xfs: fixes for realtime file truncation
cover xfs: fixes for realtime file truncation
From: Omar Sandoval <[email protected]>
Hello,
These two patches fix bugs in a corner case of truncating realtime files. We encountered this as a soft lockup in production while truncating certain files, but I found the space leak on further investigation. The lockup is caused by an interaction between the two bugs fixed by these patches. I've also sent a reproducer for xfstests.
These patches are based on v5.4. Thanks!
Omar Sandoval (2):
xfs: fix realtime file data space leak
xfs: don't check for AG deadlock for realtime files in bunmapi
fs/xfs/libxfs/xfs_bmap.c | 27 +++++++++++++++------------ 1 file changed, 15 insertions(+), 12 deletions(-)
Signature · 1 line
2.24.0
PATCH 1/2 xfs: fix realtime file data space leak
From: Omar Sandoval <[email protected]>
Realtime files in XFS allocate extents in rextsize units. However, the written/unwritten state of those extents is still tracked in blocksize units. Therefore, a realtime file can be split up into written and unwritten extents that are not necessarily aligned to the realtime extent size. __xfs_bunmapi() has some logic to handle these various corner cases. Consider how it handles the following case:
1. The last extent is unwritten.
2. The last extent is smaller than the realtime extent size.
3. startblock of the last extent is not aligned to the realtime extent size, but startblock + blockcount is.
In this case, __xfs_bunmapi() calls xfs_bmap_add_extent_unwritten_real() to set the second-to-last extent to unwritten. This should merge the last and second-to-last extents, so __xfs_bunmapi() moves on to the second-to-last extent.
However, if the size of the last and second-to-last extents combined is greater than MAXEXTLEN, xfs_bmap_add_extent_unwritten_real() does not merge the two extents. When that happens, __xfs_bunmapi() skips past the last extent without unmapping it, thus leaking the space.
Fix it by only unwriting the minimum amount needed to align the last extent to the realtime extent size, which is guaranteed to merge with the last extent.
--- fs/xfs/libxfs/xfs_bmap.c | 25 ++++++++++++++----------- 1 file changed, 14 insertions(+), 11 deletions(-)
fs/xfs/libxfs/xfs_bmap.c +14 -11
@@ -5376,16 +5376,17 @@ __xfs_bunmapi( }div_u64_rem (del.br_startblock, mp->m_sb.sb_rextsize, &mod);if (mod) { +xfs_extlen_t off = mp->m_sb.sb_rextsize - mod; +/* * Realtime extent is lined up at the end but not * at the front. We'll get rid of full extents if * we can. */ - mod = mp->m_sb.sb_rextsize - mod; -if (del.br_blockcount > mod) { - del.br_blockcount -= mod; - del.br_startoff += mod; - del.br_startblock += mod; +if (del.br_blockcount > off) { + del.br_blockcount -= off; + del.br_startoff += off; + del.br_startblock += off; }else if (del.br_startoff == start && (del.br_state ==XFS_EXT_UNWRITTEN || tp->t_blk_res ==0 )) {
@@ -5403,6 +5404,7 @@ __xfs_bunmapi(continue ; }else if (del.br_state ==XFS_EXT_UNWRITTEN ) {struct xfs_bmbt_irec prev; +xfs_fileoff_t unwrite_start;/* * This one is already unwritten.
@@ -5416,12 +5418,13 @@ __xfs_bunmapi(ASSERT (!isnullstartblock (prev.br_startblock));ASSERT (del.br_startblock == prev.br_startblock + prev.br_blockcount); -if (prev.br_startoff < start) { - mod = start - prev.br_startoff; - prev.br_blockcount -= mod; - prev.br_startblock += mod; - prev.br_startoff = start; - } + unwrite_start =max3 (start, + del.br_startoff - mod, + prev.br_startoff); + mod = unwrite_start - prev.br_startoff; + prev.br_startoff = unwrite_start; + prev.br_startblock += mod; + prev.br_blockcount -= mod; prev.br_state =XFS_EXT_UNWRITTEN ; error = xfs_bmap_add_extent_unwritten_real(tp, ip, whichfork, &icur, &cur,
Signature · 1 line
2.24.0
PATCH 2/2 xfs: don't check for AG deadlock for realtime files in bunmapi
In reply to Omar SandovalFrom: Omar Sandoval <[email protected]>
Commit 5b094d6dac04 ("xfs: fix multi-AG deadlock in xfs_bunmapi") added a check in __xfs_bunmapi() to stop early if we would touch multiple AGs in the wrong order. However, this check isn't applicable for realtime files. In most cases, it just makes us do unnecessary commits. However, without the fix from the previous commit ("xfs: fix realtime file data space leak"), if the last and second-to-last extents also happen to have different "AG numbers", then the break actually causes __xfs_bunmapi() to return without making any progress, which sends xfs_itruncate_extents_flags() into an infinite loop.
--- fs/xfs/libxfs/xfs_bmap.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
fs/xfs/libxfs/xfs_bmap.c +1 -1
@@ -5300,7 +5300,7 @@ __xfs_bunmapi(* Make sure we don't touch multiple AGF headers out of order * in a single transaction, as that could cause AB-BA deadlocks. */ -if (!wasdel) { +if (!wasdel && !isrt) { agno =XFS_FSB_TO_AGNO (mp, del.br_startblock);if (prev_agno !=NULLAGNUMBER && prev_agno > agno)break ;
Signature · 1 line
2.24.0
Re: [PATCH 2/2] xfs: don't check for AG deadlock for realtime files in bunmapi
Quoted 12 lines · Omar Sandoval
From: Omar Sandoval <[email protected]>
Commit
5b094d6dac04("xfs: fix multi-AG deadlock inxfs_bunmapi") added a check in__xfs_bunmapi()to stop early if we would touch multiple AGs in the wrong order. However, this check isn't applicable for realtime files. In most cases, it just makes us do unnecessary commits. However, without the fix from the previous commit ("xfs: fix realtime file data space leak"), if the last and second-to-last extents also happen to have different "AG numbers", then the break actually causes__xfs_bunmapi()to return without making any progress, which sendsxfs_itruncate_extents_flags()into an infinite loop.
Quoted 2 lines
Fixes 5b094d6dac04 ("xfs: fix multi-AG deadlock in xfs_bunmapi")Signed-off-by Omar Sandoval <[email protected]>
Looks pretty straightforward,
--D
Quoted 20 lines
--- fs/xfs/libxfs/xfs_bmap.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)fs/xfs/libxfs/xfs_bmap.c +1
-1@@ -5300,7 +5300,7 @@ __xfs_bunmapi(* Make sure we don't touch multiple AGF headers out of order * in a single transaction, as that could cause AB-BA deadlocks. */ -if (!wasdel) { +if (!wasdel && !isrt) { agno =XFS_FSB_TO_AGNO (mp, del.br_startblock);if (prev_agno !=NULLAGNUMBER && prev_agno > agno)break ;Signature · 1 line
2.24.0
Re: [PATCH 1/2] xfs: fix realtime file data space leak
In reply to Omar SandovalQuoted 47 lines · Omar Sandoval
From: Omar Sandoval <[email protected]>
Realtime files in XFS allocate extents in rextsize units. However, the written/unwritten state of those extents is still tracked in blocksize units. Therefore, a realtime file can be split up into written and unwritten extents that are not necessarily aligned to the realtime extent size.
__xfs_bunmapi()has some logic to handle these various corner cases. Consider how it handles the following case:1. The last extent is unwritten.
2. The last extent is smaller than the realtime extent size.
3. startblock of the last extent is not aligned to the realtime extent size, but startblock + blockcount is.In this case,
__xfs_bunmapi()callsxfs_bmap_add_extent_unwritten_real()to set the second-to-last extent to unwritten. This should merge the last and second-to-last extents, so__xfs_bunmapi()moves on to the second-to-last extent.However, if the size of the last and second-to-last extents combined is greater than MAXEXTLEN,
xfs_bmap_add_extent_unwritten_real()does not merge the two extents. When that happens,__xfs_bunmapi()skips past the last extent without unmapping it, thus leaking the space.Fix it by only unwriting the minimum amount needed to align the last extent to the realtime extent size, which is guaranteed to merge with the last extent.
Signed-off-by Omar Sandoval <[email protected]>--- fs/xfs/libxfs/xfs_bmap.c | 25 ++++++++++++++----------- 1 file changed, 14 insertions(+), 11 deletions(-)fs/xfs/libxfs/xfs_bmap.c +2
-0@@ -5376,16 +5376,17 @@ __xfs_bunmapi( }div_u64_rem (del.br_startblock, mp->m_sb.sb_rextsize, &mod);if (mod) { +xfs_extlen_t off = mp->m_sb.sb_rextsize - mod; +/* * Realtime extent is lined up at the end but not * at the front. We'll get rid of full extents if * we can.
fs/xfs/libxfs/xfs_bmap.c __xfs_bunmapi( · line 5389
Quoted 10 lines
fs/xfs/libxfs/xfs_bmap.c +4
-5*/ - mod = mp->m_sb.sb_rextsize - mod; -if (del.br_blockcount > mod) { - del.br_blockcount -= mod; - del.br_startoff += mod; - del.br_startblock += mod; +if (del.br_blockcount > off) { + del.br_blockcount -= off; + del.br_startoff += off; + del.br_startblock += off;
Ok, so we make this change so that we no longer change @mod once it's set by the div64 operation...
Quoted 12 lines
fs/xfs/libxfs/xfs_bmap.c +1
-0}else if (del.br_startoff == start && (del.br_state ==XFS_EXT_UNWRITTEN || tp->t_blk_res ==0 )) {@@ -5403,6 +5404,7 @@ __xfs_bunmapi(continue ; }else if (del.br_state ==XFS_EXT_UNWRITTEN ) {struct xfs_bmbt_irec prev; +xfs_fileoff_t unwrite_start;/* * This one is already unwritten.
fs/xfs/libxfs/xfs_bmap.c __xfs_bunmapi( · line 5424
Quoted 9 lines
fs/xfs/libxfs/xfs_bmap.c +0
-6@@ -5416,12 +5418,13 @@ __xfs_bunmapi(ASSERT (!isnullstartblock (prev.br_startblock));ASSERT (del.br_startblock == prev.br_startblock + prev.br_blockcount); -if (prev.br_startoff < start) { - mod = start - prev.br_startoff; - prev.br_blockcount -= mod; - prev.br_startblock += mod; - prev.br_startoff = start; - }
...and here, we have a @del extent that is unwritten and a @prev extent that is written. We aim to trick xfs_bmap_add_extent_unwritten_real into extending @del towards startoff==0 and returning with @icur pointing at @del (not @prev) so that the next time we go around the loop we see an rtextsize-aligned @del and simply unmap it...
fs/xfs/libxfs/xfs_bmap.c __xfs_bunmapi( · line 5423
Quoted 3 lines
fs/xfs/libxfs/xfs_bmap.c +3
-0+ unwrite_start =max3 (start, + del.br_startoff - mod, + prev.br_startoff);
...however, if @prev is too long to convert+combine with @del, the conversion routine converts @prev to unwritten and returns with @icur pointing to @prev, not @del. That's how we leak @del.
This patch fixes that by capping the conversion to the start of the rtext alignment, which means that we can always merge with @del and always return with @icur pointing at @del. Ok, that's exactly what the commit message says.
It was /really/ helpful to be able to use the test case to walk through exactly what this patch is trying to fix.
--D
Quoted 10 lines
fs/xfs/libxfs/xfs_bmap.c +4
-0+ mod = unwrite_start - prev.br_startoff; + prev.br_startoff = unwrite_start; + prev.br_startblock += mod; + prev.br_blockcount -= mod; prev.br_state =XFS_EXT_UNWRITTEN ; error = xfs_bmap_add_extent_unwritten_real(tp, ip, whichfork, &icur, &cur,Signature · 1 line
2.24.0