need help how to debug xfs crash issue xfs_iunlink_remove: xfs_inotobp() returned error 22

[Dave Chinner]

On Wed, Apr 10, 2013 at 12:10:25PM +0200, Emmanuel Florac wrote:
> Le Tue, 9 Apr 2013 23:10:03 +0800
> 符永涛 <yongtaofu at gmail.com> écrivait:
> 
> > > Apr  9 11:01:30 cqdx kernel: XFS (sdb): I/O Error Detected.
> > > Shutting down filesystem
> 
> This. I/O error detected. That means that at some point the underlying
> device (disk, RAID array, SAN volume) couldn't be reached. So this
> could very well be a case of a flakey drive, array, cable or SCSI
> driver.

You can't take that one line of output out of context and then say
it's a hardware problem - that's a generic IO-error-causes-shutdown
message, not an EIO from the storage stack.

The EINVAL error that is reported before this is the cause of the
shutdown, and that is from a corrupted unlinked list. EINVAL
indicates that we are falling off the end of the unlinked list
without finding the inode that we are trying to remove from the
unlinked list. Debug kernels will assert fail at this point.

What causes that problem is still unknown. Nobody has been able to
isolate the reproducer, so progress is slow. If someone can give me
a script that reprodcues it directly on XFS (i.e no gluster), then
it won't take long to find the bug....

Cheers,

Dave.
-- 
Dave Chinner
david at fromorbit.com