linux/fs/f2fs/segment.c

2387 lines
61 KiB
C
Raw Normal View History

/*
* fs/f2fs/segment.c
*
* Copyright (c) 2012 Samsung Electronics Co., Ltd.
* http://www.samsung.com/
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*/
#include <linux/fs.h>
#include <linux/f2fs_fs.h>
#include <linux/bio.h>
#include <linux/blkdev.h>
#include <linux/prefetch.h>
#include <linux/kthread.h>
#include <linux/vmalloc.h>
#include <linux/swap.h>
#include "f2fs.h"
#include "segment.h"
#include "node.h"
#include "trace.h"
#include <trace/events/f2fs.h>
#define __reverse_ffz(x) __reverse_ffs(~(x))
static struct kmem_cache *discard_entry_slab;
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
static struct kmem_cache *sit_entry_set_slab;
static struct kmem_cache *inmem_entry_slab;
/*
* __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
* MSB and LSB are reversed in a byte by f2fs_set_bit.
*/
static inline unsigned long __reverse_ffs(unsigned long word)
{
int num = 0;
#if BITS_PER_LONG == 64
if ((word & 0xffffffff) == 0) {
num += 32;
word >>= 32;
}
#endif
if ((word & 0xffff) == 0) {
num += 16;
word >>= 16;
}
if ((word & 0xff) == 0) {
num += 8;
word >>= 8;
}
if ((word & 0xf0) == 0)
num += 4;
else
word >>= 4;
if ((word & 0xc) == 0)
num += 2;
else
word >>= 2;
if ((word & 0x2) == 0)
num += 1;
return num;
}
/*
* __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
* f2fs_set_bit makes MSB and LSB reversed in a byte.
* Example:
* LSB <--> MSB
* f2fs_set_bit(0, bitmap) => 0000 0001
* f2fs_set_bit(7, bitmap) => 1000 0000
*/
static unsigned long __find_rev_next_bit(const unsigned long *addr,
unsigned long size, unsigned long offset)
{
while (!f2fs_test_bit(offset, (unsigned char *)addr))
offset++;
if (offset > size)
offset = size;
return offset;
#if 0
const unsigned long *p = addr + BIT_WORD(offset);
unsigned long result = offset & ~(BITS_PER_LONG - 1);
unsigned long tmp;
unsigned long mask, submask;
unsigned long quot, rest;
if (offset >= size)
return size;
size -= result;
offset %= BITS_PER_LONG;
if (!offset)
goto aligned;
tmp = *(p++);
quot = (offset >> 3) << 3;
rest = offset & 0x7;
mask = ~0UL << quot;
submask = (unsigned char)(0xff << rest) >> rest;
submask <<= quot;
mask &= submask;
tmp &= mask;
if (size < BITS_PER_LONG)
goto found_first;
if (tmp)
goto found_middle;
size -= BITS_PER_LONG;
result += BITS_PER_LONG;
aligned:
while (size & ~(BITS_PER_LONG-1)) {
tmp = *(p++);
if (tmp)
goto found_middle;
result += BITS_PER_LONG;
size -= BITS_PER_LONG;
}
if (!size)
return result;
tmp = *p;
found_first:
tmp &= (~0UL >> (BITS_PER_LONG - size));
if (tmp == 0UL) /* Are any bits set? */
return result + size; /* Nope. */
found_middle:
return result + __reverse_ffs(tmp);
#endif
}
static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
unsigned long size, unsigned long offset)
{
while (f2fs_test_bit(offset, (unsigned char *)addr))
offset++;
if (offset > size)
offset = size;
return offset;
#if 0
const unsigned long *p = addr + BIT_WORD(offset);
unsigned long result = offset & ~(BITS_PER_LONG - 1);
unsigned long tmp;
unsigned long mask, submask;
unsigned long quot, rest;
if (offset >= size)
return size;
size -= result;
offset %= BITS_PER_LONG;
if (!offset)
goto aligned;
tmp = *(p++);
quot = (offset >> 3) << 3;
rest = offset & 0x7;
mask = ~(~0UL << quot);
submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
submask <<= quot;
mask += submask;
tmp |= mask;
if (size < BITS_PER_LONG)
goto found_first;
if (~tmp)
goto found_middle;
size -= BITS_PER_LONG;
result += BITS_PER_LONG;
aligned:
while (size & ~(BITS_PER_LONG - 1)) {
tmp = *(p++);
if (~tmp)
goto found_middle;
result += BITS_PER_LONG;
size -= BITS_PER_LONG;
}
if (!size)
return result;
tmp = *p;
found_first:
tmp |= ~0UL << size;
if (tmp == ~0UL) /* Are any bits zero? */
return result + size; /* Nope. */
found_middle:
return result + __reverse_ffz(tmp);
#endif
}
void register_inmem_page(struct inode *inode, struct page *page)
{
struct f2fs_inode_info *fi = F2FS_I(inode);
struct inmem_pages *new;
int err;
SetPagePrivate(page);
f2fs_trace_pid(page);
new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
/* add atomic page indices to the list */
new->page = page;
INIT_LIST_HEAD(&new->list);
retry:
/* increase reference count with clean state */
mutex_lock(&fi->inmem_lock);
err = radix_tree_insert(&fi->inmem_root, page->index, new);
if (err == -EEXIST) {
mutex_unlock(&fi->inmem_lock);
kmem_cache_free(inmem_entry_slab, new);
return;
} else if (err) {
mutex_unlock(&fi->inmem_lock);
goto retry;
}
get_page(page);
list_add_tail(&new->list, &fi->inmem_pages);
inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
mutex_unlock(&fi->inmem_lock);
trace_f2fs_register_inmem_page(page, INMEM);
}
void commit_inmem_pages(struct inode *inode, bool abort)
{
struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
struct f2fs_inode_info *fi = F2FS_I(inode);
struct inmem_pages *cur, *tmp;
bool submit_bio = false;
struct f2fs_io_info fio = {
.sbi = sbi,
.type = DATA,
.rw = WRITE_SYNC | REQ_PRIO,
.encrypted_page = NULL,
};
/*
* The abort is true only when f2fs_evict_inode is called.
* Basically, the f2fs_evict_inode doesn't produce any data writes, so
* that we don't need to call f2fs_balance_fs.
* Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
* inode becomes free by iget_locked in f2fs_iget.
*/
if (!abort) {
f2fs_balance_fs(sbi);
f2fs_lock_op(sbi);
}
mutex_lock(&fi->inmem_lock);
list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
if (!abort) {
lock_page(cur->page);
if (cur->page->mapping == inode->i_mapping) {
f2fs_wait_on_page_writeback(cur->page, DATA);
if (clear_page_dirty_for_io(cur->page))
inode_dec_dirty_pages(inode);
trace_f2fs_commit_inmem_page(cur->page, INMEM);
fio.page = cur->page;
do_write_data_page(&fio);
submit_bio = true;
}
f2fs_put_page(cur->page, 1);
} else {
trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP);
put_page(cur->page);
}
radix_tree_delete(&fi->inmem_root, cur->page->index);
list_del(&cur->list);
kmem_cache_free(inmem_entry_slab, cur);
dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
}
mutex_unlock(&fi->inmem_lock);
if (!abort) {
f2fs_unlock_op(sbi);
if (submit_bio)
f2fs_submit_merged_bio(sbi, DATA, WRITE);
}
}
/*
* This function balances dirty node and dentry pages.
* In addition, it controls garbage collection.
*/
void f2fs_balance_fs(struct f2fs_sb_info *sbi)
{
/*
* We should do GC or end up with checkpoint, if there are so many dirty
* dir/node pages without enough free segments.
*/
if (has_not_enough_free_secs(sbi, 0)) {
mutex_lock(&sbi->gc_mutex);
f2fs_gc(sbi);
}
}
void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
{
/* try to shrink extent cache when there is no enough memory */
f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
/* check the # of cached NAT entries and prefree segments */
if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
excess_prefree_segs(sbi) ||
!available_free_memory(sbi, INO_ENTRIES))
f2fs_sync_fs(sbi->sb, true);
}
static int issue_flush_thread(void *data)
{
struct f2fs_sb_info *sbi = data;
struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
wait_queue_head_t *q = &fcc->flush_wait_queue;
repeat:
if (kthread_should_stop())
return 0;
if (!llist_empty(&fcc->issue_list)) {
struct bio *bio = bio_alloc(GFP_NOIO, 0);
struct flush_cmd *cmd, *next;
int ret;
fcc->dispatch_list = llist_del_all(&fcc->issue_list);
fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
bio->bi_bdev = sbi->sb->s_bdev;
ret = submit_bio_wait(WRITE_FLUSH, bio);
llist_for_each_entry_safe(cmd, next,
fcc->dispatch_list, llnode) {
cmd->ret = ret;
complete(&cmd->wait);
}
bio_put(bio);
fcc->dispatch_list = NULL;
}
wait_event_interruptible(*q,
kthread_should_stop() || !llist_empty(&fcc->issue_list));
goto repeat;
}
int f2fs_issue_flush(struct f2fs_sb_info *sbi)
{
struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
struct flush_cmd cmd;
trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
test_opt(sbi, FLUSH_MERGE));
if (test_opt(sbi, NOBARRIER))
return 0;
if (!test_opt(sbi, FLUSH_MERGE))
return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
init_completion(&cmd.wait);
llist_add(&cmd.llnode, &fcc->issue_list);
if (!fcc->dispatch_list)
wake_up(&fcc->flush_wait_queue);
wait_for_completion(&cmd.wait);
return cmd.ret;
}
int create_flush_cmd_control(struct f2fs_sb_info *sbi)
{
dev_t dev = sbi->sb->s_bdev->bd_dev;
struct flush_cmd_control *fcc;
int err = 0;
fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
if (!fcc)
return -ENOMEM;
init_waitqueue_head(&fcc->flush_wait_queue);
init_llist_head(&fcc->issue_list);
SM_I(sbi)->cmd_control_info = fcc;
fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
if (IS_ERR(fcc->f2fs_issue_flush)) {
err = PTR_ERR(fcc->f2fs_issue_flush);
kfree(fcc);
SM_I(sbi)->cmd_control_info = NULL;
return err;
}
return err;
}
void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
{
struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
if (fcc && fcc->f2fs_issue_flush)
kthread_stop(fcc->f2fs_issue_flush);
kfree(fcc);
SM_I(sbi)->cmd_control_info = NULL;
}
static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
enum dirty_type dirty_type)
{
struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
/* need not be added */
if (IS_CURSEG(sbi, segno))
return;
if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
dirty_i->nr_dirty[dirty_type]++;
if (dirty_type == DIRTY) {
struct seg_entry *sentry = get_seg_entry(sbi, segno);
enum dirty_type t = sentry->type;
if (unlikely(t >= DIRTY)) {
f2fs_bug_on(sbi, 1);
return;
}
if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
dirty_i->nr_dirty[t]++;
}
}
static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
enum dirty_type dirty_type)
{
struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
dirty_i->nr_dirty[dirty_type]--;
if (dirty_type == DIRTY) {
struct seg_entry *sentry = get_seg_entry(sbi, segno);
enum dirty_type t = sentry->type;
if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
dirty_i->nr_dirty[t]--;
if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
clear_bit(GET_SECNO(sbi, segno),
dirty_i->victim_secmap);
}
}
/*
* Should not occur error such as -ENOMEM.
* Adding dirty entry into seglist is not critical operation.
* If a given segment is one of current working segments, it won't be added.
*/
static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
{
struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
unsigned short valid_blocks;
if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
return;
mutex_lock(&dirty_i->seglist_lock);
valid_blocks = get_valid_blocks(sbi, segno, 0);
if (valid_blocks == 0) {
__locate_dirty_segment(sbi, segno, PRE);
__remove_dirty_segment(sbi, segno, DIRTY);
} else if (valid_blocks < sbi->blocks_per_seg) {
__locate_dirty_segment(sbi, segno, DIRTY);
} else {
/* Recovery routine with SSR needs this */
__remove_dirty_segment(sbi, segno, DIRTY);
}
mutex_unlock(&dirty_i->seglist_lock);
}
f2fs: avoid to conduct roll-forward due to the remained garbage blocks The f2fs always scans the next chain of direct node blocks. But some garbage blocks are able to be remained due to no discard support or SSR triggers. This occasionally wreaks recovering wrong inodes that were used or BUG_ONs due to reallocating node ids as follows. When mount this f2fs image: http://linuxtesting.org/downloads/f2fs_fault_image.zip BUG_ON is triggered in f2fs driver (messages below are generated on kernel 3.13.2; for other kernels output is similar): kernel BUG at fs/f2fs/node.c:215! Call Trace: [<ffffffffa032ebad>] recover_inode_page+0x1fd/0x3e0 [f2fs] [<ffffffff811446e7>] ? __lock_page+0x67/0x70 [<ffffffff81089990>] ? autoremove_wake_function+0x50/0x50 [<ffffffffa0337788>] recover_fsync_data+0x1398/0x15d0 [f2fs] [<ffffffff812b9e5c>] ? selinux_d_instantiate+0x1c/0x20 [<ffffffff811cb20b>] ? d_instantiate+0x5b/0x80 [<ffffffffa0321044>] f2fs_fill_super+0xb04/0xbf0 [f2fs] [<ffffffff811b861e>] ? mount_bdev+0x7e/0x210 [<ffffffff811b8769>] mount_bdev+0x1c9/0x210 [<ffffffffa0320540>] ? validate_superblock+0x210/0x210 [f2fs] [<ffffffffa031cf8d>] f2fs_mount+0x1d/0x30 [f2fs] [<ffffffff811b9497>] mount_fs+0x47/0x1c0 [<ffffffff81166e00>] ? __alloc_percpu+0x10/0x20 [<ffffffff811d4032>] vfs_kern_mount+0x72/0x110 [<ffffffff811d6763>] do_mount+0x493/0x910 [<ffffffff811615cb>] ? strndup_user+0x5b/0x80 [<ffffffff811d6c70>] SyS_mount+0x90/0xe0 [<ffffffff8166f8d9>] system_call_fastpath+0x16/0x1b Found by Linux File System Verification project (linuxtesting.org). Reported-by: Andrey Tsyvarev <tsyvarev@ispras.ru> Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-04-15 04:57:55 +00:00
static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
block_t blkstart, block_t blklen)
{
sector_t start = SECTOR_FROM_BLOCK(blkstart);
sector_t len = SECTOR_FROM_BLOCK(blklen);
struct seg_entry *se;
unsigned int offset;
block_t i;
for (i = blkstart; i < blkstart + blklen; i++) {
se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
offset = GET_BLKOFF_FROM_SEG0(sbi, i);
if (!f2fs_test_and_set_bit(offset, se->discard_map))
sbi->discard_blks--;
}
trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
f2fs: avoid to conduct roll-forward due to the remained garbage blocks The f2fs always scans the next chain of direct node blocks. But some garbage blocks are able to be remained due to no discard support or SSR triggers. This occasionally wreaks recovering wrong inodes that were used or BUG_ONs due to reallocating node ids as follows. When mount this f2fs image: http://linuxtesting.org/downloads/f2fs_fault_image.zip BUG_ON is triggered in f2fs driver (messages below are generated on kernel 3.13.2; for other kernels output is similar): kernel BUG at fs/f2fs/node.c:215! Call Trace: [<ffffffffa032ebad>] recover_inode_page+0x1fd/0x3e0 [f2fs] [<ffffffff811446e7>] ? __lock_page+0x67/0x70 [<ffffffff81089990>] ? autoremove_wake_function+0x50/0x50 [<ffffffffa0337788>] recover_fsync_data+0x1398/0x15d0 [f2fs] [<ffffffff812b9e5c>] ? selinux_d_instantiate+0x1c/0x20 [<ffffffff811cb20b>] ? d_instantiate+0x5b/0x80 [<ffffffffa0321044>] f2fs_fill_super+0xb04/0xbf0 [f2fs] [<ffffffff811b861e>] ? mount_bdev+0x7e/0x210 [<ffffffff811b8769>] mount_bdev+0x1c9/0x210 [<ffffffffa0320540>] ? validate_superblock+0x210/0x210 [f2fs] [<ffffffffa031cf8d>] f2fs_mount+0x1d/0x30 [f2fs] [<ffffffff811b9497>] mount_fs+0x47/0x1c0 [<ffffffff81166e00>] ? __alloc_percpu+0x10/0x20 [<ffffffff811d4032>] vfs_kern_mount+0x72/0x110 [<ffffffff811d6763>] do_mount+0x493/0x910 [<ffffffff811615cb>] ? strndup_user+0x5b/0x80 [<ffffffff811d6c70>] SyS_mount+0x90/0xe0 [<ffffffff8166f8d9>] system_call_fastpath+0x16/0x1b Found by Linux File System Verification project (linuxtesting.org). Reported-by: Andrey Tsyvarev <tsyvarev@ispras.ru> Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-04-15 04:57:55 +00:00
return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
}
void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
f2fs: avoid to conduct roll-forward due to the remained garbage blocks The f2fs always scans the next chain of direct node blocks. But some garbage blocks are able to be remained due to no discard support or SSR triggers. This occasionally wreaks recovering wrong inodes that were used or BUG_ONs due to reallocating node ids as follows. When mount this f2fs image: http://linuxtesting.org/downloads/f2fs_fault_image.zip BUG_ON is triggered in f2fs driver (messages below are generated on kernel 3.13.2; for other kernels output is similar): kernel BUG at fs/f2fs/node.c:215! Call Trace: [<ffffffffa032ebad>] recover_inode_page+0x1fd/0x3e0 [f2fs] [<ffffffff811446e7>] ? __lock_page+0x67/0x70 [<ffffffff81089990>] ? autoremove_wake_function+0x50/0x50 [<ffffffffa0337788>] recover_fsync_data+0x1398/0x15d0 [f2fs] [<ffffffff812b9e5c>] ? selinux_d_instantiate+0x1c/0x20 [<ffffffff811cb20b>] ? d_instantiate+0x5b/0x80 [<ffffffffa0321044>] f2fs_fill_super+0xb04/0xbf0 [f2fs] [<ffffffff811b861e>] ? mount_bdev+0x7e/0x210 [<ffffffff811b8769>] mount_bdev+0x1c9/0x210 [<ffffffffa0320540>] ? validate_superblock+0x210/0x210 [f2fs] [<ffffffffa031cf8d>] f2fs_mount+0x1d/0x30 [f2fs] [<ffffffff811b9497>] mount_fs+0x47/0x1c0 [<ffffffff81166e00>] ? __alloc_percpu+0x10/0x20 [<ffffffff811d4032>] vfs_kern_mount+0x72/0x110 [<ffffffff811d6763>] do_mount+0x493/0x910 [<ffffffff811615cb>] ? strndup_user+0x5b/0x80 [<ffffffff811d6c70>] SyS_mount+0x90/0xe0 [<ffffffff8166f8d9>] system_call_fastpath+0x16/0x1b Found by Linux File System Verification project (linuxtesting.org). Reported-by: Andrey Tsyvarev <tsyvarev@ispras.ru> Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-04-15 04:57:55 +00:00
{
int err = -ENOTSUPP;
if (test_opt(sbi, DISCARD)) {
struct seg_entry *se = get_seg_entry(sbi,
GET_SEGNO(sbi, blkaddr));
unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
if (f2fs_test_bit(offset, se->discard_map))
return;
err = f2fs_issue_discard(sbi, blkaddr, 1);
}
if (err) {
f2fs: avoid to conduct roll-forward due to the remained garbage blocks The f2fs always scans the next chain of direct node blocks. But some garbage blocks are able to be remained due to no discard support or SSR triggers. This occasionally wreaks recovering wrong inodes that were used or BUG_ONs due to reallocating node ids as follows. When mount this f2fs image: http://linuxtesting.org/downloads/f2fs_fault_image.zip BUG_ON is triggered in f2fs driver (messages below are generated on kernel 3.13.2; for other kernels output is similar): kernel BUG at fs/f2fs/node.c:215! Call Trace: [<ffffffffa032ebad>] recover_inode_page+0x1fd/0x3e0 [f2fs] [<ffffffff811446e7>] ? __lock_page+0x67/0x70 [<ffffffff81089990>] ? autoremove_wake_function+0x50/0x50 [<ffffffffa0337788>] recover_fsync_data+0x1398/0x15d0 [f2fs] [<ffffffff812b9e5c>] ? selinux_d_instantiate+0x1c/0x20 [<ffffffff811cb20b>] ? d_instantiate+0x5b/0x80 [<ffffffffa0321044>] f2fs_fill_super+0xb04/0xbf0 [f2fs] [<ffffffff811b861e>] ? mount_bdev+0x7e/0x210 [<ffffffff811b8769>] mount_bdev+0x1c9/0x210 [<ffffffffa0320540>] ? validate_superblock+0x210/0x210 [f2fs] [<ffffffffa031cf8d>] f2fs_mount+0x1d/0x30 [f2fs] [<ffffffff811b9497>] mount_fs+0x47/0x1c0 [<ffffffff81166e00>] ? __alloc_percpu+0x10/0x20 [<ffffffff811d4032>] vfs_kern_mount+0x72/0x110 [<ffffffff811d6763>] do_mount+0x493/0x910 [<ffffffff811615cb>] ? strndup_user+0x5b/0x80 [<ffffffff811d6c70>] SyS_mount+0x90/0xe0 [<ffffffff8166f8d9>] system_call_fastpath+0x16/0x1b Found by Linux File System Verification project (linuxtesting.org). Reported-by: Andrey Tsyvarev <tsyvarev@ispras.ru> Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-04-15 04:57:55 +00:00
struct page *page = grab_meta_page(sbi, blkaddr);
/* zero-filled page */
set_page_dirty(page);
f2fs_put_page(page, 1);
}
}
static void __add_discard_entry(struct f2fs_sb_info *sbi,
struct cp_control *cpc, struct seg_entry *se,
unsigned int start, unsigned int end)
{
struct list_head *head = &SM_I(sbi)->discard_list;
struct discard_entry *new, *last;
if (!list_empty(head)) {
last = list_last_entry(head, struct discard_entry, list);
if (START_BLOCK(sbi, cpc->trim_start) + start ==
last->blkaddr + last->len) {
last->len += end - start;
goto done;
}
}
new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
INIT_LIST_HEAD(&new->list);
new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
new->len = end - start;
list_add_tail(&new->list, head);
done:
SM_I(sbi)->nr_discards += end - start;
cpc->trimmed += end - start;
}
static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
{
int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
int max_blocks = sbi->blocks_per_seg;
struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
unsigned long *discard_map = (unsigned long *)se->discard_map;
unsigned long *dmap = SIT_I(sbi)->tmp_map;
unsigned int start = 0, end = -1;
bool force = (cpc->reason == CP_DISCARD);
int i;
if (se->valid_blocks == max_blocks)
return;
if (!force) {
if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
return;
}
/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
for (i = 0; i < entries; i++)
dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
start = __find_rev_next_bit(dmap, max_blocks, end + 1);
if (start >= max_blocks)
break;
end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
__add_discard_entry(sbi, cpc, se, start, end);
}
}
void release_discard_addrs(struct f2fs_sb_info *sbi)
{
struct list_head *head = &(SM_I(sbi)->discard_list);
struct discard_entry *entry, *this;
/* drop caches */
list_for_each_entry_safe(entry, this, head, list) {
list_del(&entry->list);
kmem_cache_free(discard_entry_slab, entry);
}
}
/*
* Should call clear_prefree_segments after checkpoint is done.
*/
static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
{
struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
unsigned int segno;
mutex_lock(&dirty_i->seglist_lock);
for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
__set_test_and_free(sbi, segno);
mutex_unlock(&dirty_i->seglist_lock);
}
void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
{
struct list_head *head = &(SM_I(sbi)->discard_list);
struct discard_entry *entry, *this;
struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
unsigned int start = 0, end = -1;
mutex_lock(&dirty_i->seglist_lock);
while (1) {
int i;
start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
if (start >= MAIN_SEGS(sbi))
break;
end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
start + 1);
for (i = start; i < end; i++)
clear_bit(i, prefree_map);
dirty_i->nr_dirty[PRE] -= end - start;
if (!test_opt(sbi, DISCARD))
continue;
f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
(end - start) << sbi->log_blocks_per_seg);
}
mutex_unlock(&dirty_i->seglist_lock);
/* send small discards */
list_for_each_entry_safe(entry, this, head, list) {
if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
goto skip;
f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
skip:
list_del(&entry->list);
SM_I(sbi)->nr_discards -= entry->len;
kmem_cache_free(discard_entry_slab, entry);
}
}
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
{
struct sit_info *sit_i = SIT_I(sbi);
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
sit_i->dirty_sentries++;
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
return false;
}
return true;
}
static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
unsigned int segno, int modified)
{
struct seg_entry *se = get_seg_entry(sbi, segno);
se->type = type;
if (modified)
__mark_sit_entry_dirty(sbi, segno);
}
static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
{
struct seg_entry *se;
unsigned int segno, offset;
long int new_vblocks;
segno = GET_SEGNO(sbi, blkaddr);
se = get_seg_entry(sbi, segno);
new_vblocks = se->valid_blocks + del;
offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
(new_vblocks > sbi->blocks_per_seg)));
se->valid_blocks = new_vblocks;
se->mtime = get_mtime(sbi);
SIT_I(sbi)->max_mtime = se->mtime;
/* Update valid block bitmap */
if (del > 0) {
if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
f2fs_bug_on(sbi, 1);
if (!f2fs_test_and_set_bit(offset, se->discard_map))
sbi->discard_blks--;
} else {
if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
f2fs_bug_on(sbi, 1);
if (f2fs_test_and_clear_bit(offset, se->discard_map))
sbi->discard_blks++;
}
if (!f2fs_test_bit(offset, se->ckpt_valid_map))
se->ckpt_valid_blocks += del;
__mark_sit_entry_dirty(sbi, segno);
/* update total number of valid blocks to be written in ckpt area */
SIT_I(sbi)->written_valid_blocks += del;
if (sbi->segs_per_sec > 1)
get_sec_entry(sbi, segno)->valid_blocks += del;
}
void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
{
update_sit_entry(sbi, new, 1);
if (GET_SEGNO(sbi, old) != NULL_SEGNO)
update_sit_entry(sbi, old, -1);
locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
}
void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
{
unsigned int segno = GET_SEGNO(sbi, addr);
struct sit_info *sit_i = SIT_I(sbi);
f2fs_bug_on(sbi, addr == NULL_ADDR);
if (addr == NEW_ADDR)
return;
/* add it into sit main buffer */
mutex_lock(&sit_i->sentry_lock);
update_sit_entry(sbi, addr, -1);
/* add it into dirty seglist */
locate_dirty_segment(sbi, segno);
mutex_unlock(&sit_i->sentry_lock);
}
/*
* This function should be resided under the curseg_mutex lock
*/
static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
struct f2fs_summary *sum)
{
struct curseg_info *curseg = CURSEG_I(sbi, type);
void *addr = curseg->sum_blk;
addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
memcpy(addr, sum, sizeof(struct f2fs_summary));
}
/*
* Calculate the number of current summary pages for writing
*/
int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
{
int valid_sum_count = 0;
int i, sum_in_page;
for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
if (sbi->ckpt->alloc_type[i] == SSR)
valid_sum_count += sbi->blocks_per_seg;
else {
if (for_ra)
valid_sum_count += le16_to_cpu(
F2FS_CKPT(sbi)->cur_data_blkoff[i]);
else
valid_sum_count += curseg_blkoff(sbi, i);
}
}
sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
SUM_FOOTER_SIZE) / SUMMARY_SIZE;
if (valid_sum_count <= sum_in_page)
return 1;
else if ((valid_sum_count - sum_in_page) <=
(PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
return 2;
return 3;
}
/*
* Caller should put this summary page
*/
struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
{
return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
}
static void write_sum_page(struct f2fs_sb_info *sbi,
struct f2fs_summary_block *sum_blk, block_t blk_addr)
{
struct page *page = grab_meta_page(sbi, blk_addr);
void *kaddr = page_address(page);
memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
set_page_dirty(page);
f2fs_put_page(page, 1);
}
static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
{
struct curseg_info *curseg = CURSEG_I(sbi, type);
unsigned int segno = curseg->segno + 1;
struct free_segmap_info *free_i = FREE_I(sbi);
if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
return !test_bit(segno, free_i->free_segmap);
return 0;
}
/*
* Find a new segment from the free segments bitmap to right order
* This function should be returned with success, otherwise BUG
*/
static void get_new_segment(struct f2fs_sb_info *sbi,
unsigned int *newseg, bool new_sec, int dir)
{
struct free_segmap_info *free_i = FREE_I(sbi);
unsigned int segno, secno, zoneno;
unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
unsigned int hint = *newseg / sbi->segs_per_sec;
unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
unsigned int left_start = hint;
bool init = true;
int go_left = 0;
int i;
spin_lock(&free_i->segmap_lock);
if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
segno = find_next_zero_bit(free_i->free_segmap,
MAIN_SEGS(sbi), *newseg + 1);
if (segno - *newseg < sbi->segs_per_sec -
(*newseg % sbi->segs_per_sec))
goto got_it;
}
find_other_zone:
secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
if (secno >= MAIN_SECS(sbi)) {
if (dir == ALLOC_RIGHT) {
secno = find_next_zero_bit(free_i->free_secmap,
MAIN_SECS(sbi), 0);
f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
} else {
go_left = 1;
left_start = hint - 1;
}
}
if (go_left == 0)
goto skip_left;
while (test_bit(left_start, free_i->free_secmap)) {
if (left_start > 0) {
left_start--;
continue;
}
left_start = find_next_zero_bit(free_i->free_secmap,
MAIN_SECS(sbi), 0);
f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
break;
}
secno = left_start;
skip_left:
hint = secno;
segno = secno * sbi->segs_per_sec;
zoneno = secno / sbi->secs_per_zone;
/* give up on finding another zone */
if (!init)
goto got_it;
if (sbi->secs_per_zone == 1)
goto got_it;
if (zoneno == old_zoneno)
goto got_it;
if (dir == ALLOC_LEFT) {
if (!go_left && zoneno + 1 >= total_zones)
goto got_it;
if (go_left && zoneno == 0)
goto got_it;
}
for (i = 0; i < NR_CURSEG_TYPE; i++)
if (CURSEG_I(sbi, i)->zone == zoneno)
break;
if (i < NR_CURSEG_TYPE) {
/* zone is in user, try another */
if (go_left)
hint = zoneno * sbi->secs_per_zone - 1;
else if (zoneno + 1 >= total_zones)
hint = 0;
else
hint = (zoneno + 1) * sbi->secs_per_zone;
init = false;
goto find_other_zone;
}
got_it:
/* set it as dirty segment in free segmap */
f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
__set_inuse(sbi, segno);
*newseg = segno;
spin_unlock(&free_i->segmap_lock);
}
static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
{
struct curseg_info *curseg = CURSEG_I(sbi, type);
struct summary_footer *sum_footer;
curseg->segno = curseg->next_segno;
curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
curseg->next_blkoff = 0;
curseg->next_segno = NULL_SEGNO;
sum_footer = &(curseg->sum_blk->footer);
memset(sum_footer, 0, sizeof(struct summary_footer));
if (IS_DATASEG(type))
SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
if (IS_NODESEG(type))
SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
__set_sit_entry_type(sbi, type, curseg->segno, modified);
}
/*
* Allocate a current working segment.
* This function always allocates a free segment in LFS manner.
*/
static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
{
struct curseg_info *curseg = CURSEG_I(sbi, type);
unsigned int segno = curseg->segno;
int dir = ALLOC_LEFT;
write_sum_page(sbi, curseg->sum_blk,
GET_SUM_BLOCK(sbi, segno));
if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
dir = ALLOC_RIGHT;
if (test_opt(sbi, NOHEAP))
dir = ALLOC_RIGHT;
get_new_segment(sbi, &segno, new_sec, dir);
curseg->next_segno = segno;
reset_curseg(sbi, type, 1);
curseg->alloc_type = LFS;
}
static void __next_free_blkoff(struct f2fs_sb_info *sbi,
struct curseg_info *seg, block_t start)
{
struct seg_entry *se = get_seg_entry(sbi, seg->segno);
int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
unsigned long *target_map = SIT_I(sbi)->tmp_map;
unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
int i, pos;
for (i = 0; i < entries; i++)
target_map[i] = ckpt_map[i] | cur_map[i];
pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
seg->next_blkoff = pos;
}
/*
* If a segment is written by LFS manner, next block offset is just obtained
* by increasing the current block offset. However, if a segment is written by
* SSR manner, next block offset obtained by calling __next_free_blkoff
*/
static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
struct curseg_info *seg)
{
if (seg->alloc_type == SSR)
__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
else
seg->next_blkoff++;
}
/*
* This function always allocates a used segment(from dirty seglist) by SSR
* manner, so it should recover the existing segment information of valid blocks
*/
static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
{
struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
struct curseg_info *curseg = CURSEG_I(sbi, type);
unsigned int new_segno = curseg->next_segno;
struct f2fs_summary_block *sum_node;
struct page *sum_page;
write_sum_page(sbi, curseg->sum_blk,
GET_SUM_BLOCK(sbi, curseg->segno));
__set_test_and_inuse(sbi, new_segno);
mutex_lock(&dirty_i->seglist_lock);
__remove_dirty_segment(sbi, new_segno, PRE);
__remove_dirty_segment(sbi, new_segno, DIRTY);
mutex_unlock(&dirty_i->seglist_lock);
reset_curseg(sbi, type, 1);
curseg->alloc_type = SSR;
__next_free_blkoff(sbi, curseg, 0);
if (reuse) {
sum_page = get_sum_page(sbi, new_segno);
sum_node = (struct f2fs_summary_block *)page_address(sum_page);
memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
f2fs_put_page(sum_page, 1);
}
}
static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
{
struct curseg_info *curseg = CURSEG_I(sbi, type);
const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
return v_ops->get_victim(sbi,
&(curseg)->next_segno, BG_GC, type, SSR);
/* For data segments, let's do SSR more intensively */
for (; type >= CURSEG_HOT_DATA; type--)
if (v_ops->get_victim(sbi, &(curseg)->next_segno,
BG_GC, type, SSR))
return 1;
return 0;
}
/*
* flush out current segment and replace it with new segment
* This function should be returned with success, otherwise BUG
*/
static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
int type, bool force)
{
struct curseg_info *curseg = CURSEG_I(sbi, type);
if (force)
new_curseg(sbi, type, true);
else if (type == CURSEG_WARM_NODE)
new_curseg(sbi, type, false);
else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
new_curseg(sbi, type, false);
else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
change_curseg(sbi, type, true);
else
new_curseg(sbi, type, false);
stat_inc_seg_type(sbi, curseg);
}
static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
{
struct curseg_info *curseg = CURSEG_I(sbi, type);
unsigned int old_segno;
old_segno = curseg->segno;
SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
locate_dirty_segment(sbi, old_segno);
}
void allocate_new_segments(struct f2fs_sb_info *sbi)
{
int i;
for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
__allocate_new_segments(sbi, i);
}
static const struct segment_allocation default_salloc_ops = {
.allocate_segment = allocate_segment_by_default,
};
int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
{
__u64 start = F2FS_BYTES_TO_BLK(range->start);
__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
unsigned int start_segno, end_segno;
struct cp_control cpc;
if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
return -EINVAL;
cpc.trimmed = 0;
if (end <= MAIN_BLKADDR(sbi))
goto out;
/* start/end segment number in main_area */
start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
GET_SEGNO(sbi, end);
cpc.reason = CP_DISCARD;
cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
/* do checkpoint to issue discard commands safely */
for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
cpc.trim_start = start_segno;
if (sbi->discard_blks == 0)
break;
else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
cpc.trim_end = end_segno;
else
cpc.trim_end = min_t(unsigned int,
rounddown(start_segno +
BATCHED_TRIM_SEGMENTS(sbi),
sbi->segs_per_sec) - 1, end_segno);
mutex_lock(&sbi->gc_mutex);
write_checkpoint(sbi, &cpc);
mutex_unlock(&sbi->gc_mutex);
}
out:
range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
return 0;
}
static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
{
struct curseg_info *curseg = CURSEG_I(sbi, type);
if (curseg->next_blkoff < sbi->blocks_per_seg)
return true;
return false;
}
static int __get_segment_type_2(struct page *page, enum page_type p_type)
{
if (p_type == DATA)
return CURSEG_HOT_DATA;
else
return CURSEG_HOT_NODE;
}
static int __get_segment_type_4(struct page *page, enum page_type p_type)
{
if (p_type == DATA) {
struct inode *inode = page->mapping->host;
if (S_ISDIR(inode->i_mode))
return CURSEG_HOT_DATA;
else
return CURSEG_COLD_DATA;
} else {
if (IS_DNODE(page) && is_cold_node(page))
return CURSEG_WARM_NODE;
else
return CURSEG_COLD_NODE;
}
}
static int __get_segment_type_6(struct page *page, enum page_type p_type)
{
if (p_type == DATA) {
struct inode *inode = page->mapping->host;
if (S_ISDIR(inode->i_mode))
return CURSEG_HOT_DATA;
else if (is_cold_data(page) || file_is_cold(inode))
return CURSEG_COLD_DATA;
else
return CURSEG_WARM_DATA;
} else {
if (IS_DNODE(page))
return is_cold_node(page) ? CURSEG_WARM_NODE :
CURSEG_HOT_NODE;
else
return CURSEG_COLD_NODE;
}
}
static int __get_segment_type(struct page *page, enum page_type p_type)
{
switch (F2FS_P_SB(page)->active_logs) {
case 2:
return __get_segment_type_2(page, p_type);
case 4:
return __get_segment_type_4(page, p_type);
}
/* NR_CURSEG_TYPE(6) logs by default */
f2fs_bug_on(F2FS_P_SB(page),
F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
return __get_segment_type_6(page, p_type);
}
void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
block_t old_blkaddr, block_t *new_blkaddr,
struct f2fs_summary *sum, int type)
{
struct sit_info *sit_i = SIT_I(sbi);
struct curseg_info *curseg;
bool direct_io = (type == CURSEG_DIRECT_IO);
type = direct_io ? CURSEG_WARM_DATA : type;
curseg = CURSEG_I(sbi, type);
mutex_lock(&curseg->curseg_mutex);
mutex_lock(&sit_i->sentry_lock);
/* direct_io'ed data is aligned to the segment for better performance */
if (direct_io && curseg->next_blkoff)
__allocate_new_segments(sbi, type);
*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
/*
* __add_sum_entry should be resided under the curseg_mutex
* because, this function updates a summary entry in the
* current summary block.
*/
__add_sum_entry(sbi, type, sum);
__refresh_next_blkoff(sbi, curseg);
stat_inc_block_count(sbi, curseg);
if (!__has_curseg_space(sbi, type))
sit_i->s_ops->allocate_segment(sbi, type, false);
/*
* SIT information should be updated before segment allocation,
* since SSR needs latest valid block information.
*/
refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
mutex_unlock(&sit_i->sentry_lock);
if (page && IS_NODESEG(type))
fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
mutex_unlock(&curseg->curseg_mutex);
}
static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
{
int type = __get_segment_type(fio->page, fio->type);
allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
&fio->blk_addr, sum, type);
/* writeout dirty page into bdev */
f2fs_submit_page_mbio(fio);
}
void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
{
struct f2fs_io_info fio = {
.sbi = sbi,
.type = META,
.rw = WRITE_SYNC | REQ_META | REQ_PRIO,
.blk_addr = page->index,
.page = page,
.encrypted_page = NULL,
};
set_page_writeback(page);
f2fs_submit_page_mbio(&fio);
}
void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
{
struct f2fs_summary sum;
set_summary(&sum, nid, 0, 0);
do_write_page(&sum, fio);
}
void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
{
struct f2fs_sb_info *sbi = fio->sbi;
struct f2fs_summary sum;
struct node_info ni;
f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
get_node_info(sbi, dn->nid, &ni);
set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
do_write_page(&sum, fio);
dn->data_blkaddr = fio->blk_addr;
}
void rewrite_data_page(struct f2fs_io_info *fio)
{
stat_inc_inplace_blocks(fio->sbi);
f2fs_submit_page_mbio(fio);
}
void f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
block_t old_blkaddr, block_t new_blkaddr,
bool recover_curseg)
{
struct sit_info *sit_i = SIT_I(sbi);
struct curseg_info *curseg;
unsigned int segno, old_cursegno;
struct seg_entry *se;
int type;
unsigned short old_blkoff;
segno = GET_SEGNO(sbi, new_blkaddr);
se = get_seg_entry(sbi, segno);
type = se->type;
if (!recover_curseg) {
/* for recovery flow */
if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
if (old_blkaddr == NULL_ADDR)
type = CURSEG_COLD_DATA;
else
type = CURSEG_WARM_DATA;
}
} else {
if (!IS_CURSEG(sbi, segno))
type = CURSEG_WARM_DATA;
}
curseg = CURSEG_I(sbi, type);
mutex_lock(&curseg->curseg_mutex);
mutex_lock(&sit_i->sentry_lock);
old_cursegno = curseg->segno;
old_blkoff = curseg->next_blkoff;
/* change the current segment */
if (segno != curseg->segno) {
curseg->next_segno = segno;
change_curseg(sbi, type, true);
}
curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
__add_sum_entry(sbi, type, sum);
refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
locate_dirty_segment(sbi, old_cursegno);
if (recover_curseg) {
if (old_cursegno != curseg->segno) {
curseg->next_segno = old_cursegno;
change_curseg(sbi, type, true);
}
curseg->next_blkoff = old_blkoff;
}
mutex_unlock(&sit_i->sentry_lock);
mutex_unlock(&curseg->curseg_mutex);
}
static inline bool is_merged_page(struct f2fs_sb_info *sbi,
struct page *page, enum page_type type)
{
enum page_type btype = PAGE_TYPE_OF_BIO(type);
struct f2fs_bio_info *io = &sbi->write_io[btype];
struct bio_vec *bvec;
struct page *target;
int i;
down_read(&io->io_rwsem);
if (!io->bio) {
up_read(&io->io_rwsem);
return false;
}
bio_for_each_segment_all(bvec, io->bio, i) {
if (bvec->bv_page->mapping) {
target = bvec->bv_page;
} else {
struct f2fs_crypto_ctx *ctx;
/* encrypted page */
ctx = (struct f2fs_crypto_ctx *)page_private(
bvec->bv_page);
target = ctx->control_page;
}
if (page == target) {
up_read(&io->io_rwsem);
return true;
}
}
up_read(&io->io_rwsem);
return false;
}
void f2fs_wait_on_page_writeback(struct page *page,
enum page_type type)
{
if (PageWriteback(page)) {
struct f2fs_sb_info *sbi = F2FS_P_SB(page);
if (is_merged_page(sbi, page, type))
f2fs_submit_merged_bio(sbi, type, WRITE);
wait_on_page_writeback(page);
}
}
static int read_compacted_summaries(struct f2fs_sb_info *sbi)
{
struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
struct curseg_info *seg_i;
unsigned char *kaddr;
struct page *page;
block_t start;
int i, j, offset;
start = start_sum_block(sbi);
page = get_meta_page(sbi, start++);
kaddr = (unsigned char *)page_address(page);
/* Step 1: restore nat cache */
seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
/* Step 2: restore sit cache */
seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
SUM_JOURNAL_SIZE);
offset = 2 * SUM_JOURNAL_SIZE;
/* Step 3: restore summary entries */
for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
unsigned short blk_off;
unsigned int segno;
seg_i = CURSEG_I(sbi, i);
segno = le32_to_cpu(ckpt->cur_data_segno[i]);
blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
seg_i->next_segno = segno;
reset_curseg(sbi, i, 0);
seg_i->alloc_type = ckpt->alloc_type[i];
seg_i->next_blkoff = blk_off;
if (seg_i->alloc_type == SSR)
blk_off = sbi->blocks_per_seg;
for (j = 0; j < blk_off; j++) {
struct f2fs_summary *s;
s = (struct f2fs_summary *)(kaddr + offset);
seg_i->sum_blk->entries[j] = *s;
offset += SUMMARY_SIZE;
if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
SUM_FOOTER_SIZE)
continue;
f2fs_put_page(page, 1);
page = NULL;
page = get_meta_page(sbi, start++);
kaddr = (unsigned char *)page_address(page);
offset = 0;
}
}
f2fs_put_page(page, 1);
return 0;
}
static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
{
struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
struct f2fs_summary_block *sum;
struct curseg_info *curseg;
struct page *new;
unsigned short blk_off;
unsigned int segno = 0;
block_t blk_addr = 0;
/* get segment number and block addr */
if (IS_DATASEG(type)) {
segno = le32_to_cpu(ckpt->cur_data_segno[type]);
blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
CURSEG_HOT_DATA]);
if (__exist_node_summaries(sbi))
blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
else
blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
} else {
segno = le32_to_cpu(ckpt->cur_node_segno[type -
CURSEG_HOT_NODE]);
blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
CURSEG_HOT_NODE]);
if (__exist_node_summaries(sbi))
blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
type - CURSEG_HOT_NODE);
else
blk_addr = GET_SUM_BLOCK(sbi, segno);
}
new = get_meta_page(sbi, blk_addr);
sum = (struct f2fs_summary_block *)page_address(new);
if (IS_NODESEG(type)) {
if (__exist_node_summaries(sbi)) {
struct f2fs_summary *ns = &sum->entries[0];
int i;
for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
ns->version = 0;
ns->ofs_in_node = 0;
}
} else {
int err;
err = restore_node_summary(sbi, segno, sum);
if (err) {
f2fs_put_page(new, 1);
return err;
}
}
}
/* set uncompleted segment to curseg */
curseg = CURSEG_I(sbi, type);
mutex_lock(&curseg->curseg_mutex);
memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
curseg->next_segno = segno;
reset_curseg(sbi, type, 0);
curseg->alloc_type = ckpt->alloc_type[type];
curseg->next_blkoff = blk_off;
mutex_unlock(&curseg->curseg_mutex);
f2fs_put_page(new, 1);
return 0;
}
static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
{
int type = CURSEG_HOT_DATA;
int err;
if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
int npages = npages_for_summary_flush(sbi, true);
if (npages >= 2)
ra_meta_pages(sbi, start_sum_block(sbi), npages,
META_CP);
/* restore for compacted data summary */
if (read_compacted_summaries(sbi))
return -EINVAL;
type = CURSEG_HOT_NODE;
}
if (__exist_node_summaries(sbi))
ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
NR_CURSEG_TYPE - type, META_CP);
for (; type <= CURSEG_COLD_NODE; type++) {
err = read_normal_summaries(sbi, type);
if (err)
return err;
}
return 0;
}
static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
{
struct page *page;
unsigned char *kaddr;
struct f2fs_summary *summary;
struct curseg_info *seg_i;
int written_size = 0;
int i, j;
page = grab_meta_page(sbi, blkaddr++);
kaddr = (unsigned char *)page_address(page);
/* Step 1: write nat cache */
seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
written_size += SUM_JOURNAL_SIZE;
/* Step 2: write sit cache */
seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
SUM_JOURNAL_SIZE);
written_size += SUM_JOURNAL_SIZE;
/* Step 3: write summary entries */
for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
unsigned short blkoff;
seg_i = CURSEG_I(sbi, i);
if (sbi->ckpt->alloc_type[i] == SSR)
blkoff = sbi->blocks_per_seg;
else
blkoff = curseg_blkoff(sbi, i);
for (j = 0; j < blkoff; j++) {
if (!page) {
page = grab_meta_page(sbi, blkaddr++);
kaddr = (unsigned char *)page_address(page);
written_size = 0;
}
summary = (struct f2fs_summary *)(kaddr + written_size);
*summary = seg_i->sum_blk->entries[j];
written_size += SUMMARY_SIZE;
if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
SUM_FOOTER_SIZE)
continue;
set_page_dirty(page);
f2fs_put_page(page, 1);
page = NULL;
}
}
if (page) {
set_page_dirty(page);
f2fs_put_page(page, 1);
}
}
static void write_normal_summaries(struct f2fs_sb_info *sbi,
block_t blkaddr, int type)
{
int i, end;
if (IS_DATASEG(type))
end = type + NR_CURSEG_DATA_TYPE;
else
end = type + NR_CURSEG_NODE_TYPE;
for (i = type; i < end; i++) {
struct curseg_info *sum = CURSEG_I(sbi, i);
mutex_lock(&sum->curseg_mutex);
write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
mutex_unlock(&sum->curseg_mutex);
}
}
void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
{
if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
write_compacted_summaries(sbi, start_blk);
else
write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
}
void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
{
write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
}
int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
unsigned int val, int alloc)
{
int i;
if (type == NAT_JOURNAL) {
for (i = 0; i < nats_in_cursum(sum); i++) {
if (le32_to_cpu(nid_in_journal(sum, i)) == val)
return i;
}
if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
return update_nats_in_cursum(sum, 1);
} else if (type == SIT_JOURNAL) {
for (i = 0; i < sits_in_cursum(sum); i++)
if (le32_to_cpu(segno_in_journal(sum, i)) == val)
return i;
if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
return update_sits_in_cursum(sum, 1);
}
return -1;
}
static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
unsigned int segno)
{
return get_meta_page(sbi, current_sit_addr(sbi, segno));
}
static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
unsigned int start)
{
struct sit_info *sit_i = SIT_I(sbi);
struct page *src_page, *dst_page;
pgoff_t src_off, dst_off;
void *src_addr, *dst_addr;
src_off = current_sit_addr(sbi, start);
dst_off = next_sit_addr(sbi, src_off);
/* get current sit block page without lock */
src_page = get_meta_page(sbi, src_off);
dst_page = grab_meta_page(sbi, dst_off);
f2fs_bug_on(sbi, PageDirty(src_page));
src_addr = page_address(src_page);
dst_addr = page_address(dst_page);
memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
set_page_dirty(dst_page);
f2fs_put_page(src_page, 1);
set_to_next_sit(sit_i, start);
return dst_page;
}
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
static struct sit_entry_set *grab_sit_entry_set(void)
{
struct sit_entry_set *ses =
f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
ses->entry_cnt = 0;
INIT_LIST_HEAD(&ses->set_list);
return ses;
}
static void release_sit_entry_set(struct sit_entry_set *ses)
{
list_del(&ses->set_list);
kmem_cache_free(sit_entry_set_slab, ses);
}
static void adjust_sit_entry_set(struct sit_entry_set *ses,
struct list_head *head)
{
struct sit_entry_set *next = ses;
if (list_is_last(&ses->set_list, head))
return;
list_for_each_entry_continue(next, head, set_list)
if (ses->entry_cnt <= next->entry_cnt)
break;
list_move_tail(&ses->set_list, &next->set_list);
}
static void add_sit_entry(unsigned int segno, struct list_head *head)
{
struct sit_entry_set *ses;
unsigned int start_segno = START_SEGNO(segno);
list_for_each_entry(ses, head, set_list) {
if (ses->start_segno == start_segno) {
ses->entry_cnt++;
adjust_sit_entry_set(ses, head);
return;
}
}
ses = grab_sit_entry_set();
ses->start_segno = start_segno;
ses->entry_cnt++;
list_add(&ses->set_list, head);
}
static void add_sits_in_set(struct f2fs_sb_info *sbi)
{
struct f2fs_sm_info *sm_info = SM_I(sbi);
struct list_head *set_list = &sm_info->sit_entry_set;
unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
unsigned int segno;
for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
add_sit_entry(segno, set_list);
}
static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
{
struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
struct f2fs_summary_block *sum = curseg->sum_blk;
int i;
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
unsigned int segno;
bool dirtied;
segno = le32_to_cpu(segno_in_journal(sum, i));
dirtied = __mark_sit_entry_dirty(sbi, segno);
if (!dirtied)
add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
}
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
update_sits_in_cursum(sum, -sits_in_cursum(sum));
}
/*
* CP calls this function, which flushes SIT entries including sit_journal,
* and moves prefree segs to free segs.
*/
void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
{
struct sit_info *sit_i = SIT_I(sbi);
unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
struct f2fs_summary_block *sum = curseg->sum_blk;
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
struct sit_entry_set *ses, *tmp;
struct list_head *head = &SM_I(sbi)->sit_entry_set;
bool to_journal = true;
struct seg_entry *se;
mutex_lock(&curseg->curseg_mutex);
mutex_lock(&sit_i->sentry_lock);
if (!sit_i->dirty_sentries)
goto out;
/*
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
* add and account sit entries of dirty bitmap in sit entry
* set temporarily
*/
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
add_sits_in_set(sbi);
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
/*
* if there are no enough space in journal to store dirty sit
* entries, remove all entries from journal and add and account
* them in sit entry set.
*/
if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
remove_sits_in_journal(sbi);
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
/*
* there are two steps to flush sit entries:
* #1, flush sit entries to journal in current cold data summary block.
* #2, flush sit entries to sit page.
*/
list_for_each_entry_safe(ses, tmp, head, set_list) {
struct page *page = NULL;
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
struct f2fs_sit_block *raw_sit = NULL;
unsigned int start_segno = ses->start_segno;
unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
(unsigned long)MAIN_SEGS(sbi));
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
unsigned int segno = start_segno;
if (to_journal &&
!__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
to_journal = false;
if (!to_journal) {
page = get_next_sit_page(sbi, start_segno);
raw_sit = page_address(page);
}
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
/* flush dirty sit entries in region of current sit set */
for_each_set_bit_from(segno, bitmap, end) {
int offset, sit_offset;
se = get_seg_entry(sbi, segno);
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
/* add discard candidates */
if (cpc->reason != CP_DISCARD) {
cpc->trim_start = segno;
add_discard_addrs(sbi, cpc);
}
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
if (to_journal) {
offset = lookup_journal_in_cursum(sum,
SIT_JOURNAL, segno, 1);
f2fs_bug_on(sbi, offset < 0);
segno_in_journal(sum, offset) =
cpu_to_le32(segno);
seg_info_to_raw_sit(se,
&sit_in_journal(sum, offset));
} else {
sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
seg_info_to_raw_sit(se,
&raw_sit->entries[sit_offset]);
}
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
__clear_bit(segno, bitmap);
sit_i->dirty_sentries--;
ses->entry_cnt--;
}
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
if (!to_journal)
f2fs_put_page(page, 1);
f2fs_bug_on(sbi, ses->entry_cnt);
release_sit_entry_set(ses);
}
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
f2fs_bug_on(sbi, !list_empty(head));
f2fs_bug_on(sbi, sit_i->dirty_sentries);
out:
if (cpc->reason == CP_DISCARD) {
for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
add_discard_addrs(sbi, cpc);
}
mutex_unlock(&sit_i->sentry_lock);
mutex_unlock(&curseg->curseg_mutex);
set_prefree_as_free_segments(sbi);
}
static int build_sit_info(struct f2fs_sb_info *sbi)
{
struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
struct sit_info *sit_i;
unsigned int sit_segs, start;
char *src_bitmap, *dst_bitmap;
unsigned int bitmap_size;
/* allocate memory for SIT information */
sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
if (!sit_i)
return -ENOMEM;
SM_I(sbi)->sit_info = sit_i;
sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry));
if (!sit_i->sentries)
return -ENOMEM;
bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
if (!sit_i->dirty_sentries_bitmap)
return -ENOMEM;
for (start = 0; start < MAIN_SEGS(sbi); start++) {
sit_i->sentries[start].cur_valid_map
= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
sit_i->sentries[start].ckpt_valid_map
= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
sit_i->sentries[start].discard_map
= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
if (!sit_i->sentries[start].cur_valid_map ||
!sit_i->sentries[start].ckpt_valid_map ||
!sit_i->sentries[start].discard_map)
return -ENOMEM;
}
sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
if (!sit_i->tmp_map)
return -ENOMEM;
if (sbi->segs_per_sec > 1) {
sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) *
sizeof(struct sec_entry));
if (!sit_i->sec_entries)
return -ENOMEM;
}
/* get information related with SIT */
sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
/* setup SIT bitmap from ckeckpoint pack */
bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
if (!dst_bitmap)
return -ENOMEM;
/* init SIT information */
sit_i->s_ops = &default_salloc_ops;
sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
sit_i->sit_bitmap = dst_bitmap;
sit_i->bitmap_size = bitmap_size;
sit_i->dirty_sentries = 0;
sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
mutex_init(&sit_i->sentry_lock);
return 0;
}
static int build_free_segmap(struct f2fs_sb_info *sbi)
{
struct free_segmap_info *free_i;
unsigned int bitmap_size, sec_bitmap_size;
/* allocate memory for free segmap information */
free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
if (!free_i)
return -ENOMEM;
SM_I(sbi)->free_info = free_i;
bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
if (!free_i->free_segmap)
return -ENOMEM;
sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
if (!free_i->free_secmap)
return -ENOMEM;
/* set all segments as dirty temporarily */
memset(free_i->free_segmap, 0xff, bitmap_size);
memset(free_i->free_secmap, 0xff, sec_bitmap_size);
/* init free segmap information */
free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
free_i->free_segments = 0;
free_i->free_sections = 0;
spin_lock_init(&free_i->segmap_lock);
return 0;
}
static int build_curseg(struct f2fs_sb_info *sbi)
{
struct curseg_info *array;
int i;
array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
if (!array)
return -ENOMEM;
SM_I(sbi)->curseg_array = array;
for (i = 0; i < NR_CURSEG_TYPE; i++) {
mutex_init(&array[i].curseg_mutex);
array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
if (!array[i].sum_blk)
return -ENOMEM;
array[i].segno = NULL_SEGNO;
array[i].next_blkoff = 0;
}
return restore_curseg_summaries(sbi);
}
static void build_sit_entries(struct f2fs_sb_info *sbi)
{
struct sit_info *sit_i = SIT_I(sbi);
struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
struct f2fs_summary_block *sum = curseg->sum_blk;
int sit_blk_cnt = SIT_BLK_CNT(sbi);
unsigned int i, start, end;
unsigned int readed, start_blk = 0;
int nrpages = MAX_BIO_BLOCKS(sbi);
do {
readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
start = start_blk * sit_i->sents_per_block;
end = (start_blk + readed) * sit_i->sents_per_block;
for (; start < end && start < MAIN_SEGS(sbi); start++) {
struct seg_entry *se = &sit_i->sentries[start];
struct f2fs_sit_block *sit_blk;
struct f2fs_sit_entry sit;
struct page *page;
mutex_lock(&curseg->curseg_mutex);
for (i = 0; i < sits_in_cursum(sum); i++) {
if (le32_to_cpu(segno_in_journal(sum, i))
== start) {
sit = sit_in_journal(sum, i);
mutex_unlock(&curseg->curseg_mutex);
goto got_it;
}
}
mutex_unlock(&curseg->curseg_mutex);
page = get_current_sit_page(sbi, start);
sit_blk = (struct f2fs_sit_block *)page_address(page);
sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
f2fs_put_page(page, 1);
got_it:
check_block_count(sbi, start, &sit);
seg_info_from_raw_sit(se, &sit);
/* build discard map only one time */
memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
if (sbi->segs_per_sec > 1) {
struct sec_entry *e = get_sec_entry(sbi, start);
e->valid_blocks += se->valid_blocks;
}
}
start_blk += readed;
} while (start_blk < sit_blk_cnt);
}
static void init_free_segmap(struct f2fs_sb_info *sbi)
{
unsigned int start;
int type;
for (start = 0; start < MAIN_SEGS(sbi); start++) {
struct seg_entry *sentry = get_seg_entry(sbi, start);
if (!sentry->valid_blocks)
__set_free(sbi, start);
}
/* set use the current segments */
for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
struct curseg_info *curseg_t = CURSEG_I(sbi, type);
__set_test_and_inuse(sbi, curseg_t->segno);
}
}
static void init_dirty_segmap(struct f2fs_sb_info *sbi)
{
struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
struct free_segmap_info *free_i = FREE_I(sbi);
unsigned int segno = 0, offset = 0;
unsigned short valid_blocks;
while (1) {
/* find dirty segment based on free segmap */
segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
if (segno >= MAIN_SEGS(sbi))
break;
offset = segno + 1;
valid_blocks = get_valid_blocks(sbi, segno, 0);
if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
continue;
if (valid_blocks > sbi->blocks_per_seg) {
f2fs_bug_on(sbi, 1);
continue;
}
mutex_lock(&dirty_i->seglist_lock);
__locate_dirty_segment(sbi, segno, DIRTY);
mutex_unlock(&dirty_i->seglist_lock);
}
}
static int init_victim_secmap(struct f2fs_sb_info *sbi)
{
struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
if (!dirty_i->victim_secmap)
return -ENOMEM;
return 0;
}
static int build_dirty_segmap(struct f2fs_sb_info *sbi)
{
struct dirty_seglist_info *dirty_i;
unsigned int bitmap_size, i;
/* allocate memory for dirty segments list information */
dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
if (!dirty_i)
return -ENOMEM;
SM_I(sbi)->dirty_info = dirty_i;
mutex_init(&dirty_i->seglist_lock);
bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
for (i = 0; i < NR_DIRTY_TYPE; i++) {
dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
if (!dirty_i->dirty_segmap[i])
return -ENOMEM;
}
init_dirty_segmap(sbi);
return init_victim_secmap(sbi);
}
/*
* Update min, max modified time for cost-benefit GC algorithm
*/
static void init_min_max_mtime(struct f2fs_sb_info *sbi)
{
struct sit_info *sit_i = SIT_I(sbi);
unsigned int segno;
mutex_lock(&sit_i->sentry_lock);
sit_i->min_mtime = LLONG_MAX;
for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
unsigned int i;
unsigned long long mtime = 0;
for (i = 0; i < sbi->segs_per_sec; i++)
mtime += get_seg_entry(sbi, segno + i)->mtime;
mtime = div_u64(mtime, sbi->segs_per_sec);
if (sit_i->min_mtime > mtime)
sit_i->min_mtime = mtime;
}
sit_i->max_mtime = get_mtime(sbi);
mutex_unlock(&sit_i->sentry_lock);
}
int build_segment_manager(struct f2fs_sb_info *sbi)
{
struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
struct f2fs_sm_info *sm_info;
int err;
sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
if (!sm_info)
return -ENOMEM;
/* init sm info */
sbi->sm_info = sm_info;
sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
sm_info->rec_prefree_segments = sm_info->main_segments *
DEF_RECLAIM_PREFREE_SEGMENTS / 100;
sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
INIT_LIST_HEAD(&sm_info->discard_list);
sm_info->nr_discards = 0;
sm_info->max_discards = 0;
sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
INIT_LIST_HEAD(&sm_info->sit_entry_set);
if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
err = create_flush_cmd_control(sbi);
if (err)
return err;
}
err = build_sit_info(sbi);
if (err)
return err;
err = build_free_segmap(sbi);
if (err)
return err;
err = build_curseg(sbi);
if (err)
return err;
/* reinit free segmap based on SIT */
build_sit_entries(sbi);
init_free_segmap(sbi);
err = build_dirty_segmap(sbi);
if (err)
return err;
init_min_max_mtime(sbi);
return 0;
}
static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
enum dirty_type dirty_type)
{
struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
mutex_lock(&dirty_i->seglist_lock);
kfree(dirty_i->dirty_segmap[dirty_type]);
dirty_i->nr_dirty[dirty_type] = 0;
mutex_unlock(&dirty_i->seglist_lock);
}
static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
{
struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
kfree(dirty_i->victim_secmap);
}
static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
{
struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
int i;
if (!dirty_i)
return;
/* discard pre-free/dirty segments list */
for (i = 0; i < NR_DIRTY_TYPE; i++)
discard_dirty_segmap(sbi, i);
destroy_victim_secmap(sbi);
SM_I(sbi)->dirty_info = NULL;
kfree(dirty_i);
}
static void destroy_curseg(struct f2fs_sb_info *sbi)
{
struct curseg_info *array = SM_I(sbi)->curseg_array;
int i;
if (!array)
return;
SM_I(sbi)->curseg_array = NULL;
for (i = 0; i < NR_CURSEG_TYPE; i++)
kfree(array[i].sum_blk);
kfree(array);
}
static void destroy_free_segmap(struct f2fs_sb_info *sbi)
{
struct free_segmap_info *free_i = SM_I(sbi)->free_info;
if (!free_i)
return;
SM_I(sbi)->free_info = NULL;
kfree(free_i->free_segmap);
kfree(free_i->free_secmap);
kfree(free_i);
}
static void destroy_sit_info(struct f2fs_sb_info *sbi)
{
struct sit_info *sit_i = SIT_I(sbi);
unsigned int start;
if (!sit_i)
return;
if (sit_i->sentries) {
for (start = 0; start < MAIN_SEGS(sbi); start++) {
kfree(sit_i->sentries[start].cur_valid_map);
kfree(sit_i->sentries[start].ckpt_valid_map);
kfree(sit_i->sentries[start].discard_map);
}
}
kfree(sit_i->tmp_map);
vfree(sit_i->sentries);
vfree(sit_i->sec_entries);
kfree(sit_i->dirty_sentries_bitmap);
SM_I(sbi)->sit_info = NULL;
kfree(sit_i->sit_bitmap);
kfree(sit_i);
}
void destroy_segment_manager(struct f2fs_sb_info *sbi)
{
struct f2fs_sm_info *sm_info = SM_I(sbi);
if (!sm_info)
return;
destroy_flush_cmd_control(sbi);
destroy_dirty_segmap(sbi);
destroy_curseg(sbi);
destroy_free_segmap(sbi);
destroy_sit_info(sbi);
sbi->sm_info = NULL;
kfree(sm_info);
}
int __init create_segment_manager_caches(void)
{
discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
sizeof(struct discard_entry));
if (!discard_entry_slab)
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
goto fail;
sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
sizeof(struct sit_entry_set));
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
if (!sit_entry_set_slab)
goto destory_discard_entry;
inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
sizeof(struct inmem_pages));
if (!inmem_entry_slab)
goto destroy_sit_entry_set;
return 0;
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
destroy_sit_entry_set:
kmem_cache_destroy(sit_entry_set_slab);
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
destory_discard_entry:
kmem_cache_destroy(discard_entry_slab);
fail:
return -ENOMEM;
}
void destroy_segment_manager_caches(void)
{
f2fs: refactor flush_sit_entries codes for reducing SIT writes In commit aec71382c681 ("f2fs: refactor flush_nat_entries codes for reducing NAT writes"), we descripte the issue as below: "Although building NAT journal in cursum reduce the read/write work for NAT block, but previous design leave us lower performance when write checkpoint frequently for these cases: 1. if journal in cursum has already full, it's a bit of waste that we flush all nat entries to page for persistence, but not to cache any entries. 2. if journal in cursum is not full, we fill nat entries to journal util journal is full, then flush the left dirty entries to disk without merge journaled entries, so these journaled entries may be flushed to disk at next checkpoint but lost chance to flushed last time." Actually, we have the same problem in using SIT journal area. In this patch, firstly we will update sit journal with dirty entries as many as possible. Secondly if there is no space in sit journal, we will remove all entries in journal and walk through the whole dirty entry bitmap of sit, accounting dirty sit entries located in same SIT block to sit entry set. All entry sets are linked to list sit_entry_set in sm_info, sorted ascending order by count of entries in set. Later we flush entries in set which have fewest entries into journal as many as we can, and then flush dense set with merged entries to disk. In this way we can use sit journal area more effectively, also we will reduce SIT update, result in gaining in performance and saving lifetime of flash device. In my testing environment, it shows this patch can help to reduce SIT block update obviously. virtual machine + hard disk: fsstress -p 20 -n 400 -l 5 sit page num cp count sit pages/cp based 2006.50 1349.75 1.486 patched 1566.25 1463.25 1.070 Our latency of merging op is small when handling a great number of dirty SIT entries in flush_sit_entries: latency(ns) dirty sit count 36038 2151 49168 2123 37174 2232 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-09-04 10:13:01 +00:00
kmem_cache_destroy(sit_entry_set_slab);
kmem_cache_destroy(discard_entry_slab);
kmem_cache_destroy(inmem_entry_slab);
}