xfs: fix inode count underrun
Subsystems (3)
cover xfs: fix inode count underrun
Hi folks,
With the conversion of XFS to use the generic per-cpu superblocks, I overlooked the fact that the update batch size is important to the accuracy of the comparison function. Using different batch sizes means percpu_counter_compare() doesn't detect when it should fall back to percpu_counter_sum() for accuracy correctly, resulting in counter comparisons being inaccurate. This leads to problems with zero threshold detection in XFS.
To fix, add __percpu_counter_compare() to take a caller supplied batch size. This fixes the XFS regression introduced in 4.1-rc1.
-Dave.
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
PATCH 1/2 percpu_counter: batch size aware __percpu_counter_compare()
From: Dave Chinner <[email protected]>
From: Dave Chinner <[email protected]>
XFS uses non-stanard batch sizes for avoiding frequent global counter updates on it's allocated inode counters, as they increment or decrement in batches of 64 inodes. Hence the standard percpu counter batch of 32 means that the counter is effectively a global counter. Currently Xfs uses a batch size of 128 so that it doesn't take the global lock on every single modification.
However, Xfs also needs to compare accurately against zero, which means we need to use percpu_counter_compare(), and that has a hard-coded batch size of 32, and hence will spuriously fail to detect when it is supposed to use precise comparisons and hence the accounting goes wrong.
Add __percpu_counter_compare() to take a custom batch size so we can use it sanely in XFS and factor percpu_counter_compare() to use it.
--- include/linux/percpu_counter.h | 13 ++++++++++++- lib/percpu_counter.c | 6 +++--- 2 files changed, 15 insertions(+), 4 deletions(-)
include/linux/percpu_counter.h +12 -1
@@ -41,7 +41,12 @@ 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 percpu_counter_compare (struct percpu_counter *fbc,s64 rhs); +int __percpu_counter_compare (struct percpu_counter *fbc,s64 rhs,s32 batch); + +static inline int percpu_counter_compare (struct percpu_counter *fbc,s64 rhs) +{ +return __percpu_counter_compare (fbc, rhs, percpu_counter_batch); +}static inline void percpu_counter_add (struct percpu_counter *fbc,s64 amount) {
@@ -116,6 +121,12 @@ static inline int percpu_counter_compare(struct percpu_counter *fbc, s64 rhs)return 0 ; } +static inline int +percpu_counter_compare (struct percpu_counter *fbc,s64 rhs,s32 batch) +{ +return percpu_counter_compare (fbc, rhs); +} +static inline void percpu_counter_add (struct percpu_counter *fbc,s64 amount) {
lib/percpu_counter.c +3 -3
@@ -197,13 +197,13 @@ static int percpu_counter_hotcpu_callback(struct notifier_block *nb,* Compare counter against given value. * Return 1 if greater, 0 if equal and -1 if less */ -int percpu_counter_compare (struct percpu_counter *fbc,s64 rhs) +int __percpu_counter_compare (struct percpu_counter *fbc,s64 rhs,s32 batch) {s64 count; count =percpu_counter_read (fbc);/* Check to see if rough count will be sufficient for comparison */ -if (abs (count - rhs) > (percpu_counter_batch*num_online_cpus ())) { +if (abs (count - rhs) > (batch *num_online_cpus ())) {if (count > rhs)return 1 ;else
@@ -218,7 +218,7 @@ int percpu_counter_compare(struct percpu_counter *fbc, s64 rhs)else return 0 ; } -EXPORT_SYMBOL (percpu_counter_compare); +EXPORT_SYMBOL (__percpu_counter_compare);static int __init percpu_counter_startup (void ) {
Signature · 1 line
2.0.0
Re: [PATCH 1/2] percpu_counter: batch size aware __percpu_counter_compare()
This looks reasonable, but I miss Ccs to the authors of the percpu_counter code and linux-kernel.
Quoted 3 lines · Dave Chinner
From: Dave Chinner <[email protected]>
From: Dave Chinner <[email protected]>
And this looks wrong, too :)
Re: [PATCH 1/2] percpu_counter: batch size aware __percpu_counter_compare()
Quoted 2 lines · Christoph Hellwig
This looks reasonable, but I miss Ccs to the authors of the
percpu_countercode and linux-kernel.
Ok, saw it on lkml. Guess I need more coffee, sorry..
Re: [PATCH 1/2] percpu_counter: batch size aware __percpu_counter_compare()
In reply to Christoph HellwigQuoted 2 lines · Christoph Hellwig
This looks reasonable, but I miss Ccs to the authors of the
percpu_countercode and linux-kernel.
linux-kernel was cc'd.
Quoted 6 lines · Dave Chinner
Quoted 3 lines
From: Dave Chinner <[email protected]>
From: Dave Chinner <[email protected]>
And this looks wrong, too :)
That looks like a change of tool behaviour. I just upgraded guilt; the old version stripped "from" lines in patch descriptions from the git commit log as goes into the author field of the commit. The new version does not appear to strip the from lines, and so when git-send-email formats it, it puts a from line in the message, and then the commit message has one too.
Thanks for pointing it out.
Cheers,
Dave.
Signature · 7 lines
Dave Chinner [email protected] _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 1/2] percpu_counter: batch size aware __percpu_counter_compare()
Quoted 11 lines · Dave Chinner
Quoted 2 lines · Christoph Hellwig
This looks reasonable, but I miss Ccs to the authors of the
percpu_countercode and linux-kernel.linux-kernel was cc'd.
Quoted 5 lines · Dave Chinner
Quoted 3 lines
From: Dave Chinner <[email protected]>
From: Dave Chinner <[email protected]>
Quoted 8 lines
Quoted 1 line
And this looks wrong, too :)
That looks like a change of tool behaviour. I just upgraded guilt; the old version stripped "from" lines in patch descriptions from the git commit log as goes into the author field of the commit. The new version does not appear to strip the from lines, and so when git-send-email formats it, it puts a from line in the message, and then the commit message has one too.
Ok, so I remembered it the wrong way around, but it is a tool bug:
$ guilt patchbomb -s d480578..cd83c77 cd83c773 xfs: inode counter needs to use __percpu_counter_compare 1e41b5ad percpu_counter: batch size aware __percpu_counter_compare() Are these what you want to send? [Y/n] ....
Note the use of "-s". from the man page:
-s
Don't add additional repository committer sign-offs to the patch. This allows the sign-off chain to be fully expressed in the commit messages and not changed by the act of sending a patchbomb.
It would appear that the "-s" option is not working properly. I'll fix that before sending the next round of patches....
Cheers,
Dave.
Signature · 7 lines
Dave Chinner [email protected] _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 1/2] percpu_counter: batch size aware __percpu_counter_compare()
Quoted 31 lines · Dave Chinner
Quoted 11 lines · Dave Chinner
Quoted 2 lines · Christoph Hellwig
This looks reasonable, but I miss Ccs to the authors of the
percpu_countercode and linux-kernel.linux-kernel was cc'd.
Quoted 5 lines · Dave Chinner
Quoted 3 lines
From: Dave Chinner <[email protected]>
From: Dave Chinner <[email protected]>
Quoted 8 lines
Quoted 1 line
And this looks wrong, too :)
That looks like a change of tool behaviour. I just upgraded guilt; the old version stripped "from" lines in patch descriptions from the git commit log as goes into the author field of the commit. The new version does not appear to strip the from lines, and so when git-send-email formats it, it puts a from line in the message, and then the commit message has one too.
Ok, so I remembered it the wrong way around, but it is a tool bug:
$ guilt patchbomb -s d480578..cd83c77 cd83c773 xfs: inode counter needs to use __percpu_counter_compare 1e41b5ad percpu_counter: batch size aware __percpu_counter_compare() Are these what you want to send? [Y/n] ....Note the use of "-s". from the man page:
Quoted 8 lines
-s
Don't add additional repository committer sign-offs to the patch. This allows the sign-off chain to be fully expressed in the commit messages and not changed by the act of sending a patchbomb.It would appear that the "-s" option is not working properly. I'll fix that before sending the next round of patches....
<sigh>
So, the packaged version of guilt on a brand new debian unstable installation is actually older than the 2 year version of guilt I had previously installed and used on a different machine. IOWs, the debian distro version has this bug in it:
http://repo.or.cz/w/guilt.git?a=commit;h=8f88f953580a0cacf111bf64c0e838014ee30c01
Yeah, that was fixed more than 3 years ago. Oh well, back to using a locally installed version...
Cheers,
Dave.
Signature · 2 lines
Dave Chinner [email protected]
PATCH 2/2 xfs: inode counter needs to use __percpu_counter_compare
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
From: Dave Chinner <[email protected]>
Because the counter uses a custom batch size, the comparison function needs to be aware of that batch size otherwise the comparison does not work correctly. This leads to ASSERT failures on generic/027 like this:
XFS: Assertion failed: 0, file: fs/xfs/xfs_mount.c, line: 1099
------------[ cut here ]------------ .... Call Trace: [<ffffffff81522a39>] xfs_mod_icount+0x99/0xc0 [<ffffffff815285cb>] xfs_trans_unreserve_and_mod_sb+0x28b/0x5b0 [<ffffffff8152f941>] xfs_log_commit_cil+0x321/0x580 [<ffffffff81528e17>] xfs_trans_commit+0xb7/0x260 [<ffffffff81503d4d>] xfs_bmap_finish+0xcd/0x1b0 [<ffffffff8151da41>] xfs_inactive_ifree+0x1e1/0x250 [<ffffffff8151dbe0>] xfs_inactive+0x130/0x200 [<ffffffff81523a21>] xfs_fs_evict_inode+0x91/0xf0 [<ffffffff811f3958>] evict+0xb8/0x190 [<ffffffff811f433b>] iput+0x18b/0x1f0 [<ffffffff811e8853>] do_unlinkat+0x1f3/0x320 [<ffffffff811d548a>] ? filp_close+0x5a/0x80 [<ffffffff811e999b>] SyS_unlinkat+0x1b/0x40 [<ffffffff81e0892e>] system_call_fastpath+0x12/0x71
This is a regression introduced by commit 501ab32 ("xfs: use generic percpu counters for inode counter").
--- fs/xfs/xfs_mount.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
fs/xfs/xfs_mount.c +3 -2
@@ -1106,8 +1106,9 @@ xfs_mod_icount( int64_t delta) {/* deltas are +/-64, hence the large batch size of 128. */ -__percpu_counter_add (&mp->m_icount, delta,128 ); -if (percpu_counter_compare (&mp->m_icount,0 ) <0 ) { +#define _ICOUNT_BATCH 128 +__percpu_counter_add (&mp->m_icount, delta, _ICOUNT_BATCH); +if (__percpu_counter_compare (&mp->m_icount,0 , _ICOUNT_BATCH) <0 ) {ASSERT (0 );percpu_counter_add (&mp->m_icount, -delta);return -EINVAL ;
Signature · 6 lines
2.0.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 2/2] xfs: inode counter needs to use __percpu_counter_compare
fs/xfs/xfs_mount.c xfs_mod_icount( · line 1110
Quoted 12 lines
fs/xfs/xfs_mount.c +2
-2@@ -1106,8 +1106,9 @@ xfs_mod_icount( int64_t delta) {/* deltas are +/-64, hence the large batch size of 128. */ -__percpu_counter_add (&mp->m_icount, delta,128 ); -if (percpu_counter_compare (&mp->m_icount,0 ) <0 ) { +#define _ICOUNT_BATCH 128 +__percpu_counter_add (&mp->m_icount, delta, _ICOUNT_BATCH);
Can you give XFS_ prefixes to the atch sizes and move them otuside the function? And fix up the instance in xfs_mod_fdblocks as well while you're at it.
Otherwise looks fine.
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 2/2] xfs: inode counter needs to use __percpu_counter_compare
Quoted 8 lines · Christoph Hellwig
Quoted 8 lines
fs/xfs/xfs_mount.c +0
-0@@ -1106,8 +1106,9 @@ xfs_mod_icount( int64_t delta) {/* deltas are +/-64, hence the large batch size of 128. */
Quoted 8 lines
fs/xfs/xfs_mount.c xfs_mod_icount( · line 1110
Quoted 4 lines
fs/xfs/xfs_mount.c +2
-2-__percpu_counter_add (&mp->m_icount, delta,128 ); -if (percpu_counter_compare (&mp->m_icount,0 ) <0 ) { +#define _ICOUNT_BATCH 128 +__percpu_counter_add (&mp->m_icount, delta, _ICOUNT_BATCH);Can you give
XFS_prefixes to the atch sizes and move them otuside the function? And fix up the instance inxfs_mod_fdblocksas well while you're at it.
Sure.
Cheers,
Dave.
Signature · 2 lines
Dave Chinner [email protected]