dirty pages are high (task_ratelimit < dirty_ratelimit), and dirty_ratelimit is low (dirty_ratelimit < balanced_dirty_ratelimit), there is no point to bring up dirty_ratelimit in a hurry only to hurt both the above two goals. So, we make use of task_ratelimit to limit the update of dirty_ratelimit in two ways: 1) avoid changing dirty rate when it's against the position control target (the adjusted rate will slow down the progress of dirty pages going back to setpoint). 2) limit the step size. task_ratelimit is changing values step by step, leaving a consistent trace comparing to the randomly jumping balanced_dirty_ratelimit. task_ratelimit also has the nice smaller errors in stable state and typically larger errors when there are big errors in rate. So it's a pretty good limiting factor for the step size of dirty_ratelimit. Note that bdi->dirty_ratelimit is always tracking balanced_dirty_ratelimit. task_ratelimit is merely used as a limiting factor. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-08-26 21:53:24 +00:00
if (dirty_ratelimit < balanced_dirty_ratelimit)
dirty_ratelimit += step;
else
dirty_ratelimit -= step;
WRITE_ONCE(wb->dirty_ratelimit, max(dirty_ratelimit, 1UL));
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
wb->balanced_dirty_ratelimit = balanced_dirty_ratelimit;
trace_bdi_dirty_ratelimit(wb, dirty_rate, task_ratelimit);
writeback: dirty rate control It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N) when there are N dd tasks. On write() syscall, use bdi->dirty_ratelimit ============================================ balance_dirty_pages(pages_dirtied) { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); pause = pages_dirtied / task_ratelimit; sleep(pause); } On every 200ms, update bdi->dirty_ratelimit =========================================== bdi_update_dirty_ratelimit() { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate; bdi->dirty_ratelimit = balanced_dirty_ratelimit } Estimation of balanced bdi->dirty_ratelimit =========================================== balanced task_ratelimit ----------------------- balance_dirty_pages() needs to throttle tasks dirtying pages such that the total amount of dirty pages stays below the specified dirty limit in order to avoid memory deadlocks. Furthermore we desire fairness in that tasks get throttled proportionally to the amount of pages they dirty. IOW we want to throttle tasks such that we match the dirty rate to the writeout bandwidth, this yields a stable amount of dirty pages: dirty_rate == write_bw (1) The fairness requirement gives us: task_ratelimit = balanced_dirty_ratelimit == write_bw / N (2) where N is the number of dd tasks. We don't know N beforehand, but still can estimate balanced_dirty_ratelimit within 200ms. Start by throttling each dd task at rate task_ratelimit = task_ratelimit_0 (3) (any non-zero initial value is OK) After 200ms, we measured dirty_rate = # of pages dirtied by all dd's / 200ms write_bw = # of pages written to the disk / 200ms For the aggressive dd dirtiers, the equality holds dirty_rate == N * task_rate == N * task_ratelimit_0 (4) Or task_ratelimit_0 == dirty_rate / N (5) Now we conclude that the balanced task ratelimit can be estimated by write_bw balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6) dirty_rate Because with (4) and (5) we can get the desired equality (1): write_bw balanced_dirty_ratelimit == (dirty_rate / N) * ---------- dirty_rate == write_bw / N Then using the balanced task ratelimit we can compute task pause times like: task_pause = task->nr_dirtied / task_ratelimit task_ratelimit with position control ------------------------------------ However, while the above gives us means of matching the dirty rate to the writeout bandwidth, it at best provides us with a stable dirty page count (assuming a static system). In order to control the dirty page count such that it is high enough to provide performance, but does not exceed the specified limit we need another control. The dirty position control works by extending (2) to task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7) where pos_ratio is a negative feedback function that subjects to 1) f(setpoint) = 1.0 2) df/dx < 0 That is, if the dirty pages are ABOVE the setpoint, we throttle each task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty pages are created less fast than they are cleaned, thus DROP to the setpoints (and the reverse). Based on (7) and the assumption that both dirty_ratelimit and pos_ratio remains CONSTANT for the past 200ms, we get task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8) Putting (8) into (6), we get the formula used in bdi_update_dirty_ratelimit(): write_bw balanced_dirty_ratelimit *= pos_ratio * ---------- (9) dirty_rate Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 16:51:31 +00:00
}
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
static void __wb_update_bandwidth(struct dirty_throttle_control *gdtc,
struct dirty_throttle_control *mdtc,
bool update_ratelimit)
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
{
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
struct bdi_writeback *wb = gdtc->wb;
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
unsigned long now = jiffies;
writeback: fix bandwidth estimate for spiky workload Michael Stapelberg has reported that for workload with short big spikes of writes (GCC linker seem to trigger this frequently) the write throughput is heavily underestimated and tends to steadily sink until it reaches zero. This has rather bad impact on writeback throttling (causing stalls). The problem is that writeback throughput estimate gets updated at most once per 200 ms. One update happens early after we submit pages for writeback (at that point writeout of only small fraction of pages is completed and thus observed throughput is tiny). Next update happens only during the next write spike (updates happen only from inode writeback and dirty throttling code) and if that is more than 1s after previous spike, we decide system was idle and just ignore whatever was written until this moment. Fix the problem by making sure writeback throughput estimate is also updated shortly after writeback completes to get reasonable estimate of throughput for spiky workloads. [jack@suse.cz: avoid division by 0 in wb_update_dirty_ratelimit()] Link: https://lore.kernel.org/lkml/20210617095309.3542373-1-stapelberg+linux@google.com Link: https://lkml.kernel.org/r/20210713104716.22868-3-jack@suse.cz Signed-off-by: Jan Kara <jack@suse.cz> Reported-by: Michael Stapelberg <stapelberg+linux@google.com> Tested-by: Michael Stapelberg <stapelberg+linux@google.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-09-02 21:53:09 +00:00
unsigned long elapsed;
writeback: dirty rate control It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N) when there are N dd tasks. On write() syscall, use bdi->dirty_ratelimit ============================================ balance_dirty_pages(pages_dirtied) { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); pause = pages_dirtied / task_ratelimit; sleep(pause); } On every 200ms, update bdi->dirty_ratelimit =========================================== bdi_update_dirty_ratelimit() { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate; bdi->dirty_ratelimit = balanced_dirty_ratelimit } Estimation of balanced bdi->dirty_ratelimit =========================================== balanced task_ratelimit ----------------------- balance_dirty_pages() needs to throttle tasks dirtying pages such that the total amount of dirty pages stays below the specified dirty limit in order to avoid memory deadlocks. Furthermore we desire fairness in that tasks get throttled proportionally to the amount of pages they dirty. IOW we want to throttle tasks such that we match the dirty rate to the writeout bandwidth, this yields a stable amount of dirty pages: dirty_rate == write_bw (1) The fairness requirement gives us: task_ratelimit = balanced_dirty_ratelimit == write_bw / N (2) where N is the number of dd tasks. We don't know N beforehand, but still can estimate balanced_dirty_ratelimit within 200ms. Start by throttling each dd task at rate task_ratelimit = task_ratelimit_0 (3) (any non-zero initial value is OK) After 200ms, we measured dirty_rate = # of pages dirtied by all dd's / 200ms write_bw = # of pages written to the disk / 200ms For the aggressive dd dirtiers, the equality holds dirty_rate == N * task_rate == N * task_ratelimit_0 (4) Or task_ratelimit_0 == dirty_rate / N (5) Now we conclude that the balanced task ratelimit can be estimated by write_bw balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6) dirty_rate Because with (4) and (5) we can get the desired equality (1): write_bw balanced_dirty_ratelimit == (dirty_rate / N) * ---------- dirty_rate == write_bw / N Then using the balanced task ratelimit we can compute task pause times like: task_pause = task->nr_dirtied / task_ratelimit task_ratelimit with position control ------------------------------------ However, while the above gives us means of matching the dirty rate to the writeout bandwidth, it at best provides us with a stable dirty page count (assuming a static system). In order to control the dirty page count such that it is high enough to provide performance, but does not exceed the specified limit we need another control. The dirty position control works by extending (2) to task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7) where pos_ratio is a negative feedback function that subjects to 1) f(setpoint) = 1.0 2) df/dx < 0 That is, if the dirty pages are ABOVE the setpoint, we throttle each task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty pages are created less fast than they are cleaned, thus DROP to the setpoints (and the reverse). Based on (7) and the assumption that both dirty_ratelimit and pos_ratio remains CONSTANT for the past 200ms, we get task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8) Putting (8) into (6), we get the formula used in bdi_update_dirty_ratelimit(): write_bw balanced_dirty_ratelimit *= pos_ratio * ---------- (9) dirty_rate Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 16:51:31 +00:00
unsigned long dirtied;
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
unsigned long written;
writeback: fix bandwidth estimate for spiky workload Michael Stapelberg has reported that for workload with short big spikes of writes (GCC linker seem to trigger this frequently) the write throughput is heavily underestimated and tends to steadily sink until it reaches zero. This has rather bad impact on writeback throttling (causing stalls). The problem is that writeback throughput estimate gets updated at most once per 200 ms. One update happens early after we submit pages for writeback (at that point writeout of only small fraction of pages is completed and thus observed throughput is tiny). Next update happens only during the next write spike (updates happen only from inode writeback and dirty throttling code) and if that is more than 1s after previous spike, we decide system was idle and just ignore whatever was written until this moment. Fix the problem by making sure writeback throughput estimate is also updated shortly after writeback completes to get reasonable estimate of throughput for spiky workloads. [jack@suse.cz: avoid division by 0 in wb_update_dirty_ratelimit()] Link: https://lore.kernel.org/lkml/20210617095309.3542373-1-stapelberg+linux@google.com Link: https://lkml.kernel.org/r/20210713104716.22868-3-jack@suse.cz Signed-off-by: Jan Kara <jack@suse.cz> Reported-by: Michael Stapelberg <stapelberg+linux@google.com> Tested-by: Michael Stapelberg <stapelberg+linux@google.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-09-02 21:53:09 +00:00
spin_lock(&wb->list_lock);
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
/*
writeback: fix bandwidth estimate for spiky workload Michael Stapelberg has reported that for workload with short big spikes of writes (GCC linker seem to trigger this frequently) the write throughput is heavily underestimated and tends to steadily sink until it reaches zero. This has rather bad impact on writeback throttling (causing stalls). The problem is that writeback throughput estimate gets updated at most once per 200 ms. One update happens early after we submit pages for writeback (at that point writeout of only small fraction of pages is completed and thus observed throughput is tiny). Next update happens only during the next write spike (updates happen only from inode writeback and dirty throttling code) and if that is more than 1s after previous spike, we decide system was idle and just ignore whatever was written until this moment. Fix the problem by making sure writeback throughput estimate is also updated shortly after writeback completes to get reasonable estimate of throughput for spiky workloads. [jack@suse.cz: avoid division by 0 in wb_update_dirty_ratelimit()] Link: https://lore.kernel.org/lkml/20210617095309.3542373-1-stapelberg+linux@google.com Link: https://lkml.kernel.org/r/20210713104716.22868-3-jack@suse.cz Signed-off-by: Jan Kara <jack@suse.cz> Reported-by: Michael Stapelberg <stapelberg+linux@google.com> Tested-by: Michael Stapelberg <stapelberg+linux@google.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-09-02 21:53:09 +00:00
* Lockless checks for elapsed time are racy and delayed update after
* IO completion doesn't do it at all (to make sure written pages are
* accounted reasonably quickly). Make sure elapsed >= 1 to avoid
* division errors.
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
*/
writeback: fix bandwidth estimate for spiky workload Michael Stapelberg has reported that for workload with short big spikes of writes (GCC linker seem to trigger this frequently) the write throughput is heavily underestimated and tends to steadily sink until it reaches zero. This has rather bad impact on writeback throttling (causing stalls). The problem is that writeback throughput estimate gets updated at most once per 200 ms. One update happens early after we submit pages for writeback (at that point writeout of only small fraction of pages is completed and thus observed throughput is tiny). Next update happens only during the next write spike (updates happen only from inode writeback and dirty throttling code) and if that is more than 1s after previous spike, we decide system was idle and just ignore whatever was written until this moment. Fix the problem by making sure writeback throughput estimate is also updated shortly after writeback completes to get reasonable estimate of throughput for spiky workloads. [jack@suse.cz: avoid division by 0 in wb_update_dirty_ratelimit()] Link: https://lore.kernel.org/lkml/20210617095309.3542373-1-stapelberg+linux@google.com Link: https://lkml.kernel.org/r/20210713104716.22868-3-jack@suse.cz Signed-off-by: Jan Kara <jack@suse.cz> Reported-by: Michael Stapelberg <stapelberg+linux@google.com> Tested-by: Michael Stapelberg <stapelberg+linux@google.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-09-02 21:53:09 +00:00
elapsed = max(now - wb->bw_time_stamp, 1UL);
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
dirtied = percpu_counter_read(&wb->stat[WB_DIRTIED]);
written = percpu_counter_read(&wb->stat[WB_WRITTEN]);
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
if (update_ratelimit) {
domain_update_dirty_limit(gdtc, now);
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
wb_update_dirty_ratelimit(gdtc, dirtied, elapsed);
/*
* @mdtc is always NULL if !CGROUP_WRITEBACK but the
* compiler has no way to figure that out. Help it.
*/
if (IS_ENABLED(CONFIG_CGROUP_WRITEBACK) && mdtc) {
domain_update_dirty_limit(mdtc, now);
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
wb_update_dirty_ratelimit(mdtc, dirtied, elapsed);
}
writeback: dirty rate control It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N) when there are N dd tasks. On write() syscall, use bdi->dirty_ratelimit ============================================ balance_dirty_pages(pages_dirtied) { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); pause = pages_dirtied / task_ratelimit; sleep(pause); } On every 200ms, update bdi->dirty_ratelimit =========================================== bdi_update_dirty_ratelimit() { task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio(); balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate; bdi->dirty_ratelimit = balanced_dirty_ratelimit } Estimation of balanced bdi->dirty_ratelimit =========================================== balanced task_ratelimit ----------------------- balance_dirty_pages() needs to throttle tasks dirtying pages such that the total amount of dirty pages stays below the specified dirty limit in order to avoid memory deadlocks. Furthermore we desire fairness in that tasks get throttled proportionally to the amount of pages they dirty. IOW we want to throttle tasks such that we match the dirty rate to the writeout bandwidth, this yields a stable amount of dirty pages: dirty_rate == write_bw (1) The fairness requirement gives us: task_ratelimit = balanced_dirty_ratelimit == write_bw / N (2) where N is the number of dd tasks. We don't know N beforehand, but still can estimate balanced_dirty_ratelimit within 200ms. Start by throttling each dd task at rate task_ratelimit = task_ratelimit_0 (3) (any non-zero initial value is OK) After 200ms, we measured dirty_rate = # of pages dirtied by all dd's / 200ms write_bw = # of pages written to the disk / 200ms For the aggressive dd dirtiers, the equality holds dirty_rate == N * task_rate == N * task_ratelimit_0 (4) Or task_ratelimit_0 == dirty_rate / N (5) Now we conclude that the balanced task ratelimit can be estimated by write_bw balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6) dirty_rate Because with (4) and (5) we can get the desired equality (1): write_bw balanced_dirty_ratelimit == (dirty_rate / N) * ---------- dirty_rate == write_bw / N Then using the balanced task ratelimit we can compute task pause times like: task_pause = task->nr_dirtied / task_ratelimit task_ratelimit with position control ------------------------------------ However, while the above gives us means of matching the dirty rate to the writeout bandwidth, it at best provides us with a stable dirty page count (assuming a static system). In order to control the dirty page count such that it is high enough to provide performance, but does not exceed the specified limit we need another control. The dirty position control works by extending (2) to task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7) where pos_ratio is a negative feedback function that subjects to 1) f(setpoint) = 1.0 2) df/dx < 0 That is, if the dirty pages are ABOVE the setpoint, we throttle each task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty pages are created less fast than they are cleaned, thus DROP to the setpoints (and the reverse). Based on (7) and the assumption that both dirty_ratelimit and pos_ratio remains CONSTANT for the past 200ms, we get task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8) Putting (8) into (6), we get the formula used in bdi_update_dirty_ratelimit(): write_bw balanced_dirty_ratelimit *= pos_ratio * ---------- (9) dirty_rate Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 16:51:31 +00:00
}
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
wb_update_write_bandwidth(wb, elapsed, written);
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
wb->dirtied_stamp = dirtied;
wb->written_stamp = written;
WRITE_ONCE(wb->bw_time_stamp, now);
writeback: fix bandwidth estimate for spiky workload Michael Stapelberg has reported that for workload with short big spikes of writes (GCC linker seem to trigger this frequently) the write throughput is heavily underestimated and tends to steadily sink until it reaches zero. This has rather bad impact on writeback throttling (causing stalls). The problem is that writeback throughput estimate gets updated at most once per 200 ms. One update happens early after we submit pages for writeback (at that point writeout of only small fraction of pages is completed and thus observed throughput is tiny). Next update happens only during the next write spike (updates happen only from inode writeback and dirty throttling code) and if that is more than 1s after previous spike, we decide system was idle and just ignore whatever was written until this moment. Fix the problem by making sure writeback throughput estimate is also updated shortly after writeback completes to get reasonable estimate of throughput for spiky workloads. [jack@suse.cz: avoid division by 0 in wb_update_dirty_ratelimit()] Link: https://lore.kernel.org/lkml/20210617095309.3542373-1-stapelberg+linux@google.com Link: https://lkml.kernel.org/r/20210713104716.22868-3-jack@suse.cz Signed-off-by: Jan Kara <jack@suse.cz> Reported-by: Michael Stapelberg <stapelberg+linux@google.com> Tested-by: Michael Stapelberg <stapelberg+linux@google.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-09-02 21:53:09 +00:00
spin_unlock(&wb->list_lock);
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
}
writeback: fix bandwidth estimate for spiky workload Michael Stapelberg has reported that for workload with short big spikes of writes (GCC linker seem to trigger this frequently) the write throughput is heavily underestimated and tends to steadily sink until it reaches zero. This has rather bad impact on writeback throttling (causing stalls). The problem is that writeback throughput estimate gets updated at most once per 200 ms. One update happens early after we submit pages for writeback (at that point writeout of only small fraction of pages is completed and thus observed throughput is tiny). Next update happens only during the next write spike (updates happen only from inode writeback and dirty throttling code) and if that is more than 1s after previous spike, we decide system was idle and just ignore whatever was written until this moment. Fix the problem by making sure writeback throughput estimate is also updated shortly after writeback completes to get reasonable estimate of throughput for spiky workloads. [jack@suse.cz: avoid division by 0 in wb_update_dirty_ratelimit()] Link: https://lore.kernel.org/lkml/20210617095309.3542373-1-stapelberg+linux@google.com Link: https://lkml.kernel.org/r/20210713104716.22868-3-jack@suse.cz Signed-off-by: Jan Kara <jack@suse.cz> Reported-by: Michael Stapelberg <stapelberg+linux@google.com> Tested-by: Michael Stapelberg <stapelberg+linux@google.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-09-02 21:53:09 +00:00
void wb_update_bandwidth(struct bdi_writeback *wb)
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
{
struct dirty_throttle_control gdtc = { GDTC_INIT(wb) };
__wb_update_bandwidth(&gdtc, NULL, false);
}
/* Interval after which we consider wb idle and don't estimate bandwidth */
#define WB_BANDWIDTH_IDLE_JIF (HZ)
static void wb_bandwidth_estimate_start(struct bdi_writeback *wb)
{
unsigned long now = jiffies;
unsigned long elapsed = now - READ_ONCE(wb->bw_time_stamp);
if (elapsed > WB_BANDWIDTH_IDLE_JIF &&
!atomic_read(&wb->writeback_inodes)) {
spin_lock(&wb->list_lock);
wb->dirtied_stamp = wb_stat(wb, WB_DIRTIED);
wb->written_stamp = wb_stat(wb, WB_WRITTEN);
WRITE_ONCE(wb->bw_time_stamp, now);
spin_unlock(&wb->list_lock);
}
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
}
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
/*
* After a task dirtied this many pages, balance_dirty_pages_ratelimited()
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
* will look to see if it needs to start dirty throttling.
*
* If dirty_poll_interval is too low, big NUMA machines will call the expensive
* global_zone_page_state() too often. So scale it near-sqrt to the safety margin
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
* (the number of pages we may dirty without exceeding the dirty limits).
*/
static unsigned long dirty_poll_interval(unsigned long dirty,
unsigned long thresh)
{
if (thresh > dirty)
return 1UL << (ilog2(thresh - dirty) >> 1);
return 1;
}
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
static unsigned long wb_max_pause(struct bdi_writeback *wb,
unsigned long wb_dirty)
{
unsigned long bw = READ_ONCE(wb->avg_write_bandwidth);
writeback: fix negative bdi max pause Toralf runs trinity on UML/i386. After some time it hangs and the last message line is BUG: soft lockup - CPU#0 stuck for 22s! [trinity-child0:1521] It's found that pages_dirtied becomes very large. More than 1000000000 pages in this case: period = HZ * pages_dirtied / task_ratelimit; BUG_ON(pages_dirtied > 2000000000); BUG_ON(pages_dirtied > 1000000000); <--------- UML debug printf shows that we got negative pause here: ick: pause : -984 ick: pages_dirtied : 0 ick: task_ratelimit: 0 pause: + if (pause < 0) { + extern int printf(char *, ...); + printf("ick : pause : %li\n", pause); + printf("ick: pages_dirtied : %lu\n", pages_dirtied); + printf("ick: task_ratelimit: %lu\n", task_ratelimit); + BUG_ON(1); + } trace_balance_dirty_pages(bdi, Since pause is bounded by [min_pause, max_pause] where min_pause is also bounded by max_pause. It's suspected and demonstrated that the max_pause calculation goes wrong: ick: pause : -717 ick: min_pause : -177 ick: max_pause : -717 ick: pages_dirtied : 14 ick: task_ratelimit: 0 The problem lies in the two "long = unsigned long" assignments in bdi_max_pause() which might go negative if the highest bit is 1, and the min_t(long, ...) check failed to protect it falling under 0. Fix all of them by using "unsigned long" throughout the function. Signed-off-by: Fengguang Wu <fengguang.wu@intel.com> Reported-by: Toralf Förster <toralf.foerster@gmx.de> Tested-by: Toralf Förster <toralf.foerster@gmx.de> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Richard Weinberger <richard@nod.at> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-10-16 20:47:03 +00:00
unsigned long t;
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
/*
* Limit pause time for small memory systems. If sleeping for too long
* time, a small pool of dirty/writeback pages may go empty and disk go
* idle.
*
* 8 serves as the safety ratio.
*/
t = wb_dirty / (1 + bw / roundup_pow_of_two(1 + HZ / 8));
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
t++;
writeback: fix negative bdi max pause Toralf runs trinity on UML/i386. After some time it hangs and the last message line is BUG: soft lockup - CPU#0 stuck for 22s! [trinity-child0:1521] It's found that pages_dirtied becomes very large. More than 1000000000 pages in this case: period = HZ * pages_dirtied / task_ratelimit; BUG_ON(pages_dirtied > 2000000000); BUG_ON(pages_dirtied > 1000000000); <--------- UML debug printf shows that we got negative pause here: ick: pause : -984 ick: pages_dirtied : 0 ick: task_ratelimit: 0 pause: + if (pause < 0) { + extern int printf(char *, ...); + printf("ick : pause : %li\n", pause); + printf("ick: pages_dirtied : %lu\n", pages_dirtied); + printf("ick: task_ratelimit: %lu\n", task_ratelimit); + BUG_ON(1); + } trace_balance_dirty_pages(bdi, Since pause is bounded by [min_pause, max_pause] where min_pause is also bounded by max_pause. It's suspected and demonstrated that the max_pause calculation goes wrong: ick: pause : -717 ick: min_pause : -177 ick: max_pause : -717 ick: pages_dirtied : 14 ick: task_ratelimit: 0 The problem lies in the two "long = unsigned long" assignments in bdi_max_pause() which might go negative if the highest bit is 1, and the min_t(long, ...) check failed to protect it falling under 0. Fix all of them by using "unsigned long" throughout the function. Signed-off-by: Fengguang Wu <fengguang.wu@intel.com> Reported-by: Toralf Förster <toralf.foerster@gmx.de> Tested-by: Toralf Förster <toralf.foerster@gmx.de> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Richard Weinberger <richard@nod.at> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-10-16 20:47:03 +00:00
return min_t(unsigned long, t, MAX_PAUSE);
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
}
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
static long wb_min_pause(struct bdi_writeback *wb,
long max_pause,
unsigned long task_ratelimit,
unsigned long dirty_ratelimit,
int *nr_dirtied_pause)
{
long hi = ilog2(READ_ONCE(wb->avg_write_bandwidth));
long lo = ilog2(READ_ONCE(wb->dirty_ratelimit));
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
long t; /* target pause */
long pause; /* estimated next pause */
int pages; /* target nr_dirtied_pause */
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
/* target for 10ms pause on 1-dd case */
t = max(1, HZ / 100);
/*
* Scale up pause time for concurrent dirtiers in order to reduce CPU
* overheads.
*
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
* (N * 10ms) on 2^N concurrent tasks.
*/
if (hi > lo)
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
t += (hi - lo) * (10 * HZ) / 1024;
/*
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
* This is a bit convoluted. We try to base the next nr_dirtied_pause
* on the much more stable dirty_ratelimit. However the next pause time
* will be computed based on task_ratelimit and the two rate limits may
* depart considerably at some time. Especially if task_ratelimit goes
* below dirty_ratelimit/2 and the target pause is max_pause, the next
* pause time will be max_pause*2 _trimmed down_ to max_pause. As a
* result task_ratelimit won't be executed faithfully, which could
* eventually bring down dirty_ratelimit.
*
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
* We apply two rules to fix it up:
* 1) try to estimate the next pause time and if necessary, use a lower
* nr_dirtied_pause so as not to exceed max_pause. When this happens,
* nr_dirtied_pause will be "dancing" with task_ratelimit.
* 2) limit the target pause time to max_pause/2, so that the normal
* small fluctuations of task_ratelimit won't trigger rule (1) and
* nr_dirtied_pause will remain as stable as dirty_ratelimit.
*/
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
t = min(t, 1 + max_pause / 2);
pages = dirty_ratelimit * t / roundup_pow_of_two(HZ);
/*
* Tiny nr_dirtied_pause is found to hurt I/O performance in the test
* case fio-mmap-randwrite-64k, which does 16*{sync read, async write}.
* When the 16 consecutive reads are often interrupted by some dirty
* throttling pause during the async writes, cfq will go into idles
* (deadline is fine). So push nr_dirtied_pause as high as possible
* until reaches DIRTY_POLL_THRESH=32 pages.
*/
if (pages < DIRTY_POLL_THRESH) {
t = max_pause;
pages = dirty_ratelimit * t / roundup_pow_of_two(HZ);
if (pages > DIRTY_POLL_THRESH) {
pages = DIRTY_POLL_THRESH;
t = HZ * DIRTY_POLL_THRESH / dirty_ratelimit;
}
}
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
pause = HZ * pages / (task_ratelimit + 1);
if (pause > max_pause) {
t = max_pause;
pages = task_ratelimit * t / roundup_pow_of_two(HZ);
}
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
*nr_dirtied_pause = pages;
/*
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
* The minimal pause time will normally be half the target pause time.
*/
return pages >= DIRTY_POLL_THRESH ? 1 + t / 2 : t;
}
static inline void wb_dirty_limits(struct dirty_throttle_control *dtc)
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
{
struct bdi_writeback *wb = dtc->wb;
unsigned long wb_reclaimable;
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
/*
* wb_thresh is not treated as some limiting factor as
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* dirty_thresh, due to reasons
* - in JBOD setup, wb_thresh can fluctuate a lot
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* - in a system with HDD and USB key, the USB key may somehow
* go into state (wb_dirty >> wb_thresh) either because
* wb_dirty starts high, or because wb_thresh drops low.
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
* In this case we don't want to hard throttle the USB key
* dirtiers for 100 seconds until wb_dirty drops under
* wb_thresh. Instead the auxiliary wb control line in
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
* wb_position_ratio() will let the dirtier task progress
* at some rate <= (write_bw / 2) for bringing down wb_dirty.
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
*/
dtc->wb_thresh = __wb_calc_thresh(dtc, dtc->thresh);
dtc->wb_bg_thresh = dtc->thresh ?
div_u64((u64)dtc->wb_thresh * dtc->bg_thresh, dtc->thresh) : 0;
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
/*
* In order to avoid the stacked BDI deadlock we need
* to ensure we accurately count the 'dirty' pages when
* the threshold is low.
*
* Otherwise it would be possible to get thresh+n pages
* reported dirty, even though there are thresh-m pages
* actually dirty; with m+n sitting in the percpu
* deltas.
*/
if (dtc->wb_thresh < 2 * wb_stat_error()) {
wb_reclaimable = wb_stat_sum(wb, WB_RECLAIMABLE);
dtc->wb_dirty = wb_reclaimable + wb_stat_sum(wb, WB_WRITEBACK);
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
} else {
wb_reclaimable = wb_stat(wb, WB_RECLAIMABLE);
dtc->wb_dirty = wb_reclaimable + wb_stat(wb, WB_WRITEBACK);
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
}
}
static unsigned long domain_poll_intv(struct dirty_throttle_control *dtc,
bool strictlimit)
{
unsigned long dirty, thresh;
if (strictlimit) {
dirty = dtc->wb_dirty;
thresh = dtc->wb_thresh;
} else {
dirty = dtc->dirty;
thresh = dtc->thresh;
}
return dirty_poll_interval(dirty, thresh);
}
/*
* Throttle it only when the background writeback cannot catch-up. This avoids
* (excessively) small writeouts when the wb limits are ramping up in case of
* !strictlimit.
*
* In strictlimit case make decision based on the wb counters and limits. Small
* writeouts when the wb limits are ramping up are the price we consciously pay
* for strictlimit-ing.
*/
static void domain_dirty_freerun(struct dirty_throttle_control *dtc,
bool strictlimit)
{
unsigned long dirty, thresh, bg_thresh;
if (unlikely(strictlimit)) {
wb_dirty_limits(dtc);
dirty = dtc->wb_dirty;
thresh = dtc->wb_thresh;
bg_thresh = dtc->wb_bg_thresh;
} else {
dirty = dtc->dirty;
thresh = dtc->thresh;
bg_thresh = dtc->bg_thresh;
}
dtc->freerun = dirty <= dirty_freerun_ceiling(thresh, bg_thresh);
}
static void balance_domain_limits(struct dirty_throttle_control *dtc,
bool strictlimit)
{
domain_dirty_avail(dtc, true);
domain_dirty_limits(dtc);
domain_dirty_freerun(dtc, strictlimit);
}
static void wb_dirty_freerun(struct dirty_throttle_control *dtc,
bool strictlimit)
{
dtc->freerun = false;
/* was already handled in domain_dirty_freerun */
if (strictlimit)
return;
wb_dirty_limits(dtc);
/*
* LOCAL_THROTTLE tasks must not be throttled when below the per-wb
* freerun ceiling.
*/
if (!(current->flags & PF_LOCAL_THROTTLE))
return;
dtc->freerun = dtc->wb_dirty <
dirty_freerun_ceiling(dtc->wb_thresh, dtc->wb_bg_thresh);
}
static inline void wb_dirty_exceeded(struct dirty_throttle_control *dtc,
bool strictlimit)
{
dtc->dirty_exceeded = (dtc->wb_dirty > dtc->wb_thresh) &&
((dtc->dirty > dtc->thresh) || strictlimit);
}
/*
* The limits fields dirty_exceeded and pos_ratio won't be updated if wb is
* in freerun state. Please don't use these invalid fields in freerun case.
*/
static void balance_wb_limits(struct dirty_throttle_control *dtc,
bool strictlimit)
{
wb_dirty_freerun(dtc, strictlimit);
if (dtc->freerun)
return;
wb_dirty_exceeded(dtc, strictlimit);
wb_position_ratio(dtc);
}
/*
* balance_dirty_pages() must be called by processes which are generating dirty
* data. It looks at the number of dirty pages in the machine and will force
writeback: IO-less balance_dirty_pages() As proposed by Chris, Dave and Jan, don't start foreground writeback IO inside balance_dirty_pages(). Instead, simply let it idle sleep for some time to throttle the dirtying task. In the mean while, kick off the per-bdi flusher thread to do background writeback IO. RATIONALS ========= - disk seeks on concurrent writeback of multiple inodes (Dave Chinner) If every thread doing writes and being throttled start foreground writeback, it leads to N IO submitters from at least N different inodes at the same time, end up with N different sets of IO being issued with potentially zero locality to each other, resulting in much lower elevator sort/merge efficiency and hence we seek the disk all over the place to service the different sets of IO. OTOH, if there is only one submission thread, it doesn't jump between inodes in the same way when congestion clears - it keeps writing to the same inode, resulting in large related chunks of sequential IOs being issued to the disk. This is more efficient than the above foreground writeback because the elevator works better and the disk seeks less. - lock contention and cache bouncing on concurrent IO submitters (Dave Chinner) With this patchset, the fs_mark benchmark on a 12-drive software RAID0 goes from CPU bound to IO bound, freeing "3-4 CPUs worth of spinlock contention". * "CPU usage has dropped by ~55%", "it certainly appears that most of the CPU time saving comes from the removal of contention on the inode_wb_list_lock" (IMHO at least 10% comes from the reduction of cacheline bouncing, because the new code is able to call much less frequently into balance_dirty_pages() and hence access the global page states) * the user space "App overhead" is reduced by 20%, by avoiding the cacheline pollution by the complex writeback code path * "for a ~5% throughput reduction", "the number of write IOs have dropped by ~25%", and the elapsed time reduced from 41:42.17 to 40:53.23. * On a simple test of 100 dd, it reduces the CPU %system time from 30% to 3%, and improves IO throughput from 38MB/s to 42MB/s. - IO size too small for fast arrays and too large for slow USB sticks The write_chunk used by current balance_dirty_pages() cannot be directly set to some large value (eg. 128MB) for better IO efficiency. Because it could lead to more than 1 second user perceivable stalls. Even the current 4MB write size may be too large for slow USB sticks. The fact that balance_dirty_pages() starts IO on itself couples the IO size to wait time, which makes it hard to do suitable IO size while keeping the wait time under control. Now it's possible to increase writeback chunk size proportional to the disk bandwidth. In a simple test of 50 dd's on XFS, 1-HDD, 3GB ram, the larger writeback size dramatically reduces the seek count to 1/10 (far beyond my expectation) and improves the write throughput by 24%. - long block time in balance_dirty_pages() hurts desktop responsiveness Many of us may have the experience: it often takes a couple of seconds or even long time to stop a heavy writing dd/cp/tar command with Ctrl-C or "kill -9". - IO pipeline broken by bumpy write() progress There are a broad class of "loop {read(buf); write(buf);}" applications whose read() pipeline will be under-utilized or even come to a stop if the write()s have long latencies _or_ don't progress in a constant rate. The current threshold based throttling inherently transfers the large low level IO completion fluctuations to bumpy application write()s, and further deteriorates with increasing number of dirtiers and/or bdi's. For example, when doing 50 dd's + 1 remote rsync to an XFS partition, the rsync progresses very bumpy in legacy kernel, and throughput is improved by 67% by this patchset. (plus the larger write chunk size, it will be 93% speedup). The new rate based throttling can support 1000+ dd's with excellent smoothness, low latency and low overheads. For the above reasons, it's much better to do IO-less and low latency pauses in balance_dirty_pages(). Jan Kara, Dave Chinner and me explored the scheme to let balance_dirty_pages() wait for enough writeback IO completions to safeguard the dirty limit. However it's found to have two problems: - in large NUMA systems, the per-cpu counters may have big accounting errors, leading to big throttle wait time and jitters. - NFS may kill large amount of unstable pages with one single COMMIT. Because NFS server serves COMMIT with expensive fsync() IOs, it is desirable to delay and reduce the number of COMMITs. So it's not likely to optimize away such kind of bursty IO completions, and the resulted large (and tiny) stall times in IO completion based throttling. So here is a pause time oriented approach, which tries to control the pause time in each balance_dirty_pages() invocations, by controlling the number of pages dirtied before calling balance_dirty_pages(), for smooth and efficient dirty throttling: - avoid useless (eg. zero pause time) balance_dirty_pages() calls - avoid too small pause time (less than 4ms, which burns CPU power) - avoid too large pause time (more than 200ms, which hurts responsiveness) - avoid big fluctuations of pause times It can control pause times at will. The default policy (in a followup patch) will be to do ~10ms pauses in 1-dd case, and increase to ~100ms in 1000-dd case. BEHAVIOR CHANGE =============== (1) dirty threshold Users will notice that the applications will get throttled once crossing the global (background + dirty)/2=15% threshold, and then balanced around 17.5%. Before patch, the behavior is to just throttle it at 20% dirtyable memory in 1-dd case. Since the task will be soft throttled earlier than before, it may be perceived by end users as performance "slow down" if his application happens to dirty more than 15% dirtyable memory. (2) smoothness/responsiveness Users will notice a more responsive system during heavy writeback. "killall dd" will take effect instantly. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-28 00:45:12 +00:00
* the caller to wait once crossing the (background_thresh + dirty_thresh) / 2.
* If we're over `background_thresh' then the writeback threads are woken to
* perform some writeout.
*/
static int balance_dirty_pages(struct bdi_writeback *wb,
unsigned long pages_dirtied, unsigned int flags)
{
struct dirty_throttle_control gdtc_stor = { GDTC_INIT(wb) };
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
struct dirty_throttle_control mdtc_stor = { MDTC_INIT(wb, &gdtc_stor) };
struct dirty_throttle_control * const gdtc = &gdtc_stor;
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
struct dirty_throttle_control * const mdtc = mdtc_valid(&mdtc_stor) ?
&mdtc_stor : NULL;
struct dirty_throttle_control *sdtc;
unsigned long nr_dirty;
long period;
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
long pause;
long max_pause;
long min_pause;
int nr_dirtied_pause;
writeback: IO-less balance_dirty_pages() As proposed by Chris, Dave and Jan, don't start foreground writeback IO inside balance_dirty_pages(). Instead, simply let it idle sleep for some time to throttle the dirtying task. In the mean while, kick off the per-bdi flusher thread to do background writeback IO. RATIONALS ========= - disk seeks on concurrent writeback of multiple inodes (Dave Chinner) If every thread doing writes and being throttled start foreground writeback, it leads to N IO submitters from at least N different inodes at the same time, end up with N different sets of IO being issued with potentially zero locality to each other, resulting in much lower elevator sort/merge efficiency and hence we seek the disk all over the place to service the different sets of IO. OTOH, if there is only one submission thread, it doesn't jump between inodes in the same way when congestion clears - it keeps writing to the same inode, resulting in large related chunks of sequential IOs being issued to the disk. This is more efficient than the above foreground writeback because the elevator works better and the disk seeks less. - lock contention and cache bouncing on concurrent IO submitters (Dave Chinner) With this patchset, the fs_mark benchmark on a 12-drive software RAID0 goes from CPU bound to IO bound, freeing "3-4 CPUs worth of spinlock contention". * "CPU usage has dropped by ~55%", "it certainly appears that most of the CPU time saving comes from the removal of contention on the inode_wb_list_lock" (IMHO at least 10% comes from the reduction of cacheline bouncing, because the new code is able to call much less frequently into balance_dirty_pages() and hence access the global page states) * the user space "App overhead" is reduced by 20%, by avoiding the cacheline pollution by the complex writeback code path * "for a ~5% throughput reduction", "the number of write IOs have dropped by ~25%", and the elapsed time reduced from 41:42.17 to 40:53.23. * On a simple test of 100 dd, it reduces the CPU %system time from 30% to 3%, and improves IO throughput from 38MB/s to 42MB/s. - IO size too small for fast arrays and too large for slow USB sticks The write_chunk used by current balance_dirty_pages() cannot be directly set to some large value (eg. 128MB) for better IO efficiency. Because it could lead to more than 1 second user perceivable stalls. Even the current 4MB write size may be too large for slow USB sticks. The fact that balance_dirty_pages() starts IO on itself couples the IO size to wait time, which makes it hard to do suitable IO size while keeping the wait time under control. Now it's possible to increase writeback chunk size proportional to the disk bandwidth. In a simple test of 50 dd's on XFS, 1-HDD, 3GB ram, the larger writeback size dramatically reduces the seek count to 1/10 (far beyond my expectation) and improves the write throughput by 24%. - long block time in balance_dirty_pages() hurts desktop responsiveness Many of us may have the experience: it often takes a couple of seconds or even long time to stop a heavy writing dd/cp/tar command with Ctrl-C or "kill -9". - IO pipeline broken by bumpy write() progress There are a broad class of "loop {read(buf); write(buf);}" applications whose read() pipeline will be under-utilized or even come to a stop if the write()s have long latencies _or_ don't progress in a constant rate. The current threshold based throttling inherently transfers the large low level IO completion fluctuations to bumpy application write()s, and further deteriorates with increasing number of dirtiers and/or bdi's. For example, when doing 50 dd's + 1 remote rsync to an XFS partition, the rsync progresses very bumpy in legacy kernel, and throughput is improved by 67% by this patchset. (plus the larger write chunk size, it will be 93% speedup). The new rate based throttling can support 1000+ dd's with excellent smoothness, low latency and low overheads. For the above reasons, it's much better to do IO-less and low latency pauses in balance_dirty_pages(). Jan Kara, Dave Chinner and me explored the scheme to let balance_dirty_pages() wait for enough writeback IO completions to safeguard the dirty limit. However it's found to have two problems: - in large NUMA systems, the per-cpu counters may have big accounting errors, leading to big throttle wait time and jitters. - NFS may kill large amount of unstable pages with one single COMMIT. Because NFS server serves COMMIT with expensive fsync() IOs, it is desirable to delay and reduce the number of COMMITs. So it's not likely to optimize away such kind of bursty IO completions, and the resulted large (and tiny) stall times in IO completion based throttling. So here is a pause time oriented approach, which tries to control the pause time in each balance_dirty_pages() invocations, by controlling the number of pages dirtied before calling balance_dirty_pages(), for smooth and efficient dirty throttling: - avoid useless (eg. zero pause time) balance_dirty_pages() calls - avoid too small pause time (less than 4ms, which burns CPU power) - avoid too large pause time (more than 200ms, which hurts responsiveness) - avoid big fluctuations of pause times It can control pause times at will. The default policy (in a followup patch) will be to do ~10ms pauses in 1-dd case, and increase to ~100ms in 1000-dd case. BEHAVIOR CHANGE =============== (1) dirty threshold Users will notice that the applications will get throttled once crossing the global (background + dirty)/2=15% threshold, and then balanced around 17.5%. Before patch, the behavior is to just throttle it at 20% dirtyable memory in 1-dd case. Since the task will be soft throttled earlier than before, it may be perceived by end users as performance "slow down" if his application happens to dirty more than 15% dirtyable memory. (2) smoothness/responsiveness Users will notice a more responsive system during heavy writeback. "killall dd" will take effect instantly. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-28 00:45:12 +00:00
unsigned long task_ratelimit;
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
unsigned long dirty_ratelimit;
struct backing_dev_info *bdi = wb->bdi;
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
bool strictlimit = bdi->capabilities & BDI_CAP_STRICTLIMIT;
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
unsigned long start_time = jiffies;
int ret = 0;
for (;;) {
unsigned long now = jiffies;
nr_dirty = global_node_page_state(NR_FILE_DIRTY);
balance_domain_limits(gdtc, strictlimit);
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
if (mdtc) {
/*
* If @wb belongs to !root memcg, repeat the same
* basic calculations for the memcg domain.
*/
balance_domain_limits(mdtc, strictlimit);
mm/page-writeback.c: add strictlimit feature The feature prevents mistrusted filesystems (ie: FUSE mounts created by unprivileged users) to grow a large number of dirty pages before throttling. For such filesystems balance_dirty_pages always check bdi counters against bdi limits. I.e. even if global "nr_dirty" is under "freerun", it's not allowed to skip bdi checks. The only use case for now is fuse: it sets bdi max_ratio to 1% by default and system administrators are supposed to expect that this limit won't be exceeded. The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A filesystem may set the flag when it initializes its BDI. The problematic scenario comes from the fact that nobody pays attention to the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse writeback). The implementation of fuse writeback releases original page (by calling end_page_writeback) almost immediately. A fuse request queued for real processing bears a copy of original page. Hence, if userspace fuse daemon doesn't finalize write requests in timely manner, an aggressive mmap writer can pollute virtually all memory by those temporary fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but nobody cares. To make further explanations shorter, let me use "NR_WRITEBACK_TEMP problem" as a shortcut for "a possibility of uncontrolled grow of amount of RAM consumed by temporary pages allocated by kernel fuse to process writeback". The problem was very easy to reproduce. There is a trivial example filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I added "sleep(1);" to the write methods, then recompiled and mounted it. Then created a huge file on the mount point and run a simple program which mmap-ed the file to a memory region, then wrote a data to the region. An hour later I observed almost all RAM consumed by fuse writeback. Since then some unrelated changes in kernel fuse made it more difficult to reproduce, but it is still possible now. Putting this theoretical happens-in-the-lab thing aside, there is another thing that really hurts real world (FUSE) users. This is write-through page cache policy FUSE currently uses. I.e. handling write(2), kernel fuse populates page cache and flushes user data to the server synchronously. This is excessively suboptimal. Pavel Emelyanov's patches ("writeback cache policy") solve the problem, but they also make resolving NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying a huge file to a fuse mount would result in memory starvation. Miklos, the maintainer of FUSE, believes strictlimit feature the way to go. And eventually putting FUSE topics aside, there is one more use-case for strictlimit feature. Using a slow USB stick (mass storage) in a machine with huge amount of RAM installed is a well-known pain. Let's make simple computations. Assuming 64GB of RAM installed, existing implementation of balance_dirty_pages will start throttling only after 9.6GB of RAM becomes dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file /media/my-usb-storage/" may return in a few seconds, but subsequent "umount /media/my-usb-storage/" will take more than two hours if effective throughput of the storage is, to say, 1MB/sec. After inclusion of strictlimit feature, it will be trivial to add a knob (e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand. Manually or via udev rule. May be I'm wrong, but it seems to be quite a natural desire to limit the amount of dirty memory for some devices we are not fully trust (in the sense of sustainable throughput). [akpm@linux-foundation.org: fix warning in page-writeback.c] Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com> Cc: Jan Kara <jack@suse.cz> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Pavel Emelyanov <xemul@parallels.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-11 21:22:46 +00:00
}
/*
* In laptop mode, we wait until hitting the higher threshold
* before starting background writeout, and then write out all
* the way down to the lower threshold. So slow writers cause
* minimal disk activity.
*
* In normal mode, we start background writeout at the lower
* background_thresh, to keep the amount of dirty memory low.
*/
if (!laptop_mode && nr_dirty > gdtc->bg_thresh &&
!writeback_in_progress(wb))
wb_start_background_writeback(wb);
/*
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
* If memcg domain is in effect, @dirty should be under
* both global and memcg freerun ceilings.
*/
if (gdtc->freerun && (!mdtc || mdtc->freerun)) {
mm/writeback: replace PF_LESS_THROTTLE with PF_LOCAL_THROTTLE PF_LESS_THROTTLE exists for loop-back nfsd (and a similar need in the loop block driver and callers of prctl(PR_SET_IO_FLUSHER)), where a daemon needs to write to one bdi (the final bdi) in order to free up writes queued to another bdi (the client bdi). The daemon sets PF_LESS_THROTTLE and gets a larger allowance of dirty pages, so that it can still dirty pages after other processses have been throttled. The purpose of this is to avoid deadlock that happen when the PF_LESS_THROTTLE process must write for any dirty pages to be freed, but it is being thottled and cannot write. This approach was designed when all threads were blocked equally, independently on which device they were writing to, or how fast it was. Since that time the writeback algorithm has changed substantially with different threads getting different allowances based on non-trivial heuristics. This means the simple "add 25%" heuristic is no longer reliable. The important issue is not that the daemon needs a *larger* dirty page allowance, but that it needs a *private* dirty page allowance, so that dirty pages for the "client" bdi that it is helping to clear (the bdi for an NFS filesystem or loop block device etc) do not affect the throttling of the daemon writing to the "final" bdi. This patch changes the heuristic so that the task is not throttled when the bdi it is writing to has a dirty page count below below (or equal to) the free-run threshold for that bdi. This ensures it will always be able to have some pages in flight, and so will not deadlock. In a steady-state, it is expected that PF_LOCAL_THROTTLE tasks might still be throttled by global threshold, but that is acceptable as it is only the deadlock state that is interesting for this flag. This approach of "only throttle when target bdi is busy" is consistent with the other use of PF_LESS_THROTTLE in current_may_throttle(), were it causes attention to be focussed only on the target bdi. So this patch - renames PF_LESS_THROTTLE to PF_LOCAL_THROTTLE, - removes the 25% bonus that that flag gives, and - If PF_LOCAL_THROTTLE is set, don't delay at all unless the global and the local free-run thresholds are exceeded. Note that previously realtime threads were treated the same as PF_LESS_THROTTLE threads. This patch does *not* change the behvaiour for real-time threads, so it is now different from the behaviour of nfsd and loop tasks. I don't know what is wanted for realtime. [akpm@linux-foundation.org: coding style fixes] Signed-off-by: NeilBrown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Jan Kara <jack@suse.cz> Acked-by: Chuck Lever <chuck.lever@oracle.com> [nfsd] Cc: Christoph Hellwig <hch@lst.de> Cc: Michal Hocko <mhocko@suse.com> Cc: Trond Myklebust <trond.myklebust@hammerspace.com> Link: http://lkml.kernel.org/r/87ftbf7gs3.fsf@notabene.neil.brown.name Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-02 04:48:18 +00:00
unsigned long intv;
unsigned long m_intv;
free_running:
intv = domain_poll_intv(gdtc, strictlimit);
mm/writeback: replace PF_LESS_THROTTLE with PF_LOCAL_THROTTLE PF_LESS_THROTTLE exists for loop-back nfsd (and a similar need in the loop block driver and callers of prctl(PR_SET_IO_FLUSHER)), where a daemon needs to write to one bdi (the final bdi) in order to free up writes queued to another bdi (the client bdi). The daemon sets PF_LESS_THROTTLE and gets a larger allowance of dirty pages, so that it can still dirty pages after other processses have been throttled. The purpose of this is to avoid deadlock that happen when the PF_LESS_THROTTLE process must write for any dirty pages to be freed, but it is being thottled and cannot write. This approach was designed when all threads were blocked equally, independently on which device they were writing to, or how fast it was. Since that time the writeback algorithm has changed substantially with different threads getting different allowances based on non-trivial heuristics. This means the simple "add 25%" heuristic is no longer reliable. The important issue is not that the daemon needs a *larger* dirty page allowance, but that it needs a *private* dirty page allowance, so that dirty pages for the "client" bdi that it is helping to clear (the bdi for an NFS filesystem or loop block device etc) do not affect the throttling of the daemon writing to the "final" bdi. This patch changes the heuristic so that the task is not throttled when the bdi it is writing to has a dirty page count below below (or equal to) the free-run threshold for that bdi. This ensures it will always be able to have some pages in flight, and so will not deadlock. In a steady-state, it is expected that PF_LOCAL_THROTTLE tasks might still be throttled by global threshold, but that is acceptable as it is only the deadlock state that is interesting for this flag. This approach of "only throttle when target bdi is busy" is consistent with the other use of PF_LESS_THROTTLE in current_may_throttle(), were it causes attention to be focussed only on the target bdi. So this patch - renames PF_LESS_THROTTLE to PF_LOCAL_THROTTLE, - removes the 25% bonus that that flag gives, and - If PF_LOCAL_THROTTLE is set, don't delay at all unless the global and the local free-run thresholds are exceeded. Note that previously realtime threads were treated the same as PF_LESS_THROTTLE threads. This patch does *not* change the behvaiour for real-time threads, so it is now different from the behaviour of nfsd and loop tasks. I don't know what is wanted for realtime. [akpm@linux-foundation.org: coding style fixes] Signed-off-by: NeilBrown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Jan Kara <jack@suse.cz> Acked-by: Chuck Lever <chuck.lever@oracle.com> [nfsd] Cc: Christoph Hellwig <hch@lst.de> Cc: Michal Hocko <mhocko@suse.com> Cc: Trond Myklebust <trond.myklebust@hammerspace.com> Link: http://lkml.kernel.org/r/87ftbf7gs3.fsf@notabene.neil.brown.name Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-02 04:48:18 +00:00
m_intv = ULONG_MAX;
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
current->dirty_paused_when = now;
current->nr_dirtied = 0;
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
if (mdtc)
m_intv = domain_poll_intv(mdtc, strictlimit);
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
current->nr_dirtied_pause = min(intv, m_intv);
break;
}
/* Start writeback even when in laptop mode */
if (unlikely(!writeback_in_progress(wb)))
wb_start_background_writeback(wb);
writeback: IO-less balance_dirty_pages() As proposed by Chris, Dave and Jan, don't start foreground writeback IO inside balance_dirty_pages(). Instead, simply let it idle sleep for some time to throttle the dirtying task. In the mean while, kick off the per-bdi flusher thread to do background writeback IO. RATIONALS ========= - disk seeks on concurrent writeback of multiple inodes (Dave Chinner) If every thread doing writes and being throttled start foreground writeback, it leads to N IO submitters from at least N different inodes at the same time, end up with N different sets of IO being issued with potentially zero locality to each other, resulting in much lower elevator sort/merge efficiency and hence we seek the disk all over the place to service the different sets of IO. OTOH, if there is only one submission thread, it doesn't jump between inodes in the same way when congestion clears - it keeps writing to the same inode, resulting in large related chunks of sequential IOs being issued to the disk. This is more efficient than the above foreground writeback because the elevator works better and the disk seeks less. - lock contention and cache bouncing on concurrent IO submitters (Dave Chinner) With this patchset, the fs_mark benchmark on a 12-drive software RAID0 goes from CPU bound to IO bound, freeing "3-4 CPUs worth of spinlock contention". * "CPU usage has dropped by ~55%", "it certainly appears that most of the CPU time saving comes from the removal of contention on the inode_wb_list_lock" (IMHO at least 10% comes from the reduction of cacheline bouncing, because the new code is able to call much less frequently into balance_dirty_pages() and hence access the global page states) * the user space "App overhead" is reduced by 20%, by avoiding the cacheline pollution by the complex writeback code path * "for a ~5% throughput reduction", "the number of write IOs have dropped by ~25%", and the elapsed time reduced from 41:42.17 to 40:53.23. * On a simple test of 100 dd, it reduces the CPU %system time from 30% to 3%, and improves IO throughput from 38MB/s to 42MB/s. - IO size too small for fast arrays and too large for slow USB sticks The write_chunk used by current balance_dirty_pages() cannot be directly set to some large value (eg. 128MB) for better IO efficiency. Because it could lead to more than 1 second user perceivable stalls. Even the current 4MB write size may be too large for slow USB sticks. The fact that balance_dirty_pages() starts IO on itself couples the IO size to wait time, which makes it hard to do suitable IO size while keeping the wait time under control. Now it's possible to increase writeback chunk size proportional to the disk bandwidth. In a simple test of 50 dd's on XFS, 1-HDD, 3GB ram, the larger writeback size dramatically reduces the seek count to 1/10 (far beyond my expectation) and improves the write throughput by 24%. - long block time in balance_dirty_pages() hurts desktop responsiveness Many of us may have the experience: it often takes a couple of seconds or even long time to stop a heavy writing dd/cp/tar command with Ctrl-C or "kill -9". - IO pipeline broken by bumpy write() progress There are a broad class of "loop {read(buf); write(buf);}" applications whose read() pipeline will be under-utilized or even come to a stop if the write()s have long latencies _or_ don't progress in a constant rate. The current threshold based throttling inherently transfers the large low level IO completion fluctuations to bumpy application write()s, and further deteriorates with increasing number of dirtiers and/or bdi's. For example, when doing 50 dd's + 1 remote rsync to an XFS partition, the rsync progresses very bumpy in legacy kernel, and throughput is improved by 67% by this patchset. (plus the larger write chunk size, it will be 93% speedup). The new rate based throttling can support 1000+ dd's with excellent smoothness, low latency and low overheads. For the above reasons, it's much better to do IO-less and low latency pauses in balance_dirty_pages(). Jan Kara, Dave Chinner and me explored the scheme to let balance_dirty_pages() wait for enough writeback IO completions to safeguard the dirty limit. However it's found to have two problems: - in large NUMA systems, the per-cpu counters may have big accounting errors, leading to big throttle wait time and jitters. - NFS may kill large amount of unstable pages with one single COMMIT. Because NFS server serves COMMIT with expensive fsync() IOs, it is desirable to delay and reduce the number of COMMITs. So it's not likely to optimize away such kind of bursty IO completions, and the resulted large (and tiny) stall times in IO completion based throttling. So here is a pause time oriented approach, which tries to control the pause time in each balance_dirty_pages() invocations, by controlling the number of pages dirtied before calling balance_dirty_pages(), for smooth and efficient dirty throttling: - avoid useless (eg. zero pause time) balance_dirty_pages() calls - avoid too small pause time (less than 4ms, which burns CPU power) - avoid too large pause time (more than 200ms, which hurts responsiveness) - avoid big fluctuations of pause times It can control pause times at will. The default policy (in a followup patch) will be to do ~10ms pauses in 1-dd case, and increase to ~100ms in 1000-dd case. BEHAVIOR CHANGE =============== (1) dirty threshold Users will notice that the applications will get throttled once crossing the global (background + dirty)/2=15% threshold, and then balanced around 17.5%. Before patch, the behavior is to just throttle it at 20% dirtyable memory in 1-dd case. Since the task will be soft throttled earlier than before, it may be perceived by end users as performance "slow down" if his application happens to dirty more than 15% dirtyable memory. (2) smoothness/responsiveness Users will notice a more responsive system during heavy writeback. "killall dd" will take effect instantly. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-28 00:45:12 +00:00
writeback, memcg: Implement foreign dirty flushing There's an inherent mismatch between memcg and writeback. The former trackes ownership per-page while the latter per-inode. This was a deliberate design decision because honoring per-page ownership in the writeback path is complicated, may lead to higher CPU and IO overheads and deemed unnecessary given that write-sharing an inode across different cgroups isn't a common use-case. Combined with inode majority-writer ownership switching, this works well enough in most cases but there are some pathological cases. For example, let's say there are two cgroups A and B which keep writing to different but confined parts of the same inode. B owns the inode and A's memory is limited far below B's. A's dirty ratio can rise enough to trigger balance_dirty_pages() sleeps but B's can be low enough to avoid triggering background writeback. A will be slowed down without a way to make writeback of the dirty pages happen. This patch implements foreign dirty recording and foreign mechanism so that when a memcg encounters a condition as above it can trigger flushes on bdi_writebacks which can clean its pages. Please see the comment on top of mem_cgroup_track_foreign_dirty_slowpath() for details. A reproducer follows. write-range.c:: #include <stdio.h> #include <stdlib.h> #include <unistd.h> #include <fcntl.h> #include <sys/types.h> static const char *usage = "write-range FILE START SIZE\n"; int main(int argc, char **argv) { int fd; unsigned long start, size, end, pos; char *endp; char buf[4096]; if (argc < 4) { fprintf(stderr, usage); return 1; } fd = open(argv[1], O_WRONLY); if (fd < 0) { perror("open"); return 1; } start = strtoul(argv[2], &endp, 0); if (*endp != '\0') { fprintf(stderr, usage); return 1; } size = strtoul(argv[3], &endp, 0); if (*endp != '\0') { fprintf(stderr, usage); return 1; } end = start + size; while (1) { for (pos = start; pos < end; ) { long bread, bwritten = 0; if (lseek(fd, pos, SEEK_SET) < 0) { perror("lseek"); return 1; } bread = read(0, buf, sizeof(buf) < end - pos ? sizeof(buf) : end - pos); if (bread < 0) { perror("read"); return 1; } if (bread == 0) return 0; while (bwritten < bread) { long this; this = write(fd, buf + bwritten, bread - bwritten); if (this < 0) { perror("write"); return 1; } bwritten += this; pos += bwritten; } } } } repro.sh:: #!/bin/bash set -e set -x sysctl -w vm.dirty_expire_centisecs=300000 sysctl -w vm.dirty_writeback_centisecs=300000 sysctl -w vm.dirtytime_expire_seconds=300000 echo 3 > /proc/sys/vm/drop_caches TEST=/sys/fs/cgroup/test A=$TEST/A B=$TEST/B mkdir -p $A $B echo "+memory +io" > $TEST/cgroup.subtree_control echo $((1<<30)) > $A/memory.high echo $((32<<30)) > $B/memory.high rm -f testfile touch testfile fallocate -l 4G testfile echo "Starting B" (echo $BASHPID > $B/cgroup.procs pv -q --rate-limit 70M < /dev/urandom | ./write-range testfile $((2<<30)) $((2<<30))) & echo "Waiting 10s to ensure B claims the testfile inode" sleep 5 sync sleep 5 sync echo "Starting A" (echo $BASHPID > $A/cgroup.procs pv < /dev/urandom | ./write-range testfile 0 $((2<<30))) v2: Added comments explaining why the specific intervals are being used. v3: Use 0 @nr when calling cgroup_writeback_by_id() to use best-effort flushing while avoding possible livelocks. v4: Use get_jiffies_64() and time_before/after64() instead of raw jiffies_64 and arthimetic comparisons as suggested by Jan. Reviewed-by: Jan Kara <jack@suse.cz> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-08-26 16:06:56 +00:00
mem_cgroup_flush_foreign(wb);
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
/*
* Calculate global domain's pos_ratio and select the
* global dtc by default.
*/
balance_wb_limits(gdtc, strictlimit);
if (gdtc->freerun)
goto free_running;
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
sdtc = gdtc;
if (mdtc) {
/*
* If memcg domain is in effect, calculate its
* pos_ratio. @wb should satisfy constraints from
* both global and memcg domains. Choose the one
* w/ lower pos_ratio.
*/
balance_wb_limits(mdtc, strictlimit);
if (mdtc->freerun)
goto free_running;
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
if (mdtc->pos_ratio < gdtc->pos_ratio)
sdtc = mdtc;
}
wb->dirty_exceeded = gdtc->dirty_exceeded ||
(mdtc && mdtc->dirty_exceeded);
if (time_is_before_jiffies(READ_ONCE(wb->bw_time_stamp) +
writeback: fix bandwidth estimate for spiky workload Michael Stapelberg has reported that for workload with short big spikes of writes (GCC linker seem to trigger this frequently) the write throughput is heavily underestimated and tends to steadily sink until it reaches zero. This has rather bad impact on writeback throttling (causing stalls). The problem is that writeback throughput estimate gets updated at most once per 200 ms. One update happens early after we submit pages for writeback (at that point writeout of only small fraction of pages is completed and thus observed throughput is tiny). Next update happens only during the next write spike (updates happen only from inode writeback and dirty throttling code) and if that is more than 1s after previous spike, we decide system was idle and just ignore whatever was written until this moment. Fix the problem by making sure writeback throughput estimate is also updated shortly after writeback completes to get reasonable estimate of throughput for spiky workloads. [jack@suse.cz: avoid division by 0 in wb_update_dirty_ratelimit()] Link: https://lore.kernel.org/lkml/20210617095309.3542373-1-stapelberg+linux@google.com Link: https://lkml.kernel.org/r/20210713104716.22868-3-jack@suse.cz Signed-off-by: Jan Kara <jack@suse.cz> Reported-by: Michael Stapelberg <stapelberg+linux@google.com> Tested-by: Michael Stapelberg <stapelberg+linux@google.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-09-02 21:53:09 +00:00
BANDWIDTH_INTERVAL))
__wb_update_bandwidth(gdtc, mdtc, true);
writeback: bdi write bandwidth estimation The estimation value will start from 100MB/s and adapt to the real bandwidth in seconds. It tries to update the bandwidth only when disk is fully utilized. Any inactive period of more than one second will be skipped. The estimated bandwidth will be reflecting how fast the device can writeout when _fully utilized_, and won't drop to 0 when it goes idle. The value will remain constant at disk idle time. At busy write time, if not considering fluctuations, it will also remain high unless be knocked down by possible concurrent reads that compete for the disk time and bandwidth with async writes. The estimation is not done purely in the flusher because there is no guarantee for write_cache_pages() to return timely to update bandwidth. The bdi->avg_write_bandwidth smoothing is very effective for filtering out sudden spikes, however may be a little biased in long term. The overheads are low because the bdi bandwidth update only occurs at 200ms intervals. The 200ms update interval is suitable, because it's not possible to get the real bandwidth for the instance at all, due to large fluctuations. The NFS commits can be as large as seconds worth of data. One XFS completion may be as large as half second worth of data if we are going to increase the write chunk to half second worth of data. In ext4, fluctuations with time period of around 5 seconds is observed. And there is another pattern of irregular periods of up to 20 seconds on SSD tests. That's why we are not only doing the estimation at 200ms intervals, but also averaging them over a period of 3 seconds and then go further to do another level of smoothing in avg_write_bandwidth. CC: Li Shaohua <shaohua.li@intel.com> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-29 17:22:30 +00:00
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
/* throttle according to the chosen dtc */
dirty_ratelimit = READ_ONCE(wb->dirty_ratelimit);
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
task_ratelimit = ((u64)dirty_ratelimit * sdtc->pos_ratio) >>
RATELIMIT_CALC_SHIFT;
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
max_pause = wb_max_pause(wb, sdtc->wb_dirty);
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
min_pause = wb_min_pause(wb, max_pause,
task_ratelimit, dirty_ratelimit,
&nr_dirtied_pause);
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
if (unlikely(task_ratelimit == 0)) {
period = max_pause;
pause = max_pause;
writeback: IO-less balance_dirty_pages() As proposed by Chris, Dave and Jan, don't start foreground writeback IO inside balance_dirty_pages(). Instead, simply let it idle sleep for some time to throttle the dirtying task. In the mean while, kick off the per-bdi flusher thread to do background writeback IO. RATIONALS ========= - disk seeks on concurrent writeback of multiple inodes (Dave Chinner) If every thread doing writes and being throttled start foreground writeback, it leads to N IO submitters from at least N different inodes at the same time, end up with N different sets of IO being issued with potentially zero locality to each other, resulting in much lower elevator sort/merge efficiency and hence we seek the disk all over the place to service the different sets of IO. OTOH, if there is only one submission thread, it doesn't jump between inodes in the same way when congestion clears - it keeps writing to the same inode, resulting in large related chunks of sequential IOs being issued to the disk. This is more efficient than the above foreground writeback because the elevator works better and the disk seeks less. - lock contention and cache bouncing on concurrent IO submitters (Dave Chinner) With this patchset, the fs_mark benchmark on a 12-drive software RAID0 goes from CPU bound to IO bound, freeing "3-4 CPUs worth of spinlock contention". * "CPU usage has dropped by ~55%", "it certainly appears that most of the CPU time saving comes from the removal of contention on the inode_wb_list_lock" (IMHO at least 10% comes from the reduction of cacheline bouncing, because the new code is able to call much less frequently into balance_dirty_pages() and hence access the global page states) * the user space "App overhead" is reduced by 20%, by avoiding the cacheline pollution by the complex writeback code path * "for a ~5% throughput reduction", "the number of write IOs have dropped by ~25%", and the elapsed time reduced from 41:42.17 to 40:53.23. * On a simple test of 100 dd, it reduces the CPU %system time from 30% to 3%, and improves IO throughput from 38MB/s to 42MB/s. - IO size too small for fast arrays and too large for slow USB sticks The write_chunk used by current balance_dirty_pages() cannot be directly set to some large value (eg. 128MB) for better IO efficiency. Because it could lead to more than 1 second user perceivable stalls. Even the current 4MB write size may be too large for slow USB sticks. The fact that balance_dirty_pages() starts IO on itself couples the IO size to wait time, which makes it hard to do suitable IO size while keeping the wait time under control. Now it's possible to increase writeback chunk size proportional to the disk bandwidth. In a simple test of 50 dd's on XFS, 1-HDD, 3GB ram, the larger writeback size dramatically reduces the seek count to 1/10 (far beyond my expectation) and improves the write throughput by 24%. - long block time in balance_dirty_pages() hurts desktop responsiveness Many of us may have the experience: it often takes a couple of seconds or even long time to stop a heavy writing dd/cp/tar command with Ctrl-C or "kill -9". - IO pipeline broken by bumpy write() progress There are a broad class of "loop {read(buf); write(buf);}" applications whose read() pipeline will be under-utilized or even come to a stop if the write()s have long latencies _or_ don't progress in a constant rate. The current threshold based throttling inherently transfers the large low level IO completion fluctuations to bumpy application write()s, and further deteriorates with increasing number of dirtiers and/or bdi's. For example, when doing 50 dd's + 1 remote rsync to an XFS partition, the rsync progresses very bumpy in legacy kernel, and throughput is improved by 67% by this patchset. (plus the larger write chunk size, it will be 93% speedup). The new rate based throttling can support 1000+ dd's with excellent smoothness, low latency and low overheads. For the above reasons, it's much better to do IO-less and low latency pauses in balance_dirty_pages(). Jan Kara, Dave Chinner and me explored the scheme to let balance_dirty_pages() wait for enough writeback IO completions to safeguard the dirty limit. However it's found to have two problems: - in large NUMA systems, the per-cpu counters may have big accounting errors, leading to big throttle wait time and jitters. - NFS may kill large amount of unstable pages with one single COMMIT. Because NFS server serves COMMIT with expensive fsync() IOs, it is desirable to delay and reduce the number of COMMITs. So it's not likely to optimize away such kind of bursty IO completions, and the resulted large (and tiny) stall times in IO completion based throttling. So here is a pause time oriented approach, which tries to control the pause time in each balance_dirty_pages() invocations, by controlling the number of pages dirtied before calling balance_dirty_pages(), for smooth and efficient dirty throttling: - avoid useless (eg. zero pause time) balance_dirty_pages() calls - avoid too small pause time (less than 4ms, which burns CPU power) - avoid too large pause time (more than 200ms, which hurts responsiveness) - avoid big fluctuations of pause times It can control pause times at will. The default policy (in a followup patch) will be to do ~10ms pauses in 1-dd case, and increase to ~100ms in 1000-dd case. BEHAVIOR CHANGE =============== (1) dirty threshold Users will notice that the applications will get throttled once crossing the global (background + dirty)/2=15% threshold, and then balanced around 17.5%. Before patch, the behavior is to just throttle it at 20% dirtyable memory in 1-dd case. Since the task will be soft throttled earlier than before, it may be perceived by end users as performance "slow down" if his application happens to dirty more than 15% dirtyable memory. (2) smoothness/responsiveness Users will notice a more responsive system during heavy writeback. "killall dd" will take effect instantly. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-28 00:45:12 +00:00
goto pause;
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
}
period = HZ * pages_dirtied / task_ratelimit;
pause = period;
if (current->dirty_paused_when)
pause -= now - current->dirty_paused_when;
/*
* For less than 1s think time (ext3/4 may block the dirtier
* for up to 800ms from time to time on 1-HDD; so does xfs,
* however at much less frequency), try to compensate it in
* future periods by updating the virtual time; otherwise just
* do a reset, as it may be a light dirtier.
*/
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
if (pause < min_pause) {
trace_balance_dirty_pages(wb,
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
sdtc->thresh,
sdtc->bg_thresh,
sdtc->dirty,
sdtc->wb_thresh,
sdtc->wb_dirty,
dirty_ratelimit,
task_ratelimit,
pages_dirtied,
period,
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
min(pause, 0L),
start_time);
if (pause < -HZ) {
current->dirty_paused_when = now;
current->nr_dirtied = 0;
} else if (period) {
current->dirty_paused_when += period;
current->nr_dirtied = 0;
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
} else if (current->nr_dirtied_pause <= pages_dirtied)
current->nr_dirtied_pause += pages_dirtied;
break;
mm: per device dirty threshold Scale writeback cache per backing device, proportional to its writeout speed. By decoupling the BDI dirty thresholds a number of problems we currently have will go away, namely: - mutual interference starvation (for any number of BDIs); - deadlocks with stacked BDIs (loop, FUSE and local NFS mounts). It might be that all dirty pages are for a single BDI while other BDIs are idling. By giving each BDI a 'fair' share of the dirty limit, each one can have dirty pages outstanding and make progress. A global threshold also creates a deadlock for stacked BDIs; when A writes to B, and A generates enough dirty pages to get throttled, B will never start writeback until the dirty pages go away. Again, by giving each BDI its own 'independent' dirty limit, this problem is avoided. So the problem is to determine how to distribute the total dirty limit across the BDIs fairly and efficiently. A DBI that has a large dirty limit but does not have any dirty pages outstanding is a waste. What is done is to keep a floating proportion between the DBIs based on writeback completions. This way faster/more active devices get a larger share than slower/idle devices. [akpm@linux-foundation.org: fix warnings] [hugh@veritas.com: Fix occasional hang when a task couldn't get out of balance_dirty_pages] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:25:50 +00:00
}
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
if (unlikely(pause > max_pause)) {
/* for occasional dropped task_ratelimit */
now += min(pause - max_pause, max_pause);
pause = max_pause;
}
writeback: IO-less balance_dirty_pages() As proposed by Chris, Dave and Jan, don't start foreground writeback IO inside balance_dirty_pages(). Instead, simply let it idle sleep for some time to throttle the dirtying task. In the mean while, kick off the per-bdi flusher thread to do background writeback IO. RATIONALS ========= - disk seeks on concurrent writeback of multiple inodes (Dave Chinner) If every thread doing writes and being throttled start foreground writeback, it leads to N IO submitters from at least N different inodes at the same time, end up with N different sets of IO being issued with potentially zero locality to each other, resulting in much lower elevator sort/merge efficiency and hence we seek the disk all over the place to service the different sets of IO. OTOH, if there is only one submission thread, it doesn't jump between inodes in the same way when congestion clears - it keeps writing to the same inode, resulting in large related chunks of sequential IOs being issued to the disk. This is more efficient than the above foreground writeback because the elevator works better and the disk seeks less. - lock contention and cache bouncing on concurrent IO submitters (Dave Chinner) With this patchset, the fs_mark benchmark on a 12-drive software RAID0 goes from CPU bound to IO bound, freeing "3-4 CPUs worth of spinlock contention". * "CPU usage has dropped by ~55%", "it certainly appears that most of the CPU time saving comes from the removal of contention on the inode_wb_list_lock" (IMHO at least 10% comes from the reduction of cacheline bouncing, because the new code is able to call much less frequently into balance_dirty_pages() and hence access the global page states) * the user space "App overhead" is reduced by 20%, by avoiding the cacheline pollution by the complex writeback code path * "for a ~5% throughput reduction", "the number of write IOs have dropped by ~25%", and the elapsed time reduced from 41:42.17 to 40:53.23. * On a simple test of 100 dd, it reduces the CPU %system time from 30% to 3%, and improves IO throughput from 38MB/s to 42MB/s. - IO size too small for fast arrays and too large for slow USB sticks The write_chunk used by current balance_dirty_pages() cannot be directly set to some large value (eg. 128MB) for better IO efficiency. Because it could lead to more than 1 second user perceivable stalls. Even the current 4MB write size may be too large for slow USB sticks. The fact that balance_dirty_pages() starts IO on itself couples the IO size to wait time, which makes it hard to do suitable IO size while keeping the wait time under control. Now it's possible to increase writeback chunk size proportional to the disk bandwidth. In a simple test of 50 dd's on XFS, 1-HDD, 3GB ram, the larger writeback size dramatically reduces the seek count to 1/10 (far beyond my expectation) and improves the write throughput by 24%. - long block time in balance_dirty_pages() hurts desktop responsiveness Many of us may have the experience: it often takes a couple of seconds or even long time to stop a heavy writing dd/cp/tar command with Ctrl-C or "kill -9". - IO pipeline broken by bumpy write() progress There are a broad class of "loop {read(buf); write(buf);}" applications whose read() pipeline will be under-utilized or even come to a stop if the write()s have long latencies _or_ don't progress in a constant rate. The current threshold based throttling inherently transfers the large low level IO completion fluctuations to bumpy application write()s, and further deteriorates with increasing number of dirtiers and/or bdi's. For example, when doing 50 dd's + 1 remote rsync to an XFS partition, the rsync progresses very bumpy in legacy kernel, and throughput is improved by 67% by this patchset. (plus the larger write chunk size, it will be 93% speedup). The new rate based throttling can support 1000+ dd's with excellent smoothness, low latency and low overheads. For the above reasons, it's much better to do IO-less and low latency pauses in balance_dirty_pages(). Jan Kara, Dave Chinner and me explored the scheme to let balance_dirty_pages() wait for enough writeback IO completions to safeguard the dirty limit. However it's found to have two problems: - in large NUMA systems, the per-cpu counters may have big accounting errors, leading to big throttle wait time and jitters. - NFS may kill large amount of unstable pages with one single COMMIT. Because NFS server serves COMMIT with expensive fsync() IOs, it is desirable to delay and reduce the number of COMMITs. So it's not likely to optimize away such kind of bursty IO completions, and the resulted large (and tiny) stall times in IO completion based throttling. So here is a pause time oriented approach, which tries to control the pause time in each balance_dirty_pages() invocations, by controlling the number of pages dirtied before calling balance_dirty_pages(), for smooth and efficient dirty throttling: - avoid useless (eg. zero pause time) balance_dirty_pages() calls - avoid too small pause time (less than 4ms, which burns CPU power) - avoid too large pause time (more than 200ms, which hurts responsiveness) - avoid big fluctuations of pause times It can control pause times at will. The default policy (in a followup patch) will be to do ~10ms pauses in 1-dd case, and increase to ~100ms in 1000-dd case. BEHAVIOR CHANGE =============== (1) dirty threshold Users will notice that the applications will get throttled once crossing the global (background + dirty)/2=15% threshold, and then balanced around 17.5%. Before patch, the behavior is to just throttle it at 20% dirtyable memory in 1-dd case. Since the task will be soft throttled earlier than before, it may be perceived by end users as performance "slow down" if his application happens to dirty more than 15% dirtyable memory. (2) smoothness/responsiveness Users will notice a more responsive system during heavy writeback. "killall dd" will take effect instantly. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2010-08-28 00:45:12 +00:00
pause:
trace_balance_dirty_pages(wb,
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
sdtc->thresh,
sdtc->bg_thresh,
sdtc->dirty,
sdtc->wb_thresh,
sdtc->wb_dirty,
dirty_ratelimit,
task_ratelimit,
pages_dirtied,
period,
pause,
start_time);
if (flags & BDP_ASYNC) {
ret = -EAGAIN;
break;
}
__set_current_state(TASK_KILLABLE);
bdi->last_bdp_sleep = jiffies;
io_schedule_timeout(pause);
current->dirty_paused_when = now + pause;
current->nr_dirtied = 0;
writeback: max, min and target dirty pause time Control the pause time and the call intervals to balance_dirty_pages() with three parameters: 1) max_pause, limited by bdi_dirty and MAX_PAUSE 2) the target pause time, grows with the number of dd tasks and is normally limited by max_pause/2 3) the minimal pause, set to half the target pause and is used to skip short sleeps and accumulate them into bigger ones The typical behaviors after patch: - if ever task_ratelimit is far below dirty_ratelimit, the pause time will remain constant at max_pause and nr_dirtied_pause will be fluctuating with task_ratelimit - in the normal cases, nr_dirtied_pause will remain stable (keep in the same pace with dirty_ratelimit) and the pause time will be fluctuating with task_ratelimit In summary, someone has to fluctuate with task_ratelimit, because task_ratelimit = nr_dirtied_pause / pause We normally prefer a stable nr_dirtied_pause, until reaching max_pause. The notable behavior changes are: - in stable workloads, there will no longer be sudden big trajectory switching of nr_dirtied_pause as concerned by Peter. It will be as smooth as dirty_ratelimit and changing proportionally with it (as always, assuming bdi bandwidth does not fluctuate across 2^N lines, otherwise nr_dirtied_pause will show up in 2+ parallel trajectories) - in the rare cases when something keeps task_ratelimit far below dirty_ratelimit, the smoothness can no longer be retained and nr_dirtied_pause will be "dancing" with task_ratelimit. This fixes a (not that destructive but still not good) bug that dirty_ratelimit gets brought down undesirably <= balanced_dirty_ratelimit is under estimated <= weakly executed task_ratelimit <= pause goes too large and gets trimmed down to max_pause <= nr_dirtied_pause (based on dirty_ratelimit) is set too large <= dirty_ratelimit being much larger than task_ratelimit - introduce min_pause to avoid small pause sleeps - when pause is trimmed down to max_pause, try to compensate it at the next pause time The "refactor" type of changes are: The max_pause equation is slightly transformed to make it slightly more efficient. We now scale target_pause by (N * 10ms) on 2^N concurrent tasks, which is effectively equal to the original scaling max_pause by (N * 20ms) because the original code does implicit target_pause ~= max_pause / 2. Based on the same implicit ratio, target_pause starts with 10ms on 1 dd. CC: Jan Kara <jack@suse.cz> CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-11-30 17:08:55 +00:00
current->nr_dirtied_pause = nr_dirtied_pause;
writeback: introduce max-pause and pass-good dirty limits The max-pause limit helps to keep the sleep time inside balance_dirty_pages() within MAX_PAUSE=200ms. The 200ms max sleep means per task rate limit of 8pages/200ms=160KB/s when dirty exceeded, which normally is enough to stop dirtiers from continue pushing the dirty pages high, unless there are a sufficient large number of slow dirtiers (eg. 500 tasks doing 160KB/s will still sum up to 80MB/s, exceeding the write bandwidth of a slow disk and hence accumulating more and more dirty pages). The pass-good limit helps to let go of the good bdi's in the presence of a blocked bdi (ie. NFS server not responding) or slow USB disk which for some reason build up a large number of initial dirty pages that refuse to go away anytime soon. For example, given two bdi's A and B and the initial state bdi_thresh_A = dirty_thresh / 2 bdi_thresh_B = dirty_thresh / 2 bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 Then A get blocked, after a dozen seconds bdi_thresh_A = 0 bdi_thresh_B = dirty_thresh bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 The (bdi_dirty_B < bdi_thresh_B) test is now useless and the dirty pages will be effectively throttled by condition (nr_dirty < dirty_thresh). This has two problems: (1) we lose the protections for light dirtiers (2) balance_dirty_pages() effectively becomes IO-less because the (bdi_nr_reclaimable > bdi_thresh) test won't be true. This is good for IO, but balance_dirty_pages() loses an important way to break out of the loop which leads to more spread out throttle delays. DIRTY_PASSGOOD_AREA can eliminate the above issues. The only problem is, DIRTY_PASSGOOD_AREA needs to be defined as 2 to fully cover the above example while this patch uses the more conservative value 8 so as not to surprise people with too many dirty pages than expected. The max-pause limit won't noticeably impact the speed dirty pages are knocked down when there is a sudden drop of global/bdi dirty thresholds. Because the heavy dirties will be throttled below 160KB/s which is slow enough. It does help to avoid long dirty throttle delays and especially will make light dirtiers more responsive. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-20 04:18:42 +00:00
/*
* This is typically equal to (dirty < thresh) and can also
* keep "1000+ dd on a slow USB stick" under control.
writeback: introduce max-pause and pass-good dirty limits The max-pause limit helps to keep the sleep time inside balance_dirty_pages() within MAX_PAUSE=200ms. The 200ms max sleep means per task rate limit of 8pages/200ms=160KB/s when dirty exceeded, which normally is enough to stop dirtiers from continue pushing the dirty pages high, unless there are a sufficient large number of slow dirtiers (eg. 500 tasks doing 160KB/s will still sum up to 80MB/s, exceeding the write bandwidth of a slow disk and hence accumulating more and more dirty pages). The pass-good limit helps to let go of the good bdi's in the presence of a blocked bdi (ie. NFS server not responding) or slow USB disk which for some reason build up a large number of initial dirty pages that refuse to go away anytime soon. For example, given two bdi's A and B and the initial state bdi_thresh_A = dirty_thresh / 2 bdi_thresh_B = dirty_thresh / 2 bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 Then A get blocked, after a dozen seconds bdi_thresh_A = 0 bdi_thresh_B = dirty_thresh bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 The (bdi_dirty_B < bdi_thresh_B) test is now useless and the dirty pages will be effectively throttled by condition (nr_dirty < dirty_thresh). This has two problems: (1) we lose the protections for light dirtiers (2) balance_dirty_pages() effectively becomes IO-less because the (bdi_nr_reclaimable > bdi_thresh) test won't be true. This is good for IO, but balance_dirty_pages() loses an important way to break out of the loop which leads to more spread out throttle delays. DIRTY_PASSGOOD_AREA can eliminate the above issues. The only problem is, DIRTY_PASSGOOD_AREA needs to be defined as 2 to fully cover the above example while this patch uses the more conservative value 8 so as not to surprise people with too many dirty pages than expected. The max-pause limit won't noticeably impact the speed dirty pages are knocked down when there is a sudden drop of global/bdi dirty thresholds. Because the heavy dirties will be throttled below 160KB/s which is slow enough. It does help to avoid long dirty throttle delays and especially will make light dirtiers more responsive. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-20 04:18:42 +00:00
*/
if (task_ratelimit)
writeback: introduce max-pause and pass-good dirty limits The max-pause limit helps to keep the sleep time inside balance_dirty_pages() within MAX_PAUSE=200ms. The 200ms max sleep means per task rate limit of 8pages/200ms=160KB/s when dirty exceeded, which normally is enough to stop dirtiers from continue pushing the dirty pages high, unless there are a sufficient large number of slow dirtiers (eg. 500 tasks doing 160KB/s will still sum up to 80MB/s, exceeding the write bandwidth of a slow disk and hence accumulating more and more dirty pages). The pass-good limit helps to let go of the good bdi's in the presence of a blocked bdi (ie. NFS server not responding) or slow USB disk which for some reason build up a large number of initial dirty pages that refuse to go away anytime soon. For example, given two bdi's A and B and the initial state bdi_thresh_A = dirty_thresh / 2 bdi_thresh_B = dirty_thresh / 2 bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 Then A get blocked, after a dozen seconds bdi_thresh_A = 0 bdi_thresh_B = dirty_thresh bdi_dirty_A = dirty_thresh / 2 bdi_dirty_B = dirty_thresh / 2 The (bdi_dirty_B < bdi_thresh_B) test is now useless and the dirty pages will be effectively throttled by condition (nr_dirty < dirty_thresh). This has two problems: (1) we lose the protections for light dirtiers (2) balance_dirty_pages() effectively becomes IO-less because the (bdi_nr_reclaimable > bdi_thresh) test won't be true. This is good for IO, but balance_dirty_pages() loses an important way to break out of the loop which leads to more spread out throttle delays. DIRTY_PASSGOOD_AREA can eliminate the above issues. The only problem is, DIRTY_PASSGOOD_AREA needs to be defined as 2 to fully cover the above example while this patch uses the more conservative value 8 so as not to surprise people with too many dirty pages than expected. The max-pause limit won't noticeably impact the speed dirty pages are knocked down when there is a sudden drop of global/bdi dirty thresholds. Because the heavy dirties will be throttled below 160KB/s which is slow enough. It does help to avoid long dirty throttle delays and especially will make light dirtiers more responsive. Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-20 04:18:42 +00:00
break;
/*
* In the case of an unresponsive NFS server and the NFS dirty
* pages exceeds dirty_thresh, give the other good wb's a pipe
* to go through, so that tasks on them still remain responsive.
*
* In theory 1 page is enough to keep the consumer-producer
* pipe going: the flusher cleans 1 page => the task dirties 1
* more page. However wb_dirty has accounting errors. So use
* the larger and more IO friendly wb_stat_error.
*/
if (sdtc->wb_dirty <= wb_stat_error())
break;
if (fatal_signal_pending(current))
break;
}
return ret;
}
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
static DEFINE_PER_CPU(int, bdp_ratelimits);
/*
* Normal tasks are throttled by
* loop {
* dirty tsk->nr_dirtied_pause pages;
* take a snap in balance_dirty_pages();
* }
* However there is a worst case. If every task exit immediately when dirtied
* (tsk->nr_dirtied_pause - 1) pages, balance_dirty_pages() will never be
* called to throttle the page dirties. The solution is to save the not yet
* throttled page dirties in dirty_throttle_leaks on task exit and charge them
* randomly into the running tasks. This works well for the above worst case,
* as the new task will pick up and accumulate the old task's leaked dirty
* count and eventually get throttled.
*/
DEFINE_PER_CPU(int, dirty_throttle_leaks) = 0;
/**
* balance_dirty_pages_ratelimited_flags - Balance dirty memory state.
* @mapping: address_space which was dirtied.
* @flags: BDP flags.
*
* Processes which are dirtying memory should call in here once for each page
* which was newly dirtied. The function will periodically check the system's
* dirty state and will initiate writeback if needed.
*
* See balance_dirty_pages_ratelimited() for details.
*
* Return: If @flags contains BDP_ASYNC, it may return -EAGAIN to
* indicate that memory is out of balance and the caller must wait
* for I/O to complete. Otherwise, it will return 0 to indicate
* that either memory was already in balance, or it was able to sleep
* until the amount of dirty memory returned to balance.
*/
int balance_dirty_pages_ratelimited_flags(struct address_space *mapping,
unsigned int flags)
{
struct inode *inode = mapping->host;
struct backing_dev_info *bdi = inode_to_bdi(inode);
struct bdi_writeback *wb = NULL;
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
int ratelimit;
int ret = 0;
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
int *p;
if (!(bdi->capabilities & BDI_CAP_WRITEBACK))
return ret;
if (inode_cgwb_enabled(inode))
wb = wb_get_create_current(bdi, GFP_KERNEL);
if (!wb)
wb = &bdi->wb;
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
ratelimit = current->nr_dirtied_pause;
writeback: move bandwidth related fields from backing_dev_info into bdi_writeback Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback) and the role of the separation is unclear. For cgroup support for writeback IOs, a bdi will be updated to host multiple wb's where each wb serves writeback IOs of a different cgroup on the bdi. To achieve that, a wb should carry all states necessary for servicing writeback IOs for a cgroup independently. This patch moves bandwidth related fields from backing_dev_info into bdi_writeback. * The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp, write_bandwidth, avg_write_bandwidth, dirty_ratelimit, balanced_dirty_ratelimit, completions and dirty_exceeded. * writeback_chunk_size() and over_bground_thresh() now take @wb instead of @bdi. * bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...) bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...) bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...) bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...) [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...) bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...) bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...) * Init/exits of the relocated fields are moved to bdi_wb_init/exit() respectively. Note that explicit zeroing is dropped in the process as wb's are cleared in entirety anyway. * As there's still only one bdi_writeback per backing_dev_info, all uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[] introducing no behavior changes. v2: Typo in description fixed as suggested by Jan. Signed-off-by: Tejun Heo <tj@kernel.org> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Jens Axboe <axboe@kernel.dk> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:28 +00:00
if (wb->dirty_exceeded)
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
ratelimit = min(ratelimit, 32 >> (PAGE_SHIFT - 10));
preempt_disable();
/*
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
* This prevents one CPU to accumulate too many dirtied pages without
* calling into balance_dirty_pages(), which can happen when there are
* 1000+ tasks, all of them start dirtying pages at exactly the same
* time, hence all honoured too large initial task->nr_dirtied_pause.
*/
p = this_cpu_ptr(&bdp_ratelimits);
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
if (unlikely(current->nr_dirtied >= ratelimit))
*p = 0;
else if (unlikely(*p >= ratelimit_pages)) {
*p = 0;
ratelimit = 0;
}
/*
* Pick up the dirtied pages by the exited tasks. This avoids lots of
* short-lived tasks (eg. gcc invocations in a kernel build) escaping
* the dirty throttling and livelock other long-run dirtiers.
*/
p = this_cpu_ptr(&dirty_throttle_leaks);
if (*p > 0 && current->nr_dirtied < ratelimit) {
unsigned long nr_pages_dirtied;
nr_pages_dirtied = min(*p, ratelimit - current->nr_dirtied);
*p -= nr_pages_dirtied;
current->nr_dirtied += nr_pages_dirtied;
}
preempt_enable();
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
if (unlikely(current->nr_dirtied >= ratelimit))
ret = balance_dirty_pages(wb, current->nr_dirtied, flags);
wb_put(wb);
return ret;
}
EXPORT_SYMBOL_GPL(balance_dirty_pages_ratelimited_flags);
/**
* balance_dirty_pages_ratelimited - balance dirty memory state.
* @mapping: address_space which was dirtied.
*
* Processes which are dirtying memory should call in here once for each page
* which was newly dirtied. The function will periodically check the system's
* dirty state and will initiate writeback if needed.
*
* Once we're over the dirty memory limit we decrease the ratelimiting
* by a lot, to prevent individual processes from overshooting the limit
* by (ratelimit_pages) each.
*/
void balance_dirty_pages_ratelimited(struct address_space *mapping)
{
balance_dirty_pages_ratelimited_flags(mapping, 0);
}
EXPORT_SYMBOL(balance_dirty_pages_ratelimited);
writeback: factor out wb_bg_dirty_limits to remove repeated code Patch series "Add helper functions to remove repeated code and improve readability of cgroup writeback", v2. This series adds a lot of helpers to remove repeated code between domain and wb; dirty limit and dirty background; global domain and wb domain. The helpers also improve readability. More details can be found in the respective patches. A simple domain hierarchy is tested: global domain (> 20G) | cgroup domain1(10G) | wb1 | fio Test steps: /* make it easy to observe */ echo 300000 > /proc/sys/vm/dirty_expire_centisecs echo 3000 > /proc/sys/vm/dirty_writeback_centisecs /* create cgroup domain */ cd /sys/fs/cgroup echo "+memory +io" > cgroup.subtree_control mkdir group1 cd group1 echo 10G > memory.high echo 10G > memory.max echo $$ > cgroup.procs mkfs.ext4 -F /dev/vdb mount /dev/vdb /bdi1/ /* run fio to generate dirty pages */ fio -name test -filename=/bdi1/file -size=xxx -ioengine=libaio -bs=4K \ -iodepth=1 -rw=write -direct=0 --time_based -runtime=600 -invalidate=0 When fio size is 1G, the wb is in freerun state and dirty pages are only written back when dirty inode is expired after 30 seconds. When fio size is 2G, the dirty pages keep being written back and bandwidth of fio is limited. This patch (of 8): Similar to wb_dirty_limits which calculates dirty and thresh of wb, wb_bg_dirty_limits calculates background dirty and background thresh of wb. With wb_bg_dirty_limits, we could remove repeated code in wb_over_bg_thresh. Link: https://lkml.kernel.org/r/20240514125254.142203-1-shikemeng@huaweicloud.com Link: https://lkml.kernel.org/r/20240514125254.142203-2-shikemeng@huaweicloud.com Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com> Acked-by: Tejun Heo <tj@kernel.org> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-05-14 12:52:47 +00:00
/*
* Similar to wb_dirty_limits, wb_bg_dirty_limits also calculates dirty
* and thresh, but it's for background writeback.
*/
static void wb_bg_dirty_limits(struct dirty_throttle_control *dtc)
{
struct bdi_writeback *wb = dtc->wb;
dtc->wb_bg_thresh = __wb_calc_thresh(dtc, dtc->bg_thresh);
if (dtc->wb_bg_thresh < 2 * wb_stat_error())
dtc->wb_dirty = wb_stat_sum(wb, WB_RECLAIMABLE);
else
dtc->wb_dirty = wb_stat(wb, WB_RECLAIMABLE);
}
static bool domain_over_bg_thresh(struct dirty_throttle_control *dtc)
{
domain_dirty_avail(dtc, false);
domain_dirty_limits(dtc);
if (dtc->dirty > dtc->bg_thresh)
return true;
wb_bg_dirty_limits(dtc);
if (dtc->wb_dirty > dtc->wb_bg_thresh)
return true;
return false;
}
/**
* wb_over_bg_thresh - does @wb need to be written back?
* @wb: bdi_writeback of interest
*
* Determines whether background writeback should keep writing @wb or it's
* clean enough.
*
* Return: %true if writeback should continue.
*/
bool wb_over_bg_thresh(struct bdi_writeback *wb)
{
struct dirty_throttle_control gdtc = { GDTC_INIT(wb) };
struct dirty_throttle_control mdtc = { MDTC_INIT(wb, &gdtc) };
if (domain_over_bg_thresh(&gdtc))
return true;
if (mdtc_valid(&mdtc))
return domain_over_bg_thresh(&mdtc);
writeback: implement memcg writeback domain based throttling While cgroup writeback support now connects memcg and blkcg so that writeback IOs are properly attributed and controlled, the IO back pressure propagation mechanism implemented in balance_dirty_pages() and its subroutines wasn't aware of cgroup writeback. Processes belonging to a memcg may have access to only subset of total memory available in the system and not factoring this into dirty throttling rendered it completely ineffective for processes under memcg limits and memcg ended up building a separate ad-hoc degenerate mechanism directly into vmscan code to limit page dirtying. The previous patches updated balance_dirty_pages() and its subroutines so that they can deal with multiple wb_domain's (writeback domains) and defined per-memcg wb_domain. Processes belonging to a non-root memcg are bound to two wb_domains, global wb_domain and memcg wb_domain, and should be throttled according to IO pressures from both domains. This patch updates dirty throttling code so that it repeats similar calculations for the two domains - the differences between the two are few and minor - and applies the lower of the two sets of resulting constraints. wb_over_bg_thresh(), which controls when background writeback terminates, is also updated to consider both global and memcg wb_domains. It returns true if dirty is over bg_thresh for either domain. This makes the dirty throttling mechanism operational for memcg domains including writeback-bandwidth-proportional dirty page distribution inside them but the ad-hoc memcg throttling mechanism in vmscan is still in place. The next patch will rip it out. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 22:23:35 +00:00
return false;
}
#ifdef CONFIG_SYSCTL
/*
* sysctl handler for /proc/sys/vm/dirty_writeback_centisecs
*/
sysctl: treewide: constify the ctl_table argument of proc_handlers const qualify the struct ctl_table argument in the proc_handler function signatures. This is a prerequisite to moving the static ctl_table structs into .rodata data which will ensure that proc_handler function pointers cannot be modified. This patch has been generated by the following coccinelle script: ``` virtual patch @r1@ identifier ctl, write, buffer, lenp, ppos; identifier func !~ "appldata_(timer|interval)_handler|sched_(rt|rr)_handler|rds_tcp_skbuf_handler|proc_sctp_do_(hmac_alg|rto_min|rto_max|udp_port|alpha_beta|auth|probe_interval)"; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int write, void *buffer, size_t *lenp, loff_t *ppos); @r2@ identifier func, ctl, write, buffer, lenp, ppos; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int write, void *buffer, size_t *lenp, loff_t *ppos) { ... } @r3@ identifier func; @@ int func( - struct ctl_table * + const struct ctl_table * ,int , void *, size_t *, loff_t *); @r4@ identifier func, ctl; @@ int func( - struct ctl_table *ctl + const struct ctl_table *ctl ,int , void *, size_t *, loff_t *); @r5@ identifier func, write, buffer, lenp, ppos; @@ int func( - struct ctl_table * + const struct ctl_table * ,int write, void *buffer, size_t *lenp, loff_t *ppos); ``` * Code formatting was adjusted in xfs_sysctl.c to comply with code conventions. The xfs_stats_clear_proc_handler, xfs_panic_mask_proc_handler and xfs_deprecated_dointvec_minmax where adjusted. * The ctl_table argument in proc_watchdog_common was const qualified. This is called from a proc_handler itself and is calling back into another proc_handler, making it necessary to change it as part of the proc_handler migration. Co-developed-by: Thomas Weißschuh <linux@weissschuh.net> Signed-off-by: Thomas Weißschuh <linux@weissschuh.net> Co-developed-by: Joel Granados <j.granados@samsung.com> Signed-off-by: Joel Granados <j.granados@samsung.com>
2024-07-24 18:59:29 +00:00
static int dirty_writeback_centisecs_handler(const struct ctl_table *table, int write,
void *buffer, size_t *length, loff_t *ppos)
{
unsigned int old_interval = dirty_writeback_interval;
int ret;
ret = proc_dointvec(table, write, buffer, length, ppos);
/*
* Writing 0 to dirty_writeback_interval will disable periodic writeback
* and a different non-zero value will wakeup the writeback threads.
* wb_wakeup_delayed() would be more appropriate, but it's a pain to
* iterate over all bdis and wbs.
* The reason we do this is to make the change take effect immediately.
*/
if (!ret && write && dirty_writeback_interval &&
dirty_writeback_interval != old_interval)
wakeup_flusher_threads(WB_REASON_PERIODIC);
return ret;
}
#endif
void laptop_mode_timer_fn(struct timer_list *t)
{
struct backing_dev_info *backing_dev_info =
from_timer(backing_dev_info, t, laptop_mode_wb_timer);
wakeup_flusher_threads_bdi(backing_dev_info, WB_REASON_LAPTOP_TIMER);
}
/*
* We've spun up the disk and we're in laptop mode: schedule writeback
* of all dirty data a few seconds from now. If the flush is already scheduled
* then push it back - the user is still using the disk.
*/
void laptop_io_completion(struct backing_dev_info *info)
{
mod_timer(&info->laptop_mode_wb_timer, jiffies + laptop_mode);
}
/*
* We're in laptop mode and we've just synced. The sync's writes will have
* caused another writeback to be scheduled by laptop_io_completion.
* Nothing needs to be written back anymore, so we unschedule the writeback.
*/
void laptop_sync_completion(void)
{
struct backing_dev_info *bdi;
rcu_read_lock();
list_for_each_entry_rcu(bdi, &bdi_list, bdi_list)
del_timer(&bdi->laptop_mode_wb_timer);
rcu_read_unlock();
}
/*
* If ratelimit_pages is too high then we can get into dirty-data overload
* if a large number of processes all perform writes at the same time.
*
* Here we set ratelimit_pages to a level which ensures that when all CPUs are
* dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
* thresholds.
*/
void writeback_set_ratelimit(void)
{
struct wb_domain *dom = &global_wb_domain;
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
unsigned long background_thresh;
unsigned long dirty_thresh;
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
global_dirty_limits(&background_thresh, &dirty_thresh);
dom->dirty_limit = dirty_thresh;
writeback: per task dirty rate limit Add two fields to task_struct. 1) account dirtied pages in the individual tasks, for accuracy 2) per-task balance_dirty_pages() call intervals, for flexibility The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will scale near-sqrt to the safety gap between dirty pages and threshold. The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying pages at exactly the same time, each task will be assigned a large initial nr_dirtied_pause, so that the dirty threshold will be exceeded long before each task reached its nr_dirtied_pause and hence call balance_dirty_pages(). The solution is to watch for the number of pages dirtied on each CPU in between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages (3% dirty threshold), force call balance_dirty_pages() for a chance to set bdi->dirty_exceeded. In normal situations, this safeguarding condition is not expected to trigger at all. On the sqrt in dirty_poll_interval(): It will serve as an initial guess when dirty pages are still in the freerun area. When dirty pages are floating inside the dirty control scope [freerun, limit], a followup patch will use some refined dirty poll interval to get the desired pause time. thresh-dirty (MB) sqrt 1 16 2 22 4 32 8 45 16 64 32 90 64 128 128 181 256 256 512 362 1024 512 The above table means, given 1MB (or 1GB) gap and the dd tasks polling balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't be exceeded as long as there are less than 16 (or 512) concurrent dd's. So sqrt naturally leads to less overheads and more safe concurrent tasks for large memory servers, which have large (thresh-freerun) gaps. peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case CC: Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by: Andrea Righi <andrea@betterlinux.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
2011-06-12 00:10:12 +00:00
ratelimit_pages = dirty_thresh / (num_online_cpus() * 32);
if (ratelimit_pages < 16)
ratelimit_pages = 16;
}
static int page_writeback_cpu_online(unsigned int cpu)
{
writeback_set_ratelimit();
return 0;
}
#ifdef CONFIG_SYSCTL
/* this is needed for the proc_doulongvec_minmax of vm_dirty_bytes */
static const unsigned long dirty_bytes_min = 2 * PAGE_SIZE;
static struct ctl_table vm_page_writeback_sysctls[] = {
{
.procname = "dirty_background_ratio",
.data = &dirty_background_ratio,
.maxlen = sizeof(dirty_background_ratio),
.mode = 0644,
.proc_handler = dirty_background_ratio_handler,
.extra1 = SYSCTL_ZERO,
.extra2 = SYSCTL_ONE_HUNDRED,
},
{
.procname = "dirty_background_bytes",
.data = &dirty_background_bytes,
.maxlen = sizeof(dirty_background_bytes),
.mode = 0644,
.proc_handler = dirty_background_bytes_handler,
.extra1 = SYSCTL_LONG_ONE,
},
{
.procname = "dirty_ratio",
.data = &vm_dirty_ratio,
.maxlen = sizeof(vm_dirty_ratio),
.mode = 0644,
.proc_handler = dirty_ratio_handler,
.extra1 = SYSCTL_ZERO,
.extra2 = SYSCTL_ONE_HUNDRED,
},
{
.procname = "dirty_bytes",
.data = &vm_dirty_bytes,
.maxlen = sizeof(vm_dirty_bytes),
.mode = 0644,
.proc_handler = dirty_bytes_handler,
.extra1 = (void *)&dirty_bytes_min,
},
{
.procname = "dirty_writeback_centisecs",
.data = &dirty_writeback_interval,
.maxlen = sizeof(dirty_writeback_interval),
.mode = 0644,
.proc_handler = dirty_writeback_centisecs_handler,
},
{
.procname = "dirty_expire_centisecs",
.data = &dirty_expire_interval,
.maxlen = sizeof(dirty_expire_interval),
.mode = 0644,
.proc_handler = proc_dointvec_minmax,
.extra1 = SYSCTL_ZERO,
},
#ifdef CONFIG_HIGHMEM
{
.procname = "highmem_is_dirtyable",
.data = &vm_highmem_is_dirtyable,
.maxlen = sizeof(vm_highmem_is_dirtyable),
.mode = 0644,
.proc_handler = proc_dointvec_minmax,
.extra1 = SYSCTL_ZERO,
.extra2 = SYSCTL_ONE,
},
#endif
{
.procname = "laptop_mode",
.data = &laptop_mode,
.maxlen = sizeof(laptop_mode),
.mode = 0644,
.proc_handler = proc_dointvec_jiffies,
},
};
#endif
/*
* Called early on to tune the page writeback dirty limits.
*
* We used to scale dirty pages according to how total memory
* related to pages that could be allocated for buffers.
*
* However, that was when we used "dirty_ratio" to scale with
* all memory, and we don't do that any more. "dirty_ratio"
* is now applied to total non-HIGHPAGE memory, and as such we can't
* get into the old insane situation any more where we had
* large amounts of dirty pages compared to a small amount of
* non-HIGHMEM memory.
*
* But we might still want to scale the dirty_ratio by how
* much memory the box has..
*/
void __init page_writeback_init(void)
{
BUG_ON(wb_domain_init(&global_wb_domain, GFP_KERNEL));
cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "mm/writeback:online",
page_writeback_cpu_online, NULL);
cpuhp_setup_state(CPUHP_MM_WRITEBACK_DEAD, "mm/writeback:dead", NULL,
page_writeback_cpu_online);
#ifdef CONFIG_SYSCTL
register_sysctl_init("vm", vm_page_writeback_sysctls);
#endif
}
/**
* tag_pages_for_writeback - tag pages to be written by writeback
* @mapping: address space structure to write
* @start: starting page index
* @end: ending page index (inclusive)
*
* This function scans the page range from @start to @end (inclusive) and tags
* all pages that have DIRTY tag set with a special TOWRITE tag. The caller
* can then use the TOWRITE tag to identify pages eligible for writeback.
* This mechanism is used to avoid livelocking of writeback by a process
* steadily creating new dirty pages in the file (thus it is important for this
* function to be quick so that it can tag pages faster than a dirtying process
* can create them).
*/
void tag_pages_for_writeback(struct address_space *mapping,
pgoff_t start, pgoff_t end)
{
XA_STATE(xas, &mapping->i_pages, start);
unsigned int tagged = 0;
void *page;
xas_lock_irq(&xas);
xas_for_each_marked(&xas, page, end, PAGECACHE_TAG_DIRTY) {
xas_set_mark(&xas, PAGECACHE_TAG_TOWRITE);
if (++tagged % XA_CHECK_SCHED)
continue;
xas_pause(&xas);
xas_unlock_irq(&xas);
cond_resched();
xas_lock_irq(&xas);
}
xas_unlock_irq(&xas);
}
EXPORT_SYMBOL(tag_pages_for_writeback);
static bool folio_prepare_writeback(struct address_space *mapping,
struct writeback_control *wbc, struct folio *folio)
{
/*
* Folio truncated or invalidated. We can freely skip it then,
* even for data integrity operations: the folio has disappeared
* concurrently, so there could be no real expectation of this
* data integrity operation even if there is now a new, dirty
* folio at the same pagecache index.
*/
if (unlikely(folio->mapping != mapping))
return false;
/*
* Did somebody else write it for us?
*/
if (!folio_test_dirty(folio))
return false;
if (folio_test_writeback(folio)) {
if (wbc->sync_mode == WB_SYNC_NONE)
return false;
folio_wait_writeback(folio);
}
BUG_ON(folio_test_writeback(folio));
if (!folio_clear_dirty_for_io(folio))
return false;
return true;
}
static xa_mark_t wbc_to_tag(struct writeback_control *wbc)
{
if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
return PAGECACHE_TAG_TOWRITE;
return PAGECACHE_TAG_DIRTY;
}
static pgoff_t wbc_end(struct writeback_control *wbc)
{
if (wbc->range_cyclic)
return -1;
return wbc->range_end >> PAGE_SHIFT;
}
static struct folio *writeback_get_folio(struct address_space *mapping,
struct writeback_control *wbc)
{
struct folio *folio;
retry:
folio = folio_batch_next(&wbc->fbatch);
if (!folio) {
folio_batch_release(&wbc->fbatch);
cond_resched();
filemap_get_folios_tag(mapping, &wbc->index, wbc_end(wbc),
wbc_to_tag(wbc), &wbc->fbatch);
folio = folio_batch_next(&wbc->fbatch);
if (!folio)
return NULL;
}
folio_lock(folio);
if (unlikely(!folio_prepare_writeback(mapping, wbc, folio))) {
folio_unlock(folio);
goto retry;
}
trace_wbc_writepage(wbc, inode_to_bdi(mapping->host));
return folio;
}
/**
* writeback_iter - iterate folio of a mapping for writeback
* @mapping: address space structure to write
* @wbc: writeback context
* @folio: previously iterated folio (%NULL to start)
* @error: in-out pointer for writeback errors (see below)
*
* This function returns the next folio for the writeback operation described by
* @wbc on @mapping and should be called in a while loop in the ->writepages
* implementation.
*
* To start the writeback operation, %NULL is passed in the @folio argument, and
* for every subsequent iteration the folio returned previously should be passed
* back in.
mm/page-writeback.c: fix range_cyclic writeback vs writepages deadlock We've recently seen a workload on XFS filesystems with a repeatable deadlock between background writeback and a multi-process application doing concurrent writes and fsyncs to a small range of a file. range_cyclic writeback Process 1 Process 2 xfs_vm_writepages write_cache_pages writeback_index = 2 cycled = 0 .... find page 2 dirty lock Page 2 ->writepage page 2 writeback page 2 clean page 2 added to bio no more pages write() locks page 1 dirties page 1 locks page 2 dirties page 1 fsync() .... xfs_vm_writepages write_cache_pages start index 0 find page 1 towrite lock Page 1 ->writepage page 1 writeback page 1 clean page 1 added to bio find page 2 towrite lock Page 2 page 2 is writeback <blocks> write() locks page 1 dirties page 1 fsync() .... xfs_vm_writepages write_cache_pages start index 0 !done && !cycled sets index to 0, restarts lookup find page 1 dirty find page 1 towrite lock Page 1 page 1 is writeback <blocks> lock Page 1 <blocks> DEADLOCK because: - process 1 needs page 2 writeback to complete to make enough progress to issue IO pending for page 1 - writeback needs page 1 writeback to complete so process 2 can progress and unlock the page it is blocked on, then it can issue the IO pending for page 2 - process 2 can't make progress until process 1 issues IO for page 1 The underlying cause of the problem here is that range_cyclic writeback is processing pages in descending index order as we hold higher index pages in a structure controlled from above write_cache_pages(). The write_cache_pages() caller needs to be able to submit these pages for IO before write_cache_pages restarts writeback at mapping index 0 to avoid wcp inverting the page lock/writeback wait order. generic_writepages() is not susceptible to this bug as it has no private context held across write_cache_pages() - filesystems using this infrastructure always submit pages in ->writepage immediately and so there is no problem with range_cyclic going back to mapping index 0. However: mpage_writepages() has a private bio context, exofs_writepages() has page_collect fuse_writepages() has fuse_fill_wb_data nfs_writepages() has nfs_pageio_descriptor xfs_vm_writepages() has xfs_writepage_ctx All of these ->writepages implementations can hold pages under writeback in their private structures until write_cache_pages() returns, and hence they are all susceptible to this deadlock. Also worth noting is that ext4 has it's own bastardised version of write_cache_pages() and so it /may/ have an equivalent deadlock. I looked at the code long enough to understand that it has a similar retry loop for range_cyclic writeback reaching the end of the file and then promptly ran away before my eyes bled too much. I'll leave it for the ext4 developers to determine if their code is actually has this deadlock and how to fix it if it has. There's a few ways I can see avoid this deadlock. There's probably more, but these are the first I've though of: 1. get rid of range_cyclic altogether 2. range_cyclic always stops at EOF, and we start again from writeback index 0 on the next call into write_cache_pages() 2a. wcp also returns EAGAIN to ->writepages implementations to indicate range cyclic has hit EOF. writepages implementations can then flush the current context and call wpc again to continue. i.e. lift the retry into the ->writepages implementation 3. range_cyclic uses trylock_page() rather than lock_page(), and it skips pages it can't lock without blocking. It will already do this for pages under writeback, so this seems like a no-brainer 3a. all non-WB_SYNC_ALL writeback uses trylock_page() to avoid blocking as per pages under writeback. I don't think #1 is an option - range_cyclic prevents frequently dirtied lower file offset from starving background writeback of rarely touched higher file offsets. #2 is simple, and I don't think it will have any impact on performance as going back to the start of the file implies an immediate seek. We'll have exactly the same number of seeks if we switch writeback to another inode, and then come back to this one later and restart from index 0. #2a is pretty much "status quo without the deadlock". Moving the retry loop up into the wcp caller means we can issue IO on the pending pages before calling wcp again, and so avoid locking or waiting on pages in the wrong order. I'm not convinced we need to do this given that we get the same thing from #2 on the next writeback call from the writeback infrastructure. #3 is really just a band-aid - it doesn't fix the access/wait inversion problem, just prevents it from becoming a deadlock situation. I'd prefer we fix the inversion, not sweep it under the carpet like this. #3a is really an optimisation that just so happens to include the band-aid fix of #3. So it seems that the simplest way to fix this issue is to implement solution #2 Link: http://lkml.kernel.org/r/20181005054526.21507-1-david@fromorbit.com Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Jan Kara <jack@suse.de> Cc: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-10-26 22:09:45 +00:00
*
* If there was an error in the per-folio writeback inside the writeback_iter()
* loop, @error should be set to the error value.
*
* Once the writeback described in @wbc has finished, this function will return
* %NULL and if there was an error in any iteration restore it to @error.
*
* Note: callers should not manually break out of the loop using break or goto
* but must keep calling writeback_iter() until it returns %NULL.
*
* Return: the folio to write or %NULL if the loop is done.
*/
struct folio *writeback_iter(struct address_space *mapping,
struct writeback_control *wbc, struct folio *folio, int *error)
{
if (!folio) {
folio_batch_init(&wbc->fbatch);
wbc->saved_err = *error = 0;
/*
* For range cyclic writeback we remember where we stopped so
* that we can continue where we stopped.
*
* For non-cyclic writeback we always start at the beginning of
* the passed in range.
*/
if (wbc->range_cyclic)
wbc->index = mapping->writeback_index;
else
wbc->index = wbc->range_start >> PAGE_SHIFT;
/*
* To avoid livelocks when other processes dirty new pages, we
* first tag pages which should be written back and only then
* start writing them.
*
* For data-integrity writeback we have to be careful so that we
* do not miss some pages (e.g., because some other process has
* cleared the TOWRITE tag we set). The rule we follow is that
* TOWRITE tag can be cleared only by the process clearing the
* DIRTY tag (and submitting the page for I/O).
*/
if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
tag_pages_for_writeback(mapping, wbc->index,
wbc_end(wbc));
} else {
wbc->nr_to_write -= folio_nr_pages(folio);
WARN_ON_ONCE(*error > 0);
/*
* For integrity writeback we have to keep going until we have
* written all the folios we tagged for writeback above, even if
* we run past wbc->nr_to_write or encounter errors.
* We stash away the first error we encounter in wbc->saved_err
* so that it can be retrieved when we're done. This is because
* the file system may still have state to clear for each folio.
*
* For background writeback we exit as soon as we run past
* wbc->nr_to_write or encounter the first error.
*/
if (wbc->sync_mode == WB_SYNC_ALL) {
if (*error && !wbc->saved_err)
wbc->saved_err = *error;
} else {
if (*error || wbc->nr_to_write <= 0)
goto done;
}
}
mm/page-writeback.c: fix range_cyclic writeback vs writepages deadlock We've recently seen a workload on XFS filesystems with a repeatable deadlock between background writeback and a multi-process application doing concurrent writes and fsyncs to a small range of a file. range_cyclic writeback Process 1 Process 2 xfs_vm_writepages write_cache_pages writeback_index = 2 cycled = 0 .... find page 2 dirty lock Page 2 ->writepage page 2 writeback page 2 clean page 2 added to bio no more pages write() locks page 1 dirties page 1 locks page 2 dirties page 1 fsync() .... xfs_vm_writepages write_cache_pages start index 0 find page 1 towrite lock Page 1 ->writepage page 1 writeback page 1 clean page 1 added to bio find page 2 towrite lock Page 2 page 2 is writeback <blocks> write() locks page 1 dirties page 1 fsync() .... xfs_vm_writepages write_cache_pages start index 0 !done && !cycled sets index to 0, restarts lookup find page 1 dirty find page 1 towrite lock Page 1 page 1 is writeback <blocks> lock Page 1 <blocks> DEADLOCK because: - process 1 needs page 2 writeback to complete to make enough progress to issue IO pending for page 1 - writeback needs page 1 writeback to complete so process 2 can progress and unlock the page it is blocked on, then it can issue the IO pending for page 2 - process 2 can't make progress until process 1 issues IO for page 1 The underlying cause of the problem here is that range_cyclic writeback is processing pages in descending index order as we hold higher index pages in a structure controlled from above write_cache_pages(). The write_cache_pages() caller needs to be able to submit these pages for IO before write_cache_pages restarts writeback at mapping index 0 to avoid wcp inverting the page lock/writeback wait order. generic_writepages() is not susceptible to this bug as it has no private context held across write_cache_pages() - filesystems using this infrastructure always submit pages in ->writepage immediately and so there is no problem with range_cyclic going back to mapping index 0. However: mpage_writepages() has a private bio context, exofs_writepages() has page_collect fuse_writepages() has fuse_fill_wb_data nfs_writepages() has nfs_pageio_descriptor xfs_vm_writepages() has xfs_writepage_ctx All of these ->writepages implementations can hold pages under writeback in their private structures until write_cache_pages() returns, and hence they are all susceptible to this deadlock. Also worth noting is that ext4 has it's own bastardised version of write_cache_pages() and so it /may/ have an equivalent deadlock. I looked at the code long enough to understand that it has a similar retry loop for range_cyclic writeback reaching the end of the file and then promptly ran away before my eyes bled too much. I'll leave it for the ext4 developers to determine if their code is actually has this deadlock and how to fix it if it has. There's a few ways I can see avoid this deadlock. There's probably more, but these are the first I've though of: 1. get rid of range_cyclic altogether 2. range_cyclic always stops at EOF, and we start again from writeback index 0 on the next call into write_cache_pages() 2a. wcp also returns EAGAIN to ->writepages implementations to indicate range cyclic has hit EOF. writepages implementations can then flush the current context and call wpc again to continue. i.e. lift the retry into the ->writepages implementation 3. range_cyclic uses trylock_page() rather than lock_page(), and it skips pages it can't lock without blocking. It will already do this for pages under writeback, so this seems like a no-brainer 3a. all non-WB_SYNC_ALL writeback uses trylock_page() to avoid blocking as per pages under writeback. I don't think #1 is an option - range_cyclic prevents frequently dirtied lower file offset from starving background writeback of rarely touched higher file offsets. #2 is simple, and I don't think it will have any impact on performance as going back to the start of the file implies an immediate seek. We'll have exactly the same number of seeks if we switch writeback to another inode, and then come back to this one later and restart from index 0. #2a is pretty much "status quo without the deadlock". Moving the retry loop up into the wcp caller means we can issue IO on the pending pages before calling wcp again, and so avoid locking or waiting on pages in the wrong order. I'm not convinced we need to do this given that we get the same thing from #2 on the next writeback call from the writeback infrastructure. #3 is really just a band-aid - it doesn't fix the access/wait inversion problem, just prevents it from becoming a deadlock situation. I'd prefer we fix the inversion, not sweep it under the carpet like this. #3a is really an optimisation that just so happens to include the band-aid fix of #3. So it seems that the simplest way to fix this issue is to implement solution #2 Link: http://lkml.kernel.org/r/20181005054526.21507-1-david@fromorbit.com Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Jan Kara <jack@suse.de> Cc: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-10-26 22:09:45 +00:00
folio = writeback_get_folio(mapping, wbc);
if (!folio) {
/*
* To avoid deadlocks between range_cyclic writeback and callers
* that hold pages in PageWriteback to aggregate I/O until
* the writeback iteration finishes, we do not loop back to the
* start of the file. Doing so causes a page lock/page
* writeback access order inversion - we should only ever lock
* multiple pages in ascending page->index order, and looping
* back to the start of the file violates that rule and causes
* deadlocks.
*/
if (wbc->range_cyclic)
mapping->writeback_index = 0;
/*
* Return the first error we encountered (if there was any) to
* the caller.
*/
*error = wbc->saved_err;
}
return folio;
done:
if (wbc->range_cyclic)
mapping->writeback_index = folio_next_index(folio);
folio_batch_release(&wbc->fbatch);
return NULL;
}
EXPORT_SYMBOL_GPL(writeback_iter);
/**
* write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
* @mapping: address space structure to write
* @wbc: subtract the number of written pages from *@wbc->nr_to_write
* @writepage: function called for each page
* @data: data passed to writepage function
*
* Return: %0 on success, negative error code otherwise
*
* Note: please use writeback_iter() instead.
*/
int write_cache_pages(struct address_space *mapping,
struct writeback_control *wbc, writepage_t writepage,
void *data)
{
struct folio *folio = NULL;
int error;
while ((folio = writeback_iter(mapping, wbc, folio, &error))) {
error = writepage(folio, wbc, data);
if (error == AOP_WRITEPAGE_ACTIVATE) {
folio_unlock(folio);
error = 0;
}
}
return error;
}
EXPORT_SYMBOL(write_cache_pages);
static int writeback_use_writepage(struct address_space *mapping,
struct writeback_control *wbc)
{
struct folio *folio = NULL;
struct blk_plug plug;
int err;
blk_start_plug(&plug);
while ((folio = writeback_iter(mapping, wbc, folio, &err))) {
err = mapping->a_ops->writepage(&folio->page, wbc);
if (err == AOP_WRITEPAGE_ACTIVATE) {
folio_unlock(folio);
err = 0;
}
mapping_set_error(mapping, err);
}
blk_finish_plug(&plug);
return err;
}
int do_writepages(struct address_space *mapping, struct writeback_control *wbc)
{
int ret;
struct bdi_writeback *wb;
if (wbc->nr_to_write <= 0)
return 0;
wb = inode_to_wb_wbc(mapping->host, wbc);
wb_bandwidth_estimate_start(wb);
while (1) {
if (mapping->a_ops->writepages) {
ret = mapping->a_ops->writepages(mapping, wbc);
} else if (mapping->a_ops->writepage) {
ret = writeback_use_writepage(mapping, wbc);
} else {
/* deal with chardevs and other special files */
ret = 0;
}
if (ret != -ENOMEM || wbc->sync_mode != WB_SYNC_ALL)
break;
/*
* Lacking an allocation context or the locality or writeback
* state of any of the inode's pages, throttle based on
* writeback activity on the local node. It's as good a
* guess as any.
*/
reclaim_throttle(NODE_DATA(numa_node_id()),
mm/vmscan: centralise timeout values for reclaim_throttle Neil Brown raised concerns about callers of reclaim_throttle specifying a timeout value. The original timeout values to congestion_wait() were probably pulled out of thin air or copy&pasted from somewhere else. This patch centralises the timeout values and selects a timeout based on the reason for reclaim throttling. These figures are also pulled out of the same thin air but better values may be derived Running a workload that is throttling for inappropriate periods and tracing mm_vmscan_throttled can be used to pick a more appropriate value. Excessive throttling would pick a lower timeout where as excessive CPU usage in reclaim context would select a larger timeout. Ideally a large value would always be used and the wakeups would occur before a timeout but that requires careful testing. Link: https://lkml.kernel.org/r/20211022144651.19914-7-mgorman@techsingularity.net Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Andreas Dilger <adilger.kernel@dilger.ca> Cc: "Darrick J . Wong" <djwong@kernel.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Matthew Wilcox <willy@infradead.org> Cc: Michal Hocko <mhocko@suse.com> Cc: NeilBrown <neilb@suse.de> Cc: Rik van Riel <riel@surriel.com> Cc: "Theodore Ts'o" <tytso@mit.edu> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-05 20:42:42 +00:00
VMSCAN_THROTTLE_WRITEBACK);
}
writeback: fix bandwidth estimate for spiky workload Michael Stapelberg has reported that for workload with short big spikes of writes (GCC linker seem to trigger this frequently) the write throughput is heavily underestimated and tends to steadily sink until it reaches zero. This has rather bad impact on writeback throttling (causing stalls). The problem is that writeback throughput estimate gets updated at most once per 200 ms. One update happens early after we submit pages for writeback (at that point writeout of only small fraction of pages is completed and thus observed throughput is tiny). Next update happens only during the next write spike (updates happen only from inode writeback and dirty throttling code) and if that is more than 1s after previous spike, we decide system was idle and just ignore whatever was written until this moment. Fix the problem by making sure writeback throughput estimate is also updated shortly after writeback completes to get reasonable estimate of throughput for spiky workloads. [jack@suse.cz: avoid division by 0 in wb_update_dirty_ratelimit()] Link: https://lore.kernel.org/lkml/20210617095309.3542373-1-stapelberg+linux@google.com Link: https://lkml.kernel.org/r/20210713104716.22868-3-jack@suse.cz Signed-off-by: Jan Kara <jack@suse.cz> Reported-by: Michael Stapelberg <stapelberg+linux@google.com> Tested-by: Michael Stapelberg <stapelberg+linux@google.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-09-02 21:53:09 +00:00
/*
* Usually few pages are written by now from those we've just submitted
* but if there's constant writeback being submitted, this makes sure
* writeback bandwidth is updated once in a while.
*/
if (time_is_before_jiffies(READ_ONCE(wb->bw_time_stamp) +
BANDWIDTH_INTERVAL))
writeback: fix bandwidth estimate for spiky workload Michael Stapelberg has reported that for workload with short big spikes of writes (GCC linker seem to trigger this frequently) the write throughput is heavily underestimated and tends to steadily sink until it reaches zero. This has rather bad impact on writeback throttling (causing stalls). The problem is that writeback throughput estimate gets updated at most once per 200 ms. One update happens early after we submit pages for writeback (at that point writeout of only small fraction of pages is completed and thus observed throughput is tiny). Next update happens only during the next write spike (updates happen only from inode writeback and dirty throttling code) and if that is more than 1s after previous spike, we decide system was idle and just ignore whatever was written until this moment. Fix the problem by making sure writeback throughput estimate is also updated shortly after writeback completes to get reasonable estimate of throughput for spiky workloads. [jack@suse.cz: avoid division by 0 in wb_update_dirty_ratelimit()] Link: https://lore.kernel.org/lkml/20210617095309.3542373-1-stapelberg+linux@google.com Link: https://lkml.kernel.org/r/20210713104716.22868-3-jack@suse.cz Signed-off-by: Jan Kara <jack@suse.cz> Reported-by: Michael Stapelberg <stapelberg+linux@google.com> Tested-by: Michael Stapelberg <stapelberg+linux@google.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-09-02 21:53:09 +00:00
wb_update_bandwidth(wb);
return ret;
}
/*
* For address_spaces which do not use buffers nor write back.
*/
bool noop_dirty_folio(struct address_space *mapping, struct folio *folio)
{
if (!folio_test_dirty(folio))
return !folio_test_set_dirty(folio);
return false;
}
EXPORT_SYMBOL(noop_dirty_folio);
/*
* Helper function for set_page_dirty family.
memcg: add per cgroup dirty page accounting When modifying PG_Dirty on cached file pages, update the new MEM_CGROUP_STAT_DIRTY counter. This is done in the same places where global NR_FILE_DIRTY is managed. The new memcg stat is visible in the per memcg memory.stat cgroupfs file. The most recent past attempt at this was http://thread.gmane.org/gmane.linux.kernel.cgroups/8632 The new accounting supports future efforts to add per cgroup dirty page throttling and writeback. It also helps an administrator break down a container's memory usage and provides evidence to understand memcg oom kills (the new dirty count is included in memcg oom kill messages). The ability to move page accounting between memcg (memory.move_charge_at_immigrate) makes this accounting more complicated than the global counter. The existing mem_cgroup_{begin,end}_page_stat() lock is used to serialize move accounting with stat updates. Typical update operation: memcg = mem_cgroup_begin_page_stat(page) if (TestSetPageDirty()) { [...] mem_cgroup_update_page_stat(memcg) } mem_cgroup_end_page_stat(memcg) Summary of mem_cgroup_end_page_stat() overhead: - Without CONFIG_MEMCG it's a no-op - With CONFIG_MEMCG and no inter memcg task movement, it's just rcu_read_lock() - With CONFIG_MEMCG and inter memcg task movement, it's rcu_read_lock() + spin_lock_irqsave() A memcg parameter is added to several routines because their callers now grab mem_cgroup_begin_page_stat() which returns the memcg later needed by for mem_cgroup_update_page_stat(). Because mem_cgroup_begin_page_stat() may disable interrupts, some adjustments are needed: - move __mark_inode_dirty() from __set_page_dirty() to its caller. __mark_inode_dirty() locking does not want interrupts disabled. - use spin_lock_irqsave(tree_lock) rather than spin_lock_irq() in __delete_from_page_cache(), replace_page_cache_page(), invalidate_complete_page2(), and __remove_mapping(). text data bss dec hex filename 8925147 1774832 1785856 12485835 be84cb vmlinux-!CONFIG_MEMCG-before 8925339 1774832 1785856 12486027 be858b vmlinux-!CONFIG_MEMCG-after +192 text bytes 8965977 1784992 1785856 12536825 bf4bf9 vmlinux-CONFIG_MEMCG-before 8966750 1784992 1785856 12537598 bf4efe vmlinux-CONFIG_MEMCG-after +773 text bytes Performance tests run on v4.0-rc1-36-g4f671fe2f952. Lower is better for all metrics, they're all wall clock or cycle counts. The read and write fault benchmarks just measure fault time, they do not include I/O time. * CONFIG_MEMCG not set: baseline patched kbuild 1m25.030000(+-0.088% 3 samples) 1m25.426667(+-0.120% 3 samples) dd write 100 MiB 0.859211561 +-15.10% 0.874162885 +-15.03% dd write 200 MiB 1.670653105 +-17.87% 1.669384764 +-11.99% dd write 1000 MiB 8.434691190 +-14.15% 8.474733215 +-14.77% read fault cycles 254.0(+-0.000% 10 samples) 253.0(+-0.000% 10 samples) write fault cycles 2021.2(+-3.070% 10 samples) 1984.5(+-1.036% 10 samples) * CONFIG_MEMCG=y root_memcg: baseline patched kbuild 1m25.716667(+-0.105% 3 samples) 1m25.686667(+-0.153% 3 samples) dd write 100 MiB 0.855650830 +-14.90% 0.887557919 +-14.90% dd write 200 MiB 1.688322953 +-12.72% 1.667682724 +-13.33% dd write 1000 MiB 8.418601605 +-14.30% 8.673532299 +-15.00% read fault cycles 266.0(+-0.000% 10 samples) 266.0(+-0.000% 10 samples) write fault cycles 2051.7(+-1.349% 10 samples) 2049.6(+-1.686% 10 samples) * CONFIG_MEMCG=y non-root_memcg: baseline patched kbuild 1m26.120000(+-0.273% 3 samples) 1m25.763333(+-0.127% 3 samples) dd write 100 MiB 0.861723964 +-15.25% 0.818129350 +-14.82% dd write 200 MiB 1.669887569 +-13.30% 1.698645885 +-13.27% dd write 1000 MiB 8.383191730 +-14.65% 8.351742280 +-14.52% read fault cycles 265.7(+-0.172% 10 samples) 267.0(+-0.000% 10 samples) write fault cycles 2070.6(+-1.512% 10 samples) 2084.4(+-2.148% 10 samples) As expected anon page faults are not affected by this patch. tj: Updated to apply on top of the recent cancel_dirty_page() changes. Signed-off-by: Sha Zhengju <handai.szj@gmail.com> Signed-off-by: Greg Thelen <gthelen@google.com> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:16 +00:00
*
* Caller must hold folio_memcg_lock().
memcg: add per cgroup dirty page accounting When modifying PG_Dirty on cached file pages, update the new MEM_CGROUP_STAT_DIRTY counter. This is done in the same places where global NR_FILE_DIRTY is managed. The new memcg stat is visible in the per memcg memory.stat cgroupfs file. The most recent past attempt at this was http://thread.gmane.org/gmane.linux.kernel.cgroups/8632 The new accounting supports future efforts to add per cgroup dirty page throttling and writeback. It also helps an administrator break down a container's memory usage and provides evidence to understand memcg oom kills (the new dirty count is included in memcg oom kill messages). The ability to move page accounting between memcg (memory.move_charge_at_immigrate) makes this accounting more complicated than the global counter. The existing mem_cgroup_{begin,end}_page_stat() lock is used to serialize move accounting with stat updates. Typical update operation: memcg = mem_cgroup_begin_page_stat(page) if (TestSetPageDirty()) { [...] mem_cgroup_update_page_stat(memcg) } mem_cgroup_end_page_stat(memcg) Summary of mem_cgroup_end_page_stat() overhead: - Without CONFIG_MEMCG it's a no-op - With CONFIG_MEMCG and no inter memcg task movement, it's just rcu_read_lock() - With CONFIG_MEMCG and inter memcg task movement, it's rcu_read_lock() + spin_lock_irqsave() A memcg parameter is added to several routines because their callers now grab mem_cgroup_begin_page_stat() which returns the memcg later needed by for mem_cgroup_update_page_stat(). Because mem_cgroup_begin_page_stat() may disable interrupts, some adjustments are needed: - move __mark_inode_dirty() from __set_page_dirty() to its caller. __mark_inode_dirty() locking does not want interrupts disabled. - use spin_lock_irqsave(tree_lock) rather than spin_lock_irq() in __delete_from_page_cache(), replace_page_cache_page(), invalidate_complete_page2(), and __remove_mapping(). text data bss dec hex filename 8925147 1774832 1785856 12485835 be84cb vmlinux-!CONFIG_MEMCG-before 8925339 1774832 1785856 12486027 be858b vmlinux-!CONFIG_MEMCG-after +192 text bytes 8965977 1784992 1785856 12536825 bf4bf9 vmlinux-CONFIG_MEMCG-before 8966750 1784992 1785856 12537598 bf4efe vmlinux-CONFIG_MEMCG-after +773 text bytes Performance tests run on v4.0-rc1-36-g4f671fe2f952. Lower is better for all metrics, they're all wall clock or cycle counts. The read and write fault benchmarks just measure fault time, they do not include I/O time. * CONFIG_MEMCG not set: baseline patched kbuild 1m25.030000(+-0.088% 3 samples) 1m25.426667(+-0.120% 3 samples) dd write 100 MiB 0.859211561 +-15.10% 0.874162885 +-15.03% dd write 200 MiB 1.670653105 +-17.87% 1.669384764 +-11.99% dd write 1000 MiB 8.434691190 +-14.15% 8.474733215 +-14.77% read fault cycles 254.0(+-0.000% 10 samples) 253.0(+-0.000% 10 samples) write fault cycles 2021.2(+-3.070% 10 samples) 1984.5(+-1.036% 10 samples) * CONFIG_MEMCG=y root_memcg: baseline patched kbuild 1m25.716667(+-0.105% 3 samples) 1m25.686667(+-0.153% 3 samples) dd write 100 MiB 0.855650830 +-14.90% 0.887557919 +-14.90% dd write 200 MiB 1.688322953 +-12.72% 1.667682724 +-13.33% dd write 1000 MiB 8.418601605 +-14.30% 8.673532299 +-15.00% read fault cycles 266.0(+-0.000% 10 samples) 266.0(+-0.000% 10 samples) write fault cycles 2051.7(+-1.349% 10 samples) 2049.6(+-1.686% 10 samples) * CONFIG_MEMCG=y non-root_memcg: baseline patched kbuild 1m26.120000(+-0.273% 3 samples) 1m25.763333(+-0.127% 3 samples) dd write 100 MiB 0.861723964 +-15.25% 0.818129350 +-14.82% dd write 200 MiB 1.669887569 +-13.30% 1.698645885 +-13.27% dd write 1000 MiB 8.383191730 +-14.65% 8.351742280 +-14.52% read fault cycles 265.7(+-0.172% 10 samples) 267.0(+-0.000% 10 samples) write fault cycles 2070.6(+-1.512% 10 samples) 2084.4(+-2.148% 10 samples) As expected anon page faults are not affected by this patch. tj: Updated to apply on top of the recent cancel_dirty_page() changes. Signed-off-by: Sha Zhengju <handai.szj@gmail.com> Signed-off-by: Greg Thelen <gthelen@google.com> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:16 +00:00
*
* NOTE: This relies on being atomic wrt interrupts.
*/
static void folio_account_dirtied(struct folio *folio,
struct address_space *mapping)
{
writeback: make backing_dev_info host cgroup-specific bdi_writebacks For the planned cgroup writeback support, on each bdi (backing_dev_info), each memcg will be served by a separate wb (bdi_writeback). This patch updates bdi so that a bdi can host multiple wbs (bdi_writebacks). On the default hierarchy, blkcg implicitly enables memcg. This allows using memcg's page ownership for attributing writeback IOs, and every memcg - blkcg combination can be served by its own wb by assigning a dedicated wb to each memcg. This means that there may be multiple wb's of a bdi mapped to the same blkcg. As congested state is per blkcg - bdi combination, those wb's should share the same congested state. This is achieved by tracking congested state via bdi_writeback_congested structs which are keyed by blkcg. bdi->wb remains unchanged and will keep serving the root cgroup. cgwb's (cgroup wb's) for non-root cgroups are created on-demand or looked up while dirtying an inode according to the memcg of the page being dirtied or current task. Each cgwb is indexed on bdi->cgwb_tree by its memcg id. Once an inode is associated with its wb, it can be retrieved using inode_to_wb(). Currently, none of the filesystems has FS_CGROUP_WRITEBACK and all pages will keep being associated with bdi->wb. v3: inode_attach_wb() in account_page_dirtied() moved inside mapping_cap_account_dirty() block where it's known to be !NULL. Also, an unnecessary NULL check before kfree() removed. Both detected by the kbuild bot. v2: Updated so that wb association is per inode and wb is per memcg rather than blkcg. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: kbuild test robot <fengguang.wu@intel.com> Cc: Dan Carpenter <dan.carpenter@oracle.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:37 +00:00
struct inode *inode = mapping->host;
trace_writeback_dirty_folio(folio, mapping);
if (mapping_can_writeback(mapping)) {
writeback: make backing_dev_info host cgroup-specific bdi_writebacks For the planned cgroup writeback support, on each bdi (backing_dev_info), each memcg will be served by a separate wb (bdi_writeback). This patch updates bdi so that a bdi can host multiple wbs (bdi_writebacks). On the default hierarchy, blkcg implicitly enables memcg. This allows using memcg's page ownership for attributing writeback IOs, and every memcg - blkcg combination can be served by its own wb by assigning a dedicated wb to each memcg. This means that there may be multiple wb's of a bdi mapped to the same blkcg. As congested state is per blkcg - bdi combination, those wb's should share the same congested state. This is achieved by tracking congested state via bdi_writeback_congested structs which are keyed by blkcg. bdi->wb remains unchanged and will keep serving the root cgroup. cgwb's (cgroup wb's) for non-root cgroups are created on-demand or looked up while dirtying an inode according to the memcg of the page being dirtied or current task. Each cgwb is indexed on bdi->cgwb_tree by its memcg id. Once an inode is associated with its wb, it can be retrieved using inode_to_wb(). Currently, none of the filesystems has FS_CGROUP_WRITEBACK and all pages will keep being associated with bdi->wb. v3: inode_attach_wb() in account_page_dirtied() moved inside mapping_cap_account_dirty() block where it's known to be !NULL. Also, an unnecessary NULL check before kfree() removed. Both detected by the kbuild bot. v2: Updated so that wb association is per inode and wb is per memcg rather than blkcg. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: kbuild test robot <fengguang.wu@intel.com> Cc: Dan Carpenter <dan.carpenter@oracle.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:37 +00:00
struct bdi_writeback *wb;
long nr = folio_nr_pages(folio);
inode_attach_wb(inode, folio);
writeback: make backing_dev_info host cgroup-specific bdi_writebacks For the planned cgroup writeback support, on each bdi (backing_dev_info), each memcg will be served by a separate wb (bdi_writeback). This patch updates bdi so that a bdi can host multiple wbs (bdi_writebacks). On the default hierarchy, blkcg implicitly enables memcg. This allows using memcg's page ownership for attributing writeback IOs, and every memcg - blkcg combination can be served by its own wb by assigning a dedicated wb to each memcg. This means that there may be multiple wb's of a bdi mapped to the same blkcg. As congested state is per blkcg - bdi combination, those wb's should share the same congested state. This is achieved by tracking congested state via bdi_writeback_congested structs which are keyed by blkcg. bdi->wb remains unchanged and will keep serving the root cgroup. cgwb's (cgroup wb's) for non-root cgroups are created on-demand or looked up while dirtying an inode according to the memcg of the page being dirtied or current task. Each cgwb is indexed on bdi->cgwb_tree by its memcg id. Once an inode is associated with its wb, it can be retrieved using inode_to_wb(). Currently, none of the filesystems has FS_CGROUP_WRITEBACK and all pages will keep being associated with bdi->wb. v3: inode_attach_wb() in account_page_dirtied() moved inside mapping_cap_account_dirty() block where it's known to be !NULL. Also, an unnecessary NULL check before kfree() removed. Both detected by the kbuild bot. v2: Updated so that wb association is per inode and wb is per memcg rather than blkcg. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: kbuild test robot <fengguang.wu@intel.com> Cc: Dan Carpenter <dan.carpenter@oracle.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:37 +00:00
wb = inode_to_wb(inode);
__lruvec_stat_mod_folio(folio, NR_FILE_DIRTY, nr);
__zone_stat_mod_folio(folio, NR_ZONE_WRITE_PENDING, nr);
__node_stat_mod_folio(folio, NR_DIRTIED, nr);
wb_stat_mod(wb, WB_RECLAIMABLE, nr);
wb_stat_mod(wb, WB_DIRTIED, nr);
task_io_account_write(nr * PAGE_SIZE);
current->nr_dirtied += nr;
__this_cpu_add(bdp_ratelimits, nr);
writeback, memcg: Implement foreign dirty flushing There's an inherent mismatch between memcg and writeback. The former trackes ownership per-page while the latter per-inode. This was a deliberate design decision because honoring per-page ownership in the writeback path is complicated, may lead to higher CPU and IO overheads and deemed unnecessary given that write-sharing an inode across different cgroups isn't a common use-case. Combined with inode majority-writer ownership switching, this works well enough in most cases but there are some pathological cases. For example, let's say there are two cgroups A and B which keep writing to different but confined parts of the same inode. B owns the inode and A's memory is limited far below B's. A's dirty ratio can rise enough to trigger balance_dirty_pages() sleeps but B's can be low enough to avoid triggering background writeback. A will be slowed down without a way to make writeback of the dirty pages happen. This patch implements foreign dirty recording and foreign mechanism so that when a memcg encounters a condition as above it can trigger flushes on bdi_writebacks which can clean its pages. Please see the comment on top of mem_cgroup_track_foreign_dirty_slowpath() for details. A reproducer follows. write-range.c:: #include <stdio.h> #include <stdlib.h> #include <unistd.h> #include <fcntl.h> #include <sys/types.h> static const char *usage = "write-range FILE START SIZE\n"; int main(int argc, char **argv) { int fd; unsigned long start, size, end, pos; char *endp; char buf[4096]; if (argc < 4) { fprintf(stderr, usage); return 1; } fd = open(argv[1], O_WRONLY); if (fd < 0) { perror("open"); return 1; } start = strtoul(argv[2], &endp, 0); if (*endp != '\0') { fprintf(stderr, usage); return 1; } size = strtoul(argv[3], &endp, 0); if (*endp != '\0') { fprintf(stderr, usage); return 1; } end = start + size; while (1) { for (pos = start; pos < end; ) { long bread, bwritten = 0; if (lseek(fd, pos, SEEK_SET) < 0) { perror("lseek"); return 1; } bread = read(0, buf, sizeof(buf) < end - pos ? sizeof(buf) : end - pos); if (bread < 0) { perror("read"); return 1; } if (bread == 0) return 0; while (bwritten < bread) { long this; this = write(fd, buf + bwritten, bread - bwritten); if (this < 0) { perror("write"); return 1; } bwritten += this; pos += bwritten; } } } } repro.sh:: #!/bin/bash set -e set -x sysctl -w vm.dirty_expire_centisecs=300000 sysctl -w vm.dirty_writeback_centisecs=300000 sysctl -w vm.dirtytime_expire_seconds=300000 echo 3 > /proc/sys/vm/drop_caches TEST=/sys/fs/cgroup/test A=$TEST/A B=$TEST/B mkdir -p $A $B echo "+memory +io" > $TEST/cgroup.subtree_control echo $((1<<30)) > $A/memory.high echo $((32<<30)) > $B/memory.high rm -f testfile touch testfile fallocate -l 4G testfile echo "Starting B" (echo $BASHPID > $B/cgroup.procs pv -q --rate-limit 70M < /dev/urandom | ./write-range testfile $((2<<30)) $((2<<30))) & echo "Waiting 10s to ensure B claims the testfile inode" sleep 5 sync sleep 5 sync echo "Starting A" (echo $BASHPID > $A/cgroup.procs pv < /dev/urandom | ./write-range testfile 0 $((2<<30))) v2: Added comments explaining why the specific intervals are being used. v3: Use 0 @nr when calling cgroup_writeback_by_id() to use best-effort flushing while avoding possible livelocks. v4: Use get_jiffies_64() and time_before/after64() instead of raw jiffies_64 and arthimetic comparisons as suggested by Jan. Reviewed-by: Jan Kara <jack@suse.cz> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-08-26 16:06:56 +00:00
mem_cgroup_track_foreign_dirty(folio, wb);
}
}
page_writeback: clean up mess around cancel_dirty_page() This patch replaces cancel_dirty_page() with a helper function account_page_cleaned() which only updates counters. It's called from truncate_complete_page() and from try_to_free_buffers() (hack for ext3). Page is locked in both cases, page-lock protects against concurrent dirtiers: see commit 2d6d7f982846 ("mm: protect set_page_dirty() from ongoing truncation"). Delete_from_page_cache() shouldn't be called for dirty pages, they must be handled by caller (either written or truncated). This patch treats final dirty accounting fixup at the end of __delete_from_page_cache() as a debug check and adds WARN_ON_ONCE() around it. If something removes dirty pages without proper handling that might be a bug and unwritten data might be lost. Hugetlbfs has no dirty pages accounting, ClearPageDirty() is enough here. cancel_dirty_page() in nfs_wb_page_cancel() is redundant. This is helper for nfs_invalidate_page() and it's called only in case complete invalidation. The mess was started in v2.6.20 after commits 46d2277c796f ("Clean up and make try_to_free_buffers() not race with dirty pages") and 3e67c0987d75 ("truncate: clear page dirtiness before running try_to_free_buffers()") first was reverted right in v2.6.20 in commit ecdfc9787fe5 ("Resurrect 'try_to_free_buffers()' VM hackery"), second in v2.6.25 commit a2b345642f53 ("Fix dirty page accounting leak with ext3 data=journal"). Custom fixes were introduced between these points. NFS in v2.6.23, commit 1b3b4a1a2deb ("NFS: Fix a write request leak in nfs_invalidate_page()"). Kludge in __delete_from_page_cache() in v2.6.24, commit 3a6927906f1b ("Do dirty page accounting when removing a page from the page cache"). Since v2.6.25 all of them are redundant. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru> Cc: Tejun Heo <tj@kernel.org> Cc: Jan Kara <jack@suse.cz> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-04-14 22:45:27 +00:00
/*
* Helper function for deaccounting dirty page without writeback.
*
* Caller must hold folio_memcg_lock().
page_writeback: clean up mess around cancel_dirty_page() This patch replaces cancel_dirty_page() with a helper function account_page_cleaned() which only updates counters. It's called from truncate_complete_page() and from try_to_free_buffers() (hack for ext3). Page is locked in both cases, page-lock protects against concurrent dirtiers: see commit 2d6d7f982846 ("mm: protect set_page_dirty() from ongoing truncation"). Delete_from_page_cache() shouldn't be called for dirty pages, they must be handled by caller (either written or truncated). This patch treats final dirty accounting fixup at the end of __delete_from_page_cache() as a debug check and adds WARN_ON_ONCE() around it. If something removes dirty pages without proper handling that might be a bug and unwritten data might be lost. Hugetlbfs has no dirty pages accounting, ClearPageDirty() is enough here. cancel_dirty_page() in nfs_wb_page_cancel() is redundant. This is helper for nfs_invalidate_page() and it's called only in case complete invalidation. The mess was started in v2.6.20 after commits 46d2277c796f ("Clean up and make try_to_free_buffers() not race with dirty pages") and 3e67c0987d75 ("truncate: clear page dirtiness before running try_to_free_buffers()") first was reverted right in v2.6.20 in commit ecdfc9787fe5 ("Resurrect 'try_to_free_buffers()' VM hackery"), second in v2.6.25 commit a2b345642f53 ("Fix dirty page accounting leak with ext3 data=journal"). Custom fixes were introduced between these points. NFS in v2.6.23, commit 1b3b4a1a2deb ("NFS: Fix a write request leak in nfs_invalidate_page()"). Kludge in __delete_from_page_cache() in v2.6.24, commit 3a6927906f1b ("Do dirty page accounting when removing a page from the page cache"). Since v2.6.25 all of them are redundant. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru> Cc: Tejun Heo <tj@kernel.org> Cc: Jan Kara <jack@suse.cz> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-04-14 22:45:27 +00:00
*/
void folio_account_cleaned(struct folio *folio, struct bdi_writeback *wb)
page_writeback: clean up mess around cancel_dirty_page() This patch replaces cancel_dirty_page() with a helper function account_page_cleaned() which only updates counters. It's called from truncate_complete_page() and from try_to_free_buffers() (hack for ext3). Page is locked in both cases, page-lock protects against concurrent dirtiers: see commit 2d6d7f982846 ("mm: protect set_page_dirty() from ongoing truncation"). Delete_from_page_cache() shouldn't be called for dirty pages, they must be handled by caller (either written or truncated). This patch treats final dirty accounting fixup at the end of __delete_from_page_cache() as a debug check and adds WARN_ON_ONCE() around it. If something removes dirty pages without proper handling that might be a bug and unwritten data might be lost. Hugetlbfs has no dirty pages accounting, ClearPageDirty() is enough here. cancel_dirty_page() in nfs_wb_page_cancel() is redundant. This is helper for nfs_invalidate_page() and it's called only in case complete invalidation. The mess was started in v2.6.20 after commits 46d2277c796f ("Clean up and make try_to_free_buffers() not race with dirty pages") and 3e67c0987d75 ("truncate: clear page dirtiness before running try_to_free_buffers()") first was reverted right in v2.6.20 in commit ecdfc9787fe5 ("Resurrect 'try_to_free_buffers()' VM hackery"), second in v2.6.25 commit a2b345642f53 ("Fix dirty page accounting leak with ext3 data=journal"). Custom fixes were introduced between these points. NFS in v2.6.23, commit 1b3b4a1a2deb ("NFS: Fix a write request leak in nfs_invalidate_page()"). Kludge in __delete_from_page_cache() in v2.6.24, commit 3a6927906f1b ("Do dirty page accounting when removing a page from the page cache"). Since v2.6.25 all of them are redundant. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru> Cc: Tejun Heo <tj@kernel.org> Cc: Jan Kara <jack@suse.cz> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-04-14 22:45:27 +00:00
{
long nr = folio_nr_pages(folio);
lruvec_stat_mod_folio(folio, NR_FILE_DIRTY, -nr);
zone_stat_mod_folio(folio, NR_ZONE_WRITE_PENDING, -nr);
wb_stat_mod(wb, WB_RECLAIMABLE, -nr);
task_io_account_cancelled_write(nr * PAGE_SIZE);
page_writeback: clean up mess around cancel_dirty_page() This patch replaces cancel_dirty_page() with a helper function account_page_cleaned() which only updates counters. It's called from truncate_complete_page() and from try_to_free_buffers() (hack for ext3). Page is locked in both cases, page-lock protects against concurrent dirtiers: see commit 2d6d7f982846 ("mm: protect set_page_dirty() from ongoing truncation"). Delete_from_page_cache() shouldn't be called for dirty pages, they must be handled by caller (either written or truncated). This patch treats final dirty accounting fixup at the end of __delete_from_page_cache() as a debug check and adds WARN_ON_ONCE() around it. If something removes dirty pages without proper handling that might be a bug and unwritten data might be lost. Hugetlbfs has no dirty pages accounting, ClearPageDirty() is enough here. cancel_dirty_page() in nfs_wb_page_cancel() is redundant. This is helper for nfs_invalidate_page() and it's called only in case complete invalidation. The mess was started in v2.6.20 after commits 46d2277c796f ("Clean up and make try_to_free_buffers() not race with dirty pages") and 3e67c0987d75 ("truncate: clear page dirtiness before running try_to_free_buffers()") first was reverted right in v2.6.20 in commit ecdfc9787fe5 ("Resurrect 'try_to_free_buffers()' VM hackery"), second in v2.6.25 commit a2b345642f53 ("Fix dirty page accounting leak with ext3 data=journal"). Custom fixes were introduced between these points. NFS in v2.6.23, commit 1b3b4a1a2deb ("NFS: Fix a write request leak in nfs_invalidate_page()"). Kludge in __delete_from_page_cache() in v2.6.24, commit 3a6927906f1b ("Do dirty page accounting when removing a page from the page cache"). Since v2.6.25 all of them are redundant. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru> Cc: Tejun Heo <tj@kernel.org> Cc: Jan Kara <jack@suse.cz> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-04-14 22:45:27 +00:00
}
/*
* Mark the folio dirty, and set it dirty in the page cache.
*
* If warn is true, then emit a warning if the folio is not uptodate and has
* not been truncated.
*
* The caller must hold folio_memcg_lock(). It is the caller's
* responsibility to prevent the folio from being truncated while
* this function is in progress, although it may have been truncated
* before this function is called. Most callers have the folio locked.
* A few have the folio blocked from truncation through other means (e.g.
* zap_vma_pages() has it mapped and is holding the page table lock).
* When called from mark_buffer_dirty(), the filesystem should hold a
* reference to the buffer_head that is being marked dirty, which causes
* try_to_free_buffers() to fail.
*/
void __folio_mark_dirty(struct folio *folio, struct address_space *mapping,
int warn)
{
unsigned long flags;
xa_lock_irqsave(&mapping->i_pages, flags);
if (folio->mapping) { /* Race with truncate? */
WARN_ON_ONCE(warn && !folio_test_uptodate(folio));
folio_account_dirtied(folio, mapping);
__xa_set_mark(&mapping->i_pages, folio_index(folio),
PAGECACHE_TAG_DIRTY);
}
xa_unlock_irqrestore(&mapping->i_pages, flags);
}
/**
* filemap_dirty_folio - Mark a folio dirty for filesystems which do not use buffer_heads.
* @mapping: Address space this folio belongs to.
* @folio: Folio to be marked as dirty.
*
* Filesystems which do not use buffer heads should call this function
* from their dirty_folio address space operation. It ignores the
* contents of folio_get_private(), so if the filesystem marks individual
* blocks as dirty, the filesystem should handle that itself.
*
* This is also sometimes used by filesystems which use buffer_heads when
* a single buffer is being dirtied: we want to set the folio dirty in
* that case, but not all the buffers. This is a "bottom-up" dirtying,
* whereas block_dirty_folio() is a "top-down" dirtying.
*
* The caller must ensure this doesn't race with truncation. Most will
* simply hold the folio lock, but e.g. zap_pte_range() calls with the
* folio mapped and the pte lock held, which also locks out truncation.
*/
bool filemap_dirty_folio(struct address_space *mapping, struct folio *folio)
{
folio_memcg_lock(folio);
if (folio_test_set_dirty(folio)) {
folio_memcg_unlock(folio);
return false;
}
__folio_mark_dirty(folio, mapping, !folio_test_private(folio));
folio_memcg_unlock(folio);
memcg: add per cgroup dirty page accounting When modifying PG_Dirty on cached file pages, update the new MEM_CGROUP_STAT_DIRTY counter. This is done in the same places where global NR_FILE_DIRTY is managed. The new memcg stat is visible in the per memcg memory.stat cgroupfs file. The most recent past attempt at this was http://thread.gmane.org/gmane.linux.kernel.cgroups/8632 The new accounting supports future efforts to add per cgroup dirty page throttling and writeback. It also helps an administrator break down a container's memory usage and provides evidence to understand memcg oom kills (the new dirty count is included in memcg oom kill messages). The ability to move page accounting between memcg (memory.move_charge_at_immigrate) makes this accounting more complicated than the global counter. The existing mem_cgroup_{begin,end}_page_stat() lock is used to serialize move accounting with stat updates. Typical update operation: memcg = mem_cgroup_begin_page_stat(page) if (TestSetPageDirty()) { [...] mem_cgroup_update_page_stat(memcg) } mem_cgroup_end_page_stat(memcg) Summary of mem_cgroup_end_page_stat() overhead: - Without CONFIG_MEMCG it's a no-op - With CONFIG_MEMCG and no inter memcg task movement, it's just rcu_read_lock() - With CONFIG_MEMCG and inter memcg task movement, it's rcu_read_lock() + spin_lock_irqsave() A memcg parameter is added to several routines because their callers now grab mem_cgroup_begin_page_stat() which returns the memcg later needed by for mem_cgroup_update_page_stat(). Because mem_cgroup_begin_page_stat() may disable interrupts, some adjustments are needed: - move __mark_inode_dirty() from __set_page_dirty() to its caller. __mark_inode_dirty() locking does not want interrupts disabled. - use spin_lock_irqsave(tree_lock) rather than spin_lock_irq() in __delete_from_page_cache(), replace_page_cache_page(), invalidate_complete_page2(), and __remove_mapping(). text data bss dec hex filename 8925147 1774832 1785856 12485835 be84cb vmlinux-!CONFIG_MEMCG-before 8925339 1774832 1785856 12486027 be858b vmlinux-!CONFIG_MEMCG-after +192 text bytes 8965977 1784992 1785856 12536825 bf4bf9 vmlinux-CONFIG_MEMCG-before 8966750 1784992 1785856 12537598 bf4efe vmlinux-CONFIG_MEMCG-after +773 text bytes Performance tests run on v4.0-rc1-36-g4f671fe2f952. Lower is better for all metrics, they're all wall clock or cycle counts. The read and write fault benchmarks just measure fault time, they do not include I/O time. * CONFIG_MEMCG not set: baseline patched kbuild 1m25.030000(+-0.088% 3 samples) 1m25.426667(+-0.120% 3 samples) dd write 100 MiB 0.859211561 +-15.10% 0.874162885 +-15.03% dd write 200 MiB 1.670653105 +-17.87% 1.669384764 +-11.99% dd write 1000 MiB 8.434691190 +-14.15% 8.474733215 +-14.77% read fault cycles 254.0(+-0.000% 10 samples) 253.0(+-0.000% 10 samples) write fault cycles 2021.2(+-3.070% 10 samples) 1984.5(+-1.036% 10 samples) * CONFIG_MEMCG=y root_memcg: baseline patched kbuild 1m25.716667(+-0.105% 3 samples) 1m25.686667(+-0.153% 3 samples) dd write 100 MiB 0.855650830 +-14.90% 0.887557919 +-14.90% dd write 200 MiB 1.688322953 +-12.72% 1.667682724 +-13.33% dd write 1000 MiB 8.418601605 +-14.30% 8.673532299 +-15.00% read fault cycles 266.0(+-0.000% 10 samples) 266.0(+-0.000% 10 samples) write fault cycles 2051.7(+-1.349% 10 samples) 2049.6(+-1.686% 10 samples) * CONFIG_MEMCG=y non-root_memcg: baseline patched kbuild 1m26.120000(+-0.273% 3 samples) 1m25.763333(+-0.127% 3 samples) dd write 100 MiB 0.861723964 +-15.25% 0.818129350 +-14.82% dd write 200 MiB 1.669887569 +-13.30% 1.698645885 +-13.27% dd write 1000 MiB 8.383191730 +-14.65% 8.351742280 +-14.52% read fault cycles 265.7(+-0.172% 10 samples) 267.0(+-0.000% 10 samples) write fault cycles 2070.6(+-1.512% 10 samples) 2084.4(+-2.148% 10 samples) As expected anon page faults are not affected by this patch. tj: Updated to apply on top of the recent cancel_dirty_page() changes. Signed-off-by: Sha Zhengju <handai.szj@gmail.com> Signed-off-by: Greg Thelen <gthelen@google.com> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:16 +00:00
if (mapping->host) {
/* !PageAnon && !swapper_space */
__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
}
return true;
}
EXPORT_SYMBOL(filemap_dirty_folio);
/**
* folio_redirty_for_writepage - Decline to write a dirty folio.
* @wbc: The writeback control.
* @folio: The folio.
*
* When a writepage implementation decides that it doesn't want to write
* @folio for some reason, it should call this function, unlock @folio and
* return 0.
*
* Return: True if we redirtied the folio. False if someone else dirtied
* it first.
*/
bool folio_redirty_for_writepage(struct writeback_control *wbc,
struct folio *folio)
{
struct address_space *mapping = folio->mapping;
long nr = folio_nr_pages(folio);
bool ret;
wbc->pages_skipped += nr;
ret = filemap_dirty_folio(mapping, folio);
if (mapping && mapping_can_writeback(mapping)) {
struct inode *inode = mapping->host;
struct bdi_writeback *wb;
struct wb_lock_cookie cookie = {};
wb = unlocked_inode_to_wb_begin(inode, &cookie);
current->nr_dirtied -= nr;
node_stat_mod_folio(folio, NR_DIRTIED, -nr);
wb_stat_mod(wb, WB_DIRTIED, -nr);
unlocked_inode_to_wb_end(inode, &cookie);
}
return ret;
}
EXPORT_SYMBOL(folio_redirty_for_writepage);
/**
* folio_mark_dirty - Mark a folio as being modified.
* @folio: The folio.
*
* The folio may not be truncated while this function is running.
* Holding the folio lock is sufficient to prevent truncation, but some
* callers cannot acquire a sleeping lock. These callers instead hold
* the page table lock for a page table which contains at least one page
* in this folio. Truncation will block on the page table lock as it
* unmaps pages before removing the folio from its mapping.
*
* Return: True if the folio was newly dirtied, false if it was already dirty.
*/
bool folio_mark_dirty(struct folio *folio)
{
struct address_space *mapping = folio_mapping(folio);
if (likely(mapping)) {
mm: reclaim invalidated page ASAP invalidate_mapping_pages is very big hint to reclaimer. It means user doesn't want to use the page any more. So in order to prevent working set page eviction, this patch move the page into tail of inactive list by PG_reclaim. Please, remember that pages in inactive list are working set as well as active list. If we don't move pages into inactive list's tail, pages near by tail of inactive list can be evicted although we have a big clue about useless pages. It's totally bad. Now PG_readahead/PG_reclaim is shared. fe3cba17 added ClearPageReclaim into clear_page_dirty_for_io for preventing fast reclaiming readahead marker page. In this series, PG_reclaim is used by invalidated page, too. If VM find the page is invalidated and it's dirty, it sets PG_reclaim to reclaim asap. Then, when the dirty page will be writeback, clear_page_dirty_for_io will clear PG_reclaim unconditionally. It disturbs this serie's goal. I think it's okay to clear PG_readahead when the page is dirty, not writeback time. So this patch moves ClearPageReadahead. In v4, ClearPageReadahead in set_page_dirty has a problem which is reported by Steven Barrett. It's due to compound page. Some driver(ex, audio) calls set_page_dirty with compound page which isn't on LRU. but my patch does ClearPageRelcaim on compound page. In non-CONFIG_PAGEFLAGS_EXTENDED, it breaks PageTail flag. I think it doesn't affect THP and pass my test with THP enabling but Cced Andrea for double check. Signed-off-by: Minchan Kim <minchan.kim@gmail.com> Reported-by: Steven Barrett <damentz@liquorix.net> Reviewed-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Andrea Arcangeli <aarcange@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-22 23:32:54 +00:00
/*
* readahead/folio_deactivate could remain
* PG_readahead/PG_reclaim due to race with folio_end_writeback
* About readahead, if the folio is written, the flags would be
mm: reclaim invalidated page ASAP invalidate_mapping_pages is very big hint to reclaimer. It means user doesn't want to use the page any more. So in order to prevent working set page eviction, this patch move the page into tail of inactive list by PG_reclaim. Please, remember that pages in inactive list are working set as well as active list. If we don't move pages into inactive list's tail, pages near by tail of inactive list can be evicted although we have a big clue about useless pages. It's totally bad. Now PG_readahead/PG_reclaim is shared. fe3cba17 added ClearPageReclaim into clear_page_dirty_for_io for preventing fast reclaiming readahead marker page. In this series, PG_reclaim is used by invalidated page, too. If VM find the page is invalidated and it's dirty, it sets PG_reclaim to reclaim asap. Then, when the dirty page will be writeback, clear_page_dirty_for_io will clear PG_reclaim unconditionally. It disturbs this serie's goal. I think it's okay to clear PG_readahead when the page is dirty, not writeback time. So this patch moves ClearPageReadahead. In v4, ClearPageReadahead in set_page_dirty has a problem which is reported by Steven Barrett. It's due to compound page. Some driver(ex, audio) calls set_page_dirty with compound page which isn't on LRU. but my patch does ClearPageRelcaim on compound page. In non-CONFIG_PAGEFLAGS_EXTENDED, it breaks PageTail flag. I think it doesn't affect THP and pass my test with THP enabling but Cced Andrea for double check. Signed-off-by: Minchan Kim <minchan.kim@gmail.com> Reported-by: Steven Barrett <damentz@liquorix.net> Reviewed-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Andrea Arcangeli <aarcange@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-22 23:32:54 +00:00
* reset. So no problem.
* About folio_deactivate, if the folio is redirtied,
* the flag will be reset. So no problem. but if the
* folio is used by readahead it will confuse readahead
* and make it restart the size rampup process. But it's
* a trivial problem.
mm: reclaim invalidated page ASAP invalidate_mapping_pages is very big hint to reclaimer. It means user doesn't want to use the page any more. So in order to prevent working set page eviction, this patch move the page into tail of inactive list by PG_reclaim. Please, remember that pages in inactive list are working set as well as active list. If we don't move pages into inactive list's tail, pages near by tail of inactive list can be evicted although we have a big clue about useless pages. It's totally bad. Now PG_readahead/PG_reclaim is shared. fe3cba17 added ClearPageReclaim into clear_page_dirty_for_io for preventing fast reclaiming readahead marker page. In this series, PG_reclaim is used by invalidated page, too. If VM find the page is invalidated and it's dirty, it sets PG_reclaim to reclaim asap. Then, when the dirty page will be writeback, clear_page_dirty_for_io will clear PG_reclaim unconditionally. It disturbs this serie's goal. I think it's okay to clear PG_readahead when the page is dirty, not writeback time. So this patch moves ClearPageReadahead. In v4, ClearPageReadahead in set_page_dirty has a problem which is reported by Steven Barrett. It's due to compound page. Some driver(ex, audio) calls set_page_dirty with compound page which isn't on LRU. but my patch does ClearPageRelcaim on compound page. In non-CONFIG_PAGEFLAGS_EXTENDED, it breaks PageTail flag. I think it doesn't affect THP and pass my test with THP enabling but Cced Andrea for double check. Signed-off-by: Minchan Kim <minchan.kim@gmail.com> Reported-by: Steven Barrett <damentz@liquorix.net> Reviewed-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Andrea Arcangeli <aarcange@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-22 23:32:54 +00:00
*/
if (folio_test_reclaim(folio))
folio_clear_reclaim(folio);
return mapping->a_ops->dirty_folio(mapping, folio);
}
return noop_dirty_folio(mapping, folio);
}
EXPORT_SYMBOL(folio_mark_dirty);
/*
* set_page_dirty() is racy if the caller has no reference against
* page->mapping->host, and if the page is unlocked. This is because another
* CPU could truncate the page off the mapping and then free the mapping.
*
* Usually, the page _is_ locked, or the caller is a user-space process which
* holds a reference on the inode by having an open file.
*
* In other cases, the page should be locked before running set_page_dirty().
*/
int set_page_dirty_lock(struct page *page)
{
int ret;
lock_page(page);
ret = set_page_dirty(page);
unlock_page(page);
return ret;
}
EXPORT_SYMBOL(set_page_dirty_lock);
/*
* This cancels just the dirty bit on the kernel page itself, it does NOT
* actually remove dirty bits on any mmap's that may be around. It also
* leaves the page tagged dirty, so any sync activity will still find it on
* the dirty lists, and in particular, clear_page_dirty_for_io() will still
* look at the dirty bits in the VM.
*
* Doing this should *normally* only ever be done when a page is truncated,
* and is not actually mapped anywhere at all. However, fs/buffer.c does
* this when it notices that somebody has cleaned out all the buffers on a
* page without actually doing it through the VM. Can you say "ext3 is
* horribly ugly"? Thought you could.
*/
void __folio_cancel_dirty(struct folio *folio)
{
struct address_space *mapping = folio_mapping(folio);
memcg: add per cgroup dirty page accounting When modifying PG_Dirty on cached file pages, update the new MEM_CGROUP_STAT_DIRTY counter. This is done in the same places where global NR_FILE_DIRTY is managed. The new memcg stat is visible in the per memcg memory.stat cgroupfs file. The most recent past attempt at this was http://thread.gmane.org/gmane.linux.kernel.cgroups/8632 The new accounting supports future efforts to add per cgroup dirty page throttling and writeback. It also helps an administrator break down a container's memory usage and provides evidence to understand memcg oom kills (the new dirty count is included in memcg oom kill messages). The ability to move page accounting between memcg (memory.move_charge_at_immigrate) makes this accounting more complicated than the global counter. The existing mem_cgroup_{begin,end}_page_stat() lock is used to serialize move accounting with stat updates. Typical update operation: memcg = mem_cgroup_begin_page_stat(page) if (TestSetPageDirty()) { [...] mem_cgroup_update_page_stat(memcg) } mem_cgroup_end_page_stat(memcg) Summary of mem_cgroup_end_page_stat() overhead: - Without CONFIG_MEMCG it's a no-op - With CONFIG_MEMCG and no inter memcg task movement, it's just rcu_read_lock() - With CONFIG_MEMCG and inter memcg task movement, it's rcu_read_lock() + spin_lock_irqsave() A memcg parameter is added to several routines because their callers now grab mem_cgroup_begin_page_stat() which returns the memcg later needed by for mem_cgroup_update_page_stat(). Because mem_cgroup_begin_page_stat() may disable interrupts, some adjustments are needed: - move __mark_inode_dirty() from __set_page_dirty() to its caller. __mark_inode_dirty() locking does not want interrupts disabled. - use spin_lock_irqsave(tree_lock) rather than spin_lock_irq() in __delete_from_page_cache(), replace_page_cache_page(), invalidate_complete_page2(), and __remove_mapping(). text data bss dec hex filename 8925147 1774832 1785856 12485835 be84cb vmlinux-!CONFIG_MEMCG-before 8925339 1774832 1785856 12486027 be858b vmlinux-!CONFIG_MEMCG-after +192 text bytes 8965977 1784992 1785856 12536825 bf4bf9 vmlinux-CONFIG_MEMCG-before 8966750 1784992 1785856 12537598 bf4efe vmlinux-CONFIG_MEMCG-after +773 text bytes Performance tests run on v4.0-rc1-36-g4f671fe2f952. Lower is better for all metrics, they're all wall clock or cycle counts. The read and write fault benchmarks just measure fault time, they do not include I/O time. * CONFIG_MEMCG not set: baseline patched kbuild 1m25.030000(+-0.088% 3 samples) 1m25.426667(+-0.120% 3 samples) dd write 100 MiB 0.859211561 +-15.10% 0.874162885 +-15.03% dd write 200 MiB 1.670653105 +-17.87% 1.669384764 +-11.99% dd write 1000 MiB 8.434691190 +-14.15% 8.474733215 +-14.77% read fault cycles 254.0(+-0.000% 10 samples) 253.0(+-0.000% 10 samples) write fault cycles 2021.2(+-3.070% 10 samples) 1984.5(+-1.036% 10 samples) * CONFIG_MEMCG=y root_memcg: baseline patched kbuild 1m25.716667(+-0.105% 3 samples) 1m25.686667(+-0.153% 3 samples) dd write 100 MiB 0.855650830 +-14.90% 0.887557919 +-14.90% dd write 200 MiB 1.688322953 +-12.72% 1.667682724 +-13.33% dd write 1000 MiB 8.418601605 +-14.30% 8.673532299 +-15.00% read fault cycles 266.0(+-0.000% 10 samples) 266.0(+-0.000% 10 samples) write fault cycles 2051.7(+-1.349% 10 samples) 2049.6(+-1.686% 10 samples) * CONFIG_MEMCG=y non-root_memcg: baseline patched kbuild 1m26.120000(+-0.273% 3 samples) 1m25.763333(+-0.127% 3 samples) dd write 100 MiB 0.861723964 +-15.25% 0.818129350 +-14.82% dd write 200 MiB 1.669887569 +-13.30% 1.698645885 +-13.27% dd write 1000 MiB 8.383191730 +-14.65% 8.351742280 +-14.52% read fault cycles 265.7(+-0.172% 10 samples) 267.0(+-0.000% 10 samples) write fault cycles 2070.6(+-1.512% 10 samples) 2084.4(+-2.148% 10 samples) As expected anon page faults are not affected by this patch. tj: Updated to apply on top of the recent cancel_dirty_page() changes. Signed-off-by: Sha Zhengju <handai.szj@gmail.com> Signed-off-by: Greg Thelen <gthelen@google.com> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:16 +00:00
if (mapping_can_writeback(mapping)) {
writeback: implement unlocked_inode_to_wb transaction and use it for stat updates The mechanism for detecting whether an inode should switch its wb (bdi_writeback) association is now in place. This patch build the framework for the actual switching. This patch adds a new inode flag I_WB_SWITCHING, which has two functions. First, the easy one, it ensures that there's only one switching in progress for a give inode. Second, it's used as a mechanism to synchronize wb stat updates. The two stats, WB_RECLAIMABLE and WB_WRITEBACK, aren't event counters but track the current number of dirty pages and pages under writeback respectively. As such, when an inode is moved from one wb to another, the inode's portion of those stats have to be transferred together; unfortunately, this is a bit tricky as those stat updates are percpu operations which are performed without holding any lock in some places. This patch solves the problem in a similar way as memcg. Each such lockless stat updates are wrapped in transaction surrounded by unlocked_inode_to_wb_begin/end(). During normal operation, they map to rcu_read_lock/unlock(); however, if I_WB_SWITCHING is asserted, mapping->tree_lock is grabbed across the transaction. In turn, the switching path sets I_WB_SWITCHING and waits for a RCU grace period to pass before actually starting to switch, which guarantees that all stat update paths are synchronizing against mapping->tree_lock. This patch still doesn't implement the actual switching. v3: Updated on top of the recent cancel_dirty_page() updates. unlocked_inode_to_wb_begin() now nests inside mem_cgroup_begin_page_stat() to match the locking order. v2: The i_wb access transaction will be used for !stat accesses too. Function names and comments updated accordingly. s/inode_wb_stat_unlocked_{begin|end}/unlocked_inode_to_wb_{begin|end}/ s/switch_wb/switch_wbs/ Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-28 18:50:53 +00:00
struct inode *inode = mapping->host;
struct bdi_writeback *wb;
writeback: safer lock nesting lock_page_memcg()/unlock_page_memcg() use spin_lock_irqsave/restore() if the page's memcg is undergoing move accounting, which occurs when a process leaves its memcg for a new one that has memory.move_charge_at_immigrate set. unlocked_inode_to_wb_begin,end() use spin_lock_irq/spin_unlock_irq() if the given inode is switching writeback domains. Switches occur when enough writes are issued from a new domain. This existing pattern is thus suspicious: lock_page_memcg(page); unlocked_inode_to_wb_begin(inode, &locked); ... unlocked_inode_to_wb_end(inode, locked); unlock_page_memcg(page); If both inode switch and process memcg migration are both in-flight then unlocked_inode_to_wb_end() will unconditionally enable interrupts while still holding the lock_page_memcg() irq spinlock. This suggests the possibility of deadlock if an interrupt occurs before unlock_page_memcg(). truncate __cancel_dirty_page lock_page_memcg unlocked_inode_to_wb_begin unlocked_inode_to_wb_end <interrupts mistakenly enabled> <interrupt> end_page_writeback test_clear_page_writeback lock_page_memcg <deadlock> unlock_page_memcg Due to configuration limitations this deadlock is not currently possible because we don't mix cgroup writeback (a cgroupv2 feature) and memory.move_charge_at_immigrate (a cgroupv1 feature). If the kernel is hacked to always claim inode switching and memcg moving_account, then this script triggers lockup in less than a minute: cd /mnt/cgroup/memory mkdir a b echo 1 > a/memory.move_charge_at_immigrate echo 1 > b/memory.move_charge_at_immigrate ( echo $BASHPID > a/cgroup.procs while true; do dd if=/dev/zero of=/mnt/big bs=1M count=256 done ) & while true; do sync done & sleep 1h & SLEEP=$! while true; do echo $SLEEP > a/cgroup.procs echo $SLEEP > b/cgroup.procs done The deadlock does not seem possible, so it's debatable if there's any reason to modify the kernel. I suggest we should to prevent future surprises. And Wang Long said "this deadlock occurs three times in our environment", so there's more reason to apply this, even to stable. Stable 4.4 has minor conflicts applying this patch. For a clean 4.4 patch see "[PATCH for-4.4] writeback: safer lock nesting" https://lkml.org/lkml/2018/4/11/146 Wang Long said "this deadlock occurs three times in our environment" [gthelen@google.com: v4] Link: http://lkml.kernel.org/r/20180411084653.254724-1-gthelen@google.com [akpm@linux-foundation.org: comment tweaks, struct initialization simplification] Change-Id: Ibb773e8045852978f6207074491d262f1b3fb613 Link: http://lkml.kernel.org/r/20180410005908.167976-1-gthelen@google.com Fixes: 682aa8e1a6a1 ("writeback: implement unlocked_inode_to_wb transaction and use it for stat updates") Signed-off-by: Greg Thelen <gthelen@google.com> Reported-by: Wang Long <wanglong19@meituan.com> Acked-by: Wang Long <wanglong19@meituan.com> Acked-by: Michal Hocko <mhocko@suse.com> Reviewed-by: Andrew Morton <akpm@linux-foundation.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Tejun Heo <tj@kernel.org> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: <stable@vger.kernel.org> [v4.2+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-04-20 21:55:42 +00:00
struct wb_lock_cookie cookie = {};
memcg: add per cgroup dirty page accounting When modifying PG_Dirty on cached file pages, update the new MEM_CGROUP_STAT_DIRTY counter. This is done in the same places where global NR_FILE_DIRTY is managed. The new memcg stat is visible in the per memcg memory.stat cgroupfs file. The most recent past attempt at this was http://thread.gmane.org/gmane.linux.kernel.cgroups/8632 The new accounting supports future efforts to add per cgroup dirty page throttling and writeback. It also helps an administrator break down a container's memory usage and provides evidence to understand memcg oom kills (the new dirty count is included in memcg oom kill messages). The ability to move page accounting between memcg (memory.move_charge_at_immigrate) makes this accounting more complicated than the global counter. The existing mem_cgroup_{begin,end}_page_stat() lock is used to serialize move accounting with stat updates. Typical update operation: memcg = mem_cgroup_begin_page_stat(page) if (TestSetPageDirty()) { [...] mem_cgroup_update_page_stat(memcg) } mem_cgroup_end_page_stat(memcg) Summary of mem_cgroup_end_page_stat() overhead: - Without CONFIG_MEMCG it's a no-op - With CONFIG_MEMCG and no inter memcg task movement, it's just rcu_read_lock() - With CONFIG_MEMCG and inter memcg task movement, it's rcu_read_lock() + spin_lock_irqsave() A memcg parameter is added to several routines because their callers now grab mem_cgroup_begin_page_stat() which returns the memcg later needed by for mem_cgroup_update_page_stat(). Because mem_cgroup_begin_page_stat() may disable interrupts, some adjustments are needed: - move __mark_inode_dirty() from __set_page_dirty() to its caller. __mark_inode_dirty() locking does not want interrupts disabled. - use spin_lock_irqsave(tree_lock) rather than spin_lock_irq() in __delete_from_page_cache(), replace_page_cache_page(), invalidate_complete_page2(), and __remove_mapping(). text data bss dec hex filename 8925147 1774832 1785856 12485835 be84cb vmlinux-!CONFIG_MEMCG-before 8925339 1774832 1785856 12486027 be858b vmlinux-!CONFIG_MEMCG-after +192 text bytes 8965977 1784992 1785856 12536825 bf4bf9 vmlinux-CONFIG_MEMCG-before 8966750 1784992 1785856 12537598 bf4efe vmlinux-CONFIG_MEMCG-after +773 text bytes Performance tests run on v4.0-rc1-36-g4f671fe2f952. Lower is better for all metrics, they're all wall clock or cycle counts. The read and write fault benchmarks just measure fault time, they do not include I/O time. * CONFIG_MEMCG not set: baseline patched kbuild 1m25.030000(+-0.088% 3 samples) 1m25.426667(+-0.120% 3 samples) dd write 100 MiB 0.859211561 +-15.10% 0.874162885 +-15.03% dd write 200 MiB 1.670653105 +-17.87% 1.669384764 +-11.99% dd write 1000 MiB 8.434691190 +-14.15% 8.474733215 +-14.77% read fault cycles 254.0(+-0.000% 10 samples) 253.0(+-0.000% 10 samples) write fault cycles 2021.2(+-3.070% 10 samples) 1984.5(+-1.036% 10 samples) * CONFIG_MEMCG=y root_memcg: baseline patched kbuild 1m25.716667(+-0.105% 3 samples) 1m25.686667(+-0.153% 3 samples) dd write 100 MiB 0.855650830 +-14.90% 0.887557919 +-14.90% dd write 200 MiB 1.688322953 +-12.72% 1.667682724 +-13.33% dd write 1000 MiB 8.418601605 +-14.30% 8.673532299 +-15.00% read fault cycles 266.0(+-0.000% 10 samples) 266.0(+-0.000% 10 samples) write fault cycles 2051.7(+-1.349% 10 samples) 2049.6(+-1.686% 10 samples) * CONFIG_MEMCG=y non-root_memcg: baseline patched kbuild 1m26.120000(+-0.273% 3 samples) 1m25.763333(+-0.127% 3 samples) dd write 100 MiB 0.861723964 +-15.25% 0.818129350 +-14.82% dd write 200 MiB 1.669887569 +-13.30% 1.698645885 +-13.27% dd write 1000 MiB 8.383191730 +-14.65% 8.351742280 +-14.52% read fault cycles 265.7(+-0.172% 10 samples) 267.0(+-0.000% 10 samples) write fault cycles 2070.6(+-1.512% 10 samples) 2084.4(+-2.148% 10 samples) As expected anon page faults are not affected by this patch. tj: Updated to apply on top of the recent cancel_dirty_page() changes. Signed-off-by: Sha Zhengju <handai.szj@gmail.com> Signed-off-by: Greg Thelen <gthelen@google.com> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:16 +00:00
folio_memcg_lock(folio);
writeback: safer lock nesting lock_page_memcg()/unlock_page_memcg() use spin_lock_irqsave/restore() if the page's memcg is undergoing move accounting, which occurs when a process leaves its memcg for a new one that has memory.move_charge_at_immigrate set. unlocked_inode_to_wb_begin,end() use spin_lock_irq/spin_unlock_irq() if the given inode is switching writeback domains. Switches occur when enough writes are issued from a new domain. This existing pattern is thus suspicious: lock_page_memcg(page); unlocked_inode_to_wb_begin(inode, &locked); ... unlocked_inode_to_wb_end(inode, locked); unlock_page_memcg(page); If both inode switch and process memcg migration are both in-flight then unlocked_inode_to_wb_end() will unconditionally enable interrupts while still holding the lock_page_memcg() irq spinlock. This suggests the possibility of deadlock if an interrupt occurs before unlock_page_memcg(). truncate __cancel_dirty_page lock_page_memcg unlocked_inode_to_wb_begin unlocked_inode_to_wb_end <interrupts mistakenly enabled> <interrupt> end_page_writeback test_clear_page_writeback lock_page_memcg <deadlock> unlock_page_memcg Due to configuration limitations this deadlock is not currently possible because we don't mix cgroup writeback (a cgroupv2 feature) and memory.move_charge_at_immigrate (a cgroupv1 feature). If the kernel is hacked to always claim inode switching and memcg moving_account, then this script triggers lockup in less than a minute: cd /mnt/cgroup/memory mkdir a b echo 1 > a/memory.move_charge_at_immigrate echo 1 > b/memory.move_charge_at_immigrate ( echo $BASHPID > a/cgroup.procs while true; do dd if=/dev/zero of=/mnt/big bs=1M count=256 done ) & while true; do sync done & sleep 1h & SLEEP=$! while true; do echo $SLEEP > a/cgroup.procs echo $SLEEP > b/cgroup.procs done The deadlock does not seem possible, so it's debatable if there's any reason to modify the kernel. I suggest we should to prevent future surprises. And Wang Long said "this deadlock occurs three times in our environment", so there's more reason to apply this, even to stable. Stable 4.4 has minor conflicts applying this patch. For a clean 4.4 patch see "[PATCH for-4.4] writeback: safer lock nesting" https://lkml.org/lkml/2018/4/11/146 Wang Long said "this deadlock occurs three times in our environment" [gthelen@google.com: v4] Link: http://lkml.kernel.org/r/20180411084653.254724-1-gthelen@google.com [akpm@linux-foundation.org: comment tweaks, struct initialization simplification] Change-Id: Ibb773e8045852978f6207074491d262f1b3fb613 Link: http://lkml.kernel.org/r/20180410005908.167976-1-gthelen@google.com Fixes: 682aa8e1a6a1 ("writeback: implement unlocked_inode_to_wb transaction and use it for stat updates") Signed-off-by: Greg Thelen <gthelen@google.com> Reported-by: Wang Long <wanglong19@meituan.com> Acked-by: Wang Long <wanglong19@meituan.com> Acked-by: Michal Hocko <mhocko@suse.com> Reviewed-by: Andrew Morton <akpm@linux-foundation.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Tejun Heo <tj@kernel.org> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: <stable@vger.kernel.org> [v4.2+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-04-20 21:55:42 +00:00
wb = unlocked_inode_to_wb_begin(inode, &cookie);
memcg: add per cgroup dirty page accounting When modifying PG_Dirty on cached file pages, update the new MEM_CGROUP_STAT_DIRTY counter. This is done in the same places where global NR_FILE_DIRTY is managed. The new memcg stat is visible in the per memcg memory.stat cgroupfs file. The most recent past attempt at this was http://thread.gmane.org/gmane.linux.kernel.cgroups/8632 The new accounting supports future efforts to add per cgroup dirty page throttling and writeback. It also helps an administrator break down a container's memory usage and provides evidence to understand memcg oom kills (the new dirty count is included in memcg oom kill messages). The ability to move page accounting between memcg (memory.move_charge_at_immigrate) makes this accounting more complicated than the global counter. The existing mem_cgroup_{begin,end}_page_stat() lock is used to serialize move accounting with stat updates. Typical update operation: memcg = mem_cgroup_begin_page_stat(page) if (TestSetPageDirty()) { [...] mem_cgroup_update_page_stat(memcg) } mem_cgroup_end_page_stat(memcg) Summary of mem_cgroup_end_page_stat() overhead: - Without CONFIG_MEMCG it's a no-op - With CONFIG_MEMCG and no inter memcg task movement, it's just rcu_read_lock() - With CONFIG_MEMCG and inter memcg task movement, it's rcu_read_lock() + spin_lock_irqsave() A memcg parameter is added to several routines because their callers now grab mem_cgroup_begin_page_stat() which returns the memcg later needed by for mem_cgroup_update_page_stat(). Because mem_cgroup_begin_page_stat() may disable interrupts, some adjustments are needed: - move __mark_inode_dirty() from __set_page_dirty() to its caller. __mark_inode_dirty() locking does not want interrupts disabled. - use spin_lock_irqsave(tree_lock) rather than spin_lock_irq() in __delete_from_page_cache(), replace_page_cache_page(), invalidate_complete_page2(), and __remove_mapping(). text data bss dec hex filename 8925147 1774832 1785856 12485835 be84cb vmlinux-!CONFIG_MEMCG-before 8925339 1774832 1785856 12486027 be858b vmlinux-!CONFIG_MEMCG-after +192 text bytes 8965977 1784992 1785856 12536825 bf4bf9 vmlinux-CONFIG_MEMCG-before 8966750 1784992 1785856 12537598 bf4efe vmlinux-CONFIG_MEMCG-after +773 text bytes Performance tests run on v4.0-rc1-36-g4f671fe2f952. Lower is better for all metrics, they're all wall clock or cycle counts. The read and write fault benchmarks just measure fault time, they do not include I/O time. * CONFIG_MEMCG not set: baseline patched kbuild 1m25.030000(+-0.088% 3 samples) 1m25.426667(+-0.120% 3 samples) dd write 100 MiB 0.859211561 +-15.10% 0.874162885 +-15.03% dd write 200 MiB 1.670653105 +-17.87% 1.669384764 +-11.99% dd write 1000 MiB 8.434691190 +-14.15% 8.474733215 +-14.77% read fault cycles 254.0(+-0.000% 10 samples) 253.0(+-0.000% 10 samples) write fault cycles 2021.2(+-3.070% 10 samples) 1984.5(+-1.036% 10 samples) * CONFIG_MEMCG=y root_memcg: baseline patched kbuild 1m25.716667(+-0.105% 3 samples) 1m25.686667(+-0.153% 3 samples) dd write 100 MiB 0.855650830 +-14.90% 0.887557919 +-14.90% dd write 200 MiB 1.688322953 +-12.72% 1.667682724 +-13.33% dd write 1000 MiB 8.418601605 +-14.30% 8.673532299 +-15.00% read fault cycles 266.0(+-0.000% 10 samples) 266.0(+-0.000% 10 samples) write fault cycles 2051.7(+-1.349% 10 samples) 2049.6(+-1.686% 10 samples) * CONFIG_MEMCG=y non-root_memcg: baseline patched kbuild 1m26.120000(+-0.273% 3 samples) 1m25.763333(+-0.127% 3 samples) dd write 100 MiB 0.861723964 +-15.25% 0.818129350 +-14.82% dd write 200 MiB 1.669887569 +-13.30% 1.698645885 +-13.27% dd write 1000 MiB 8.383191730 +-14.65% 8.351742280 +-14.52% read fault cycles 265.7(+-0.172% 10 samples) 267.0(+-0.000% 10 samples) write fault cycles 2070.6(+-1.512% 10 samples) 2084.4(+-2.148% 10 samples) As expected anon page faults are not affected by this patch. tj: Updated to apply on top of the recent cancel_dirty_page() changes. Signed-off-by: Sha Zhengju <handai.szj@gmail.com> Signed-off-by: Greg Thelen <gthelen@google.com> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:16 +00:00
if (folio_test_clear_dirty(folio))
folio_account_cleaned(folio, wb);
memcg: add per cgroup dirty page accounting When modifying PG_Dirty on cached file pages, update the new MEM_CGROUP_STAT_DIRTY counter. This is done in the same places where global NR_FILE_DIRTY is managed. The new memcg stat is visible in the per memcg memory.stat cgroupfs file. The most recent past attempt at this was http://thread.gmane.org/gmane.linux.kernel.cgroups/8632 The new accounting supports future efforts to add per cgroup dirty page throttling and writeback. It also helps an administrator break down a container's memory usage and provides evidence to understand memcg oom kills (the new dirty count is included in memcg oom kill messages). The ability to move page accounting between memcg (memory.move_charge_at_immigrate) makes this accounting more complicated than the global counter. The existing mem_cgroup_{begin,end}_page_stat() lock is used to serialize move accounting with stat updates. Typical update operation: memcg = mem_cgroup_begin_page_stat(page) if (TestSetPageDirty()) { [...] mem_cgroup_update_page_stat(memcg) } mem_cgroup_end_page_stat(memcg) Summary of mem_cgroup_end_page_stat() overhead: - Without CONFIG_MEMCG it's a no-op - With CONFIG_MEMCG and no inter memcg task movement, it's just rcu_read_lock() - With CONFIG_MEMCG and inter memcg task movement, it's rcu_read_lock() + spin_lock_irqsave() A memcg parameter is added to several routines because their callers now grab mem_cgroup_begin_page_stat() which returns the memcg later needed by for mem_cgroup_update_page_stat(). Because mem_cgroup_begin_page_stat() may disable interrupts, some adjustments are needed: - move __mark_inode_dirty() from __set_page_dirty() to its caller. __mark_inode_dirty() locking does not want interrupts disabled. - use spin_lock_irqsave(tree_lock) rather than spin_lock_irq() in __delete_from_page_cache(), replace_page_cache_page(), invalidate_complete_page2(), and __remove_mapping(). text data bss dec hex filename 8925147 1774832 1785856 12485835 be84cb vmlinux-!CONFIG_MEMCG-before 8925339 1774832 1785856 12486027 be858b vmlinux-!CONFIG_MEMCG-after +192 text bytes 8965977 1784992 1785856 12536825 bf4bf9 vmlinux-CONFIG_MEMCG-before 8966750 1784992 1785856 12537598 bf4efe vmlinux-CONFIG_MEMCG-after +773 text bytes Performance tests run on v4.0-rc1-36-g4f671fe2f952. Lower is better for all metrics, they're all wall clock or cycle counts. The read and write fault benchmarks just measure fault time, they do not include I/O time. * CONFIG_MEMCG not set: baseline patched kbuild 1m25.030000(+-0.088% 3 samples) 1m25.426667(+-0.120% 3 samples) dd write 100 MiB 0.859211561 +-15.10% 0.874162885 +-15.03% dd write 200 MiB 1.670653105 +-17.87% 1.669384764 +-11.99% dd write 1000 MiB 8.434691190 +-14.15% 8.474733215 +-14.77% read fault cycles 254.0(+-0.000% 10 samples) 253.0(+-0.000% 10 samples) write fault cycles 2021.2(+-3.070% 10 samples) 1984.5(+-1.036% 10 samples) * CONFIG_MEMCG=y root_memcg: baseline patched kbuild 1m25.716667(+-0.105% 3 samples) 1m25.686667(+-0.153% 3 samples) dd write 100 MiB 0.855650830 +-14.90% 0.887557919 +-14.90% dd write 200 MiB 1.688322953 +-12.72% 1.667682724 +-13.33% dd write 1000 MiB 8.418601605 +-14.30% 8.673532299 +-15.00% read fault cycles 266.0(+-0.000% 10 samples) 266.0(+-0.000% 10 samples) write fault cycles 2051.7(+-1.349% 10 samples) 2049.6(+-1.686% 10 samples) * CONFIG_MEMCG=y non-root_memcg: baseline patched kbuild 1m26.120000(+-0.273% 3 samples) 1m25.763333(+-0.127% 3 samples) dd write 100 MiB 0.861723964 +-15.25% 0.818129350 +-14.82% dd write 200 MiB 1.669887569 +-13.30% 1.698645885 +-13.27% dd write 1000 MiB 8.383191730 +-14.65% 8.351742280 +-14.52% read fault cycles 265.7(+-0.172% 10 samples) 267.0(+-0.000% 10 samples) write fault cycles 2070.6(+-1.512% 10 samples) 2084.4(+-2.148% 10 samples) As expected anon page faults are not affected by this patch. tj: Updated to apply on top of the recent cancel_dirty_page() changes. Signed-off-by: Sha Zhengju <handai.szj@gmail.com> Signed-off-by: Greg Thelen <gthelen@google.com> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:16 +00:00
writeback: safer lock nesting lock_page_memcg()/unlock_page_memcg() use spin_lock_irqsave/restore() if the page's memcg is undergoing move accounting, which occurs when a process leaves its memcg for a new one that has memory.move_charge_at_immigrate set. unlocked_inode_to_wb_begin,end() use spin_lock_irq/spin_unlock_irq() if the given inode is switching writeback domains. Switches occur when enough writes are issued from a new domain. This existing pattern is thus suspicious: lock_page_memcg(page); unlocked_inode_to_wb_begin(inode, &locked); ... unlocked_inode_to_wb_end(inode, locked); unlock_page_memcg(page); If both inode switch and process memcg migration are both in-flight then unlocked_inode_to_wb_end() will unconditionally enable interrupts while still holding the lock_page_memcg() irq spinlock. This suggests the possibility of deadlock if an interrupt occurs before unlock_page_memcg(). truncate __cancel_dirty_page lock_page_memcg unlocked_inode_to_wb_begin unlocked_inode_to_wb_end <interrupts mistakenly enabled> <interrupt> end_page_writeback test_clear_page_writeback lock_page_memcg <deadlock> unlock_page_memcg Due to configuration limitations this deadlock is not currently possible because we don't mix cgroup writeback (a cgroupv2 feature) and memory.move_charge_at_immigrate (a cgroupv1 feature). If the kernel is hacked to always claim inode switching and memcg moving_account, then this script triggers lockup in less than a minute: cd /mnt/cgroup/memory mkdir a b echo 1 > a/memory.move_charge_at_immigrate echo 1 > b/memory.move_charge_at_immigrate ( echo $BASHPID > a/cgroup.procs while true; do dd if=/dev/zero of=/mnt/big bs=1M count=256 done ) & while true; do sync done & sleep 1h & SLEEP=$! while true; do echo $SLEEP > a/cgroup.procs echo $SLEEP > b/cgroup.procs done The deadlock does not seem possible, so it's debatable if there's any reason to modify the kernel. I suggest we should to prevent future surprises. And Wang Long said "this deadlock occurs three times in our environment", so there's more reason to apply this, even to stable. Stable 4.4 has minor conflicts applying this patch. For a clean 4.4 patch see "[PATCH for-4.4] writeback: safer lock nesting" https://lkml.org/lkml/2018/4/11/146 Wang Long said "this deadlock occurs three times in our environment" [gthelen@google.com: v4] Link: http://lkml.kernel.org/r/20180411084653.254724-1-gthelen@google.com [akpm@linux-foundation.org: comment tweaks, struct initialization simplification] Change-Id: Ibb773e8045852978f6207074491d262f1b3fb613 Link: http://lkml.kernel.org/r/20180410005908.167976-1-gthelen@google.com Fixes: 682aa8e1a6a1 ("writeback: implement unlocked_inode_to_wb transaction and use it for stat updates") Signed-off-by: Greg Thelen <gthelen@google.com> Reported-by: Wang Long <wanglong19@meituan.com> Acked-by: Wang Long <wanglong19@meituan.com> Acked-by: Michal Hocko <mhocko@suse.com> Reviewed-by: Andrew Morton <akpm@linux-foundation.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Tejun Heo <tj@kernel.org> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: <stable@vger.kernel.org> [v4.2+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-04-20 21:55:42 +00:00
unlocked_inode_to_wb_end(inode, &cookie);
folio_memcg_unlock(folio);
memcg: add per cgroup dirty page accounting When modifying PG_Dirty on cached file pages, update the new MEM_CGROUP_STAT_DIRTY counter. This is done in the same places where global NR_FILE_DIRTY is managed. The new memcg stat is visible in the per memcg memory.stat cgroupfs file. The most recent past attempt at this was http://thread.gmane.org/gmane.linux.kernel.cgroups/8632 The new accounting supports future efforts to add per cgroup dirty page throttling and writeback. It also helps an administrator break down a container's memory usage and provides evidence to understand memcg oom kills (the new dirty count is included in memcg oom kill messages). The ability to move page accounting between memcg (memory.move_charge_at_immigrate) makes this accounting more complicated than the global counter. The existing mem_cgroup_{begin,end}_page_stat() lock is used to serialize move accounting with stat updates. Typical update operation: memcg = mem_cgroup_begin_page_stat(page) if (TestSetPageDirty()) { [...] mem_cgroup_update_page_stat(memcg) } mem_cgroup_end_page_stat(memcg) Summary of mem_cgroup_end_page_stat() overhead: - Without CONFIG_MEMCG it's a no-op - With CONFIG_MEMCG and no inter memcg task movement, it's just rcu_read_lock() - With CONFIG_MEMCG and inter memcg task movement, it's rcu_read_lock() + spin_lock_irqsave() A memcg parameter is added to several routines because their callers now grab mem_cgroup_begin_page_stat() which returns the memcg later needed by for mem_cgroup_update_page_stat(). Because mem_cgroup_begin_page_stat() may disable interrupts, some adjustments are needed: - move __mark_inode_dirty() from __set_page_dirty() to its caller. __mark_inode_dirty() locking does not want interrupts disabled. - use spin_lock_irqsave(tree_lock) rather than spin_lock_irq() in __delete_from_page_cache(), replace_page_cache_page(), invalidate_complete_page2(), and __remove_mapping(). text data bss dec hex filename 8925147 1774832 1785856 12485835 be84cb vmlinux-!CONFIG_MEMCG-before 8925339 1774832 1785856 12486027 be858b vmlinux-!CONFIG_MEMCG-after +192 text bytes 8965977 1784992 1785856 12536825 bf4bf9 vmlinux-CONFIG_MEMCG-before 8966750 1784992 1785856 12537598 bf4efe vmlinux-CONFIG_MEMCG-after +773 text bytes Performance tests run on v4.0-rc1-36-g4f671fe2f952. Lower is better for all metrics, they're all wall clock or cycle counts. The read and write fault benchmarks just measure fault time, they do not include I/O time. * CONFIG_MEMCG not set: baseline patched kbuild 1m25.030000(+-0.088% 3 samples) 1m25.426667(+-0.120% 3 samples) dd write 100 MiB 0.859211561 +-15.10% 0.874162885 +-15.03% dd write 200 MiB 1.670653105 +-17.87% 1.669384764 +-11.99% dd write 1000 MiB 8.434691190 +-14.15% 8.474733215 +-14.77% read fault cycles 254.0(+-0.000% 10 samples) 253.0(+-0.000% 10 samples) write fault cycles 2021.2(+-3.070% 10 samples) 1984.5(+-1.036% 10 samples) * CONFIG_MEMCG=y root_memcg: baseline patched kbuild 1m25.716667(+-0.105% 3 samples) 1m25.686667(+-0.153% 3 samples) dd write 100 MiB 0.855650830 +-14.90% 0.887557919 +-14.90% dd write 200 MiB 1.688322953 +-12.72% 1.667682724 +-13.33% dd write 1000 MiB 8.418601605 +-14.30% 8.673532299 +-15.00% read fault cycles 266.0(+-0.000% 10 samples) 266.0(+-0.000% 10 samples) write fault cycles 2051.7(+-1.349% 10 samples) 2049.6(+-1.686% 10 samples) * CONFIG_MEMCG=y non-root_memcg: baseline patched kbuild 1m26.120000(+-0.273% 3 samples) 1m25.763333(+-0.127% 3 samples) dd write 100 MiB 0.861723964 +-15.25% 0.818129350 +-14.82% dd write 200 MiB 1.669887569 +-13.30% 1.698645885 +-13.27% dd write 1000 MiB 8.383191730 +-14.65% 8.351742280 +-14.52% read fault cycles 265.7(+-0.172% 10 samples) 267.0(+-0.000% 10 samples) write fault cycles 2070.6(+-1.512% 10 samples) 2084.4(+-2.148% 10 samples) As expected anon page faults are not affected by this patch. tj: Updated to apply on top of the recent cancel_dirty_page() changes. Signed-off-by: Sha Zhengju <handai.szj@gmail.com> Signed-off-by: Greg Thelen <gthelen@google.com> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:16 +00:00
} else {
folio_clear_dirty(folio);
memcg: add per cgroup dirty page accounting When modifying PG_Dirty on cached file pages, update the new MEM_CGROUP_STAT_DIRTY counter. This is done in the same places where global NR_FILE_DIRTY is managed. The new memcg stat is visible in the per memcg memory.stat cgroupfs file. The most recent past attempt at this was http://thread.gmane.org/gmane.linux.kernel.cgroups/8632 The new accounting supports future efforts to add per cgroup dirty page throttling and writeback. It also helps an administrator break down a container's memory usage and provides evidence to understand memcg oom kills (the new dirty count is included in memcg oom kill messages). The ability to move page accounting between memcg (memory.move_charge_at_immigrate) makes this accounting more complicated than the global counter. The existing mem_cgroup_{begin,end}_page_stat() lock is used to serialize move accounting with stat updates. Typical update operation: memcg = mem_cgroup_begin_page_stat(page) if (TestSetPageDirty()) { [...] mem_cgroup_update_page_stat(memcg) } mem_cgroup_end_page_stat(memcg) Summary of mem_cgroup_end_page_stat() overhead: - Without CONFIG_MEMCG it's a no-op - With CONFIG_MEMCG and no inter memcg task movement, it's just rcu_read_lock() - With CONFIG_MEMCG and inter memcg task movement, it's rcu_read_lock() + spin_lock_irqsave() A memcg parameter is added to several routines because their callers now grab mem_cgroup_begin_page_stat() which returns the memcg later needed by for mem_cgroup_update_page_stat(). Because mem_cgroup_begin_page_stat() may disable interrupts, some adjustments are needed: - move __mark_inode_dirty() from __set_page_dirty() to its caller. __mark_inode_dirty() locking does not want interrupts disabled. - use spin_lock_irqsave(tree_lock) rather than spin_lock_irq() in __delete_from_page_cache(), replace_page_cache_page(), invalidate_complete_page2(), and __remove_mapping(). text data bss dec hex filename 8925147 1774832 1785856 12485835 be84cb vmlinux-!CONFIG_MEMCG-before 8925339 1774832 1785856 12486027 be858b vmlinux-!CONFIG_MEMCG-after +192 text bytes 8965977 1784992 1785856 12536825 bf4bf9 vmlinux-CONFIG_MEMCG-before 8966750 1784992 1785856 12537598 bf4efe vmlinux-CONFIG_MEMCG-after +773 text bytes Performance tests run on v4.0-rc1-36-g4f671fe2f952. Lower is better for all metrics, they're all wall clock or cycle counts. The read and write fault benchmarks just measure fault time, they do not include I/O time. * CONFIG_MEMCG not set: baseline patched kbuild 1m25.030000(+-0.088% 3 samples) 1m25.426667(+-0.120% 3 samples) dd write 100 MiB 0.859211561 +-15.10% 0.874162885 +-15.03% dd write 200 MiB 1.670653105 +-17.87% 1.669384764 +-11.99% dd write 1000 MiB 8.434691190 +-14.15% 8.474733215 +-14.77% read fault cycles 254.0(+-0.000% 10 samples) 253.0(+-0.000% 10 samples) write fault cycles 2021.2(+-3.070% 10 samples) 1984.5(+-1.036% 10 samples) * CONFIG_MEMCG=y root_memcg: baseline patched kbuild 1m25.716667(+-0.105% 3 samples) 1m25.686667(+-0.153% 3 samples) dd write 100 MiB 0.855650830 +-14.90% 0.887557919 +-14.90% dd write 200 MiB 1.688322953 +-12.72% 1.667682724 +-13.33% dd write 1000 MiB 8.418601605 +-14.30% 8.673532299 +-15.00% read fault cycles 266.0(+-0.000% 10 samples) 266.0(+-0.000% 10 samples) write fault cycles 2051.7(+-1.349% 10 samples) 2049.6(+-1.686% 10 samples) * CONFIG_MEMCG=y non-root_memcg: baseline patched kbuild 1m26.120000(+-0.273% 3 samples) 1m25.763333(+-0.127% 3 samples) dd write 100 MiB 0.861723964 +-15.25% 0.818129350 +-14.82% dd write 200 MiB 1.669887569 +-13.30% 1.698645885 +-13.27% dd write 1000 MiB 8.383191730 +-14.65% 8.351742280 +-14.52% read fault cycles 265.7(+-0.172% 10 samples) 267.0(+-0.000% 10 samples) write fault cycles 2070.6(+-1.512% 10 samples) 2084.4(+-2.148% 10 samples) As expected anon page faults are not affected by this patch. tj: Updated to apply on top of the recent cancel_dirty_page() changes. Signed-off-by: Sha Zhengju <handai.szj@gmail.com> Signed-off-by: Greg Thelen <gthelen@google.com> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:16 +00:00
}
}
EXPORT_SYMBOL(__folio_cancel_dirty);
/*
* Clear a folio's dirty flag, while caring for dirty memory accounting.
* Returns true if the folio was previously dirty.
*
* This is for preparing to put the folio under writeout. We leave
* the folio tagged as dirty in the xarray so that a concurrent
* write-for-sync can discover it via a PAGECACHE_TAG_DIRTY walk.
* The ->writepage implementation will run either folio_start_writeback()
* or folio_mark_dirty(), at which stage we bring the folio's dirty flag
* and xarray dirty tag back into sync.
*
* This incoherency between the folio's dirty flag and xarray tag is
* unfortunate, but it only exists while the folio is locked.
*/
bool folio_clear_dirty_for_io(struct folio *folio)
{
struct address_space *mapping = folio_mapping(folio);
bool ret = false;
VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
if (mapping && mapping_can_writeback(mapping)) {
writeback: implement unlocked_inode_to_wb transaction and use it for stat updates The mechanism for detecting whether an inode should switch its wb (bdi_writeback) association is now in place. This patch build the framework for the actual switching. This patch adds a new inode flag I_WB_SWITCHING, which has two functions. First, the easy one, it ensures that there's only one switching in progress for a give inode. Second, it's used as a mechanism to synchronize wb stat updates. The two stats, WB_RECLAIMABLE and WB_WRITEBACK, aren't event counters but track the current number of dirty pages and pages under writeback respectively. As such, when an inode is moved from one wb to another, the inode's portion of those stats have to be transferred together; unfortunately, this is a bit tricky as those stat updates are percpu operations which are performed without holding any lock in some places. This patch solves the problem in a similar way as memcg. Each such lockless stat updates are wrapped in transaction surrounded by unlocked_inode_to_wb_begin/end(). During normal operation, they map to rcu_read_lock/unlock(); however, if I_WB_SWITCHING is asserted, mapping->tree_lock is grabbed across the transaction. In turn, the switching path sets I_WB_SWITCHING and waits for a RCU grace period to pass before actually starting to switch, which guarantees that all stat update paths are synchronizing against mapping->tree_lock. This patch still doesn't implement the actual switching. v3: Updated on top of the recent cancel_dirty_page() updates. unlocked_inode_to_wb_begin() now nests inside mem_cgroup_begin_page_stat() to match the locking order. v2: The i_wb access transaction will be used for !stat accesses too. Function names and comments updated accordingly. s/inode_wb_stat_unlocked_{begin|end}/unlocked_inode_to_wb_{begin|end}/ s/switch_wb/switch_wbs/ Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-28 18:50:53 +00:00
struct inode *inode = mapping->host;
struct bdi_writeback *wb;
writeback: safer lock nesting lock_page_memcg()/unlock_page_memcg() use spin_lock_irqsave/restore() if the page's memcg is undergoing move accounting, which occurs when a process leaves its memcg for a new one that has memory.move_charge_at_immigrate set. unlocked_inode_to_wb_begin,end() use spin_lock_irq/spin_unlock_irq() if the given inode is switching writeback domains. Switches occur when enough writes are issued from a new domain. This existing pattern is thus suspicious: lock_page_memcg(page); unlocked_inode_to_wb_begin(inode, &locked); ... unlocked_inode_to_wb_end(inode, locked); unlock_page_memcg(page); If both inode switch and process memcg migration are both in-flight then unlocked_inode_to_wb_end() will unconditionally enable interrupts while still holding the lock_page_memcg() irq spinlock. This suggests the possibility of deadlock if an interrupt occurs before unlock_page_memcg(). truncate __cancel_dirty_page lock_page_memcg unlocked_inode_to_wb_begin unlocked_inode_to_wb_end <interrupts mistakenly enabled> <interrupt> end_page_writeback test_clear_page_writeback lock_page_memcg <deadlock> unlock_page_memcg Due to configuration limitations this deadlock is not currently possible because we don't mix cgroup writeback (a cgroupv2 feature) and memory.move_charge_at_immigrate (a cgroupv1 feature). If the kernel is hacked to always claim inode switching and memcg moving_account, then this script triggers lockup in less than a minute: cd /mnt/cgroup/memory mkdir a b echo 1 > a/memory.move_charge_at_immigrate echo 1 > b/memory.move_charge_at_immigrate ( echo $BASHPID > a/cgroup.procs while true; do dd if=/dev/zero of=/mnt/big bs=1M count=256 done ) & while true; do sync done & sleep 1h & SLEEP=$! while true; do echo $SLEEP > a/cgroup.procs echo $SLEEP > b/cgroup.procs done The deadlock does not seem possible, so it's debatable if there's any reason to modify the kernel. I suggest we should to prevent future surprises. And Wang Long said "this deadlock occurs three times in our environment", so there's more reason to apply this, even to stable. Stable 4.4 has minor conflicts applying this patch. For a clean 4.4 patch see "[PATCH for-4.4] writeback: safer lock nesting" https://lkml.org/lkml/2018/4/11/146 Wang Long said "this deadlock occurs three times in our environment" [gthelen@google.com: v4] Link: http://lkml.kernel.org/r/20180411084653.254724-1-gthelen@google.com [akpm@linux-foundation.org: comment tweaks, struct initialization simplification] Change-Id: Ibb773e8045852978f6207074491d262f1b3fb613 Link: http://lkml.kernel.org/r/20180410005908.167976-1-gthelen@google.com Fixes: 682aa8e1a6a1 ("writeback: implement unlocked_inode_to_wb transaction and use it for stat updates") Signed-off-by: Greg Thelen <gthelen@google.com> Reported-by: Wang Long <wanglong19@meituan.com> Acked-by: Wang Long <wanglong19@meituan.com> Acked-by: Michal Hocko <mhocko@suse.com> Reviewed-by: Andrew Morton <akpm@linux-foundation.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Tejun Heo <tj@kernel.org> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: <stable@vger.kernel.org> [v4.2+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-04-20 21:55:42 +00:00
struct wb_lock_cookie cookie = {};
writeback: implement unlocked_inode_to_wb transaction and use it for stat updates The mechanism for detecting whether an inode should switch its wb (bdi_writeback) association is now in place. This patch build the framework for the actual switching. This patch adds a new inode flag I_WB_SWITCHING, which has two functions. First, the easy one, it ensures that there's only one switching in progress for a give inode. Second, it's used as a mechanism to synchronize wb stat updates. The two stats, WB_RECLAIMABLE and WB_WRITEBACK, aren't event counters but track the current number of dirty pages and pages under writeback respectively. As such, when an inode is moved from one wb to another, the inode's portion of those stats have to be transferred together; unfortunately, this is a bit tricky as those stat updates are percpu operations which are performed without holding any lock in some places. This patch solves the problem in a similar way as memcg. Each such lockless stat updates are wrapped in transaction surrounded by unlocked_inode_to_wb_begin/end(). During normal operation, they map to rcu_read_lock/unlock(); however, if I_WB_SWITCHING is asserted, mapping->tree_lock is grabbed across the transaction. In turn, the switching path sets I_WB_SWITCHING and waits for a RCU grace period to pass before actually starting to switch, which guarantees that all stat update paths are synchronizing against mapping->tree_lock. This patch still doesn't implement the actual switching. v3: Updated on top of the recent cancel_dirty_page() updates. unlocked_inode_to_wb_begin() now nests inside mem_cgroup_begin_page_stat() to match the locking order. v2: The i_wb access transaction will be used for !stat accesses too. Function names and comments updated accordingly. s/inode_wb_stat_unlocked_{begin|end}/unlocked_inode_to_wb_{begin|end}/ s/switch_wb/switch_wbs/ Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Jan Kara <jack@suse.cz> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Greg Thelen <gthelen@google.com> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-28 18:50:53 +00:00
VM: Fix nasty and subtle race in shared mmap'ed page writeback The VM layer (on the face of it, fairly reasonably) expected that when it does a ->writepage() call to the filesystem, it would write out the full page at that point in time. Especially since it had earlier marked the whole page dirty with "set_page_dirty()". But that isn't actually the case: ->writepage() does not actually write a page, it writes the parts of the page that have been explicitly marked dirty before, *and* that had not got written out for other reasons since the last time we told it they were dirty. That last caveat is the important one. Which _most_ of the time ends up being the whole page (since we had called "set_page_dirty()" on the page earlier), but if the filesystem had done any dirty flushing of its own (for example, to honor some internal write ordering guarantees), it might end up doing only a partial page IO (or none at all) when ->writepage() is actually called. That is the correct thing in general (since we actually often _want_ only the known-dirty parts of the page to be written out), but the shared dirty page handling had implicitly forgotten about these details, and had a number of cases where it was doing just the "->writepage()" part, without telling the low-level filesystem that the whole page might have been re-dirtied as part of being mapped writably into user space. Since most of the time the FS did actually write out the full page, we didn't notice this for a loong time, and this needed some really odd patterns to trigger. But it caused occasional corruption with rtorrent and with the Debian "apt" database, because both use shared mmaps to update the end result. This fixes it. Finally. After way too much hair-pulling. Acked-by: Nick Piggin <nickpiggin@yahoo.com.au> Acked-by: Martin J. Bligh <mbligh@google.com> Acked-by: Martin Michlmayr <tbm@cyrius.com> Acked-by: Martin Johansson <martin@fatbob.nu> Acked-by: Ingo Molnar <mingo@elte.hu> Acked-by: Andrei Popa <andrei.popa@i-neo.ro> Cc: High Dickins <hugh@veritas.com> Cc: Andrew Morton <akpm@osdl.org>, Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Segher Boessenkool <segher@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Arjan van de Ven <arjan@infradead.org> Cc: Gordon Farquharson <gordonfarquharson@gmail.com> Cc: Guillaume Chazarain <guichaz@yahoo.fr> Cc: Theodore Tso <tytso@mit.edu> Cc: Kenneth Cheng <kenneth.w.chen@intel.com> Cc: Tobias Diedrich <ranma@tdiedrich.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-29 18:00:58 +00:00
/*
* Yes, Virginia, this is indeed insane.
*
* We use this sequence to make sure that
* (a) we account for dirty stats properly
* (b) we tell the low-level filesystem to
* mark the whole folio dirty if it was
VM: Fix nasty and subtle race in shared mmap'ed page writeback The VM layer (on the face of it, fairly reasonably) expected that when it does a ->writepage() call to the filesystem, it would write out the full page at that point in time. Especially since it had earlier marked the whole page dirty with "set_page_dirty()". But that isn't actually the case: ->writepage() does not actually write a page, it writes the parts of the page that have been explicitly marked dirty before, *and* that had not got written out for other reasons since the last time we told it they were dirty. That last caveat is the important one. Which _most_ of the time ends up being the whole page (since we had called "set_page_dirty()" on the page earlier), but if the filesystem had done any dirty flushing of its own (for example, to honor some internal write ordering guarantees), it might end up doing only a partial page IO (or none at all) when ->writepage() is actually called. That is the correct thing in general (since we actually often _want_ only the known-dirty parts of the page to be written out), but the shared dirty page handling had implicitly forgotten about these details, and had a number of cases where it was doing just the "->writepage()" part, without telling the low-level filesystem that the whole page might have been re-dirtied as part of being mapped writably into user space. Since most of the time the FS did actually write out the full page, we didn't notice this for a loong time, and this needed some really odd patterns to trigger. But it caused occasional corruption with rtorrent and with the Debian "apt" database, because both use shared mmaps to update the end result. This fixes it. Finally. After way too much hair-pulling. Acked-by: Nick Piggin <nickpiggin@yahoo.com.au> Acked-by: Martin J. Bligh <mbligh@google.com> Acked-by: Martin Michlmayr <tbm@cyrius.com> Acked-by: Martin Johansson <martin@fatbob.nu> Acked-by: Ingo Molnar <mingo@elte.hu> Acked-by: Andrei Popa <andrei.popa@i-neo.ro> Cc: High Dickins <hugh@veritas.com> Cc: Andrew Morton <akpm@osdl.org>, Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Segher Boessenkool <segher@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Arjan van de Ven <arjan@infradead.org> Cc: Gordon Farquharson <gordonfarquharson@gmail.com> Cc: Guillaume Chazarain <guichaz@yahoo.fr> Cc: Theodore Tso <tytso@mit.edu> Cc: Kenneth Cheng <kenneth.w.chen@intel.com> Cc: Tobias Diedrich <ranma@tdiedrich.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-29 18:00:58 +00:00
* dirty in a pagetable. Only to then
* (c) clean the folio again and return 1 to
VM: Fix nasty and subtle race in shared mmap'ed page writeback The VM layer (on the face of it, fairly reasonably) expected that when it does a ->writepage() call to the filesystem, it would write out the full page at that point in time. Especially since it had earlier marked the whole page dirty with "set_page_dirty()". But that isn't actually the case: ->writepage() does not actually write a page, it writes the parts of the page that have been explicitly marked dirty before, *and* that had not got written out for other reasons since the last time we told it they were dirty. That last caveat is the important one. Which _most_ of the time ends up being the whole page (since we had called "set_page_dirty()" on the page earlier), but if the filesystem had done any dirty flushing of its own (for example, to honor some internal write ordering guarantees), it might end up doing only a partial page IO (or none at all) when ->writepage() is actually called. That is the correct thing in general (since we actually often _want_ only the known-dirty parts of the page to be written out), but the shared dirty page handling had implicitly forgotten about these details, and had a number of cases where it was doing just the "->writepage()" part, without telling the low-level filesystem that the whole page might have been re-dirtied as part of being mapped writably into user space. Since most of the time the FS did actually write out the full page, we didn't notice this for a loong time, and this needed some really odd patterns to trigger. But it caused occasional corruption with rtorrent and with the Debian "apt" database, because both use shared mmaps to update the end result. This fixes it. Finally. After way too much hair-pulling. Acked-by: Nick Piggin <nickpiggin@yahoo.com.au> Acked-by: Martin J. Bligh <mbligh@google.com> Acked-by: Martin Michlmayr <tbm@cyrius.com> Acked-by: Martin Johansson <martin@fatbob.nu> Acked-by: Ingo Molnar <mingo@elte.hu> Acked-by: Andrei Popa <andrei.popa@i-neo.ro> Cc: High Dickins <hugh@veritas.com> Cc: Andrew Morton <akpm@osdl.org>, Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Segher Boessenkool <segher@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Arjan van de Ven <arjan@infradead.org> Cc: Gordon Farquharson <gordonfarquharson@gmail.com> Cc: Guillaume Chazarain <guichaz@yahoo.fr> Cc: Theodore Tso <tytso@mit.edu> Cc: Kenneth Cheng <kenneth.w.chen@intel.com> Cc: Tobias Diedrich <ranma@tdiedrich.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-29 18:00:58 +00:00
* cause the writeback.
*
* This way we avoid all nasty races with the
* dirty bit in multiple places and clearing
* them concurrently from different threads.
*
* Note! Normally the "folio_mark_dirty(folio)"
VM: Fix nasty and subtle race in shared mmap'ed page writeback The VM layer (on the face of it, fairly reasonably) expected that when it does a ->writepage() call to the filesystem, it would write out the full page at that point in time. Especially since it had earlier marked the whole page dirty with "set_page_dirty()". But that isn't actually the case: ->writepage() does not actually write a page, it writes the parts of the page that have been explicitly marked dirty before, *and* that had not got written out for other reasons since the last time we told it they were dirty. That last caveat is the important one. Which _most_ of the time ends up being the whole page (since we had called "set_page_dirty()" on the page earlier), but if the filesystem had done any dirty flushing of its own (for example, to honor some internal write ordering guarantees), it might end up doing only a partial page IO (or none at all) when ->writepage() is actually called. That is the correct thing in general (since we actually often _want_ only the known-dirty parts of the page to be written out), but the shared dirty page handling had implicitly forgotten about these details, and had a number of cases where it was doing just the "->writepage()" part, without telling the low-level filesystem that the whole page might have been re-dirtied as part of being mapped writably into user space. Since most of the time the FS did actually write out the full page, we didn't notice this for a loong time, and this needed some really odd patterns to trigger. But it caused occasional corruption with rtorrent and with the Debian "apt" database, because both use shared mmaps to update the end result. This fixes it. Finally. After way too much hair-pulling. Acked-by: Nick Piggin <nickpiggin@yahoo.com.au> Acked-by: Martin J. Bligh <mbligh@google.com> Acked-by: Martin Michlmayr <tbm@cyrius.com> Acked-by: Martin Johansson <martin@fatbob.nu> Acked-by: Ingo Molnar <mingo@elte.hu> Acked-by: Andrei Popa <andrei.popa@i-neo.ro> Cc: High Dickins <hugh@veritas.com> Cc: Andrew Morton <akpm@osdl.org>, Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Segher Boessenkool <segher@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Arjan van de Ven <arjan@infradead.org> Cc: Gordon Farquharson <gordonfarquharson@gmail.com> Cc: Guillaume Chazarain <guichaz@yahoo.fr> Cc: Theodore Tso <tytso@mit.edu> Cc: Kenneth Cheng <kenneth.w.chen@intel.com> Cc: Tobias Diedrich <ranma@tdiedrich.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-29 18:00:58 +00:00
* has no effect on the actual dirty bit - since
* that will already usually be set. But we
* need the side effects, and it can help us
* avoid races.
*
* We basically use the folio "master dirty bit"
VM: Fix nasty and subtle race in shared mmap'ed page writeback The VM layer (on the face of it, fairly reasonably) expected that when it does a ->writepage() call to the filesystem, it would write out the full page at that point in time. Especially since it had earlier marked the whole page dirty with "set_page_dirty()". But that isn't actually the case: ->writepage() does not actually write a page, it writes the parts of the page that have been explicitly marked dirty before, *and* that had not got written out for other reasons since the last time we told it they were dirty. That last caveat is the important one. Which _most_ of the time ends up being the whole page (since we had called "set_page_dirty()" on the page earlier), but if the filesystem had done any dirty flushing of its own (for example, to honor some internal write ordering guarantees), it might end up doing only a partial page IO (or none at all) when ->writepage() is actually called. That is the correct thing in general (since we actually often _want_ only the known-dirty parts of the page to be written out), but the shared dirty page handling had implicitly forgotten about these details, and had a number of cases where it was doing just the "->writepage()" part, without telling the low-level filesystem that the whole page might have been re-dirtied as part of being mapped writably into user space. Since most of the time the FS did actually write out the full page, we didn't notice this for a loong time, and this needed some really odd patterns to trigger. But it caused occasional corruption with rtorrent and with the Debian "apt" database, because both use shared mmaps to update the end result. This fixes it. Finally. After way too much hair-pulling. Acked-by: Nick Piggin <nickpiggin@yahoo.com.au> Acked-by: Martin J. Bligh <mbligh@google.com> Acked-by: Martin Michlmayr <tbm@cyrius.com> Acked-by: Martin Johansson <martin@fatbob.nu> Acked-by: Ingo Molnar <mingo@elte.hu> Acked-by: Andrei Popa <andrei.popa@i-neo.ro> Cc: High Dickins <hugh@veritas.com> Cc: Andrew Morton <akpm@osdl.org>, Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Segher Boessenkool <segher@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Arjan van de Ven <arjan@infradead.org> Cc: Gordon Farquharson <gordonfarquharson@gmail.com> Cc: Guillaume Chazarain <guichaz@yahoo.fr> Cc: Theodore Tso <tytso@mit.edu> Cc: Kenneth Cheng <kenneth.w.chen@intel.com> Cc: Tobias Diedrich <ranma@tdiedrich.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-29 18:00:58 +00:00
* as a serialization point for all the different
* threads doing their things.
*/
if (folio_mkclean(folio))
folio_mark_dirty(folio);
/*
* We carefully synchronise fault handlers against
* installing a dirty pte and marking the folio dirty
mm: protect set_page_dirty() from ongoing truncation Tejun, while reviewing the code, spotted the following race condition between the dirtying and truncation of a page: __set_page_dirty_nobuffers() __delete_from_page_cache() if (TestSetPageDirty(page)) page->mapping = NULL if (PageDirty()) dec_zone_page_state(page, NR_FILE_DIRTY); dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); if (page->mapping) account_page_dirtied(page) __inc_zone_page_state(page, NR_FILE_DIRTY); __inc_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); which results in an imbalance of NR_FILE_DIRTY and BDI_RECLAIMABLE. Dirtiers usually lock out truncation, either by holding the page lock directly, or in case of zap_pte_range(), by pinning the mapcount with the page table lock held. The notable exception to this rule, though, is do_wp_page(), for which this race exists. However, do_wp_page() already waits for a locked page to unlock before setting the dirty bit, in order to prevent a race where clear_page_dirty() misses the page bit in the presence of dirty ptes. Upgrade that wait to a fully locked set_page_dirty() to also cover the situation explained above. Afterwards, the code in set_page_dirty() dealing with a truncation race is no longer needed. Remove it. Reported-by: Tejun Heo <tj@kernel.org> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Reviewed-by: Jan Kara <jack@suse.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-01-08 22:32:18 +00:00
* at this point. We do this by having them hold the
* page lock while dirtying the folio, and folios are
mm: protect set_page_dirty() from ongoing truncation Tejun, while reviewing the code, spotted the following race condition between the dirtying and truncation of a page: __set_page_dirty_nobuffers() __delete_from_page_cache() if (TestSetPageDirty(page)) page->mapping = NULL if (PageDirty()) dec_zone_page_state(page, NR_FILE_DIRTY); dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); if (page->mapping) account_page_dirtied(page) __inc_zone_page_state(page, NR_FILE_DIRTY); __inc_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); which results in an imbalance of NR_FILE_DIRTY and BDI_RECLAIMABLE. Dirtiers usually lock out truncation, either by holding the page lock directly, or in case of zap_pte_range(), by pinning the mapcount with the page table lock held. The notable exception to this rule, though, is do_wp_page(), for which this race exists. However, do_wp_page() already waits for a locked page to unlock before setting the dirty bit, in order to prevent a race where clear_page_dirty() misses the page bit in the presence of dirty ptes. Upgrade that wait to a fully locked set_page_dirty() to also cover the situation explained above. Afterwards, the code in set_page_dirty() dealing with a truncation race is no longer needed. Remove it. Reported-by: Tejun Heo <tj@kernel.org> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Reviewed-by: Jan Kara <jack@suse.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-01-08 22:32:18 +00:00
* always locked coming in here, so we get the desired
* exclusion.
*/
writeback: safer lock nesting lock_page_memcg()/unlock_page_memcg() use spin_lock_irqsave/restore() if the page's memcg is undergoing move accounting, which occurs when a process leaves its memcg for a new one that has memory.move_charge_at_immigrate set. unlocked_inode_to_wb_begin,end() use spin_lock_irq/spin_unlock_irq() if the given inode is switching writeback domains. Switches occur when enough writes are issued from a new domain. This existing pattern is thus suspicious: lock_page_memcg(page); unlocked_inode_to_wb_begin(inode, &locked); ... unlocked_inode_to_wb_end(inode, locked); unlock_page_memcg(page); If both inode switch and process memcg migration are both in-flight then unlocked_inode_to_wb_end() will unconditionally enable interrupts while still holding the lock_page_memcg() irq spinlock. This suggests the possibility of deadlock if an interrupt occurs before unlock_page_memcg(). truncate __cancel_dirty_page lock_page_memcg unlocked_inode_to_wb_begin unlocked_inode_to_wb_end <interrupts mistakenly enabled> <interrupt> end_page_writeback test_clear_page_writeback lock_page_memcg <deadlock> unlock_page_memcg Due to configuration limitations this deadlock is not currently possible because we don't mix cgroup writeback (a cgroupv2 feature) and memory.move_charge_at_immigrate (a cgroupv1 feature). If the kernel is hacked to always claim inode switching and memcg moving_account, then this script triggers lockup in less than a minute: cd /mnt/cgroup/memory mkdir a b echo 1 > a/memory.move_charge_at_immigrate echo 1 > b/memory.move_charge_at_immigrate ( echo $BASHPID > a/cgroup.procs while true; do dd if=/dev/zero of=/mnt/big bs=1M count=256 done ) & while true; do sync done & sleep 1h & SLEEP=$! while true; do echo $SLEEP > a/cgroup.procs echo $SLEEP > b/cgroup.procs done The deadlock does not seem possible, so it's debatable if there's any reason to modify the kernel. I suggest we should to prevent future surprises. And Wang Long said "this deadlock occurs three times in our environment", so there's more reason to apply this, even to stable. Stable 4.4 has minor conflicts applying this patch. For a clean 4.4 patch see "[PATCH for-4.4] writeback: safer lock nesting" https://lkml.org/lkml/2018/4/11/146 Wang Long said "this deadlock occurs three times in our environment" [gthelen@google.com: v4] Link: http://lkml.kernel.org/r/20180411084653.254724-1-gthelen@google.com [akpm@linux-foundation.org: comment tweaks, struct initialization simplification] Change-Id: Ibb773e8045852978f6207074491d262f1b3fb613 Link: http://lkml.kernel.org/r/20180410005908.167976-1-gthelen@google.com Fixes: 682aa8e1a6a1 ("writeback: implement unlocked_inode_to_wb transaction and use it for stat updates") Signed-off-by: Greg Thelen <gthelen@google.com> Reported-by: Wang Long <wanglong19@meituan.com> Acked-by: Wang Long <wanglong19@meituan.com> Acked-by: Michal Hocko <mhocko@suse.com> Reviewed-by: Andrew Morton <akpm@linux-foundation.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Tejun Heo <tj@kernel.org> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: <stable@vger.kernel.org> [v4.2+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-04-20 21:55:42 +00:00
wb = unlocked_inode_to_wb_begin(inode, &cookie);
if (folio_test_clear_dirty(folio)) {
long nr = folio_nr_pages(folio);
lruvec_stat_mod_folio(folio, NR_FILE_DIRTY, -nr);
zone_stat_mod_folio(folio, NR_ZONE_WRITE_PENDING, -nr);
wb_stat_mod(wb, WB_RECLAIMABLE, -nr);
ret = true;
}
writeback: safer lock nesting lock_page_memcg()/unlock_page_memcg() use spin_lock_irqsave/restore() if the page's memcg is undergoing move accounting, which occurs when a process leaves its memcg for a new one that has memory.move_charge_at_immigrate set. unlocked_inode_to_wb_begin,end() use spin_lock_irq/spin_unlock_irq() if the given inode is switching writeback domains. Switches occur when enough writes are issued from a new domain. This existing pattern is thus suspicious: lock_page_memcg(page); unlocked_inode_to_wb_begin(inode, &locked); ... unlocked_inode_to_wb_end(inode, locked); unlock_page_memcg(page); If both inode switch and process memcg migration are both in-flight then unlocked_inode_to_wb_end() will unconditionally enable interrupts while still holding the lock_page_memcg() irq spinlock. This suggests the possibility of deadlock if an interrupt occurs before unlock_page_memcg(). truncate __cancel_dirty_page lock_page_memcg unlocked_inode_to_wb_begin unlocked_inode_to_wb_end <interrupts mistakenly enabled> <interrupt> end_page_writeback test_clear_page_writeback lock_page_memcg <deadlock> unlock_page_memcg Due to configuration limitations this deadlock is not currently possible because we don't mix cgroup writeback (a cgroupv2 feature) and memory.move_charge_at_immigrate (a cgroupv1 feature). If the kernel is hacked to always claim inode switching and memcg moving_account, then this script triggers lockup in less than a minute: cd /mnt/cgroup/memory mkdir a b echo 1 > a/memory.move_charge_at_immigrate echo 1 > b/memory.move_charge_at_immigrate ( echo $BASHPID > a/cgroup.procs while true; do dd if=/dev/zero of=/mnt/big bs=1M count=256 done ) & while true; do sync done & sleep 1h & SLEEP=$! while true; do echo $SLEEP > a/cgroup.procs echo $SLEEP > b/cgroup.procs done The deadlock does not seem possible, so it's debatable if there's any reason to modify the kernel. I suggest we should to prevent future surprises. And Wang Long said "this deadlock occurs three times in our environment", so there's more reason to apply this, even to stable. Stable 4.4 has minor conflicts applying this patch. For a clean 4.4 patch see "[PATCH for-4.4] writeback: safer lock nesting" https://lkml.org/lkml/2018/4/11/146 Wang Long said "this deadlock occurs three times in our environment" [gthelen@google.com: v4] Link: http://lkml.kernel.org/r/20180411084653.254724-1-gthelen@google.com [akpm@linux-foundation.org: comment tweaks, struct initialization simplification] Change-Id: Ibb773e8045852978f6207074491d262f1b3fb613 Link: http://lkml.kernel.org/r/20180410005908.167976-1-gthelen@google.com Fixes: 682aa8e1a6a1 ("writeback: implement unlocked_inode_to_wb transaction and use it for stat updates") Signed-off-by: Greg Thelen <gthelen@google.com> Reported-by: Wang Long <wanglong19@meituan.com> Acked-by: Wang Long <wanglong19@meituan.com> Acked-by: Michal Hocko <mhocko@suse.com> Reviewed-by: Andrew Morton <akpm@linux-foundation.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Tejun Heo <tj@kernel.org> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: <stable@vger.kernel.org> [v4.2+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-04-20 21:55:42 +00:00
unlocked_inode_to_wb_end(inode, &cookie);
memcg: add per cgroup dirty page accounting When modifying PG_Dirty on cached file pages, update the new MEM_CGROUP_STAT_DIRTY counter. This is done in the same places where global NR_FILE_DIRTY is managed. The new memcg stat is visible in the per memcg memory.stat cgroupfs file. The most recent past attempt at this was http://thread.gmane.org/gmane.linux.kernel.cgroups/8632 The new accounting supports future efforts to add per cgroup dirty page throttling and writeback. It also helps an administrator break down a container's memory usage and provides evidence to understand memcg oom kills (the new dirty count is included in memcg oom kill messages). The ability to move page accounting between memcg (memory.move_charge_at_immigrate) makes this accounting more complicated than the global counter. The existing mem_cgroup_{begin,end}_page_stat() lock is used to serialize move accounting with stat updates. Typical update operation: memcg = mem_cgroup_begin_page_stat(page) if (TestSetPageDirty()) { [...] mem_cgroup_update_page_stat(memcg) } mem_cgroup_end_page_stat(memcg) Summary of mem_cgroup_end_page_stat() overhead: - Without CONFIG_MEMCG it's a no-op - With CONFIG_MEMCG and no inter memcg task movement, it's just rcu_read_lock() - With CONFIG_MEMCG and inter memcg task movement, it's rcu_read_lock() + spin_lock_irqsave() A memcg parameter is added to several routines because their callers now grab mem_cgroup_begin_page_stat() which returns the memcg later needed by for mem_cgroup_update_page_stat(). Because mem_cgroup_begin_page_stat() may disable interrupts, some adjustments are needed: - move __mark_inode_dirty() from __set_page_dirty() to its caller. __mark_inode_dirty() locking does not want interrupts disabled. - use spin_lock_irqsave(tree_lock) rather than spin_lock_irq() in __delete_from_page_cache(), replace_page_cache_page(), invalidate_complete_page2(), and __remove_mapping(). text data bss dec hex filename 8925147 1774832 1785856 12485835 be84cb vmlinux-!CONFIG_MEMCG-before 8925339 1774832 1785856 12486027 be858b vmlinux-!CONFIG_MEMCG-after +192 text bytes 8965977 1784992 1785856 12536825 bf4bf9 vmlinux-CONFIG_MEMCG-before 8966750 1784992 1785856 12537598 bf4efe vmlinux-CONFIG_MEMCG-after +773 text bytes Performance tests run on v4.0-rc1-36-g4f671fe2f952. Lower is better for all metrics, they're all wall clock or cycle counts. The read and write fault benchmarks just measure fault time, they do not include I/O time. * CONFIG_MEMCG not set: baseline patched kbuild 1m25.030000(+-0.088% 3 samples) 1m25.426667(+-0.120% 3 samples) dd write 100 MiB 0.859211561 +-15.10% 0.874162885 +-15.03% dd write 200 MiB 1.670653105 +-17.87% 1.669384764 +-11.99% dd write 1000 MiB 8.434691190 +-14.15% 8.474733215 +-14.77% read fault cycles 254.0(+-0.000% 10 samples) 253.0(+-0.000% 10 samples) write fault cycles 2021.2(+-3.070% 10 samples) 1984.5(+-1.036% 10 samples) * CONFIG_MEMCG=y root_memcg: baseline patched kbuild 1m25.716667(+-0.105% 3 samples) 1m25.686667(+-0.153% 3 samples) dd write 100 MiB 0.855650830 +-14.90% 0.887557919 +-14.90% dd write 200 MiB 1.688322953 +-12.72% 1.667682724 +-13.33% dd write 1000 MiB 8.418601605 +-14.30% 8.673532299 +-15.00% read fault cycles 266.0(+-0.000% 10 samples) 266.0(+-0.000% 10 samples) write fault cycles 2051.7(+-1.349% 10 samples) 2049.6(+-1.686% 10 samples) * CONFIG_MEMCG=y non-root_memcg: baseline patched kbuild 1m26.120000(+-0.273% 3 samples) 1m25.763333(+-0.127% 3 samples) dd write 100 MiB 0.861723964 +-15.25% 0.818129350 +-14.82% dd write 200 MiB 1.669887569 +-13.30% 1.698645885 +-13.27% dd write 1000 MiB 8.383191730 +-14.65% 8.351742280 +-14.52% read fault cycles 265.7(+-0.172% 10 samples) 267.0(+-0.000% 10 samples) write fault cycles 2070.6(+-1.512% 10 samples) 2084.4(+-2.148% 10 samples) As expected anon page faults are not affected by this patch. tj: Updated to apply on top of the recent cancel_dirty_page() changes. Signed-off-by: Sha Zhengju <handai.szj@gmail.com> Signed-off-by: Greg Thelen <gthelen@google.com> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: Jens Axboe <axboe@fb.com>
2015-05-22 21:13:16 +00:00
return ret;
}
return folio_test_clear_dirty(folio);
}
EXPORT_SYMBOL(folio_clear_dirty_for_io);
static void wb_inode_writeback_start(struct bdi_writeback *wb)
{
atomic_inc(&wb->writeback_inodes);
}
static void wb_inode_writeback_end(struct bdi_writeback *wb)
{
unsigned long flags;
atomic_dec(&wb->writeback_inodes);
writeback: fix bandwidth estimate for spiky workload Michael Stapelberg has reported that for workload with short big spikes of writes (GCC linker seem to trigger this frequently) the write throughput is heavily underestimated and tends to steadily sink until it reaches zero. This has rather bad impact on writeback throttling (causing stalls). The problem is that writeback throughput estimate gets updated at most once per 200 ms. One update happens early after we submit pages for writeback (at that point writeout of only small fraction of pages is completed and thus observed throughput is tiny). Next update happens only during the next write spike (updates happen only from inode writeback and dirty throttling code) and if that is more than 1s after previous spike, we decide system was idle and just ignore whatever was written until this moment. Fix the problem by making sure writeback throughput estimate is also updated shortly after writeback completes to get reasonable estimate of throughput for spiky workloads. [jack@suse.cz: avoid division by 0 in wb_update_dirty_ratelimit()] Link: https://lore.kernel.org/lkml/20210617095309.3542373-1-stapelberg+linux@google.com Link: https://lkml.kernel.org/r/20210713104716.22868-3-jack@suse.cz Signed-off-by: Jan Kara <jack@suse.cz> Reported-by: Michael Stapelberg <stapelberg+linux@google.com> Tested-by: Michael Stapelberg <stapelberg+linux@google.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-09-02 21:53:09 +00:00
/*
* Make sure estimate of writeback throughput gets updated after
* writeback completed. We delay the update by BANDWIDTH_INTERVAL
* (which is the interval other bandwidth updates use for batching) so
* that if multiple inodes end writeback at a similar time, they get
* batched into one bandwidth update.
*/
spin_lock_irqsave(&wb->work_lock, flags);
if (test_bit(WB_registered, &wb->state))
queue_delayed_work(bdi_wq, &wb->bw_dwork, BANDWIDTH_INTERVAL);
spin_unlock_irqrestore(&wb->work_lock, flags);
}
bool __folio_end_writeback(struct folio *folio)
{
long nr = folio_nr_pages(folio);
struct address_space *mapping = folio_mapping(folio);
bool ret;
folio_memcg_lock(folio);
mm: don't use radix tree writeback tags for pages in swap cache File pages use a set of radix tree tags (DIRTY, TOWRITE, WRITEBACK, etc.) to accelerate finding the pages with a specific tag in the radix tree during inode writeback. But for anonymous pages in the swap cache, there is no inode writeback. So there is no need to find the pages with some writeback tags in the radix tree. It is not necessary to touch radix tree writeback tags for pages in the swap cache. Per Rik van Riel's suggestion, a new flag AS_NO_WRITEBACK_TAGS is introduced for address spaces which don't need to update the writeback tags. The flag is set for swap caches. It may be used for DAX file systems, etc. With this patch, the swap out bandwidth improved 22.3% (from ~1.2GB/s to ~1.48GBps) in the vm-scalability swap-w-seq test case with 8 processes. The test is done on a Xeon E5 v3 system. The swap device used is a RAM simulated PMEM (persistent memory) device. The improvement comes from the reduced contention on the swap cache radix tree lock. To test sequential swapping out, the test case uses 8 processes, which sequentially allocate and write to the anonymous pages until RAM and part of the swap device is used up. Details of comparison is as follow, base base+patch ---------------- -------------------------- %stddev %change %stddev \ | \ 2506952 ± 2% +28.1% 3212076 ± 7% vm-scalability.throughput 1207402 ± 7% +22.3% 1476578 ± 6% vmstat.swap.so 10.86 ± 12% -23.4% 8.31 ± 16% perf-profile.cycles-pp._raw_spin_lock_irq.__add_to_swap_cache.add_to_swap_cache.add_to_swap.shrink_page_list 10.82 ± 13% -33.1% 7.24 ± 14% perf-profile.cycles-pp._raw_spin_lock_irqsave.__remove_mapping.shrink_page_list.shrink_inactive_list.shrink_zone_memcg 10.36 ± 11% -100.0% 0.00 ± -1% perf-profile.cycles-pp._raw_spin_lock_irqsave.__test_set_page_writeback.bdev_write_page.__swap_writepage.swap_writepage 10.52 ± 12% -100.0% 0.00 ± -1% perf-profile.cycles-pp._raw_spin_lock_irqsave.test_clear_page_writeback.end_page_writeback.page_endio.pmem_rw_page Link: http://lkml.kernel.org/r/1472578089-5560-1-git-send-email-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Acked-by: Rik van Riel <riel@redhat.com> Cc: Hugh Dickins <hughd@google.com> Cc: Shaohua Li <shli@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Tejun Heo <tj@kernel.org> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-07 23:59:30 +00:00
if (mapping && mapping_use_writeback_tags(mapping)) {
struct inode *inode = mapping->host;
struct backing_dev_info *bdi = inode_to_bdi(inode);
unsigned long flags;
xa_lock_irqsave(&mapping->i_pages, flags);
ret = folio_xor_flags_has_waiters(folio, 1 << PG_writeback);
__xa_clear_mark(&mapping->i_pages, folio_index(folio),
PAGECACHE_TAG_WRITEBACK);
if (bdi->capabilities & BDI_CAP_WRITEBACK_ACCT) {
struct bdi_writeback *wb = inode_to_wb(inode);
wb_stat_mod(wb, WB_WRITEBACK, -nr);
__wb_writeout_add(wb, nr);
if (!mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK))
wb_inode_writeback_end(wb);
}
fs/fs-writeback.c: add a new writeback list for sync wait_sb_inodes() currently does a walk of all inodes in the filesystem to find dirty one to wait on during sync. This is highly inefficient and wastes a lot of CPU when there are lots of clean cached inodes that we don't need to wait on. To avoid this "all inode" walk, we need to track inodes that are currently under writeback that we need to wait for. We do this by adding inodes to a writeback list on the sb when the mapping is first tagged as having pages under writeback. wait_sb_inodes() can then walk this list of "inodes under IO" and wait specifically just for the inodes that the current sync(2) needs to wait for. Define a couple helpers to add/remove an inode from the writeback list and call them when the overall mapping is tagged for or cleared from writeback. Update wait_sb_inodes() to walk only the inodes under writeback due to the sync. With this change, filesystem sync times are significantly reduced for fs' with largely populated inode caches and otherwise no other work to do. For example, on a 16xcpu 2GHz x86-64 server, 10TB XFS filesystem with a ~10m entry inode cache, sync times are reduced from ~7.3s to less than 0.1s when the filesystem is fully clean. Link: http://lkml.kernel.org/r/1466594593-6757-2-git-send-email-bfoster@redhat.com Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Josef Bacik <jbacik@fb.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Jan Kara <jack@suse.cz> Tested-by: Holger Hoffstätte <holger.hoffstaette@applied-asynchrony.com> Cc: Al Viro <viro@ZenIV.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-26 22:21:50 +00:00
if (mapping->host && !mapping_tagged(mapping,
PAGECACHE_TAG_WRITEBACK))
sb_clear_inode_writeback(mapping->host);
xa_unlock_irqrestore(&mapping->i_pages, flags);
} else {
ret = folio_xor_flags_has_waiters(folio, 1 << PG_writeback);
}
lruvec_stat_mod_folio(folio, NR_WRITEBACK, -nr);
zone_stat_mod_folio(folio, NR_ZONE_WRITE_PENDING, -nr);
node_stat_mod_folio(folio, NR_WRITTEN, nr);
folio_memcg_unlock(folio);
return ret;
}
void __folio_start_writeback(struct folio *folio, bool keep_write)
{
long nr = folio_nr_pages(folio);
struct address_space *mapping = folio_mapping(folio);
int access_ret;
VM_BUG_ON_FOLIO(folio_test_writeback(folio), folio);
folio_memcg_lock(folio);
mm: don't use radix tree writeback tags for pages in swap cache File pages use a set of radix tree tags (DIRTY, TOWRITE, WRITEBACK, etc.) to accelerate finding the pages with a specific tag in the radix tree during inode writeback. But for anonymous pages in the swap cache, there is no inode writeback. So there is no need to find the pages with some writeback tags in the radix tree. It is not necessary to touch radix tree writeback tags for pages in the swap cache. Per Rik van Riel's suggestion, a new flag AS_NO_WRITEBACK_TAGS is introduced for address spaces which don't need to update the writeback tags. The flag is set for swap caches. It may be used for DAX file systems, etc. With this patch, the swap out bandwidth improved 22.3% (from ~1.2GB/s to ~1.48GBps) in the vm-scalability swap-w-seq test case with 8 processes. The test is done on a Xeon E5 v3 system. The swap device used is a RAM simulated PMEM (persistent memory) device. The improvement comes from the reduced contention on the swap cache radix tree lock. To test sequential swapping out, the test case uses 8 processes, which sequentially allocate and write to the anonymous pages until RAM and part of the swap device is used up. Details of comparison is as follow, base base+patch ---------------- -------------------------- %stddev %change %stddev \ | \ 2506952 ± 2% +28.1% 3212076 ± 7% vm-scalability.throughput 1207402 ± 7% +22.3% 1476578 ± 6% vmstat.swap.so 10.86 ± 12% -23.4% 8.31 ± 16% perf-profile.cycles-pp._raw_spin_lock_irq.__add_to_swap_cache.add_to_swap_cache.add_to_swap.shrink_page_list 10.82 ± 13% -33.1% 7.24 ± 14% perf-profile.cycles-pp._raw_spin_lock_irqsave.__remove_mapping.shrink_page_list.shrink_inactive_list.shrink_zone_memcg 10.36 ± 11% -100.0% 0.00 ± -1% perf-profile.cycles-pp._raw_spin_lock_irqsave.__test_set_page_writeback.bdev_write_page.__swap_writepage.swap_writepage 10.52 ± 12% -100.0% 0.00 ± -1% perf-profile.cycles-pp._raw_spin_lock_irqsave.test_clear_page_writeback.end_page_writeback.page_endio.pmem_rw_page Link: http://lkml.kernel.org/r/1472578089-5560-1-git-send-email-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Acked-by: Rik van Riel <riel@redhat.com> Cc: Hugh Dickins <hughd@google.com> Cc: Shaohua Li <shli@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Tejun Heo <tj@kernel.org> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-07 23:59:30 +00:00
if (mapping && mapping_use_writeback_tags(mapping)) {
XA_STATE(xas, &mapping->i_pages, folio_index(folio));
struct inode *inode = mapping->host;
struct backing_dev_info *bdi = inode_to_bdi(inode);
unsigned long flags;
bool on_wblist;
xas_lock_irqsave(&xas, flags);
xas_load(&xas);
folio_test_set_writeback(folio);
fs/fs-writeback.c: add a new writeback list for sync wait_sb_inodes() currently does a walk of all inodes in the filesystem to find dirty one to wait on during sync. This is highly inefficient and wastes a lot of CPU when there are lots of clean cached inodes that we don't need to wait on. To avoid this "all inode" walk, we need to track inodes that are currently under writeback that we need to wait for. We do this by adding inodes to a writeback list on the sb when the mapping is first tagged as having pages under writeback. wait_sb_inodes() can then walk this list of "inodes under IO" and wait specifically just for the inodes that the current sync(2) needs to wait for. Define a couple helpers to add/remove an inode from the writeback list and call them when the overall mapping is tagged for or cleared from writeback. Update wait_sb_inodes() to walk only the inodes under writeback due to the sync. With this change, filesystem sync times are significantly reduced for fs' with largely populated inode caches and otherwise no other work to do. For example, on a 16xcpu 2GHz x86-64 server, 10TB XFS filesystem with a ~10m entry inode cache, sync times are reduced from ~7.3s to less than 0.1s when the filesystem is fully clean. Link: http://lkml.kernel.org/r/1466594593-6757-2-git-send-email-bfoster@redhat.com Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Josef Bacik <jbacik@fb.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Jan Kara <jack@suse.cz> Tested-by: Holger Hoffstätte <holger.hoffstaette@applied-asynchrony.com> Cc: Al Viro <viro@ZenIV.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-26 22:21:50 +00:00
on_wblist = mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK);
fs/fs-writeback.c: add a new writeback list for sync wait_sb_inodes() currently does a walk of all inodes in the filesystem to find dirty one to wait on during sync. This is highly inefficient and wastes a lot of CPU when there are lots of clean cached inodes that we don't need to wait on. To avoid this "all inode" walk, we need to track inodes that are currently under writeback that we need to wait for. We do this by adding inodes to a writeback list on the sb when the mapping is first tagged as having pages under writeback. wait_sb_inodes() can then walk this list of "inodes under IO" and wait specifically just for the inodes that the current sync(2) needs to wait for. Define a couple helpers to add/remove an inode from the writeback list and call them when the overall mapping is tagged for or cleared from writeback. Update wait_sb_inodes() to walk only the inodes under writeback due to the sync. With this change, filesystem sync times are significantly reduced for fs' with largely populated inode caches and otherwise no other work to do. For example, on a 16xcpu 2GHz x86-64 server, 10TB XFS filesystem with a ~10m entry inode cache, sync times are reduced from ~7.3s to less than 0.1s when the filesystem is fully clean. Link: http://lkml.kernel.org/r/1466594593-6757-2-git-send-email-bfoster@redhat.com Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Josef Bacik <jbacik@fb.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Jan Kara <jack@suse.cz> Tested-by: Holger Hoffstätte <holger.hoffstaette@applied-asynchrony.com> Cc: Al Viro <viro@ZenIV.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-26 22:21:50 +00:00
xas_set_mark(&xas, PAGECACHE_TAG_WRITEBACK);
if (bdi->capabilities & BDI_CAP_WRITEBACK_ACCT) {
struct bdi_writeback *wb = inode_to_wb(inode);
fs/fs-writeback.c: add a new writeback list for sync wait_sb_inodes() currently does a walk of all inodes in the filesystem to find dirty one to wait on during sync. This is highly inefficient and wastes a lot of CPU when there are lots of clean cached inodes that we don't need to wait on. To avoid this "all inode" walk, we need to track inodes that are currently under writeback that we need to wait for. We do this by adding inodes to a writeback list on the sb when the mapping is first tagged as having pages under writeback. wait_sb_inodes() can then walk this list of "inodes under IO" and wait specifically just for the inodes that the current sync(2) needs to wait for. Define a couple helpers to add/remove an inode from the writeback list and call them when the overall mapping is tagged for or cleared from writeback. Update wait_sb_inodes() to walk only the inodes under writeback due to the sync. With this change, filesystem sync times are significantly reduced for fs' with largely populated inode caches and otherwise no other work to do. For example, on a 16xcpu 2GHz x86-64 server, 10TB XFS filesystem with a ~10m entry inode cache, sync times are reduced from ~7.3s to less than 0.1s when the filesystem is fully clean. Link: http://lkml.kernel.org/r/1466594593-6757-2-git-send-email-bfoster@redhat.com Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Josef Bacik <jbacik@fb.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Jan Kara <jack@suse.cz> Tested-by: Holger Hoffstätte <holger.hoffstaette@applied-asynchrony.com> Cc: Al Viro <viro@ZenIV.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-26 22:21:50 +00:00
wb_stat_mod(wb, WB_WRITEBACK, nr);
if (!on_wblist)
wb_inode_writeback_start(wb);
}
/*
* We can come through here when swapping anonymous
* folios, so we don't necessarily have an inode to
* track for sync.
*/
if (mapping->host && !on_wblist)
sb_mark_inode_writeback(mapping->host);
if (!folio_test_dirty(folio))
xas_clear_mark(&xas, PAGECACHE_TAG_DIRTY);
ext4: fix data integrity sync in ordered mode When we perform a data integrity sync we tag all the dirty pages with PAGECACHE_TAG_TOWRITE at start of ext4_da_writepages. Later we check for this tag in write_cache_pages_da and creates a struct mpage_da_data containing contiguously indexed pages tagged with this tag and sync these pages with a call to mpage_da_map_and_submit. This process is done in while loop until all the PAGECACHE_TAG_TOWRITE pages are synced. We also do journal start and stop in each iteration. journal_stop could initiate journal commit which would call ext4_writepage which in turn will call ext4_bio_write_page even for delayed OR unwritten buffers. When ext4_bio_write_page is called for such buffers, even though it does not sync them but it clears the PAGECACHE_TAG_TOWRITE of the corresponding page and hence these pages are also not synced by the currently running data integrity sync. We will end up with dirty pages although sync is completed. This could cause a potential data loss when the sync call is followed by a truncate_pagecache call, which is exactly the case in collapse_range. (It will cause generic/127 failure in xfstests) To avoid this issue, we can use set_page_writeback_keepwrite instead of set_page_writeback, which doesn't clear TOWRITE tag. Cc: stable@vger.kernel.org Signed-off-by: Namjae Jeon <namjae.jeon@samsung.com> Signed-off-by: Ashish Sangwan <a.sangwan@samsung.com> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu> Reviewed-by: Jan Kara <jack@suse.cz>
2014-05-12 12:12:25 +00:00
if (!keep_write)
xas_clear_mark(&xas, PAGECACHE_TAG_TOWRITE);
xas_unlock_irqrestore(&xas, flags);
} else {
folio_test_set_writeback(folio);
}
lruvec_stat_mod_folio(folio, NR_WRITEBACK, nr);
zone_stat_mod_folio(folio, NR_ZONE_WRITE_PENDING, nr);
folio_memcg_unlock(folio);
access_ret = arch_make_folio_accessible(folio);
mm/gup/writeback: add callbacks for inaccessible pages With the introduction of protected KVM guests on s390 there is now a concept of inaccessible pages. These pages need to be made accessible before the host can access them. While cpu accesses will trigger a fault that can be resolved, I/O accesses will just fail. We need to add a callback into architecture code for places that will do I/O, namely when writeback is started or when a page reference is taken. This is not only to enable paging, file backing etc, it is also necessary to protect the host against a malicious user space. For example a bad QEMU could simply start direct I/O on such protected memory. We do not want userspace to be able to trigger I/O errors and thus the logic is "whenever somebody accesses that page (gup) or does I/O, make sure that this page can be accessed". When the guest tries to access that page we will wait in the page fault handler for writeback to have finished and for the page_ref to be the expected value. On s390x the function is not supposed to fail, so it is ok to use a WARN_ON on failure. If we ever need some more finegrained handling we can tackle this when we know the details. Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: David Hildenbrand <david@redhat.com> Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com> Reviewed-by: John Hubbard <jhubbard@nvidia.com> Acked-by: Will Deacon <will@kernel.org> Cc: Jan Kara <jack@suse.cz> Cc: Matthew Wilcox <willy@infradead.org> Cc: Ira Weiny <ira.weiny@intel.com> Cc: Jérôme Glisse <jglisse@redhat.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Chinner <david@fromorbit.com> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Shuah Khan <shuah@kernel.org> Cc: Vlastimil Babka <vbabka@suse.cz> Link: http://lkml.kernel.org/r/20200306132537.783769-3-imbrenda@linux.ibm.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-04-02 04:05:56 +00:00
/*
* If writeback has been triggered on a page that cannot be made
* accessible, it is too late to recover here.
*/
VM_BUG_ON_FOLIO(access_ret != 0, folio);
}
EXPORT_SYMBOL(__folio_start_writeback);
/**
* folio_wait_writeback - Wait for a folio to finish writeback.
* @folio: The folio to wait for.
*
* If the folio is currently being written back to storage, wait for the
* I/O to complete.
*
* Context: Sleeps. Must be called in process context and with
* no spinlocks held. Caller should hold a reference on the folio.
* If the folio is not locked, writeback may start again after writeback
* has finished.
*/
void folio_wait_writeback(struct folio *folio)
{
while (folio_test_writeback(folio)) {
trace_folio_wait_writeback(folio, folio_mapping(folio));
folio_wait_bit(folio, PG_writeback);
}
}
EXPORT_SYMBOL_GPL(folio_wait_writeback);
/**
* folio_wait_writeback_killable - Wait for a folio to finish writeback.
* @folio: The folio to wait for.
*
* If the folio is currently being written back to storage, wait for the
* I/O to complete or a fatal signal to arrive.
*
* Context: Sleeps. Must be called in process context and with
* no spinlocks held. Caller should hold a reference on the folio.
* If the folio is not locked, writeback may start again after writeback
* has finished.
* Return: 0 on success, -EINTR if we get a fatal signal while waiting.
*/
int folio_wait_writeback_killable(struct folio *folio)
{
while (folio_test_writeback(folio)) {
trace_folio_wait_writeback(folio, folio_mapping(folio));
if (folio_wait_bit_killable(folio, PG_writeback))
return -EINTR;
}
return 0;
}
EXPORT_SYMBOL_GPL(folio_wait_writeback_killable);
mm: only enforce stable page writes if the backing device requires it Create a helper function to check if a backing device requires stable page writes and, if so, performs the necessary wait. Then, make it so that all points in the memory manager that handle making pages writable use the helper function. This should provide stable page write support to most filesystems, while eliminating unnecessary waiting for devices that don't require the feature. Before this patchset, all filesystems would block, regardless of whether or not it was necessary. ext3 would wait, but still generate occasional checksum errors. The network filesystems were left to do their own thing, so they'd wait too. After this patchset, all the disk filesystems except ext3 and btrfs will wait only if the hardware requires it. ext3 (if necessary) snapshots pages instead of blocking, and btrfs provides its own bdi so the mm will never wait. Network filesystems haven't been touched, so either they provide their own stable page guarantees or they don't block at all. The blocking behavior is back to what it was before 3.0 if you don't have a disk requiring stable page writes. Here's the result of using dbench to test latency on ext2: 3.8.0-rc3: Operation Count AvgLat MaxLat ---------------------------------------- WriteX 109347 0.028 59.817 ReadX 347180 0.004 3.391 Flush 15514 29.828 287.283 Throughput 57.429 MB/sec 4 clients 4 procs max_latency=287.290 ms 3.8.0-rc3 + patches: WriteX 105556 0.029 4.273 ReadX 335004 0.005 4.112 Flush 14982 30.540 298.634 Throughput 55.4496 MB/sec 4 clients 4 procs max_latency=298.650 ms As you can see, the maximum write latency drops considerably with this patch enabled. The other filesystems (ext3/ext4/xfs/btrfs) behave similarly, but see the cover letter for those results. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Acked-by: Steven Whitehouse <swhiteho@redhat.com> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Artem Bityutskiy <dedekind1@gmail.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Mark Fasheh <mfasheh@suse.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Eric Van Hensbergen <ericvh@gmail.com> Cc: Ron Minnich <rminnich@sandia.gov> Cc: Latchesar Ionkov <lucho@ionkov.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-22 00:42:51 +00:00
/**
* folio_wait_stable() - wait for writeback to finish, if necessary.
* @folio: The folio to wait on.
mm: only enforce stable page writes if the backing device requires it Create a helper function to check if a backing device requires stable page writes and, if so, performs the necessary wait. Then, make it so that all points in the memory manager that handle making pages writable use the helper function. This should provide stable page write support to most filesystems, while eliminating unnecessary waiting for devices that don't require the feature. Before this patchset, all filesystems would block, regardless of whether or not it was necessary. ext3 would wait, but still generate occasional checksum errors. The network filesystems were left to do their own thing, so they'd wait too. After this patchset, all the disk filesystems except ext3 and btrfs will wait only if the hardware requires it. ext3 (if necessary) snapshots pages instead of blocking, and btrfs provides its own bdi so the mm will never wait. Network filesystems haven't been touched, so either they provide their own stable page guarantees or they don't block at all. The blocking behavior is back to what it was before 3.0 if you don't have a disk requiring stable page writes. Here's the result of using dbench to test latency on ext2: 3.8.0-rc3: Operation Count AvgLat MaxLat ---------------------------------------- WriteX 109347 0.028 59.817 ReadX 347180 0.004 3.391 Flush 15514 29.828 287.283 Throughput 57.429 MB/sec 4 clients 4 procs max_latency=287.290 ms 3.8.0-rc3 + patches: WriteX 105556 0.029 4.273 ReadX 335004 0.005 4.112 Flush 14982 30.540 298.634 Throughput 55.4496 MB/sec 4 clients 4 procs max_latency=298.650 ms As you can see, the maximum write latency drops considerably with this patch enabled. The other filesystems (ext3/ext4/xfs/btrfs) behave similarly, but see the cover letter for those results. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Acked-by: Steven Whitehouse <swhiteho@redhat.com> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Artem Bityutskiy <dedekind1@gmail.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Mark Fasheh <mfasheh@suse.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Eric Van Hensbergen <ericvh@gmail.com> Cc: Ron Minnich <rminnich@sandia.gov> Cc: Latchesar Ionkov <lucho@ionkov.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-22 00:42:51 +00:00
*
* This function determines if the given folio is related to a backing
* device that requires folio contents to be held stable during writeback.
* If so, then it will wait for any pending writeback to complete.
*
* Context: Sleeps. Must be called in process context and with
* no spinlocks held. Caller should hold a reference on the folio.
* If the folio is not locked, writeback may start again after writeback
* has finished.
mm: only enforce stable page writes if the backing device requires it Create a helper function to check if a backing device requires stable page writes and, if so, performs the necessary wait. Then, make it so that all points in the memory manager that handle making pages writable use the helper function. This should provide stable page write support to most filesystems, while eliminating unnecessary waiting for devices that don't require the feature. Before this patchset, all filesystems would block, regardless of whether or not it was necessary. ext3 would wait, but still generate occasional checksum errors. The network filesystems were left to do their own thing, so they'd wait too. After this patchset, all the disk filesystems except ext3 and btrfs will wait only if the hardware requires it. ext3 (if necessary) snapshots pages instead of blocking, and btrfs provides its own bdi so the mm will never wait. Network filesystems haven't been touched, so either they provide their own stable page guarantees or they don't block at all. The blocking behavior is back to what it was before 3.0 if you don't have a disk requiring stable page writes. Here's the result of using dbench to test latency on ext2: 3.8.0-rc3: Operation Count AvgLat MaxLat ---------------------------------------- WriteX 109347 0.028 59.817 ReadX 347180 0.004 3.391 Flush 15514 29.828 287.283 Throughput 57.429 MB/sec 4 clients 4 procs max_latency=287.290 ms 3.8.0-rc3 + patches: WriteX 105556 0.029 4.273 ReadX 335004 0.005 4.112 Flush 14982 30.540 298.634 Throughput 55.4496 MB/sec 4 clients 4 procs max_latency=298.650 ms As you can see, the maximum write latency drops considerably with this patch enabled. The other filesystems (ext3/ext4/xfs/btrfs) behave similarly, but see the cover letter for those results. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Acked-by: Steven Whitehouse <swhiteho@redhat.com> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Artem Bityutskiy <dedekind1@gmail.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Mark Fasheh <mfasheh@suse.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Eric Van Hensbergen <ericvh@gmail.com> Cc: Ron Minnich <rminnich@sandia.gov> Cc: Latchesar Ionkov <lucho@ionkov.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-22 00:42:51 +00:00
*/
void folio_wait_stable(struct folio *folio)
mm: only enforce stable page writes if the backing device requires it Create a helper function to check if a backing device requires stable page writes and, if so, performs the necessary wait. Then, make it so that all points in the memory manager that handle making pages writable use the helper function. This should provide stable page write support to most filesystems, while eliminating unnecessary waiting for devices that don't require the feature. Before this patchset, all filesystems would block, regardless of whether or not it was necessary. ext3 would wait, but still generate occasional checksum errors. The network filesystems were left to do their own thing, so they'd wait too. After this patchset, all the disk filesystems except ext3 and btrfs will wait only if the hardware requires it. ext3 (if necessary) snapshots pages instead of blocking, and btrfs provides its own bdi so the mm will never wait. Network filesystems haven't been touched, so either they provide their own stable page guarantees or they don't block at all. The blocking behavior is back to what it was before 3.0 if you don't have a disk requiring stable page writes. Here's the result of using dbench to test latency on ext2: 3.8.0-rc3: Operation Count AvgLat MaxLat ---------------------------------------- WriteX 109347 0.028 59.817 ReadX 347180 0.004 3.391 Flush 15514 29.828 287.283 Throughput 57.429 MB/sec 4 clients 4 procs max_latency=287.290 ms 3.8.0-rc3 + patches: WriteX 105556 0.029 4.273 ReadX 335004 0.005 4.112 Flush 14982 30.540 298.634 Throughput 55.4496 MB/sec 4 clients 4 procs max_latency=298.650 ms As you can see, the maximum write latency drops considerably with this patch enabled. The other filesystems (ext3/ext4/xfs/btrfs) behave similarly, but see the cover letter for those results. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Acked-by: Steven Whitehouse <swhiteho@redhat.com> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Artem Bityutskiy <dedekind1@gmail.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Mark Fasheh <mfasheh@suse.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Eric Van Hensbergen <ericvh@gmail.com> Cc: Ron Minnich <rminnich@sandia.gov> Cc: Latchesar Ionkov <lucho@ionkov.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-22 00:42:51 +00:00
{
if (mapping_stable_writes(folio_mapping(folio)))
folio_wait_writeback(folio);
mm: only enforce stable page writes if the backing device requires it Create a helper function to check if a backing device requires stable page writes and, if so, performs the necessary wait. Then, make it so that all points in the memory manager that handle making pages writable use the helper function. This should provide stable page write support to most filesystems, while eliminating unnecessary waiting for devices that don't require the feature. Before this patchset, all filesystems would block, regardless of whether or not it was necessary. ext3 would wait, but still generate occasional checksum errors. The network filesystems were left to do their own thing, so they'd wait too. After this patchset, all the disk filesystems except ext3 and btrfs will wait only if the hardware requires it. ext3 (if necessary) snapshots pages instead of blocking, and btrfs provides its own bdi so the mm will never wait. Network filesystems haven't been touched, so either they provide their own stable page guarantees or they don't block at all. The blocking behavior is back to what it was before 3.0 if you don't have a disk requiring stable page writes. Here's the result of using dbench to test latency on ext2: 3.8.0-rc3: Operation Count AvgLat MaxLat ---------------------------------------- WriteX 109347 0.028 59.817 ReadX 347180 0.004 3.391 Flush 15514 29.828 287.283 Throughput 57.429 MB/sec 4 clients 4 procs max_latency=287.290 ms 3.8.0-rc3 + patches: WriteX 105556 0.029 4.273 ReadX 335004 0.005 4.112 Flush 14982 30.540 298.634 Throughput 55.4496 MB/sec 4 clients 4 procs max_latency=298.650 ms As you can see, the maximum write latency drops considerably with this patch enabled. The other filesystems (ext3/ext4/xfs/btrfs) behave similarly, but see the cover letter for those results. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Acked-by: Steven Whitehouse <swhiteho@redhat.com> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Artem Bityutskiy <dedekind1@gmail.com> Cc: Joel Becker <jlbec@evilplan.org> Cc: Mark Fasheh <mfasheh@suse.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Eric Van Hensbergen <ericvh@gmail.com> Cc: Ron Minnich <rminnich@sandia.gov> Cc: Latchesar Ionkov <lucho@ionkov.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-22 00:42:51 +00:00
}
EXPORT_SYMBOL_GPL(folio_wait_stable);