Re: [2/3] 2.6.22-rc1: known regressions v2 - XFS

[David Chinner <dgc@sgi.com>]

On Wed, May 16, 2007 at 04:40:20PM -0700, Jeremy Fitzhardinge wrote:
> David Chinner wrote:
> > Jeremy has tentatively indicated that the patch has fixed the problem.
> > Have you seen any more problems since applying the patch, Jeremy?
> >   
> 
> No, it continues to seem sound with casual use; I would have expected to
> see the problem reoccur by now.  I'd like to rerun the full set of tests
> I did before to be sure, but so far so good.  No other apparent
> regressions either.

Good to here. I think the problem is fixed, then.

> Also, the match between the observed symptoms and the bugfix is very
> good, which adds confidence (ie, no element of "it works now but we
> don't know why").  I guess the only remaining concern is whether there
> are any other paths which fail to dirty the inode.

There aren't any that I can see - if more come up we'll deal with
them then.

> Did you manage to repro the problem?

xfs_io is my friend ;)

Without patch:

# touch /mnt/scratch/fred
# xfs_io -c "pwrite 0 5" -c "s" -c "pwrite 5 5" /mnt/scratch/fred
wrote 5/5 bytes at offset 0
5.000000 bytes, 1 ops; 0.0000 sec (78.755 KiB/sec and 16129.0323 ops/sec)
wrote 5/5 bytes at offset 5
5.000000 bytes, 1 ops; 0.0000 sec (542.535 KiB/sec and 111111.1111 ops/sec)
# umount /mnt/scratch; mount /mnt/scratch; ls -l /mnt/scratch/fred
-rw-r--r-- 1 root      root      5 May 17 10:04 fred
#

So the second 5 byte write didn't change the file size.

With patch:

# touch /mnt/scratch/fred
# xfs_io -c "pwrite 0 5" -c "s" -c "pwrite 5 5" /mnt/scratch/fred
wrote 5/5 bytes at offset 0
5.000000 bytes, 1 ops; 0.0000 sec (76 KiB/sec and 15625.0000 ops/sec)
wrote 5/5 bytes at offset 5
5.000000 bytes, 1 ops; 0.0000 sec (610 KiB/sec and 125000.0000 ops/sec)
# umount /mnt/scratch; mount /mnt/scratch; ls -l /mnt/scratch/fred
-rw-r--r-- 1 root      root     10 May 17 09:53 fred
#

So yes, I've reproduced it and confirmed the patch fixes the problem.

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group