Skip to contentSkip to conversation

sync livelock fixes v2

PATCH In review Wu Fengguang 3 participants 19 messages May 2, 2011 – May 5, 2011

Subsystems (4)
Memory management VFS ext4 file system Page cache
lore
Share & tools
lore

Wu Fengguang

author
#0 lore raw

cover sync livelock fixes v2

Andrew,

changes from v1:
- retain tagged sync behavior for all WB_SYNC_ALL callers (Dave Chinner)
- rename wbc.for_sync to wbc.tagged_sync
- remove the changes to trace events

The below sync livelock test script is working fine after this patchset:

sync time: 2
Dirty: 26492 kB
Writeback: 30260 kB
NFS_Unstable: 0 kB
WritebackTmp: 0 kB
sync NOT livelocked

In particular patch 2 fixes the sync livelock problem introduced by patch "writeback: try more writeback as long as something was written".

Thanks,
Fengguang
---

#!/bin/sh

umount /dev/sda7

# mkfs.xfs -f /dev/sda7
mkfs.ext4 /dev/sda7
mount /dev/sda7 /fs

echo $((50<<20)) > /proc/sys/vm/dirty_bytes

pid=
for i in seq 10
do

        dd if=/dev/zero of=/fs/zero-$i bs=1M count=1000 &
        pid="$pid $!"

done

sleep 1

tic=$(date +'%s')
sync
tac=$(date +'%s')

echo
echo sync time: $((tac-tic))
egrep '(Dirty|Writeback|NFS_Unstable)' /proc/meminfo

pidof dd > /dev/null && { kill -9 $pid; echo sync NOT livelocked; }

author
#1 lore raw

PATCH 1/3 writeback: introduce wbc.tagged_sync for the WB_SYNC_NONE sync stage

sync(2) is performed in two stages: the WB_SYNC_NONE sync and the WB_SYNC_ALL sync. Tag the first stage with wbc.tagged_sync and do livelock prevention for it, too.

Note that writeback_inodes_sb() is called by not only sync(), they are treated the same because the other callers need also need livelock prevention.

Impacts:

- it changes the order in which pages/inodes are synced to disk. Now in the WB_SYNC_NONE stage, it won't proceed to write the next inode until finished with the current inode.

- this adds a new field to the writeback trace events and may possibly break some scripts.

CC Jan Kara <[email protected]>
CC Dave Chinner <[email protected]>
Signed-off-by Wu Fengguang <[email protected]>
---
 fs/ext4/inode.c           |    4 ++--
 fs/fs-writeback.c         |    9 +++++----
 include/linux/writeback.h |    1 +
 mm/page-writeback.c       |    4 ++--
 4 files changed, 10 insertions(+), 8 deletions(-)
fs/fs-writeback.c +5 -4
@@ -36,6 +36,7 @@ struct wb_writeback_work {
 	long nr_pages;
 	struct super_block *sb;
 	enum writeback_sync_modes sync_mode;
+	unsigned int tagged_sync:1;
 	unsigned int for_kupdate:1;
 	unsigned int range_cyclic:1;
 	unsigned int for_background:1;

@@ -644,6 +645,7 @@ static long wb_writeback(struct bdi_writ
 {
 	struct writeback_control wbc = {
 		.sync_mode		= work->sync_mode,
+		.tagged_sync		= work->tagged_sync,
 		.older_than_this	= NULL,
 		.for_kupdate		= work->for_kupdate,
 		.for_background		= work->for_background,

@@ -651,7 +653,7 @@ static long wb_writeback(struct bdi_writ
 	};
 	unsigned long oldest_jif;
 	long wrote = 0;
-	long write_chunk;
+	long write_chunk = MAX_WRITEBACK_PAGES;
 	struct inode *inode;
 
 	if (!wbc.range_cyclic) {

@@ -672,9 +674,7 @@ static long wb_writeback(struct bdi_writ
 	 *                   (quickly) tag currently dirty pages
 	 *                   (maybe slowly) sync all tagged pages
 	 */
-	if (wbc.sync_mode == WB_SYNC_NONE)
-		write_chunk = MAX_WRITEBACK_PAGES;
-	else
+	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync)
 		write_chunk = LONG_MAX;
 
 	wbc.wb_start = jiffies; /* livelock avoidance */

@@ -1209,6 +1209,7 @@ void writeback_inodes_sb_nr(struct super
 	struct wb_writeback_work work = {
 		.sb		= sb,
 		.sync_mode	= WB_SYNC_NONE,
+		.tagged_sync	= 1,
 		.done		= &done,
 		.nr_pages	= nr,
 	};
include/linux/writeback.h +1 -0
@@ -46,6 +46,7 @@ struct writeback_control {
 	unsigned encountered_congestion:1; /* An output: a queue is full */
 	unsigned for_kupdate:1;		/* A kupdate writeback */
 	unsigned for_background:1;	/* A background writeback */
+	unsigned tagged_sync:1;		/* do livelock prevention for sync */
 	unsigned for_reclaim:1;		/* Invoked from the page allocator */
 	unsigned range_cyclic:1;	/* range_start is cyclic */
 	unsigned more_io:1;		/* more io to be dispatched */
mm/page-writeback.c +2 -2
@@ -892,12 +892,12 @@ int write_cache_pages(struct address_spa
 			range_whole = 1;
 		cycled = 1; /* ignore range_cyclic tests */
 	}
-	if (wbc->sync_mode == WB_SYNC_ALL)
+	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_sync)
 		tag = PAGECACHE_TAG_TOWRITE;
 	else
 		tag = PAGECACHE_TAG_DIRTY;
 retry:
-	if (wbc->sync_mode == WB_SYNC_ALL)
+	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_sync)
 		tag_pages_for_writeback(mapping, index, end);
 	done_index = index;
 	while (!done && (index <= end)) {
fs/ext4/inode.c +2 -2
@@ -2741,7 +2741,7 @@ static int write_cache_pages_da(struct a
 	index = wbc->range_start >> PAGE_CACHE_SHIFT;
 	end = wbc->range_end >> PAGE_CACHE_SHIFT;
 
-	if (wbc->sync_mode == WB_SYNC_ALL)
+	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_sync)
 		tag = PAGECACHE_TAG_TOWRITE;
 	else
 		tag = PAGECACHE_TAG_DIRTY;

@@ -2975,7 +2975,7 @@ static int ext4_da_writepages(struct add
 	}
 
 retry:
-	if (wbc->sync_mode == WB_SYNC_ALL)
+	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_sync)
 		tag_pages_for_writeback(mapping, index, end);
 
 	while (!ret && wbc->nr_to_write > 0) {
maintainer
#2 lore raw

Re: [PATCH 1/3] writeback: introduce wbc.tagged_sync for the WB_SYNC_NONE sync stage

Quoted 7 lines · Wu Fengguang

sync(2) is performed in two stages: the WB_SYNC_NONE sync and the WB_SYNC_ALL sync. Tag the first stage with wbc.tagged_sync and do livelock prevention for it, too.

Note that writeback_inodes_sb() is called by not only sync(), they are treated the same because the other callers need also need livelock prevention.

I was thinking about this and could not find any - which other callers
of writeback_inodes_sb() need the livelock prevention?

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author
#3 lore raw

Re: [PATCH 1/3] writeback: introduce wbc.tagged_sync for the WB_SYNC_NONE sync stage

Quoted 10 lines · Jan Kara
Quoted 7 lines · Wu Fengguang

sync(2) is performed in two stages: the WB_SYNC_NONE sync and the WB_SYNC_ALL sync. Tag the first stage with wbc.tagged_sync and do livelock prevention for it, too.

Note that writeback_inodes_sb() is called by not only sync(), they are treated the same because the other callers need also need livelock prevention.

I was thinking about this and could not find any - which other callers
of writeback_inodes_sb() need the livelock prevention?

For example, the writeback_inodes_sb_if_idle() call from ext4. In general anyone that pass get_nr_dirty_pages() as work->nr_pages may be highly over-estimating the work set.

It won't be directly livelocked since ext4 won't wait for completion, however there is possibility the works queued behind are delayed and livelocked.

Ideally simple ->nr_pages works should be given lower priority and even may be merged with each other, and that would be future work.

Thanks,
Fengguang

maintainer
#4 lore raw

Re: [PATCH 1/3] writeback: introduce wbc.tagged_sync for the WB_SYNC_NONE sync stage

Quoted 9 lines · Wu Fengguang
Quoted 8 lines · Jan Kara
Quoted 7 lines · Wu Fengguang

sync(2) is performed in two stages: the WB_SYNC_NONE sync and the WB_SYNC_ALL sync. Tag the first stage with wbc.tagged_sync and do livelock prevention for it, too.

Note that writeback_inodes_sb() is called by not only sync(), they are treated the same because the other callers need also need livelock prevention.

Quoted 6 lines
Quoted 2 lines

I was thinking about this and could not find any - which other callers
of writeback_inodes_sb() need the livelock prevention?

For example, the writeback_inodes_sb_if_idle() call from ext4. In general anyone that pass get_nr_dirty_pages() as work->nr_pages may be highly over-estimating the work set.

OK, I see what you mean. I agree using tagging in these cases probably
makes sense.

Quoted 3 lines

It won't be directly livelocked since ext4 won't wait for completion, however there is possibility the works queued behind are delayed and livelocked.

Actually it will, writeback_inodes_sb() waits for completion (because of
s_umount locking). ext4 should use writeback_inodes_sb_if_idle_nr() (with some relatively small number) anyway to avoid the pauses.

Quoted 2 lines

Ideally simple ->nr_pages works should be given lower priority and even may be merged with each other, and that would be future work.

Merging works wanting to do the same thing would be nice. I though about
it some time ago for a while but getting all the combinations right and making the merging code resonably simple was hard so I postponed it for later because it was not urgent.

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author
#5 lore raw

Re: [PATCH 1/3] writeback: introduce wbc.tagged_sync for the WB_SYNC_NONE sync stage

Quoted 19 lines · Jan Kara
Quoted 9 lines · Wu Fengguang
Quoted 8 lines · Jan Kara
Quoted 7 lines · Wu Fengguang

sync(2) is performed in two stages: the WB_SYNC_NONE sync and the WB_SYNC_ALL sync. Tag the first stage with wbc.tagged_sync and do livelock prevention for it, too.

Note that writeback_inodes_sb() is called by not only sync(), they are treated the same because the other callers need also need livelock prevention.

Quoted 6 lines
Quoted 2 lines

I was thinking about this and could not find any - which other callers
of writeback_inodes_sb() need the livelock prevention?

For example, the writeback_inodes_sb_if_idle() call from ext4. In general anyone that pass get_nr_dirty_pages() as work->nr_pages may be highly over-estimating the work set.

OK, I see what you mean. I agree using tagging in these cases probably
makes sense.

Quoted 6 lines
Quoted 3 lines

It won't be directly livelocked since ext4 won't wait for completion, however there is possibility the works queued behind are delayed and livelocked.

Actually it will, writeback_inodes_sb() waits for completion (because of
s_umount locking). ext4 should use writeback_inodes_sb_if_idle_nr() (with some relatively small number) anyway to avoid the pauses.

Ah yes!

Quoted 6 lines
Quoted 2 lines

Ideally simple ->nr_pages works should be given lower priority and even may be merged with each other, and that would be future work.

Merging works wanting to do the same thing would be nice. I though about
it some time ago for a while but getting all the combinations right and making the merging code resonably simple was hard so I postponed it for later because it was not urgent.

Agreed.

Thanks,
Fengguang

author
#6 lore raw

PATCH 2/3 writeback: update dirtied_when for synced inode to prevent livelock

In reply to Wu Fengguang

With the more aggressive "keep writeback as long as we wrote something" logic in wb_writeback(), the "use LONG_MAX .nr_to_write" trick in commit b9543dac5bbc ("writeback: avoid livelocking WB_SYNC_ALL writeback") is no longer enough to stop sync livelock.

The fix is to explicitly update .dirtied_when on synced inodes, so that they are no longer considered for writeback in the next round.

CC Jan Kara <[email protected]>
Signed-off-by Wu Fengguang <[email protected]>
---
 fs/fs-writeback.c |    9 +++++++++
 1 file changed, 9 insertions(+)

ext3/ext4 are working fine now, however tests show that XFS may still livelock inside the XFS routines:

[ 3581.181253] sync            D ffff8800b6ca15d8  4560  4403   4392 0x00000000
[ 3581.181734]  ffff88006f775bc8 0000000000000046 ffff8800b6ca12b8 00000001b6ca1938
[ 3581.182411]  ffff88006f774000 00000000001d2e40 00000000001d2e40 ffff8800b6ca1280
[ 3581.183088]  00000000001d2e40 ffff88006f775fd8 00000340af111ef2 00000000001d2e40
[ 3581.183765] Call Trace:
[ 3581.184008]  [<ffffffff8109be73>] ? lock_release_holdtime+0xa3/0xab
[ 3581.184392]  [<ffffffff8108cc0d>] ? prepare_to_wait+0x6c/0x79
[ 3581.184756]  [<ffffffff8108cc0d>] ? prepare_to_wait+0x6c/0x79
[ 3581.185120]  [<ffffffff812ed520>] xfs_ioend_wait+0x87/0x9f
[ 3581.185474]  [<ffffffff8108c97a>] ? wake_up_bit+0x2a/0x2a
[ 3581.185827]  [<ffffffff812f742a>] xfs_sync_inode_data+0x92/0x9d
[ 3581.186198]  [<ffffffff812f76e2>] xfs_inode_ag_walk+0x1a5/0x287
[ 3581.186569]  [<ffffffff812f779b>] ? xfs_inode_ag_walk+0x25e/0x287
[ 3581.186946]  [<ffffffff812f7398>] ? xfs_sync_worker+0x69/0x69
[ 3581.187311]  [<ffffffff812e2354>] ? xfs_perag_get+0x68/0xd0
[ 3581.187669]  [<ffffffff81092175>] ? local_clock+0x41/0x5a
[ 3581.188020]  [<ffffffff8109be73>] ? lock_release_holdtime+0xa3/0xab
[ 3581.188403]  [<ffffffff812e22ec>] ? xfs_check_sizes+0x160/0x160
[ 3581.188773]  [<ffffffff812e2354>] ? xfs_perag_get+0x68/0xd0
[ 3581.189130]  [<ffffffff812e236c>] ? xfs_perag_get+0x80/0xd0
[ 3581.189488]  [<ffffffff812e22ec>] ? xfs_check_sizes+0x160/0x160
[ 3581.189858]  [<ffffffff812f7831>] ? xfs_inode_ag_iterator+0x6d/0x8f
[ 3581.190241]  [<ffffffff812f7398>] ? xfs_sync_worker+0x69/0x69
[ 3581.190606]  [<ffffffff812f780b>] xfs_inode_ag_iterator+0x47/0x8f
[ 3581.190982]  [<ffffffff811611f5>] ? __sync_filesystem+0x7a/0x7a
[ 3581.191352]  [<ffffffff812f7877>] xfs_sync_data+0x24/0x43
[ 3581.191703]  [<ffffffff812f7911>] xfs_quiesce_data+0x2c/0x88
[ 3581.192065]  [<ffffffff812f5556>] xfs_fs_sync_fs+0x21/0x48
[ 3581.192419]  [<ffffffff811611e1>] __sync_filesystem+0x66/0x7a
[ 3581.192783]  [<ffffffff8116120b>] sync_one_sb+0x16/0x18
[ 3581.193128]  [<ffffffff8113e3e3>] iterate_supers+0x72/0xce
[ 3581.193482]  [<ffffffff81161140>] sync_filesystems+0x20/0x22
[ 3581.193842]  [<ffffffff8116127e>] sys_sync+0x21/0x33
[ 3581.194177]  [<ffffffff819016c2>] system_call_fastpath+0x16/0x1b
fs/fs-writeback.c +9 -0
@@ -431,6 +431,15 @@ writeback_single_inode(struct inode *ino
 				requeue_io(inode, wb);
 			} else {
 				/*
+				 * sync livelock prevention: each inode is
+				 * tagged and synced in one shot. If still
+				 * dirty, move it back to s_dirty with updated
+				 * dirty time to prevent syncing it again.
+				 */
+				if (wbc->sync_mode == WB_SYNC_ALL ||
+				    wbc->tagged_sync)
+					inode->dirtied_when = jiffies;
+				/*
 				 * Writeback blocked by something other than
 				 * congestion. Delay the inode for some time to
 				 * avoid spinning on the CPU (100% iowait)
maintainer
#7 lore raw

Re: [PATCH 2/3] writeback: update dirtied_when for synced inode to prevent livelock

Quoted 7 lines · Wu Fengguang

With the more aggressive "keep writeback as long as we wrote something" logic in wb_writeback(), the "use LONG_MAX .nr_to_write" trick in commit b9543dac5bbc ("writeback: avoid livelocking WB_SYNC_ALL writeback") is no longer enough to stop sync livelock.

The fix is to explicitly update .dirtied_when on synced inodes, so that they are no longer considered for writeback in the next round.

Looks good. But shouldn't you also update i_dirtied_when in the case where !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) but inode->i_state & I_DIRTY? Mostly for consistency since the chances of livelock because of this are really close to zero.

Honza

Quoted 65 lines
CC Jan Kara <[email protected]>
Signed-off-by Wu Fengguang <[email protected]>
---
 fs/fs-writeback.c |    9 +++++++++
 1 file changed, 9 insertions(+)

ext3/ext4 are working fine now, however tests show that XFS may still livelock inside the XFS routines:

[ 3581.181253] sync            D ffff8800b6ca15d8  4560  4403   4392 0x00000000
[ 3581.181734]  ffff88006f775bc8 0000000000000046 ffff8800b6ca12b8 00000001b6ca1938
[ 3581.182411]  ffff88006f774000 00000000001d2e40 00000000001d2e40 ffff8800b6ca1280
[ 3581.183088]  00000000001d2e40 ffff88006f775fd8 00000340af111ef2 00000000001d2e40
[ 3581.183765] Call Trace:
[ 3581.184008]  [<ffffffff8109be73>] ? lock_release_holdtime+0xa3/0xab
[ 3581.184392]  [<ffffffff8108cc0d>] ? prepare_to_wait+0x6c/0x79
[ 3581.184756]  [<ffffffff8108cc0d>] ? prepare_to_wait+0x6c/0x79
[ 3581.185120]  [<ffffffff812ed520>] xfs_ioend_wait+0x87/0x9f
[ 3581.185474]  [<ffffffff8108c97a>] ? wake_up_bit+0x2a/0x2a
[ 3581.185827]  [<ffffffff812f742a>] xfs_sync_inode_data+0x92/0x9d
[ 3581.186198]  [<ffffffff812f76e2>] xfs_inode_ag_walk+0x1a5/0x287
[ 3581.186569]  [<ffffffff812f779b>] ? xfs_inode_ag_walk+0x25e/0x287
[ 3581.186946]  [<ffffffff812f7398>] ? xfs_sync_worker+0x69/0x69
[ 3581.187311]  [<ffffffff812e2354>] ? xfs_perag_get+0x68/0xd0
[ 3581.187669]  [<ffffffff81092175>] ? local_clock+0x41/0x5a
[ 3581.188020]  [<ffffffff8109be73>] ? lock_release_holdtime+0xa3/0xab
[ 3581.188403]  [<ffffffff812e22ec>] ? xfs_check_sizes+0x160/0x160
[ 3581.188773]  [<ffffffff812e2354>] ? xfs_perag_get+0x68/0xd0
[ 3581.189130]  [<ffffffff812e236c>] ? xfs_perag_get+0x80/0xd0
[ 3581.189488]  [<ffffffff812e22ec>] ? xfs_check_sizes+0x160/0x160
[ 3581.189858]  [<ffffffff812f7831>] ? xfs_inode_ag_iterator+0x6d/0x8f
[ 3581.190241]  [<ffffffff812f7398>] ? xfs_sync_worker+0x69/0x69
[ 3581.190606]  [<ffffffff812f780b>] xfs_inode_ag_iterator+0x47/0x8f
[ 3581.190982]  [<ffffffff811611f5>] ? __sync_filesystem+0x7a/0x7a
[ 3581.191352]  [<ffffffff812f7877>] xfs_sync_data+0x24/0x43
[ 3581.191703]  [<ffffffff812f7911>] xfs_quiesce_data+0x2c/0x88
[ 3581.192065]  [<ffffffff812f5556>] xfs_fs_sync_fs+0x21/0x48
[ 3581.192419]  [<ffffffff811611e1>] __sync_filesystem+0x66/0x7a
[ 3581.192783]  [<ffffffff8116120b>] sync_one_sb+0x16/0x18
[ 3581.193128]  [<ffffffff8113e3e3>] iterate_supers+0x72/0xce
[ 3581.193482]  [<ffffffff81161140>] sync_filesystems+0x20/0x22
[ 3581.193842]  [<ffffffff8116127e>] sys_sync+0x21/0x33
[ 3581.194177]  [<ffffffff819016c2>] system_call_fastpath+0x16/0x1b
fs/fs-writeback.c +9 -0
@@ -431,6 +431,15 @@ writeback_single_inode(struct inode *ino
 				requeue_io(inode, wb);
 			} else {
 				/*
+				 * sync livelock prevention: each inode is
+				 * tagged and synced in one shot. If still
+				 * dirty, move it back to s_dirty with updated
+				 * dirty time to prevent syncing it again.
+				 */
+				if (wbc->sync_mode == WB_SYNC_ALL ||
+				    wbc->tagged_sync)
+					inode->dirtied_when = jiffies;
+				/*
 				 * Writeback blocked by something other than
 				 * congestion. Delay the inode for some time to
 				 * avoid spinning on the CPU (100% iowait)
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author
#8 lore raw

Re: [PATCH 2/3] writeback: update dirtied_when for synced inode to prevent livelock

Quoted 6 lines · Jan Kara
Quoted 5 lines · Wu Fengguang

With the more aggressive "keep writeback as long as we wrote something" logic in wb_writeback(), the "use LONG_MAX .nr_to_write" trick in commit b9543dac5bbc ("writeback: avoid livelocking WB_SYNC_ALL writeback") is no longer enough to stop sync livelock.

Quoted 6 lines
Quoted 2 lines

The fix is to explicitly update .dirtied_when on synced inodes, so that they are no longer considered for writeback in the next round.

Looks good. But shouldn't you also update i_dirtied_when in the case where !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) but inode->i_state & I_DIRTY? Mostly for consistency since the chances of livelock because of this are really close to zero.

Yes the I_DIRTY_SYNC may well be redirtied by the FS. It's very unlikely to cause live locks, so I'm more than glad to kill the optional code :)

Thanks,
Fengguang

Quoted 67 lines
Quoted 64 lines
CC Jan Kara <[email protected]>
Signed-off-by Wu Fengguang <[email protected]>
---
 fs/fs-writeback.c |    9 +++++++++
 1 file changed, 9 insertions(+)

ext3/ext4 are working fine now, however tests show that XFS may still livelock inside the XFS routines:

[ 3581.181253] sync            D ffff8800b6ca15d8  4560  4403   4392 0x00000000
[ 3581.181734]  ffff88006f775bc8 0000000000000046 ffff8800b6ca12b8 00000001b6ca1938
[ 3581.182411]  ffff88006f774000 00000000001d2e40 00000000001d2e40 ffff8800b6ca1280
[ 3581.183088]  00000000001d2e40 ffff88006f775fd8 00000340af111ef2 00000000001d2e40
[ 3581.183765] Call Trace:
[ 3581.184008]  [<ffffffff8109be73>] ? lock_release_holdtime+0xa3/0xab
[ 3581.184392]  [<ffffffff8108cc0d>] ? prepare_to_wait+0x6c/0x79
[ 3581.184756]  [<ffffffff8108cc0d>] ? prepare_to_wait+0x6c/0x79
[ 3581.185120]  [<ffffffff812ed520>] xfs_ioend_wait+0x87/0x9f
[ 3581.185474]  [<ffffffff8108c97a>] ? wake_up_bit+0x2a/0x2a
[ 3581.185827]  [<ffffffff812f742a>] xfs_sync_inode_data+0x92/0x9d
[ 3581.186198]  [<ffffffff812f76e2>] xfs_inode_ag_walk+0x1a5/0x287
[ 3581.186569]  [<ffffffff812f779b>] ? xfs_inode_ag_walk+0x25e/0x287
[ 3581.186946]  [<ffffffff812f7398>] ? xfs_sync_worker+0x69/0x69
[ 3581.187311]  [<ffffffff812e2354>] ? xfs_perag_get+0x68/0xd0
[ 3581.187669]  [<ffffffff81092175>] ? local_clock+0x41/0x5a
[ 3581.188020]  [<ffffffff8109be73>] ? lock_release_holdtime+0xa3/0xab
[ 3581.188403]  [<ffffffff812e22ec>] ? xfs_check_sizes+0x160/0x160
[ 3581.188773]  [<ffffffff812e2354>] ? xfs_perag_get+0x68/0xd0
[ 3581.189130]  [<ffffffff812e236c>] ? xfs_perag_get+0x80/0xd0
[ 3581.189488]  [<ffffffff812e22ec>] ? xfs_check_sizes+0x160/0x160
[ 3581.189858]  [<ffffffff812f7831>] ? xfs_inode_ag_iterator+0x6d/0x8f
[ 3581.190241]  [<ffffffff812f7398>] ? xfs_sync_worker+0x69/0x69
[ 3581.190606]  [<ffffffff812f780b>] xfs_inode_ag_iterator+0x47/0x8f
[ 3581.190982]  [<ffffffff811611f5>] ? __sync_filesystem+0x7a/0x7a
[ 3581.191352]  [<ffffffff812f7877>] xfs_sync_data+0x24/0x43
[ 3581.191703]  [<ffffffff812f7911>] xfs_quiesce_data+0x2c/0x88
[ 3581.192065]  [<ffffffff812f5556>] xfs_fs_sync_fs+0x21/0x48
[ 3581.192419]  [<ffffffff811611e1>] __sync_filesystem+0x66/0x7a
[ 3581.192783]  [<ffffffff8116120b>] sync_one_sb+0x16/0x18
[ 3581.193128]  [<ffffffff8113e3e3>] iterate_supers+0x72/0xce
[ 3581.193482]  [<ffffffff81161140>] sync_filesystems+0x20/0x22
[ 3581.193842]  [<ffffffff8116127e>] sys_sync+0x21/0x33
[ 3581.194177]  [<ffffffff819016c2>] system_call_fastpath+0x16/0x1b
fs/fs-writeback.c +9 -0
@@ -431,6 +431,15 @@ writeback_single_inode(struct inode *ino
 				requeue_io(inode, wb);
 			} else {
 				/*
+				 * sync livelock prevention: each inode is
+				 * tagged and synced in one shot. If still
+				 * dirty, move it back to s_dirty with updated
+				 * dirty time to prevent syncing it again.
+				 */
+				if (wbc->sync_mode == WB_SYNC_ALL ||
+				    wbc->tagged_sync)
+					inode->dirtied_when = jiffies;
+				/*
 				 * Writeback blocked by something other than
 				 * congestion. Delay the inode for some time to
 				 * avoid spinning on the CPU (100% iowait)
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author
#9 lore raw

PATCH 3/3 writeback: avoid extra sync work at enqueue time

In reply to Wu Fengguang

This removes writeback_control.wb_start and does more straightforward sync livelock prevention by setting .older_than_this to prevent extra inodes from being enqueued in the first place.

CC Jan Kara <[email protected]>
Signed-off-by Wu Fengguang <[email protected]>
---
 fs/fs-writeback.c         |   17 ++++-------------
 include/linux/writeback.h |    3 ---
 2 files changed, 4 insertions(+), 16 deletions(-)
fs/fs-writeback.c +4 -13
@@ -544,15 +544,6 @@ static int writeback_sb_inodes(struct su
 			continue;
 		}
 
-		/*
-		 * Was this inode dirtied after sync_sb_inodes was called?
-		 * This keeps sync from extra jobs and livelock.
-		 */
-		if (inode_dirtied_after(inode, wbc->wb_start)) {
-			spin_unlock(&inode->i_lock);
-			return 1;
-		}
-
 		__iget(inode);
 
 		pages_skipped = wbc->pages_skipped;

@@ -585,8 +576,6 @@ static void __writeback_inodes_wb(struct
 {
 	int ret = 0;
 
-	if (!wbc->wb_start)
-		wbc->wb_start = jiffies; /* livelock avoidance */
 
 	while (!list_empty(&wb->b_io)) {
 		struct inode *inode = wb_inode(wb->b_io.prev);

@@ -683,10 +672,12 @@ static long wb_writeback(struct bdi_writ
 	 *                   (quickly) tag currently dirty pages
 	 *                   (maybe slowly) sync all tagged pages
 	 */
-	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync)
+	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync) {
 		write_chunk = LONG_MAX;
+		oldest_jif = jiffies;
+		wbc.older_than_this = &oldest_jif;
+	}
 
-	wbc.wb_start = jiffies; /* livelock avoidance */
 	for (;;) {
 		/*
 		 * Stop writeback when nr_pages has been consumed
include/linux/writeback.h +0 -3
@@ -26,9 +26,6 @@ struct writeback_control {
 	enum writeback_sync_modes sync_mode;
 	unsigned long *older_than_this;	/* If !NULL, only write back inodes
 					   older than this */
-	unsigned long wb_start;         /* Time writeback_inodes_wb was
-					   called. This is needed to avoid
-					   extra jobs and livelock */
 	long nr_to_write;		/* Write this many pages, and decrement
 					   this for each page written */
 	long pages_skipped;		/* Pages which were not written */
maintainer
#10 lore raw

Re: [PATCH 3/3] writeback: avoid extra sync work at enqueue time

Quoted 7 lines · Wu Fengguang

This removes writeback_control.wb_start and does more straightforward sync livelock prevention by setting .older_than_this to prevent extra inodes from being enqueued in the first place.

fs/fs-writeback.c static long wb_writeback(struct bdi_writ · line 679

Quoted 9 lines
fs/fs-writeback.c +4 -1
@@ -683,10 +672,12 @@ static long wb_writeback(struct bdi_writ
 	 *                   (quickly) tag currently dirty pages
 	 *                   (maybe slowly) sync all tagged pages
 	 */
-	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync)
+	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync) {
 		write_chunk = LONG_MAX;
+		oldest_jif = jiffies;
+		wbc.older_than_this = &oldest_jif;
+	}

What are the implications of not doing dirty-time livelock avoidance for
other types of writeback? Is that a mistake? I'd prefer to have in wb_writeback():
if (wbc.for_kupdate)

	oldest_jif = jiffies - msecs_to_jiffies(dirty_expire_interval * 10);
else
	oldest_jif = jiffies;

wbc.older_than_this = &oldest_jif;

And when you have this, you can make wbc.older_than_this just a plain number and remove all those checks for wbc.older_than_this == NULL.

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author
#11 lore raw

Re: [PATCH 3/3] writeback: avoid extra sync work at enqueue time

Quoted 21 lines · Jan Kara
Quoted 7 lines · Wu Fengguang

This removes writeback_control.wb_start and does more straightforward sync livelock prevention by setting .older_than_this to prevent extra inodes from being enqueued in the first place.

fs/fs-writeback.c static long wb_writeback(struct bdi_writ · line 679

Quoted 9 lines
fs/fs-writeback.c +4 -1
@@ -683,10 +672,12 @@ static long wb_writeback(struct bdi_writ
 	 *                   (quickly) tag currently dirty pages
 	 *                   (maybe slowly) sync all tagged pages
 	 */
-	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync)
+	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync) {
 		write_chunk = LONG_MAX;
+		oldest_jif = jiffies;
+		wbc.older_than_this = &oldest_jif;
+	}

What are the implications of not doing dirty-time livelock avoidance for
other types of writeback? Is that a mistake? I'd prefer to have in wb_writeback():
if (wbc.for_kupdate)

Quoted 7 lines
	oldest_jif = jiffies - msecs_to_jiffies(dirty_expire_interval * 10);
else
	oldest_jif = jiffies;

wbc.older_than_this = &oldest_jif;

And when you have this, you can make wbc.older_than_this just a plain number and remove all those checks for wbc.older_than_this == NULL.

Good point. Here is the fixed patch. Will you send the patch to change the type when the current patches are settled down?

Thanks,
Fengguang
---
Subject: writeback: avoid extra sync work at enqueue time
Date: Sat Apr 23 11:26:07 CST 2011

This removes writeback_control.wb_start and does more straightforward sync livelock prevention by setting .older_than_this to prevent extra inodes from being enqueued in the first place.

CC Jan Kara <[email protected]>
Signed-off-by Wu Fengguang <[email protected]>
---
 fs/fs-writeback.c         |   16 +++-------------
 include/linux/writeback.h |    3 ---
 2 files changed, 3 insertions(+), 16 deletions(-)
fs/fs-writeback.c +3 -13
@@ -544,15 +544,6 @@ static int writeback_sb_inodes(struct su
 			continue;
 		}
 
-		/*
-		 * Was this inode dirtied after sync_sb_inodes was called?
-		 * This keeps sync from extra jobs and livelock.
-		 */
-		if (inode_dirtied_after(inode, wbc->wb_start)) {
-			spin_unlock(&inode->i_lock);
-			return 1;
-		}
-
 		__iget(inode);
 
 		pages_skipped = wbc->pages_skipped;

@@ -585,9 +576,6 @@ static void __writeback_inodes_wb(struct
 {
 	int ret = 0;
 
-	if (!wbc->wb_start)
-		wbc->wb_start = jiffies; /* livelock avoidance */
-
 	while (!list_empty(&wb->b_io)) {
 		struct inode *inode = wb_inode(wb->b_io.prev);
 		struct super_block *sb = inode->i_sb;

@@ -686,7 +674,9 @@ static long wb_writeback(struct bdi_writ
 	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync)
 		write_chunk = LONG_MAX;
 
-	wbc.wb_start = jiffies; /* livelock avoidance */
+	oldest_jif = jiffies;
+	wbc.older_than_this = &oldest_jif;
+
 	for (;;) {
 		/*
 		 * Stop writeback when nr_pages has been consumed
include/linux/writeback.h +0 -3
@@ -26,9 +26,6 @@ struct writeback_control {
 	enum writeback_sync_modes sync_mode;
 	unsigned long *older_than_this;	/* If !NULL, only write back inodes
 					   older than this */
-	unsigned long wb_start;         /* Time writeback_inodes_wb was
-					   called. This is needed to avoid
-					   extra jobs and livelock */
 	long nr_to_write;		/* Write this many pages, and decrement
 					   this for each page written */
 	long pages_skipped;		/* Pages which were not written */
#12 lore raw

Re: [PATCH 3/3] writeback: avoid extra sync work at enqueue time

fs/fs-writeback.c static long wb_writeback(struct bdi_writ · line 678

Quoted 3 lines · Wu Fengguang
fs/fs-writeback.c +2 -1
-	wbc.wb_start = jiffies; /* livelock avoidance */
+	oldest_jif = jiffies;
+	wbc.older_than_this = &oldest_jif;

Can't you kill the oldest_jif variable now?

#13 lore raw

Re: [PATCH 3/3] writeback: avoid extra sync work at enqueue time

Quoted 6 lines · Christoph Hellwig

fs/fs-writeback.c static long wb_writeback(struct bdi_writ · line 678

Quoted 3 lines · Wu Fengguang
fs/fs-writeback.c +2 -1
-	wbc.wb_start = jiffies; /* livelock avoidance */
+	oldest_jif = jiffies;
+	wbc.older_than_this = &oldest_jif;

Can't you kill the oldest_jif variable now?

Err, -ENOCOFFEE. You could only kill it if you implement Jan's suggestion of not making older_than_this a pointer..

author
#14 lore raw

Re: [PATCH 3/3] writeback: avoid extra sync work at enqueue time

Quoted 10 lines · Christoph Hellwig
Quoted 6 lines · Christoph Hellwig

fs/fs-writeback.c static long wb_writeback(struct bdi_writ · line 678

Quoted 3 lines · Wu Fengguang
fs/fs-writeback.c +2 -1
-	wbc.wb_start = jiffies; /* livelock avoidance */
+	oldest_jif = jiffies;
+	wbc.older_than_this = &oldest_jif;

Can't you kill the oldest_jif variable now?

Err, -ENOCOFFEE. You could only kill it if you implement Jan's suggestion of not making older_than_this a pointer..

Yes. And that won't be a straightforward code refactor. It will have very small side effects, so we'd better kill it in a standalone patch.

Thanks,
Fengguang

maintainer
#15 lore raw

Re: [PATCH 3/3] writeback: avoid extra sync work at enqueue time

In reply to Wu Fengguang
Quoted 27 lines · Wu Fengguang
Quoted 26 lines · Jan Kara
Quoted 7 lines · Wu Fengguang

This removes writeback_control.wb_start and does more straightforward sync livelock prevention by setting .older_than_this to prevent extra inodes from being enqueued in the first place.

fs/fs-writeback.c static long wb_writeback(struct bdi_writ · line 679

Quoted 9 lines
fs/fs-writeback.c +4 -1
@@ -683,10 +672,12 @@ static long wb_writeback(struct bdi_writ
 	 *                   (quickly) tag currently dirty pages
 	 *                   (maybe slowly) sync all tagged pages
 	 */
-	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync)
+	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync) {
 		write_chunk = LONG_MAX;
+		oldest_jif = jiffies;
+		wbc.older_than_this = &oldest_jif;
+	}

What are the implications of not doing dirty-time livelock avoidance for
other types of writeback? Is that a mistake? I'd prefer to have in wb_writeback():
if (wbc.for_kupdate)

	oldest_jif = jiffies - msecs_to_jiffies(dirty_expire_interval * 10);
else
	oldest_jif = jiffies;

wbc.older_than_this = &oldest_jif;

Quoted 5 lines
Quoted 2 lines

And when you have this, you can make wbc.older_than_this just a plain number and remove all those checks for wbc.older_than_this == NULL.

Good point. Here is the fixed patch. Will you send the patch to change the type when the current patches are settled down?

OK, I will do that.

fs/fs-writeback.c static long wb_writeback(struct bdi_writ · line 679

Quoted 8 lines
fs/fs-writeback.c +3 -1
@@ -686,7 +674,9 @@ static long wb_writeback(struct bdi_writ
 	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync)
 		write_chunk = LONG_MAX;
 
-	wbc.wb_start = jiffies; /* livelock avoidance */
+	oldest_jif = jiffies;
+	wbc.older_than_this = &oldest_jif;
+

I might be already confused with all the code moving around but won't
this overwrite the value set for the for_kupdate case?

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author
#16 lore raw

Re: [PATCH 3/3] writeback: avoid extra sync work at enqueue time

Quoted 28 lines · Jan Kara
Quoted 27 lines · Wu Fengguang
Quoted 26 lines · Jan Kara
Quoted 7 lines · Wu Fengguang

This removes writeback_control.wb_start and does more straightforward sync livelock prevention by setting .older_than_this to prevent extra inodes from being enqueued in the first place.

fs/fs-writeback.c static long wb_writeback(struct bdi_writ · line 679

Quoted 9 lines
fs/fs-writeback.c +4 -1
@@ -683,10 +672,12 @@ static long wb_writeback(struct bdi_writ
 	 *                   (quickly) tag currently dirty pages
 	 *                   (maybe slowly) sync all tagged pages
 	 */
-	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync)
+	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync) {
 		write_chunk = LONG_MAX;
+		oldest_jif = jiffies;
+		wbc.older_than_this = &oldest_jif;
+	}

What are the implications of not doing dirty-time livelock avoidance for
other types of writeback? Is that a mistake? I'd prefer to have in wb_writeback():
if (wbc.for_kupdate)

	oldest_jif = jiffies - msecs_to_jiffies(dirty_expire_interval * 10);
else
	oldest_jif = jiffies;

wbc.older_than_this = &oldest_jif;

Quoted 6 lines
Quoted 5 lines
Quoted 2 lines

And when you have this, you can make wbc.older_than_this just a plain number and remove all those checks for wbc.older_than_this == NULL.

Good point. Here is the fixed patch. Will you send the patch to change the type when the current patches are settled down?

OK, I will do that.

Thank you.

Quoted 10 lines

fs/fs-writeback.c static long wb_writeback(struct bdi_writ · line 679

Quoted 8 lines
fs/fs-writeback.c +3 -1
@@ -686,7 +674,9 @@ static long wb_writeback(struct bdi_writ
 	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync)
 		write_chunk = LONG_MAX;
 
-	wbc.wb_start = jiffies; /* livelock avoidance */
+	oldest_jif = jiffies;
+	wbc.older_than_this = &oldest_jif;
+

I might be already confused with all the code moving around but won't
this overwrite the value set for the for_kupdate case?

It's the opposite -- it will be overwritten inside the loop by for_kupdate, which may run for long time and hence need to update oldest_jif from time to time.

Thanks,
Fengguang

author
#17 lore raw

Re: [PATCH 3/3] writeback: avoid extra sync work at enqueue time

Quoted 49 lines · Wu Fengguang
Quoted 28 lines · Jan Kara
Quoted 27 lines · Wu Fengguang
Quoted 26 lines · Jan Kara
Quoted 7 lines · Wu Fengguang

This removes writeback_control.wb_start and does more straightforward sync livelock prevention by setting .older_than_this to prevent extra inodes from being enqueued in the first place.

fs/fs-writeback.c static long wb_writeback(struct bdi_writ · line 679

Quoted 9 lines
fs/fs-writeback.c +4 -1
@@ -683,10 +672,12 @@ static long wb_writeback(struct bdi_writ
 	 *                   (quickly) tag currently dirty pages
 	 *                   (maybe slowly) sync all tagged pages
 	 */
-	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync)
+	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync) {
 		write_chunk = LONG_MAX;
+		oldest_jif = jiffies;
+		wbc.older_than_this = &oldest_jif;
+	}

What are the implications of not doing dirty-time livelock avoidance for
other types of writeback? Is that a mistake? I'd prefer to have in wb_writeback():
if (wbc.for_kupdate)

	oldest_jif = jiffies - msecs_to_jiffies(dirty_expire_interval * 10);
else
	oldest_jif = jiffies;

wbc.older_than_this = &oldest_jif;

Quoted 6 lines
Quoted 5 lines
Quoted 2 lines

And when you have this, you can make wbc.older_than_this just a plain number and remove all those checks for wbc.older_than_this == NULL.

Good point. Here is the fixed patch. Will you send the patch to change the type when the current patches are settled down?

OK, I will do that.

Thank you.

Quoted 10 lines

fs/fs-writeback.c static long wb_writeback(struct bdi_writ · line 679

Quoted 8 lines
fs/fs-writeback.c +3 -1
@@ -686,7 +674,9 @@ static long wb_writeback(struct bdi_writ
 	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync)
 		write_chunk = LONG_MAX;
 
-	wbc.wb_start = jiffies; /* livelock avoidance */
+	oldest_jif = jiffies;
+	wbc.older_than_this = &oldest_jif;
+

I might be already confused with all the code moving around but won't
this overwrite the value set for the for_kupdate case?

Quoted 3 lines

It's the opposite -- it will be overwritten inside the loop by for_kupdate, which may run for long time and hence need to update oldest_jif from time to time.

The code is now:

        oldest_jif = jiffies;
        work->older_than_this = &oldest_jif;

        for (;;) {
                // ...
               
                if (work->for_kupdate || work->for_background) {
                        oldest_jif = jiffies -
                                msecs_to_jiffies(dirty_expire_interval * 10);
                        work->older_than_this = &oldest_jif;
                }

retry:

                // ...
               
                /*
                 * background writeback will start with expired inodes, and
                 * if none is found, fallback to all inodes. This order helps
                 * reduce the number of dirty pages reaching the end of LRU
                 * lists and cause trouble to the page reclaim.
                 */
                if (work->for_background &&
                    work->older_than_this &&
                    list_empty(&wb->b_io) &&
                    list_empty(&wb->b_more_io)) {
                        work->older_than_this = NULL;
                        goto retry;
                }

                // ...
        }

Thanks,
Fengguang

maintainer
#18 lore raw

Re: [PATCH 3/3] writeback: avoid extra sync work at enqueue time

In reply to Wu Fengguang
Quoted 49 lines · Wu Fengguang
Quoted 28 lines · Jan Kara
Quoted 27 lines · Wu Fengguang
Quoted 26 lines · Jan Kara
Quoted 7 lines · Wu Fengguang

This removes writeback_control.wb_start and does more straightforward sync livelock prevention by setting .older_than_this to prevent extra inodes from being enqueued in the first place.

fs/fs-writeback.c static long wb_writeback(struct bdi_writ · line 679

Quoted 9 lines
fs/fs-writeback.c +4 -1
@@ -683,10 +672,12 @@ static long wb_writeback(struct bdi_writ
 	 *                   (quickly) tag currently dirty pages
 	 *                   (maybe slowly) sync all tagged pages
 	 */
-	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync)
+	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync) {
 		write_chunk = LONG_MAX;
+		oldest_jif = jiffies;
+		wbc.older_than_this = &oldest_jif;
+	}

What are the implications of not doing dirty-time livelock avoidance for
other types of writeback? Is that a mistake? I'd prefer to have in wb_writeback():
if (wbc.for_kupdate)

	oldest_jif = jiffies - msecs_to_jiffies(dirty_expire_interval * 10);
else
	oldest_jif = jiffies;

wbc.older_than_this = &oldest_jif;

Quoted 6 lines
Quoted 5 lines
Quoted 2 lines

And when you have this, you can make wbc.older_than_this just a plain number and remove all those checks for wbc.older_than_this == NULL.

Good point. Here is the fixed patch. Will you send the patch to change the type when the current patches are settled down?

OK, I will do that.

Thank you.

Quoted 10 lines

fs/fs-writeback.c static long wb_writeback(struct bdi_writ · line 679

Quoted 8 lines
fs/fs-writeback.c +3 -1
@@ -686,7 +674,9 @@ static long wb_writeback(struct bdi_writ
 	if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_sync)
 		write_chunk = LONG_MAX;
 
-	wbc.wb_start = jiffies; /* livelock avoidance */
+	oldest_jif = jiffies;
+	wbc.older_than_this = &oldest_jif;
+

I might be already confused with all the code moving around but won't
this overwrite the value set for the for_kupdate case?

Quoted 3 lines

It's the opposite -- it will be overwritten inside the loop by for_kupdate, which may run for long time and hence need to update oldest_jif from time to time.

Oh, right. I guess I'll setup that git tree (at least for myself if for
nothing else) so that I can see what is actually the current code state. Because I think I've already lost the track of all the patches...

I've looked at mmotm on the web but it seems to have older versions of the patches and also some of the patches which I think Andrew took (like this one) are not there. Can you please make latest versions of all the patches available somewhere so that I don't have dig them from the mailing list archives?

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR