Greetings,
Using the kernel 2.6.27.45 with an ext4 filesystem, the same deadlock as
was reported earlier in kernel bugzilla #12579 can occur. Simply running
"dbench 500" on an ext4 filesystem can cause the deadlock in one to two
minutes.
In later stable kernels, the deadlock was fixed by commit 2acf2c26
("ext4: Implement range_cyclic in ext4_da_writepages instead of
write_cache_pages"). I can confirm that the same commit fixes the
deadlock on 2.6.27.y. However, as the code in ext4_da_writepages is
different in 2.6.27.y than in later kernels, it was necessary to add
some other mainline patches before 2acf2c26.
Alltogether, I've added 11 mainline patches including 2acf2c26 to my
local 2.6.27.y kernel, and confirmed that the deadlock is fixed only
when the patch 2acf2c26 is applied.
The 11 patches are:
Aneesh Kumar K.V (10):
ext4: invalidate pages if delalloc block allocation fails.
ext4: Make sure all the block allocation paths reserve blocks
ext4: Add percpu dirty block accounting.
ext4: Retry block reservation
ext4: Retry block allocation if we have free blocks left
ext4: Use tag dirty lookup during mpage_da_submit_io
vfs: Remove the range_cont writeback mode.
vfs: Add no_nrwrite_index_update writeback control flag
ext4: Fix file fragmentation during large file write.
ext4: Implement range_cyclic in ext4_da_writepages instead of write_cache_pages
Mingming Cao (1):
percpu counter: clean up percpu_counter_sum_and_set()
All will be posted in a reply to this message, in the order that they apply.
I think these patches should be committed to stable 2.6.27.y.
Most of these patches applied cleanly to 2.6.27.45 with no changes other
than some context lines, but two of the patches needed some very slight
modifications. I'll note the changes that I made to them above my
signed-off line for each.
Jayson
From: Aneesh Kumar K.V <[email protected]>
Date: Tue Aug 19 21:08:18 2008 -0400
Subject: ext4: invalidate pages if delalloc block allocation fails.
commit c4a0c46ec92c194c873232b88debce4e1a448483 upstream.
We are a bit agressive in invalidating all the pages. But
it is ok because we really don't know why the block allocation
failed and it is better to come of the writeback path
so that user can look for more info.
Signed-off-by: Aneesh Kumar K.V <[email protected]>
Signed-off-by: Jayson R. King <[email protected]>
---
fs/ext4/inode.c | 85 +++++++++++++++++++++++++++++++++++++++-------
1 file changed, 73 insertions(+), 12 deletions(-)
diff -urNp linux-2.6.27.orig/fs/ext4/inode.c linux-2.6.27/fs/ext4/inode.c
--- linux-2.6.27.orig/fs/ext4/inode.c 2010-02-26 14:10:41.168252288 -0600
+++ linux-2.6.27/fs/ext4/inode.c 2010-02-26 14:14:55.379252600 -0600
@@ -1821,6 +1821,39 @@ static inline void __unmap_underlying_bl
unmap_underlying_metadata(bdev, bh->b_blocknr + i);
}
+static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
+ sector_t logical, long blk_cnt)
+{
+ int nr_pages, i;
+ pgoff_t index, end;
+ struct pagevec pvec;
+ struct inode *inode = mpd->inode;
+ struct address_space *mapping = inode->i_mapping;
+
+ index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
+ end = (logical + blk_cnt - 1) >>
+ (PAGE_CACHE_SHIFT - inode->i_blkbits);
+ while (index <= end) {
+ nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
+ if (nr_pages == 0)
+ break;
+ for (i = 0; i < nr_pages; i++) {
+ struct page *page = pvec.pages[i];
+ index = page->index;
+ if (index > end)
+ break;
+ index++;
+
+ BUG_ON(!PageLocked(page));
+ BUG_ON(PageWriteback(page));
+ block_invalidatepage(page, 0);
+ ClearPageUptodate(page);
+ unlock_page(page);
+ }
+ }
+ return;
+}
+
/*
* mpage_da_map_blocks - go through given space
*
@@ -1830,7 +1863,7 @@ static inline void __unmap_underlying_bl
* The function skips space we know is already mapped to disk blocks.
*
*/
-static void mpage_da_map_blocks(struct mpage_da_data *mpd)
+static int mpage_da_map_blocks(struct mpage_da_data *mpd)
{
int err = 0;
struct buffer_head *lbh = &mpd->lbh;
@@ -1841,7 +1874,7 @@ static void mpage_da_map_blocks(struct m
* We consider only non-mapped and non-allocated blocks
*/
if (buffer_mapped(lbh) && !buffer_delay(lbh))
- return;
+ return 0;
new.b_state = lbh->b_state;
new.b_blocknr = 0;
@@ -1852,10 +1885,38 @@ static void mpage_da_map_blocks(struct m
* to write simply return
*/
if (!new.b_size)
- return;
+ return 0;
err = mpd->get_block(mpd->inode, next, &new, 1);
- if (err)
- return;
+ if (err) {
+
+ /* If get block returns with error
+ * we simply return. Later writepage
+ * will redirty the page and writepages
+ * will find the dirty page again
+ */
+ if (err == -EAGAIN)
+ return 0;
+ /*
+ * get block failure will cause us
+ * to loop in writepages. Because
+ * a_ops->writepage won't be able to
+ * make progress. The page will be redirtied
+ * by writepage and writepages will again
+ * try to write the same.
+ */
+ printk(KERN_EMERG "%s block allocation failed for inode %lu "
+ "at logical offset %llu with max blocks "
+ "%zd with error %d\n",
+ __func__, mpd->inode->i_ino,
+ (unsigned long long)next,
+ lbh->b_size >> mpd->inode->i_blkbits, err);
+ printk(KERN_EMERG "This should not happen.!! "
+ "Data will be lost\n");
+ /* invlaidate all the pages */
+ ext4_da_block_invalidatepages(mpd, next,
+ lbh->b_size >> mpd->inode->i_blkbits);
+ return err;
+ }
BUG_ON(new.b_size == 0);
if (buffer_new(&new))
@@ -1868,7 +1929,7 @@ static void mpage_da_map_blocks(struct m
if (buffer_delay(lbh) || buffer_unwritten(lbh))
mpage_put_bnr_to_bhs(mpd, next, &new);
- return;
+ return 0;
}
#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
@@ -1937,8 +1998,8 @@ flush_it:
* We couldn't merge the block to our extent, so we
* need to flush current extent and start new one
*/
- mpage_da_map_blocks(mpd);
- mpage_da_submit_io(mpd);
+ if (mpage_da_map_blocks(mpd) == 0)
+ mpage_da_submit_io(mpd);
mpd->io_done = 1;
return;
}
@@ -1980,8 +2041,8 @@ static int __mpage_da_writepage(struct p
* and start IO on them using writepage()
*/
if (mpd->next_page != mpd->first_page) {
- mpage_da_map_blocks(mpd);
- mpage_da_submit_io(mpd);
+ if (mpage_da_map_blocks(mpd) == 0)
+ mpage_da_submit_io(mpd);
/*
* skip rest of the page in the page_vec
*/
@@ -2102,8 +2163,8 @@ static int mpage_da_writepages(struct ad
* Handle last extent of pages
*/
if (!mpd.io_done && mpd.next_page != mpd.first_page) {
- mpage_da_map_blocks(&mpd);
- mpage_da_submit_io(&mpd);
+ if (mpage_da_map_blocks(&mpd) == 0)
+ mpage_da_submit_io(&mpd);
}
wbc->nr_to_write = to_write - mpd.pages_written;
From: Mingming Cao <[email protected]>
Date: Thu Oct 9 12:50:59 2008 -0400
Subject: percpu counter: clean up percpu_counter_sum_and_set()
commit 1f7c14c62ce63805f9574664a6c6de3633d4a354 upstream.
percpu_counter_sum_and_set() and percpu_counter_sum() is the same except
the former updates the global counter after accounting. Since we are
taking the fbc->lock to calculate the precise value of the counter in
percpu_counter_sum() anyway, it should simply set fbc->count too, as the
percpu_counter_sum_and_set() does.
This patch merges these two interfaces into one.
Signed-off-by: Mingming Cao <[email protected]>
Acked-by: Peter Zijlstra <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: "Theodore Ts'o" <[email protected]>
Signed-off-by: Jayson R. King <[email protected]>
---
fs/ext4/balloc.c | 2 +-
include/linux/percpu_counter.h | 12 +++---------
lib/percpu_counter.c | 8 +++-----
3 files changed, 7 insertions(+), 15 deletions(-)
diff -urNp linux-2.6.27.orig/fs/ext4/balloc.c linux-2.6.27/fs/ext4/balloc.c
--- linux-2.6.27.orig/fs/ext4/balloc.c 2010-02-26 14:10:41.167252303 -0600
+++ linux-2.6.27/fs/ext4/balloc.c 2010-02-26 14:18:30.811252658 -0600
@@ -1778,7 +1778,7 @@ ext4_fsblk_t ext4_has_free_blocks(struct
#ifdef CONFIG_SMP
if (free_blocks - root_blocks < FBC_BATCH)
free_blocks =
- percpu_counter_sum_and_set(&sbi->s_freeblocks_counter);
+ percpu_counter_sum(&sbi->s_freeblocks_counter);
#endif
if (free_blocks <= root_blocks)
/* we don't have free space */
diff -urNp linux-2.6.27.orig/include/linux/percpu_counter.h linux-2.6.27/include/linux/percpu_counter.h
--- linux-2.6.27.orig/include/linux/percpu_counter.h 2010-02-26 14:10:41.961252665 -0600
+++ linux-2.6.27/include/linux/percpu_counter.h 2010-02-26 14:18:30.811252658 -0600
@@ -35,7 +35,7 @@ int percpu_counter_init_irq(struct percp
void percpu_counter_destroy(struct percpu_counter *fbc);
void percpu_counter_set(struct percpu_counter *fbc, s64 amount);
void __percpu_counter_add(struct percpu_counter *fbc, s64 amount, s32 batch);
-s64 __percpu_counter_sum(struct percpu_counter *fbc, int set);
+s64 __percpu_counter_sum(struct percpu_counter *fbc);
static inline void percpu_counter_add(struct percpu_counter *fbc, s64 amount)
{
@@ -44,19 +44,13 @@ static inline void percpu_counter_add(st
static inline s64 percpu_counter_sum_positive(struct percpu_counter *fbc)
{
- s64 ret = __percpu_counter_sum(fbc, 0);
+ s64 ret = __percpu_counter_sum(fbc);
return ret < 0 ? 0 : ret;
}
-static inline s64 percpu_counter_sum_and_set(struct percpu_counter *fbc)
-{
- return __percpu_counter_sum(fbc, 1);
-}
-
-
static inline s64 percpu_counter_sum(struct percpu_counter *fbc)
{
- return __percpu_counter_sum(fbc, 0);
+ return __percpu_counter_sum(fbc);
}
static inline s64 percpu_counter_read(struct percpu_counter *fbc)
diff -urNp linux-2.6.27.orig/lib/percpu_counter.c linux-2.6.27/lib/percpu_counter.c
--- linux-2.6.27.orig/lib/percpu_counter.c 2010-02-26 14:10:42.176252671 -0600
+++ linux-2.6.27/lib/percpu_counter.c 2010-02-26 14:18:30.811252658 -0600
@@ -52,7 +52,7 @@ EXPORT_SYMBOL(__percpu_counter_add);
* Add up all the per-cpu counts, return the result. This is a more accurate
* but much slower version of percpu_counter_read_positive()
*/
-s64 __percpu_counter_sum(struct percpu_counter *fbc, int set)
+s64 __percpu_counter_sum(struct percpu_counter *fbc)
{
s64 ret;
int cpu;
@@ -62,11 +62,9 @@ s64 __percpu_counter_sum(struct percpu_c
for_each_online_cpu(cpu) {
s32 *pcount = per_cpu_ptr(fbc->counters, cpu);
ret += *pcount;
- if (set)
- *pcount = 0;
+ *pcount = 0;
}
- if (set)
- fbc->count = ret;
+ fbc->count = ret;
spin_unlock(&fbc->lock);
return ret;
From: Aneesh Kumar K.V <[email protected]>
Date: Mon Sep 8 23:14:50 2008 -0400
Subject: ext4: Retry block reservation
commit 030ba6bc67b4f2bc5cd174f57785a1745c929abe upstream.
During block reservation if we don't have enough blocks left, retry
block reservation with smaller block counts. This makes sure we try
fallocate and DIO with smaller request size and don't fail early. The
delayed allocation reservation cannot try with smaller block count. So
retry block reservation to handle temporary disk full conditions. Also
print free blocks details if we fail block allocation during writepages.
Signed-off-by: Aneesh Kumar K.V <[email protected]>
Signed-off-by: Mingming Cao <[email protected]>
Signed-off-by: "Theodore Ts'o" <[email protected]>
Signed-off-by: Jayson R. King <[email protected]>
---
fs/ext4/balloc.c | 8 +++++++-
fs/ext4/inode.c | 14 +++++++++++---
fs/ext4/mballoc.c | 7 ++++++-
3 files changed, 24 insertions(+), 5 deletions(-)
diff -urNp linux-2.6.27.orig/fs/ext4/balloc.c linux-2.6.27/fs/ext4/balloc.c
--- linux-2.6.27.orig/fs/ext4/balloc.c 2010-02-26 14:22:58.236252972 -0600
+++ linux-2.6.27/fs/ext4/balloc.c 2010-02-26 14:23:05.964253171 -0600
@@ -1907,10 +1907,16 @@ ext4_fsblk_t ext4_old_new_blocks(handle_
/*
* With delalloc we already reserved the blocks
*/
- if (ext4_claim_free_blocks(sbi, *count)) {
+ while (*count && ext4_claim_free_blocks(sbi, *count)) {
+ /* let others to free the space */
+ yield();
+ *count = *count >> 1;
+ }
+ if (!*count) {
*errp = -ENOSPC;
return 0; /*return with ENOSPC error */
}
+ num = *count;
}
/*
* Check quota for allocation of this block.
diff -urNp linux-2.6.27.orig/fs/ext4/inode.c linux-2.6.27/fs/ext4/inode.c
--- linux-2.6.27.orig/fs/ext4/inode.c 2010-02-26 14:22:58.237252599 -0600
+++ linux-2.6.27/fs/ext4/inode.c 2010-02-26 14:23:05.965252990 -0600
@@ -1549,6 +1549,7 @@ static int ext4_journalled_write_end(str
static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
{
+ int retries = 0;
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
unsigned long md_needed, mdblocks, total = 0;
@@ -1557,6 +1558,7 @@ static int ext4_da_reserve_space(struct
* in order to allocate nrblocks
* worse case is one extent per block
*/
+repeat:
spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
mdblocks = ext4_calc_metadata_amount(inode, total);
@@ -1567,6 +1569,10 @@ static int ext4_da_reserve_space(struct
if (ext4_claim_free_blocks(sbi, total)) {
spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
+ if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
+ yield();
+ goto repeat;
+ }
return -ENOSPC;
}
EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
@@ -1864,20 +1870,18 @@ static void ext4_da_block_invalidatepage
static int mpage_da_map_blocks(struct mpage_da_data *mpd)
{
int err = 0;
+ struct buffer_head new;
struct buffer_head *lbh = &mpd->lbh;
sector_t next = lbh->b_blocknr;
- struct buffer_head new;
/*
* We consider only non-mapped and non-allocated blocks
*/
if (buffer_mapped(lbh) && !buffer_delay(lbh))
return 0;
-
new.b_state = lbh->b_state;
new.b_blocknr = 0;
new.b_size = lbh->b_size;
-
/*
* If we didn't accumulate anything
* to write simply return
@@ -1910,6 +1914,10 @@ static int mpage_da_map_blocks(struct m
lbh->b_size >> mpd->inode->i_blkbits, err);
printk(KERN_EMERG "This should not happen.!! "
"Data will be lost\n");
+ if (err == -ENOSPC) {
+ printk(KERN_CRIT "Total free blocks count %lld\n",
+ ext4_count_free_blocks(mpd->inode->i_sb));
+ }
/* invlaidate all the pages */
ext4_da_block_invalidatepages(mpd, next,
lbh->b_size >> mpd->inode->i_blkbits);
diff -urNp linux-2.6.27.orig/fs/ext4/mballoc.c linux-2.6.27/fs/ext4/mballoc.c
--- linux-2.6.27.orig/fs/ext4/mballoc.c 2010-02-26 14:22:58.238252631 -0600
+++ linux-2.6.27/fs/ext4/mballoc.c 2010-02-26 14:23:05.967252672 -0600
@@ -4651,7 +4651,12 @@ ext4_fsblk_t ext4_mb_new_blocks(handle_t
/*
* With delalloc we already reserved the blocks
*/
- if (ext4_claim_free_blocks(sbi, ar->len)) {
+ while (ar->len && ext4_claim_free_blocks(sbi, ar->len)) {
+ /* let others to free the space */
+ yield();
+ ar->len = ar->len >> 1;
+ }
+ if (!ar->len) {
*errp = -ENOSPC;
return 0;
}
From: Aneesh Kumar K.V <[email protected]>
Date: Fri Oct 10 09:39:00 2008 -0400
Subject: ext4: Add percpu dirty block accounting.
commit 6bc6e63fcd7dac9e633ea29f1fddd9580ab28f3f upstream.
This patch adds dirty block accounting using percpu_counters. Delayed
allocation block reservation is now done by updating dirty block
counter. In a later patch we switch to non delalloc mode if the
filesystem free blocks is greater than 150% of total filesystem dirty
blocks
Signed-off-by: Aneesh Kumar K.V <[email protected]>
Signed-off-by: Mingming Cao<[email protected]>
Signed-off-by: "Theodore Ts'o" <[email protected]>
Signed-off-by: Jayson R. King <[email protected]>
---
fs/ext4/balloc.c | 62 ++++++++++++++++++++++++++++----------------
fs/ext4/ext4_sb.h | 1
fs/ext4/inode.c | 22 +++++++--------
fs/ext4/mballoc.c | 31 +++++++++-------------
fs/ext4/super.c | 8 ++++-
5 files changed, 73 insertions(+), 51 deletions(-)
diff -urNp linux-2.6.27.orig/fs/ext4/balloc.c linux-2.6.27/fs/ext4/balloc.c
--- linux-2.6.27.orig/fs/ext4/balloc.c 2010-02-26 14:21:03.148252808 -0600
+++ linux-2.6.27/fs/ext4/balloc.c 2010-02-26 14:21:14.292252864 -0600
@@ -1757,26 +1757,38 @@ out:
int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
ext4_fsblk_t nblocks)
{
- s64 free_blocks;
+ s64 free_blocks, dirty_blocks;
ext4_fsblk_t root_blocks = 0;
struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
+ struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
- free_blocks = percpu_counter_read(fbc);
+ free_blocks = percpu_counter_read_positive(fbc);
+ dirty_blocks = percpu_counter_read_positive(dbc);
if (!capable(CAP_SYS_RESOURCE) &&
sbi->s_resuid != current->fsuid &&
(sbi->s_resgid == 0 || !in_group_p(sbi->s_resgid)))
root_blocks = ext4_r_blocks_count(sbi->s_es);
- if (free_blocks - (nblocks + root_blocks) < EXT4_FREEBLOCKS_WATERMARK)
- free_blocks = percpu_counter_sum(&sbi->s_freeblocks_counter);
-
- if (free_blocks < (root_blocks + nblocks))
+ if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
+ EXT4_FREEBLOCKS_WATERMARK) {
+ free_blocks = percpu_counter_sum(fbc);
+ dirty_blocks = percpu_counter_sum(dbc);
+ if (dirty_blocks < 0) {
+ printk(KERN_CRIT "Dirty block accounting "
+ "went wrong %lld\n",
+ dirty_blocks);
+ }
+ }
+ /* Check whether we have space after
+ * accounting for current dirty blocks
+ */
+ if (free_blocks < ((s64)(root_blocks + nblocks) + dirty_blocks))
/* we don't have free space */
return -ENOSPC;
- /* reduce fs free blocks counter */
- percpu_counter_sub(fbc, nblocks);
+ /* Add the blocks to nblocks */
+ percpu_counter_add(dbc, nblocks);
return 0;
}
@@ -1792,23 +1804,28 @@ int ext4_claim_free_blocks(struct ext4_s
ext4_fsblk_t ext4_has_free_blocks(struct ext4_sb_info *sbi,
ext4_fsblk_t nblocks)
{
- ext4_fsblk_t free_blocks;
+ ext4_fsblk_t free_blocks, dirty_blocks;
ext4_fsblk_t root_blocks = 0;
+ struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
+ struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
- free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
+ free_blocks = percpu_counter_read_positive(fbc);
+ dirty_blocks = percpu_counter_read_positive(dbc);
if (!capable(CAP_SYS_RESOURCE) &&
sbi->s_resuid != current->fsuid &&
(sbi->s_resgid == 0 || !in_group_p(sbi->s_resgid)))
root_blocks = ext4_r_blocks_count(sbi->s_es);
- if (free_blocks - (nblocks + root_blocks) < EXT4_FREEBLOCKS_WATERMARK)
- free_blocks = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
-
- if (free_blocks <= root_blocks)
+ if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
+ EXT4_FREEBLOCKS_WATERMARK) {
+ free_blocks = percpu_counter_sum_positive(fbc);
+ dirty_blocks = percpu_counter_sum_positive(dbc);
+ }
+ if (free_blocks <= (root_blocks + dirty_blocks))
/* we don't have free space */
return 0;
- if (free_blocks - root_blocks < nblocks)
+ if (free_blocks - (root_blocks + dirty_blocks) < nblocks)
return free_blocks - root_blocks;
return nblocks;
}
@@ -2089,13 +2106,14 @@ allocated:
le16_add_cpu(&gdp->bg_free_blocks_count, -num);
gdp->bg_checksum = ext4_group_desc_csum(sbi, group_no, gdp);
spin_unlock(sb_bgl_lock(sbi, group_no));
- if (!EXT4_I(inode)->i_delalloc_reserved_flag && (*count != num)) {
- /*
- * we allocated less blocks than we
- * claimed. Add the difference back.
- */
- percpu_counter_add(&sbi->s_freeblocks_counter, *count - num);
- }
+ percpu_counter_sub(&sbi->s_freeblocks_counter, num);
+ /*
+ * Now reduce the dirty block count also. Should not go negative
+ */
+ if (!EXT4_I(inode)->i_delalloc_reserved_flag)
+ percpu_counter_sub(&sbi->s_dirtyblocks_counter, *count);
+ else
+ percpu_counter_sub(&sbi->s_dirtyblocks_counter, num);
if (sbi->s_log_groups_per_flex) {
ext4_group_t flex_group = ext4_flex_group(sbi, group_no);
spin_lock(sb_bgl_lock(sbi, flex_group));
diff -urNp linux-2.6.27.orig/fs/ext4/ext4_sb.h linux-2.6.27/fs/ext4/ext4_sb.h
--- linux-2.6.27.orig/fs/ext4/ext4_sb.h 2010-02-26 14:10:41.167252303 -0600
+++ linux-2.6.27/fs/ext4/ext4_sb.h 2010-02-26 14:21:14.292252864 -0600
@@ -60,6 +60,7 @@ struct ext4_sb_info {
struct percpu_counter s_freeblocks_counter;
struct percpu_counter s_freeinodes_counter;
struct percpu_counter s_dirs_counter;
+ struct percpu_counter s_dirtyblocks_counter;
struct blockgroup_lock s_blockgroup_lock;
/* root of the per fs reservation window tree */
diff -urNp linux-2.6.27.orig/fs/ext4/inode.c linux-2.6.27/fs/ext4/inode.c
--- linux-2.6.27.orig/fs/ext4/inode.c 2010-02-26 14:21:03.149252723 -0600
+++ linux-2.6.27/fs/ext4/inode.c 2010-02-26 14:21:14.293252633 -0600
@@ -1032,19 +1032,20 @@ static void ext4_da_update_reserve_space
BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
- /* Account for allocated meta_blocks */
- mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
-
- /* update fs free blocks counter for truncate case */
- percpu_counter_add(&sbi->s_freeblocks_counter, mdb_free);
+ if (mdb_free) {
+ /* Account for allocated meta_blocks */
+ mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
+
+ /* update fs dirty blocks counter */
+ percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
+ EXT4_I(inode)->i_allocated_meta_blocks = 0;
+ EXT4_I(inode)->i_reserved_meta_blocks = mdb;
+ }
/* update per-inode reservations */
BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
EXT4_I(inode)->i_reserved_data_blocks -= used;
- BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
- EXT4_I(inode)->i_reserved_meta_blocks = mdb;
- EXT4_I(inode)->i_allocated_meta_blocks = 0;
spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
/*
@@ -1609,8 +1610,8 @@ static void ext4_da_release_space(struct
release = to_free + mdb_free;
- /* update fs free blocks counter for truncate case */
- percpu_counter_add(&sbi->s_freeblocks_counter, release);
+ /* update fs dirty blocks counter for truncate case */
+ percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
/* update per-inode reservations */
BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
@@ -2546,7 +2547,6 @@ static int ext4_da_write_begin(struct fi
index = pos >> PAGE_CACHE_SHIFT;
from = pos & (PAGE_CACHE_SIZE - 1);
to = from + len;
-
retry:
/*
* With delayed allocation, we don't log the i_disksize update
diff -urNp linux-2.6.27.orig/fs/ext4/mballoc.c linux-2.6.27/fs/ext4/mballoc.c
--- linux-2.6.27.orig/fs/ext4/mballoc.c 2010-02-26 14:21:03.150252678 -0600
+++ linux-2.6.27/fs/ext4/mballoc.c 2010-02-26 14:21:14.294252546 -0600
@@ -3100,7 +3100,7 @@ void exit_ext4_mballoc(void)
*/
static noinline_for_stack int
ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
- handle_t *handle)
+ handle_t *handle, unsigned long reserv_blks)
{
struct buffer_head *bitmap_bh = NULL;
struct ext4_super_block *es;
@@ -3188,21 +3188,16 @@ ext4_mb_mark_diskspace_used(struct ext4_
le16_add_cpu(&gdp->bg_free_blocks_count, -ac->ac_b_ex.fe_len);
gdp->bg_checksum = ext4_group_desc_csum(sbi, ac->ac_b_ex.fe_group, gdp);
spin_unlock(sb_bgl_lock(sbi, ac->ac_b_ex.fe_group));
-
+ percpu_counter_sub(&sbi->s_freeblocks_counter, ac->ac_b_ex.fe_len);
/*
- * free blocks account has already be reduced/reserved
- * at write_begin() time for delayed allocation
- * do not double accounting
+ * Now reduce the dirty block count also. Should not go negative
*/
- if (!(ac->ac_flags & EXT4_MB_DELALLOC_RESERVED) &&
- ac->ac_o_ex.fe_len != ac->ac_b_ex.fe_len) {
- /*
- * we allocated less blocks than we calimed
- * Add the difference back
- */
- percpu_counter_add(&sbi->s_freeblocks_counter,
- ac->ac_o_ex.fe_len - ac->ac_b_ex.fe_len);
- }
+ if (!(ac->ac_flags & EXT4_MB_DELALLOC_RESERVED))
+ /* release all the reserved blocks if non delalloc */
+ percpu_counter_sub(&sbi->s_dirtyblocks_counter, reserv_blks);
+ else
+ percpu_counter_sub(&sbi->s_dirtyblocks_counter,
+ ac->ac_b_ex.fe_len);
if (sbi->s_log_groups_per_flex) {
ext4_group_t flex_group = ext4_flex_group(sbi,
@@ -4636,12 +4631,13 @@ static int ext4_mb_discard_preallocation
ext4_fsblk_t ext4_mb_new_blocks(handle_t *handle,
struct ext4_allocation_request *ar, int *errp)
{
+ int freed;
struct ext4_allocation_context *ac = NULL;
struct ext4_sb_info *sbi;
struct super_block *sb;
ext4_fsblk_t block = 0;
- int freed;
- int inquota;
+ unsigned long inquota;
+ unsigned long reserv_blks = 0;
sb = ar->inode->i_sb;
sbi = EXT4_SB(sb);
@@ -4659,6 +4655,7 @@ ext4_fsblk_t ext4_mb_new_blocks(handle_t
*errp = -ENOSPC;
return 0;
}
+ reserv_blks = ar->len;
}
while (ar->len && DQUOT_ALLOC_BLOCK(ar->inode, ar->len)) {
ar->flags |= EXT4_MB_HINT_NOPREALLOC;
@@ -4704,7 +4701,7 @@ repeat:
ext4_mb_new_preallocation(ac);
}
if (likely(ac->ac_status == AC_STATUS_FOUND)) {
- *errp = ext4_mb_mark_diskspace_used(ac, handle);
+ *errp = ext4_mb_mark_diskspace_used(ac, handle, reserv_blks);
if (*errp == -EAGAIN) {
/*
* drop the reference that we took
diff -urNp linux-2.6.27.orig/fs/ext4/super.c linux-2.6.27/fs/ext4/super.c
--- linux-2.6.27.orig/fs/ext4/super.c 2010-02-26 14:10:41.170252357 -0600
+++ linux-2.6.27/fs/ext4/super.c 2010-02-26 14:21:14.295252319 -0600
@@ -520,6 +520,7 @@ static void ext4_put_super(struct super_
percpu_counter_destroy(&sbi->s_freeblocks_counter);
percpu_counter_destroy(&sbi->s_freeinodes_counter);
percpu_counter_destroy(&sbi->s_dirs_counter);
+ percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
brelse(sbi->s_sbh);
#ifdef CONFIG_QUOTA
for (i = 0; i < MAXQUOTAS; i++)
@@ -2279,6 +2280,9 @@ static int ext4_fill_super(struct super_
err = percpu_counter_init(&sbi->s_dirs_counter,
ext4_count_dirs(sb));
}
+ if (!err) {
+ err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
+ }
if (err) {
printk(KERN_ERR "EXT4-fs: insufficient memory\n");
goto failed_mount3;
@@ -2516,6 +2520,7 @@ failed_mount3:
percpu_counter_destroy(&sbi->s_freeblocks_counter);
percpu_counter_destroy(&sbi->s_freeinodes_counter);
percpu_counter_destroy(&sbi->s_dirs_counter);
+ percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
failed_mount2:
for (i = 0; i < db_count; i++)
brelse(sbi->s_group_desc[i]);
@@ -3207,7 +3212,8 @@ static int ext4_statfs(struct dentry *de
buf->f_type = EXT4_SUPER_MAGIC;
buf->f_bsize = sb->s_blocksize;
buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
- buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
+ buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
+ percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
ext4_free_blocks_count_set(es, buf->f_bfree);
buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
if (buf->f_bfree < ext4_r_blocks_count(es))
From: Aneesh Kumar K.V <[email protected]>
Date: Mon Sep 8 23:05:34 2008 -0400
Subject: ext4: Retry block allocation if we have free blocks left
commit df22291ff0fde0d350cf15dac3e5cc33ac528875 upstream.
When we truncate files, the meta-data blocks released are not reused
untill we commit the truncate transaction. That means delayed get_block
request will return ENOSPC even if we have free blocks left. Force a
journal commit and retry block allocation if we get ENOSPC with free
blocks left.
Signed-off-by: Aneesh Kumar K.V <[email protected]>
Signed-off-by: Mingming Cao <[email protected]>
Signed-off-by: "Theodore Ts'o" <[email protected]>
Signed-off-by: Jayson R. King <[email protected]>
---
fs/ext4/inode.c | 81 ++++++++++++++++++++++++++++++++--------------
1 file changed, 57 insertions(+), 24 deletions(-)
diff -urNp linux-2.6.27.orig/fs/ext4/inode.c linux-2.6.27/fs/ext4/inode.c
--- linux-2.6.27.orig/fs/ext4/inode.c 2010-02-26 14:23:46.597252740 -0600
+++ linux-2.6.27/fs/ext4/inode.c 2010-02-26 14:24:08.405252691 -0600
@@ -1661,6 +1661,7 @@ struct mpage_da_data {
struct writeback_control *wbc;
int io_done;
long pages_written;
+ int retval;
};
/*
@@ -1858,6 +1859,24 @@ static void ext4_da_block_invalidatepage
return;
}
+static void ext4_print_free_blocks(struct inode *inode)
+{
+ struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
+ printk(KERN_EMERG "Total free blocks count %lld\n",
+ ext4_count_free_blocks(inode->i_sb));
+ printk(KERN_EMERG "Free/Dirty block details\n");
+ printk(KERN_EMERG "free_blocks=%lld\n",
+ percpu_counter_sum(&sbi->s_freeblocks_counter));
+ printk(KERN_EMERG "dirty_blocks=%lld\n",
+ percpu_counter_sum(&sbi->s_dirtyblocks_counter));
+ printk(KERN_EMERG "Block reservation details\n");
+ printk(KERN_EMERG "i_reserved_data_blocks=%lu\n",
+ EXT4_I(inode)->i_reserved_data_blocks);
+ printk(KERN_EMERG "i_reserved_meta_blocks=%lu\n",
+ EXT4_I(inode)->i_reserved_meta_blocks);
+ return;
+}
+
/*
* mpage_da_map_blocks - go through given space
*
@@ -1872,7 +1891,7 @@ static int mpage_da_map_blocks(struct m
int err = 0;
struct buffer_head new;
struct buffer_head *lbh = &mpd->lbh;
- sector_t next = lbh->b_blocknr;
+ sector_t next;
/*
* We consider only non-mapped and non-allocated blocks
@@ -1882,6 +1901,7 @@ static int mpage_da_map_blocks(struct m
new.b_state = lbh->b_state;
new.b_blocknr = 0;
new.b_size = lbh->b_size;
+ next = lbh->b_blocknr;
/*
* If we didn't accumulate anything
* to write simply return
@@ -1898,6 +1918,13 @@ static int mpage_da_map_blocks(struct m
*/
if (err == -EAGAIN)
return 0;
+
+ if (err == -ENOSPC &&
+ ext4_count_free_blocks(mpd->inode->i_sb)) {
+ mpd->retval = err;
+ return 0;
+ }
+
/*
* get block failure will cause us
* to loop in writepages. Because
@@ -1915,8 +1942,7 @@ static int mpage_da_map_blocks(struct m
printk(KERN_EMERG "This should not happen.!! "
"Data will be lost\n");
if (err == -ENOSPC) {
- printk(KERN_CRIT "Total free blocks count %lld\n",
- ext4_count_free_blocks(mpd->inode->i_sb));
+ ext4_print_free_blocks(mpd->inode);
}
/* invlaidate all the pages */
ext4_da_block_invalidatepages(mpd, next,
@@ -2141,39 +2167,36 @@ static int __mpage_da_writepage(struct p
*/
static int mpage_da_writepages(struct address_space *mapping,
struct writeback_control *wbc,
- get_block_t get_block)
+ struct mpage_da_data *mpd)
{
- struct mpage_da_data mpd;
long to_write;
int ret;
- if (!get_block)
+ if (!mpd->get_block)
return generic_writepages(mapping, wbc);
- mpd.wbc = wbc;
- mpd.inode = mapping->host;
- mpd.lbh.b_size = 0;
- mpd.lbh.b_state = 0;
- mpd.lbh.b_blocknr = 0;
- mpd.first_page = 0;
- mpd.next_page = 0;
- mpd.get_block = get_block;
- mpd.io_done = 0;
- mpd.pages_written = 0;
+ mpd->lbh.b_size = 0;
+ mpd->lbh.b_state = 0;
+ mpd->lbh.b_blocknr = 0;
+ mpd->first_page = 0;
+ mpd->next_page = 0;
+ mpd->io_done = 0;
+ mpd->pages_written = 0;
+ mpd->retval = 0;
to_write = wbc->nr_to_write;
- ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, &mpd);
+ ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, mpd);
/*
* Handle last extent of pages
*/
- if (!mpd.io_done && mpd.next_page != mpd.first_page) {
- if (mpage_da_map_blocks(&mpd) == 0)
- mpage_da_submit_io(&mpd);
+ if (!mpd->io_done && mpd->next_page != mpd->first_page) {
+ if (mpage_da_map_blocks(mpd) == 0)
+ mpage_da_submit_io(mpd);
}
- wbc->nr_to_write = to_write - mpd.pages_written;
+ wbc->nr_to_write = to_write - mpd->pages_written;
return ret;
}
@@ -2420,6 +2443,7 @@ static int ext4_da_writepages(struct add
{
handle_t *handle = NULL;
loff_t range_start = 0;
+ struct mpage_da_data mpd;
struct inode *inode = mapping->host;
int needed_blocks, ret = 0, nr_to_writebump = 0;
long to_write, pages_skipped = 0;
@@ -2467,6 +2491,9 @@ static int ext4_da_writepages(struct add
range_start = wbc->range_start;
pages_skipped = wbc->pages_skipped;
+ mpd.wbc = wbc;
+ mpd.inode = mapping->host;
+
restart_loop:
to_write = wbc->nr_to_write;
while (!ret && to_write > 0) {
@@ -2502,11 +2529,17 @@ restart_loop:
goto out_writepages;
}
}
From: Aneesh Kumar K.V <[email protected]>
Date: Tue Oct 14 09:20:19 2008 -0400
Subject: ext4: Use tag dirty lookup during mpage_da_submit_io
commit af6f029d3836eb7264cd3fbb13a6baf0e5fdb5ea upstream.
This enables us to drop the range_cont writeback mode
use from ext4_da_writepages.
Signed-off-by: Aneesh Kumar K.V <[email protected]>
Signed-off-by: Jayson R. King <[email protected]>
---
fs/ext4/inode.c | 30 +++++++++++++-----------------
1 file changed, 13 insertions(+), 17 deletions(-)
diff -urNp linux-2.6.27.orig/fs/ext4/inode.c linux-2.6.27/fs/ext4/inode.c
--- linux-2.6.27.orig/fs/ext4/inode.c 2010-02-26 14:24:53.269252744 -0600
+++ linux-2.6.27/fs/ext4/inode.c 2010-02-26 14:25:01.277252724 -0600
@@ -1699,17 +1699,23 @@ static int mpage_da_submit_io(struct mpa
pagevec_init(&pvec, 0);
while (index <= end) {
- nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
+ /*
+ * We can use PAGECACHE_TAG_DIRTY lookup here because
+ * even though we have cleared the dirty flag on the page
+ * We still keep the page in the radix tree with tag
+ * PAGECACHE_TAG_DIRTY. See clear_page_dirty_for_io.
+ * The PAGECACHE_TAG_DIRTY is cleared in set_page_writeback
+ * which is called via the below writepage callback.
+ */
+ nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
+ PAGECACHE_TAG_DIRTY,
+ min(end - index,
+ (pgoff_t)PAGEVEC_SIZE-1) + 1);
if (nr_pages == 0)
break;
for (i = 0; i < nr_pages; i++) {
struct page *page = pvec.pages[i];
- index = page->index;
- if (index > end)
- break;
- index++;
-
BUG_ON(!PageLocked(page));
BUG_ON(PageWriteback(page));
@@ -2442,7 +2448,6 @@ static int ext4_da_writepages(struct add
struct writeback_control *wbc)
{
handle_t *handle = NULL;
- loff_t range_start = 0;
struct mpage_da_data mpd;
struct inode *inode = mapping->host;
int needed_blocks, ret = 0, nr_to_writebump = 0;
@@ -2481,14 +2486,7 @@ static int ext4_da_writepages(struct add
wbc->nr_to_write = sbi->s_mb_stream_request;
}
- if (!wbc->range_cyclic)
- /*
- * If range_cyclic is not set force range_cont
- * and save the old writeback_index
- */
- wbc->range_cont = 1;
- range_start = wbc->range_start;
pages_skipped = wbc->pages_skipped;
mpd.wbc = wbc;
@@ -2559,9 +2557,8 @@ restart_loop:
wbc->nr_to_write = to_write;
}
- if (wbc->range_cont && (pages_skipped != wbc->pages_skipped)) {
+ if (!wbc->range_cyclic && (pages_skipped != wbc->pages_skipped)) {
/* We skipped pages in this loop */
- wbc->range_start = range_start;
wbc->nr_to_write = to_write +
wbc->pages_skipped - pages_skipped;
wbc->pages_skipped = pages_skipped;
@@ -2570,7 +2567,6 @@ restart_loop:
out_writepages:
wbc->nr_to_write = to_write - nr_to_writebump;
- wbc->range_start = range_start;
return ret;
}
From: Aneesh Kumar K.V <[email protected]>
Date: Thu Oct 16 10:10:36 2008 -0400
Subject: ext4: Fix file fragmentation during large file write.
commit 22208dedbd7626e5fc4339c417f8d24cc21f79d7 upstream.
The range_cyclic writeback mode uses the address_space writeback_index
as the start index for writeback. With delayed allocation we were
updating writeback_index wrongly resulting in highly fragmented file.
This patch reduces the number of extents reduced from 4000 to 27 for a
3GB file.
Signed-off-by: Aneesh Kumar K.V <[email protected]>
Signed-off-by: Theodore Ts'o <[email protected]>
[[email protected]: Some changed lines from the original version of this patch were dropped, since they were rolled up with another cherry-picked patch applied to 2.6.27.y earlier.]
Signed-off-by: Jayson R. King <[email protected]>
---
fs/ext4/inode.c | 88 ++++++++++++++++++++++++++++------------------
1 file changed, 55 insertions(+), 33 deletions(-)
diff -urNp linux-2.6.27.orig/fs/ext4/inode.c linux-2.6.27/fs/ext4/inode.c
--- linux-2.6.27.orig/fs/ext4/inode.c 2010-02-26 14:25:37.157252759 -0600
+++ linux-2.6.27/fs/ext4/inode.c 2010-02-26 14:27:31.342252417 -0600
@@ -1721,7 +1721,11 @@ static int mpage_da_submit_io(struct mpa
pages_skipped = mpd->wbc->pages_skipped;
err = mapping->a_ops->writepage(page, mpd->wbc);
- if (!err)
+ if (!err && (pages_skipped == mpd->wbc->pages_skipped))
+ /*
+ * have successfully written the page
+ * without skipping the same
+ */
mpd->pages_written++;
/*
* In error case, we have to continue because
@@ -2175,7 +2179,6 @@ static int mpage_da_writepages(struct ad
struct writeback_control *wbc,
struct mpage_da_data *mpd)
{
- long to_write;
int ret;
if (!mpd->get_block)
@@ -2190,19 +2193,18 @@ static int mpage_da_writepages(struct ad
mpd->pages_written = 0;
mpd->retval = 0;
- to_write = wbc->nr_to_write;
-
ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, mpd);
-
/*
* Handle last extent of pages
*/
if (!mpd->io_done && mpd->next_page != mpd->first_page) {
if (mpage_da_map_blocks(mpd) == 0)
mpage_da_submit_io(mpd);
- }
- wbc->nr_to_write = to_write - mpd->pages_written;
+ mpd->io_done = 1;
+ ret = MPAGE_DA_EXTENT_TAIL;
+ }
+ wbc->nr_to_write -= mpd->pages_written;
return ret;
}
@@ -2447,11 +2449,14 @@ static int ext4_da_writepages_trans_bloc
static int ext4_da_writepages(struct address_space *mapping,
struct writeback_control *wbc)
{
+ pgoff_t index;
+ int range_whole = 0;
handle_t *handle = NULL;
struct mpage_da_data mpd;
struct inode *inode = mapping->host;
+ int no_nrwrite_index_update;
+ long pages_written = 0, pages_skipped;
int needed_blocks, ret = 0, nr_to_writebump = 0;
- long to_write, pages_skipped = 0;
struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
/*
@@ -2485,16 +2490,26 @@ static int ext4_da_writepages(struct add
nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
wbc->nr_to_write = sbi->s_mb_stream_request;
}
+ if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
+ range_whole = 1;
-
- pages_skipped = wbc->pages_skipped;
+ if (wbc->range_cyclic)
+ index = mapping->writeback_index;
+ else
+ index = wbc->range_start >> PAGE_CACHE_SHIFT;
mpd.wbc = wbc;
mpd.inode = mapping->host;
-restart_loop:
- to_write = wbc->nr_to_write;
- while (!ret && to_write > 0) {
+ /*
+ * we don't want write_cache_pages to update
+ * nr_to_write and writeback_index
+ */
+ no_nrwrite_index_update = wbc->no_nrwrite_index_update;
+ wbc->no_nrwrite_index_update = 1;
+ pages_skipped = wbc->pages_skipped;
+
+ while (!ret && wbc->nr_to_write > 0) {
/*
* we insert one extent at a time. So we need
@@ -2527,46 +2542,53 @@ restart_loop:
goto out_writepages;
}
}
- to_write -= wbc->nr_to_write;
-
mpd.get_block = ext4_da_get_block_write;
ret = mpage_da_writepages(mapping, wbc, &mpd);
ext4_journal_stop(handle);
- if (mpd.retval == -ENOSPC)
+ if (mpd.retval == -ENOSPC) {
+ /* commit the transaction which would
+ * free blocks released in the transaction
+ * and try again
+ */
jbd2_journal_force_commit_nested(sbi->s_journal);
-
- /* reset the retry count */
- if (ret == MPAGE_DA_EXTENT_TAIL) {
+ wbc->pages_skipped = pages_skipped;
+ ret = 0;
+ } else if (ret == MPAGE_DA_EXTENT_TAIL) {
/*
* got one extent now try with
* rest of the pages
*/
- to_write += wbc->nr_to_write;
+ pages_written += mpd.pages_written;
+ wbc->pages_skipped = pages_skipped;
ret = 0;
- } else if (wbc->nr_to_write) {
+ } else if (wbc->nr_to_write)
/*
* There is no more writeout needed
* or we requested for a noblocking writeout
* and we found the device congested
*/
- to_write += wbc->nr_to_write;
break;
- }
- wbc->nr_to_write = to_write;
- }
-
- if (!wbc->range_cyclic && (pages_skipped != wbc->pages_skipped)) {
- /* We skipped pages in this loop */
- wbc->nr_to_write = to_write +
- wbc->pages_skipped - pages_skipped;
- wbc->pages_skipped = pages_skipped;
- goto restart_loop;
}
+ if (pages_skipped != wbc->pages_skipped)
+ printk(KERN_EMERG "This should not happen leaving %s "
+ "with nr_to_write = %ld ret = %d\n",
+ __func__, wbc->nr_to_write, ret);
+
+ /* Update index */
+ index += pages_written;
+ if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
+ /*
+ * set the writeback_index so that range_cyclic
+ * mode will write it back later
+ */
+ mapping->writeback_index = index;
out_writepages:
- wbc->nr_to_write = to_write - nr_to_writebump;
+ if (!no_nrwrite_index_update)
+ wbc->no_nrwrite_index_update = 0;
+ wbc->nr_to_write -= nr_to_writebump;
return ret;
}
From: Aneesh Kumar K.V <[email protected]>
Date: Tue Oct 14 09:21:02 2008 -0400
Subject: vfs: Remove the range_cont writeback mode.
commit 74baaaaec8b4f22e1ae279f5ecca4ff705b28912 upstream.
Ext4 was the only user of range_cont writeback mode and ext4 switched
to a different method. So remove the range_cont mode which is not used
in the kernel.
Signed-off-by: Aneesh Kumar K.V <[email protected]>
Signed-off-by: "Theodore Ts'o" <[email protected]>
CC: [email protected]
Signed-off-by: Jayson R. King <[email protected]>
---
include/linux/writeback.h | 1 -
mm/page-writeback.c | 2 --
2 files changed, 3 deletions(-)
diff -urNp linux-2.6.27.orig/include/linux/writeback.h linux-2.6.27/include/linux/writeback.h
--- linux-2.6.27.orig/include/linux/writeback.h 2010-02-26 14:10:42.106252410 -0600
+++ linux-2.6.27/include/linux/writeback.h 2010-02-26 14:25:58.868253029 -0600
@@ -62,7 +62,6 @@ struct writeback_control {
unsigned for_writepages:1; /* This is a writepages() call */
unsigned range_cyclic:1; /* range_start is cyclic */
unsigned more_io:1; /* more io to be dispatched */
- unsigned range_cont:1;
};
/*
diff -urNp linux-2.6.27.orig/mm/page-writeback.c linux-2.6.27/mm/page-writeback.c
--- linux-2.6.27.orig/mm/page-writeback.c 2010-02-26 14:10:42.197252340 -0600
+++ linux-2.6.27/mm/page-writeback.c 2010-02-26 14:25:58.869252611 -0600
@@ -1027,8 +1027,6 @@ continue_unlock:
if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
mapping->writeback_index = done_index;
- if (wbc->range_cont)
- wbc->range_start = index << PAGE_CACHE_SHIFT;
return ret;
}
EXPORT_SYMBOL(write_cache_pages);
From: Aneesh Kumar K.V <[email protected]>
Date: Thu Oct 9 10:56:23 2008 -0400
Subject: ext4: Make sure all the block allocation paths reserve blocks
commit a30d542a0035b886ffaafd0057ced0a2b28c3a4f upstream.
With delayed allocation we need to make sure block are reserved before
we attempt to allocate them. Otherwise we get block allocation failure
(ENOSPC) during writepages which cannot be handled. This would mean
silent data loss (We do a printk stating data will be lost). This patch
updates the DIO and fallocate code path to do block reservation before
block allocation. This is needed to make sure parallel DIO and fallocate
request doesn't take block out of delayed reserve space.
When free blocks count go below a threshold we switch to a slow patch
which looks at other CPU's accumulated percpu counter values.
Signed-off-by: Aneesh Kumar K.V <[email protected]>
Signed-off-by: "Theodore Ts'o" <[email protected]>
Signed-off-by: Jayson R. King <[email protected]>
---
fs/ext4/balloc.c | 58 +++++++++++++++++++++++++++++++-------------
fs/ext4/ext4.h | 13 +++++++++
fs/ext4/inode.c | 5 ---
fs/ext4/mballoc.c | 23 +++++++++--------
4 files changed, 69 insertions(+), 30 deletions(-)
diff -urNp linux-2.6.27.orig/fs/ext4/balloc.c linux-2.6.27/fs/ext4/balloc.c
--- linux-2.6.27.orig/fs/ext4/balloc.c 2010-02-26 14:19:37.651252924 -0600
+++ linux-2.6.27/fs/ext4/balloc.c 2010-02-26 14:20:04.172252779 -0600
@@ -1754,6 +1754,32 @@ out:
return ret;
}
+int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
+ ext4_fsblk_t nblocks)
+{
+ s64 free_blocks;
+ ext4_fsblk_t root_blocks = 0;
+ struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
+
+ free_blocks = percpu_counter_read(fbc);
+
+ if (!capable(CAP_SYS_RESOURCE) &&
+ sbi->s_resuid != current->fsuid &&
+ (sbi->s_resgid == 0 || !in_group_p(sbi->s_resgid)))
+ root_blocks = ext4_r_blocks_count(sbi->s_es);
+
+ if (free_blocks - (nblocks + root_blocks) < EXT4_FREEBLOCKS_WATERMARK)
+ free_blocks = percpu_counter_sum(&sbi->s_freeblocks_counter);
+
+ if (free_blocks < (root_blocks + nblocks))
+ /* we don't have free space */
+ return -ENOSPC;
+
+ /* reduce fs free blocks counter */
+ percpu_counter_sub(fbc, nblocks);
+ return 0;
+}
+
/**
* ext4_has_free_blocks()
* @sbi: in-core super block structure.
@@ -1775,18 +1801,17 @@ ext4_fsblk_t ext4_has_free_blocks(struct
sbi->s_resuid != current->fsuid &&
(sbi->s_resgid == 0 || !in_group_p(sbi->s_resgid)))
root_blocks = ext4_r_blocks_count(sbi->s_es);
-#ifdef CONFIG_SMP
- if (free_blocks - root_blocks < FBC_BATCH)
- free_blocks =
- percpu_counter_sum(&sbi->s_freeblocks_counter);
-#endif
+
+ if (free_blocks - (nblocks + root_blocks) < EXT4_FREEBLOCKS_WATERMARK)
+ free_blocks = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
+
if (free_blocks <= root_blocks)
/* we don't have free space */
return 0;
if (free_blocks - root_blocks < nblocks)
return free_blocks - root_blocks;
return nblocks;
- }
+}
/**
@@ -1865,14 +1890,11 @@ ext4_fsblk_t ext4_old_new_blocks(handle_
/*
* With delalloc we already reserved the blocks
*/
- *count = ext4_has_free_blocks(sbi, *count);
- }
- if (*count == 0) {
- *errp = -ENOSPC;
- return 0; /*return with ENOSPC error */
+ if (ext4_claim_free_blocks(sbi, *count)) {
+ *errp = -ENOSPC;
+ return 0; /*return with ENOSPC error */
+ }
}
- num = *count;
-
/*
* Check quota for allocation of this block.
*/
@@ -2067,9 +2089,13 @@ allocated:
le16_add_cpu(&gdp->bg_free_blocks_count, -num);
gdp->bg_checksum = ext4_group_desc_csum(sbi, group_no, gdp);
spin_unlock(sb_bgl_lock(sbi, group_no));
- if (!EXT4_I(inode)->i_delalloc_reserved_flag)
- percpu_counter_sub(&sbi->s_freeblocks_counter, num);
-
+ if (!EXT4_I(inode)->i_delalloc_reserved_flag && (*count != num)) {
+ /*
+ * we allocated less blocks than we
+ * claimed. Add the difference back.
+ */
+ percpu_counter_add(&sbi->s_freeblocks_counter, *count - num);
+ }
if (sbi->s_log_groups_per_flex) {
ext4_group_t flex_group = ext4_flex_group(sbi, group_no);
spin_lock(sb_bgl_lock(sbi, flex_group));
diff -urNp linux-2.6.27.orig/fs/ext4/ext4.h linux-2.6.27/fs/ext4/ext4.h
--- linux-2.6.27.orig/fs/ext4/ext4.h 2010-02-26 14:10:41.167252303 -0600
+++ linux-2.6.27/fs/ext4/ext4.h 2010-02-26 14:20:04.172252779 -0600
@@ -1015,6 +1015,8 @@ extern ext4_fsblk_t ext4_new_blocks(hand
unsigned long *count, int *errp);
extern ext4_fsblk_t ext4_old_new_blocks(handle_t *handle, struct inode *inode,
ext4_fsblk_t goal, unsigned long *count, int *errp);
+extern int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
+ ext4_fsblk_t nblocks);
extern ext4_fsblk_t ext4_has_free_blocks(struct ext4_sb_info *sbi,
ext4_fsblk_t nblocks);
extern void ext4_free_blocks (handle_t *handle, struct inode *inode,
@@ -1245,6 +1247,17 @@ do { \
__ext4_std_error((sb), __func__, (errno)); \
} while (0)
+#ifdef CONFIG_SMP
+/* Each CPU can accumulate FBC_BATCH blocks in their local
+ * counters. So we need to make sure we have free blocks more
+ * than FBC_BATCH * nr_cpu_ids. Also add a window of 4 times.
+ */
+#define EXT4_FREEBLOCKS_WATERMARK (4 * (FBC_BATCH * nr_cpu_ids))
+#else
+#define EXT4_FREEBLOCKS_WATERMARK 0
+#endif
+
+
/*
* Inodes and files operations
*/
diff -urNp linux-2.6.27.orig/fs/ext4/inode.c linux-2.6.27/fs/ext4/inode.c
--- linux-2.6.27.orig/fs/ext4/inode.c 2010-02-26 14:15:33.187252503 -0600
+++ linux-2.6.27/fs/ext4/inode.c 2010-02-26 14:20:04.173252623 -0600
@@ -1564,13 +1564,10 @@ static int ext4_da_reserve_space(struct
md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
total = md_needed + nrblocks;
- if (ext4_has_free_blocks(sbi, total) < total) {
+ if (ext4_claim_free_blocks(sbi, total)) {
spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
return -ENOSPC;
}
- /* reduce fs free blocks counter */
- percpu_counter_sub(&sbi->s_freeblocks_counter, total);
-
EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
diff -urNp linux-2.6.27.orig/fs/ext4/mballoc.c linux-2.6.27/fs/ext4/mballoc.c
--- linux-2.6.27.orig/fs/ext4/mballoc.c 2010-02-26 14:10:41.169252348 -0600
+++ linux-2.6.27/fs/ext4/mballoc.c 2010-02-26 14:20:04.174252546 -0600
@@ -3194,9 +3194,15 @@ ext4_mb_mark_diskspace_used(struct ext4_
* at write_begin() time for delayed allocation
* do not double accounting
*/
- if (!(ac->ac_flags & EXT4_MB_DELALLOC_RESERVED))
- percpu_counter_sub(&sbi->s_freeblocks_counter,
- ac->ac_b_ex.fe_len);
+ if (!(ac->ac_flags & EXT4_MB_DELALLOC_RESERVED) &&
+ ac->ac_o_ex.fe_len != ac->ac_b_ex.fe_len) {
+ /*
+ * we allocated less blocks than we calimed
+ * Add the difference back
+ */
+ percpu_counter_add(&sbi->s_freeblocks_counter,
+ ac->ac_o_ex.fe_len - ac->ac_b_ex.fe_len);
+ }
if (sbi->s_log_groups_per_flex) {
ext4_group_t flex_group = ext4_flex_group(sbi,
@@ -4649,14 +4655,11 @@ ext4_fsblk_t ext4_mb_new_blocks(handle_t
/*
* With delalloc we already reserved the blocks
*/
- ar->len = ext4_has_free_blocks(sbi, ar->len);
- }
-
- if (ar->len == 0) {
- *errp = -ENOSPC;
- return 0;
+ if (ext4_claim_free_blocks(sbi, ar->len)) {
+ *errp = -ENOSPC;
+ return 0;
+ }
}
From: Aneesh Kumar K.V <[email protected]>
Date: Thu Oct 16 10:09:17 2008 -0400
Subject: vfs: Add no_nrwrite_index_update writeback control flag
commit 17bc6c30cf6bfffd816bdc53682dd46fc34a2cf4 upstream.
If no_nrwrite_index_update is set we don't update nr_to_write and
address space writeback_index in write_cache_pages. This change
enables a file system to skip these updates in write_cache_pages and do
them in the writepages() callback. This patch will be followed by an
ext4 patch that make use of these new flags.
Signed-off-by: Aneesh Kumar K.V <[email protected]>
Signed-off-by: "Theodore Ts'o" <[email protected]>
CC: [email protected]
[[email protected]: Modified the patch to account for subsequent changes in mainline being cherry-picked earlier for 2.6.27.y.]
Signed-off-by: Jayson R. King <[email protected]>
---
include/linux/writeback.h | 9 +++++++++
mm/page-writeback.c | 14 +++++++++-----
2 files changed, 18 insertions(+), 5 deletions(-)
diff -urNp linux-2.6.27.orig/include/linux/writeback.h linux-2.6.27/include/linux/writeback.h
--- linux-2.6.27.orig/include/linux/writeback.h 2010-02-26 14:26:43.084252730 -0600
+++ linux-2.6.27/include/linux/writeback.h 2010-02-26 14:26:48.468252521 -0600
@@ -62,6 +62,15 @@ struct writeback_control {
unsigned for_writepages:1; /* This is a writepages() call */
unsigned range_cyclic:1; /* range_start is cyclic */
unsigned more_io:1; /* more io to be dispatched */
+ /*
+ * write_cache_pages() won't update wbc->nr_to_write and
+ * mapping->writeback_index if no_nrwrite_index_update
+ * is set. write_cache_pages() may write more than we
+ * requested and we want to make sure nr_to_write and
+ * writeback_index are updated in a consistent manner
+ * so we use a single control to update them
+ */
+ unsigned no_nrwrite_index_update:1;
};
/*
diff -urNp linux-2.6.27.orig/mm/page-writeback.c linux-2.6.27/mm/page-writeback.c
--- linux-2.6.27.orig/mm/page-writeback.c 2010-02-26 14:26:43.085252646 -0600
+++ linux-2.6.27/mm/page-writeback.c 2010-02-26 14:26:48.469252852 -0600
@@ -878,6 +878,7 @@ int write_cache_pages(struct address_spa
pgoff_t done_index;
int cycled;
int range_whole = 0;
+ long nr_to_write = wbc->nr_to_write;
if (wbc->nonblocking && bdi_write_congested(bdi)) {
wbc->encountered_congestion = 1;
@@ -985,9 +986,9 @@ continue_unlock:
}
}
- if (wbc->nr_to_write > 0) {
- wbc->nr_to_write--;
- if (wbc->nr_to_write == 0 &&
+ if (nr_to_write > 0) {
+ nr_to_write--;
+ if (nr_to_write == 0 &&
wbc->sync_mode == WB_SYNC_NONE) {
/*
* We stop writing back only if we are
@@ -1024,8 +1025,11 @@ continue_unlock:
end = writeback_index - 1;
goto retry;
}
- if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
- mapping->writeback_index = done_index;
+ if (!wbc->no_nrwrite_index_update) {
+ if (wbc->range_cyclic || (range_whole && nr_to_write > 0))
+ mapping->writeback_index = done_index;
+ wbc->nr_to_write = nr_to_write;
+ }
return ret;
}
From: Aneesh Kumar K.V <[email protected]>
Date: Sat Feb 14 10:42:58 2009 -0500
Subject: ext4: Implement range_cyclic in ext4_da_writepages instead of write_cache_pages
commit 2acf2c261b823d9d9ed954f348b97620297a36b5 upstream.
With delayed allocation we lock the page in write_cache_pages() and
try to build an in memory extent of contiguous blocks. This is needed
so that we can get large contiguous blocks request. If range_cyclic
mode is enabled, write_cache_pages() will loop back to the 0 index if
no I/O has been done yet, and try to start writing from the beginning
of the range. That causes an attempt to take the page lock of lower
index page while holding the page lock of higher index page, which can
cause a dead lock with another writeback thread.
The solution is to implement the range_cyclic behavior in
ext4_da_writepages() instead.
http://bugzilla.kernel.org/show_bug.cgi?id=12579
Signed-off-by: Aneesh Kumar K.V <[email protected]>
Signed-off-by: "Theodore Ts'o" <[email protected]>
Signed-off-by: Jayson R. King <[email protected]>
---
fs/ext4/inode.c | 21 +++++++++++++++++++--
1 file changed, 19 insertions(+), 2 deletions(-)
diff -urNp linux-2.6.27.orig/fs/ext4/inode.c linux-2.6.27/fs/ext4/inode.c
--- linux-2.6.27.orig/fs/ext4/inode.c 2010-02-26 14:28:11.222252883 -0600
+++ linux-2.6.27/fs/ext4/inode.c 2010-02-26 14:28:16.190252556 -0600
@@ -2456,6 +2456,7 @@ static int ext4_da_writepages(struct add
struct inode *inode = mapping->host;
int no_nrwrite_index_update;
long pages_written = 0, pages_skipped;
+ int range_cyclic, cycled = 1, io_done = 0;
int needed_blocks, ret = 0, nr_to_writebump = 0;
struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
@@ -2493,9 +2494,15 @@ static int ext4_da_writepages(struct add
if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
range_whole = 1;
- if (wbc->range_cyclic)
+ range_cyclic = wbc->range_cyclic;
+ if (wbc->range_cyclic) {
index = mapping->writeback_index;
- else
+ if (index)
+ cycled = 0;
+ wbc->range_start = index << PAGE_CACHE_SHIFT;
+ wbc->range_end = LLONG_MAX;
+ wbc->range_cyclic = 0;
+ } else
index = wbc->range_start >> PAGE_CACHE_SHIFT;
mpd.wbc = wbc;
@@ -2509,6 +2516,7 @@ static int ext4_da_writepages(struct add
wbc->no_nrwrite_index_update = 1;
pages_skipped = wbc->pages_skipped;
+retry:
while (!ret && wbc->nr_to_write > 0) {
/*
@@ -2563,6 +2571,7 @@ static int ext4_da_writepages(struct add
pages_written += mpd.pages_written;
wbc->pages_skipped = pages_skipped;
ret = 0;
+ io_done = 1;
} else if (wbc->nr_to_write)
/*
* There is no more writeout needed
@@ -2571,6 +2580,13 @@ static int ext4_da_writepages(struct add
*/
break;
}
+ if (!io_done && !cycled) {
+ cycled = 1;
+ index = 0;
+ wbc->range_start = index << PAGE_CACHE_SHIFT;
+ wbc->range_end = mapping->writeback_index - 1;
+ goto retry;
+ }
if (pages_skipped != wbc->pages_skipped)
printk(KERN_EMERG "This should not happen leaving %s "
"with nr_to_write = %ld ret = %d\n",
@@ -2578,6 +2594,7 @@ static int ext4_da_writepages(struct add
/* Update index */
index += pages_written;
+ wbc->range_cyclic = range_cyclic;
if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
/*
* set the writeback_index so that range_cyclic
On Sat, Feb 27, 2010 at 12:22:19AM -0600, Jayson R. King wrote:
> Greetings,
>
> Using the kernel 2.6.27.45 with an ext4 filesystem, the same deadlock as
> was reported earlier in kernel bugzilla #12579 can occur. Simply running
> "dbench 500" on an ext4 filesystem can cause the deadlock in one to two
> minutes.
>
> In later stable kernels, the deadlock was fixed by commit 2acf2c26
> ("ext4: Implement range_cyclic in ext4_da_writepages instead of
> write_cache_pages"). I can confirm that the same commit fixes the
> deadlock on 2.6.27.y. However, as the code in ext4_da_writepages is
> different in 2.6.27.y than in later kernels, it was necessary to add
> some other mainline patches before 2acf2c26.
>
> Alltogether, I've added 11 mainline patches including 2acf2c26 to my
> local 2.6.27.y kernel, and confirmed that the deadlock is fixed only
> when the patch 2acf2c26 is applied.
>
> The 11 patches are:
>
> Aneesh Kumar K.V (10):
> ext4: invalidate pages if delalloc block allocation fails.
> ext4: Make sure all the block allocation paths reserve blocks
> ext4: Add percpu dirty block accounting.
> ext4: Retry block reservation
> ext4: Retry block allocation if we have free blocks left
> ext4: Use tag dirty lookup during mpage_da_submit_io
> vfs: Remove the range_cont writeback mode.
> vfs: Add no_nrwrite_index_update writeback control flag
> ext4: Fix file fragmentation during large file write.
> ext4: Implement range_cyclic in ext4_da_writepages instead of write_cache_pages
>
> Mingming Cao (1):
> percpu counter: clean up percpu_counter_sum_and_set()
>
> All will be posted in a reply to this message, in the order that they apply.
>
> I think these patches should be committed to stable 2.6.27.y.
What do the ext4 maintainers think about this? If they ack this, I'll
be glad to queue them up in the next release.
thanks,
greg k-h
On Sat, Feb 27, 2010 at 07:16:35PM -0800, Greg KH wrote:
> On Sat, Feb 27, 2010 at 12:22:19AM -0600, Jayson R. King wrote:
> > Greetings,
> >
> > Using the kernel 2.6.27.45 with an ext4 filesystem, the same deadlock as
> > was reported earlier in kernel bugzilla #12579 can occur. Simply running
> > "dbench 500" on an ext4 filesystem can cause the deadlock in one to two
> > minutes.
> >
> > In later stable kernels, the deadlock was fixed by commit 2acf2c26
> > ("ext4: Implement range_cyclic in ext4_da_writepages instead of
> > write_cache_pages"). I can confirm that the same commit fixes the
> > deadlock on 2.6.27.y. However, as the code in ext4_da_writepages is
> > different in 2.6.27.y than in later kernels, it was necessary to add
> > some other mainline patches before 2acf2c26.
> >
> > Alltogether, I've added 11 mainline patches including 2acf2c26 to my
> > local 2.6.27.y kernel, and confirmed that the deadlock is fixed only
> > when the patch 2acf2c26 is applied.
> >
> > The 11 patches are:
> >
> > Aneesh Kumar K.V (10):
> > ext4: invalidate pages if delalloc block allocation fails.
> > ext4: Make sure all the block allocation paths reserve blocks
> > ext4: Add percpu dirty block accounting.
> > ext4: Retry block reservation
> > ext4: Retry block allocation if we have free blocks left
> > ext4: Use tag dirty lookup during mpage_da_submit_io
> > vfs: Remove the range_cont writeback mode.
> > vfs: Add no_nrwrite_index_update writeback control flag
> > ext4: Fix file fragmentation during large file write.
> > ext4: Implement range_cyclic in ext4_da_writepages instead of write_cache_pages
> >
> > Mingming Cao (1):
> > percpu counter: clean up percpu_counter_sum_and_set()
> >
> > All will be posted in a reply to this message, in the order that they apply.
> >
> > I think these patches should be committed to stable 2.6.27.y.
>
> What do the ext4 maintainers think about this? If they ack this, I'll
> be glad to queue them up in the next release.
As no one has responded, I'm dropping these from my email queue. If you
get an answer, please feel free to resend them.
thanks,
greg k-h