aops->write_begin may allocate a new page and make it visible only to have
mark_page_accessed called almost immediately after. Once the page is
visible the atomic operations are necessary which is noticable overhead
when writing to an in-memory filesystem like tmpfs but should also be
noticable with fast storage. The objective of the patch is to initialse
the accessed information with non-atomic operations before the page is
visible.
The bulk of filesystems directly or indirectly use
grab_cache_page_write_begin or find_or_create_page for the initial
allocation of a page cache page. This patch adds an init_page_accessed()
helper which behaves like the first call to mark_page_accessed() but may
called before the page is visible and can be done non-atomically.
The primary APIs of concern in this care are the following and are used
by most filesystems.
find_get_page
find_lock_page
find_or_create_page
grab_cache_page_nowait
grab_cache_page_write_begin
All of them are very similar in detail to the patch creates a core helper
pagecache_get_page() which takes a flags parameter that affects its
behavior such as whether the page should be marked accessed or not. Then
old API is preserved but is basically a thin wrapper around this core
function.
Each of the filesystems are then updated to avoid calling
mark_page_accessed when it is known that the VM interfaces have already
done the job. There is a slight snag in that the timing of the
mark_page_accessed() has now changed so in rare cases it's possible a page
gets to the end of the LRU as PageReferenced where as previously it might
have been repromoted. This is expected to be rare but it's worth the
filesystem people thinking about it in case they see a problem with the
timing change. It is also the case that some filesystems may be marking
pages accessed that previously did not but it makes sense that filesystems
have consistent behaviour in this regard.
The test case used to evaulate this is a simple dd of a large file done
multiple times with the file deleted on each iterations. The size of the
file is 1/10th physical memory to avoid dirty page balancing. In the
async case it will be possible that the workload completes without even
hitting the disk and will have variable results but highlight the impact
of mark_page_accessed for async IO. The sync results are expected to be
more stable. The exception is tmpfs where the normal case is for the "IO"
to not hit the disk.
The test machine was single socket and UMA to avoid any scheduling or NUMA
artifacts. Throughput and wall times are presented for sync IO, only wall
times are shown for async as the granularity reported by dd and the
variability is unsuitable for comparison. As async results were variable
do to writback timings, I'm only reporting the maximum figures. The sync
results were stable enough to make the mean and stddev uninteresting.
The performance results are reported based on a run with no profiling.
Profile data is based on a separate run with oprofile running.
async dd
3.15.0-rc3 3.15.0-rc3
vanilla accessed-v2
ext3 Max elapsed 13.9900 ( 0.00%) 11.5900 ( 17.16%)
tmpfs Max elapsed 0.5100 ( 0.00%) 0.4900 ( 3.92%)
btrfs Max elapsed 12.8100 ( 0.00%) 12.7800 ( 0.23%)
ext4 Max elapsed 18.6000 ( 0.00%) 13.3400 ( 28.28%)
xfs Max elapsed 12.5600 ( 0.00%) 2.0900 ( 83.36%)
The XFS figure is a bit strange as it managed to avoid a worst case by
sheer luck but the average figures looked reasonable.
samples percentage
ext3 86107 0.9783 vmlinux-3.15.0-rc4-vanilla mark_page_accessed
ext3 23833 0.2710 vmlinux-3.15.0-rc4-accessed-v3r25 mark_page_accessed
ext3 5036 0.0573 vmlinux-3.15.0-rc4-accessed-v3r25 init_page_accessed
ext4 64566 0.8961 vmlinux-3.15.0-rc4-vanilla mark_page_accessed
ext4 5322 0.0713 vmlinux-3.15.0-rc4-accessed-v3r25 mark_page_accessed
ext4 2869 0.0384 vmlinux-3.15.0-rc4-accessed-v3r25 init_page_accessed
xfs 62126 1.7675 vmlinux-3.15.0-rc4-vanilla mark_page_accessed
xfs 1904 0.0554 vmlinux-3.15.0-rc4-accessed-v3r25 init_page_accessed
xfs 103 0.0030 vmlinux-3.15.0-rc4-accessed-v3r25 mark_page_accessed
btrfs 10655 0.1338 vmlinux-3.15.0-rc4-vanilla mark_page_accessed
btrfs 2020 0.0273 vmlinux-3.15.0-rc4-accessed-v3r25 init_page_accessed
btrfs 587 0.0079 vmlinux-3.15.0-rc4-accessed-v3r25 mark_page_accessed
tmpfs 59562 3.2628 vmlinux-3.15.0-rc4-vanilla mark_page_accessed
tmpfs 1210 0.0696 vmlinux-3.15.0-rc4-accessed-v3r25 init_page_accessed
tmpfs 94 0.0054 vmlinux-3.15.0-rc4-accessed-v3r25 mark_page_accessed
[akpm@linux-foundation.org: don't run init_page_accessed() against an uninitialised pointer]
Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jan Kara <jack@suse.cz>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Tested-by: Prabhakar Lad <prabhakar.csengg@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch use list_for_each_entry{_safe} instead of list_for_each{_safe} for
simplfying code.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Move kmem_cache_free out of spinlock protection region for better performance.
Change log from v1:
o remove spinlock protection for kmem_cache_free in destroy_node_manager
suggested by Jaegeuk Kim.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
If a page is on writeback, f2fs can face with deadlock due to under writepages.
This is caused by merging IOs inside f2fs, so if it comes to detect, let's throw
merged IOs, which is implemented by f2fs_wait_on_page_writeback.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch introduces nr_pages_to_write to align page writes to the segment
or other operational unit size, which can be tuned according to the system
environment.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch introduces nr_pages_to_skip(sbi, type) to determine writepages can
be skipped.
The dentry, node, and meta pages can be conrolled by F2FS without breaking the
FS consistency.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
We should de-account dirty counters for page when redirty in ->writepage().
Wu Fengguang described in 'commit 971767caf632190f77a40b4011c19948232eed75':
"writeback: fix dirtied pages accounting on redirty
De-account the accumulative dirty counters on page redirty.
Page redirties (very common in ext4) will introduce mismatch between
counters (a) and (b)
a) NR_DIRTIED, BDI_DIRTIED, tsk->nr_dirtied
b) NR_WRITTEN, BDI_WRITTEN
This will introduce systematic errors in balanced_rate and result in
dirty page position errors (ie. the dirty pages are no longer balanced
around the global/bdi setpoints)."
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
If there are multi segments in one section, we will read those SSA blocks which
have contiguous address one by one in f2fs_gc. It may lost performance, let's
read ahead SSA blocks by merge multi read request.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch shows the counts of checkpoint in f2fs' status.
Signed-off-by: Changman Lee <cm224.lee@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch help us to cleanup the readahead code by merging ra_{sit,nat}_pages
function into ra_meta_pages.
Additionally the new function is used to readahead cp block in
recover_orphan_inodes.
Change log from v1:
o fix a deadloop bug pointed by Jaegeuk Kim.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
If f2fs entered errorneous checkpoint status, it should skip writing meta
pages instead of redirtying the pages out.
Otherwise, it cannot unmount the partition even though f2fs is under read-only
status.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch adds NODE_MAPPING which is similar as META_MAPPING introduced by
Gu Zheng.
Cc: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
As the orphan_blocks may be max to 504, so it is not security
and rigorous to store such a large array in the kernel stack
as Dan Carpenter said.
In fact, grab_meta_page has locked the page in the page cache,
and we can use find_get_page() to fetch the page safely in the
downstream, so we can remove the page array directly.
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Introduce help function META_MAPPING() to get the cache meta blocks'
address space.
Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
With the 2 previous changes, all the long time operations are moved out
of the protection region, so here we can use spinlock rather than mutex
(orphan_inode_mutex) for lower overhead.
Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Move alloc new orphan node out of lock protection region.
Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Previously, we need to calculate the max orphan num when we try to acquire an
orphan inode, but it's a stable value since the super block was inited. So
converting it to a field of f2fs_sb_info and use it directly when needed seems
a better choose.
Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch enhances writing dirty meta pages collectively in background.
During the file data writes, it'd better avoid to write small dirty meta pages
frequently.
So let's give a chance to collect a number of dirty meta pages for a while.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch introduces f2fs_io_info to mitigate the complex parameter list.
struct f2fs_io_info {
enum page_type type; /* contains DATA/NODE/META/META_FLUSH */
int rw; /* contains R/RS/W/WS */
int rw_flag; /* contains REQ_META/REQ_PRIO */
}
1. f2fs_write_data_pages
- DATA
- WRITE_SYNC is set when wbc->WB_SYNC_ALL.
2. sync_node_pages
- NODE
- WRITE_SYNC all the time
3. sync_meta_pages
- META
- WRITE_SYNC all the time
- REQ_META | REQ_PRIO all the time
** f2fs_submit_merged_bio() handles META_FLUSH.
4. ra_nat_pages, ra_sit_pages, ra_sum_pages
- META
- READ_SYNC
Cc: Fan Li <fanofcode.li@samsung.com>
Cc: Changman Lee <cm224.lee@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch adds unlikely() macro into the most of codes.
The basic rule is to add that when:
- checking unusual errors,
- checking page mappings,
- and the other unlikely conditions.
Change log from v1:
- Don't add unlikely for the NULL test and error test: advised by Andi Kleen.
Cc: Chao Yu <chao2.yu@samsung.com>
Cc: Andi Kleen <andi@firstfloor.org>
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
As we know, some of our branch condition will rarely be true. So we could add
'unlikely' to let compiler optimize these code, by this way we could drop
unneeded 'jump' assemble code to improve performance.
change log:
o add *unlikely* as many as possible across the whole source files at once
suggested by Jaegeuk Kim.
Suggested-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch integrates redundant bio operations on read and write IOs.
1. Move bio-related codes to the top of data.c.
2. Replace f2fs_submit_bio with f2fs_submit_merged_bio, which handles read
bios additionally.
3. Introduce __submit_merged_bio to submit the merged bio.
4. Change f2fs_readpage to f2fs_submit_page_bio.
5. Introduce f2fs_submit_page_mbio to integrate previous submit_read_page and
submit_write_page.
Reviewed-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Reviewed-by: Chao Yu <chao2.yu@samsung.com >
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
The recover_orphan_inodes() returns no error all the time, so we don't need to
check its errors.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
[Jaegeuk Kim: add description]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Because we will write node summaries when do_checkpoint with umount flag,
our number of max orphan blocks should minus NR_CURSEG_NODE_TYPE additional.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Shu Tan <shu.tan@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch fixes some bit overflows by the shift operations.
Dan Carpenter reported potential bugs on bit overflows as follows.
fs/f2fs/segment.c:910 submit_write_page()
warn: should 'blk_addr << ((sbi)->log_blocksize - 9)' be a 64 bit type?
fs/f2fs/checkpoint.c:429 get_valid_checkpoint()
warn: should '1 << ()' be a 64 bit type?
fs/f2fs/data.c:408 f2fs_readpage()
warn: should 'blk_addr << ((sbi)->log_blocksize - 9)' be a 64 bit type?
fs/f2fs/data.c:457 submit_read_page()
warn: should 'blk_addr << ((sbi)->log_blocksize - 9)' be a 64 bit type?
fs/f2fs/data.c:525 get_data_block_ro()
warn: should 'i << blkbits' be a 64 bit type?
Bug-Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Fix a potential out of range issue introduced by commit:
22fb72225a
f2fs: simplify write_orphan_inodes for better readable
Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Let's send REQ_META or REQ_PRIO when reading meta area such as NAT/SIT
etc.
Signed-off-by: Changman Lee <cm224.lee@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Simplify write_orphan_inodes for better readable. Because we hold the
orphan_inode_mutex, so it's safe to use list_for_each_entry instead of
list_for_each_safe.
Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
use genernal method supported by kernel
o changes from v1
If any waiter exists at end io, wake up it.
Signed-off-by: Changman Lee <cm224.lee@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
If you want to remove unnecessary BUG_ONs, you can just turn off F2FS_CHECK_FS
in your kernel config.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Introduce the unfailed version of kmem_cache_alloc named f2fs_kmem_cache_alloc
to hide the retry routine and make the code a bit cleaner.
v2:
Fix the wrong use of 'retry' tag pointed out by Gao feng.
Use more neat code to remove redundant tag suggested by Haicheng Li.
Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch enhances the recovery routine not to write any data/node/meta until
its completion.
If any writes are sent to the disk, it could contaminate the written history
that will be used for further recovery.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Previously, do_checkpoint() will call congestion_wait() for waiting the pages
(previous submitted node/meta/data pages) to be written back.
Because congestion_wait() will set a regular period (e.g. HZ / 50 ) for waiting, and
no additional wake up mechanism was introduced if IO ends up before regular period costed.
Yuan Zhong found there is a situation that after the pages have been written back,
but the checkpoint thread still wait for congestion_wait to exit.
So here we store checkpoint task into f2fs_sb when doing checkpoint, it'll wait for IO completes
if there's IO going on, and in the end IO path, wake up checkpoint task when IO ends up.
Thanks to Yuan Zhong's pre work about this problem.
Reported-by: Yuan Zhong <yuan.mark.zhong@samsung.com>
Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
The fs_locks is used to block other ops(ex, recovery) when doing checkpoint.
And each other operate routine(besides checkpoint) needs to acquire a fs_lock,
there is a terrible problem here, if these are too many concurrency threads acquiring
fs_lock, so that they will block each other and may lead to some performance problem,
but this is not the phenomenon we want to see.
Though there are some optimization patches introduced to enhance the usage of fs_lock,
but the thorough solution is using a *rw_sem* to replace the fs_lock.
Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block
other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other,
this can avoid the problem described above completely.
Because of the weakness of rw_sem, the above change may introduce a potential problem
that the checkpoint thread might get starved if other threads are intensively locking
the read semaphore for I/O.(Pointed out by Xu Jin)
In order to avoid this, a wait_list is introduced, the appending read semaphore ops
will be dropped into the wait_list if checkpoint thread is waiting for write semaphore,
and will be waked up when checkpoint thread gives up write semaphore.
Thanks to Kim's previous review and test, and will be very glad to see other guys'
performance tests about this patch.
V2:
-fix the potential starvation problem.
-use more suitable func name suggested by Xu Jin.
Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
[Jaegeuk Kim: adjust minor coding standard]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Accounting errors from buggy code calling the acquire/release/remove
orphan inode interfaces can cause n_orphans to underflow, which will
then cause acquire_orphan_inode() to return -ENOSPC on the next
operation. This commit guards against that condition.
Signed-off-by: Russ Knize <rknize@motorola.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch fixes mishandling of the sbi->n_orphans variable.
If users request lots of f2fs_unlink(), check_orphan_space() could be contended.
In such the case, sbi->n_orphans can be read incorrectly so that f2fs_unlink()
would fall into the wrong state which results in the failure of
add_orphan_inode().
So, let's increment sbi->n_orphans virtually prior to the actual orphan inode
stuffs. After that, let's release sbi->n_orphans by calling release_orphan_inode
or remove_orphan_inode.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
As we remove the target single node, so list_for_each is enought, in order to
clean up, we use list_for_each_entry instead.
Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
While calculating CRC for the checkpoint block, we use __u32, but when storing
the crc value to the disk, we use __le32.
Let's fix the inconsistency.
Reported-and-Tested-by: Oded Gabbay <ogabbay@advaoptical.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
It is possible that iput is skipped after iget during the recovery.
In recover_dentry(),
dir = f2fs_iget();
...
if (de && inode->i_ino == le32_to_cpu(de->ino))
goto out;
In this case, this dir is not able to be added in dirty_dir_inode_list.
The actual linking is done only when set_page_dirty() is called.
So let's add this newly got inode into the list explicitly, and put it at the
end of the recovery routine.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
- iget/iput flow in the dentry recovery process
1. *dir* = f2fs_iget
2. set FI_DELAY_IPUT to *dir*
3. add *dir* to the dirty_dir_list
- __f2fs_add_link
- recover_dentry)
4. iput *dir* by remove_dirty_dir_inode
- sync_dirty_dir_inodes
- write_chekcpoint
If *dir*'s i_count is not 1 (i.e., root dir), remove_dirty_dir_inode is called
later and then iput is triggered again due to the FI_DELAY_IPUT flag.
So, let's unset the flag properly once iput is triggered.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
