Assertion fail in xfs_reclaim_inode [2.6.33.2]

[Dave Chinner]

On Mon, Apr 05, 2010 at 05:47:30PM -0700, Simon Kirby wrote:
> We're seeing crashing from XFS likely resulting from some storage
> corruption.  This seems to be getting tickled from backups every two
> days as it happens at the same time of day each time on whichever box
> is active for these storage volumes.  We've seen it about 4 times now.
> 
> This was captured over a serial console over IPMI serial-over-LAN, and
> it seemed the serial console loglevel was not low enough to see the
> priority-less printk() in assfail() in fs/xfs/support/debug.c, so all we
> got this time was this backtrace:
.....
> 
> Seems to be from the only assert in xfs_reclaim_inode:
> 
>         ASSERT_ALWAYS(__xfs_iflags_test(ip, XFS_IRECLAIMABLE));
> 
> Would any of the patches in the stable queue-2.6.32 be missing from
> 2.6.33 but perhaps fix this issue?

Yes, there is - commit f1f724e4b523d444c5a598d74505aefa3d6844d2.
I've been a bit slack in doing the backport and push to 2.6.33.x.
I'll get the ball rolling later this afternoon....

Cheers,

Dave.
-- 
Dave Chinner
david at fromorbit.com