Re: XFS hung task in xfs_ail_push_all_sync() when unmounting FS after disk failure/recovery

[Carlos Maiolino <cmaiolino@redhat.com>]

On Thu, Mar 24, 2016 at 09:37:47AM +1100, Dave Chinner wrote:
> On Wed, Mar 23, 2016 at 04:32:21PM +0100, Carlos Maiolino wrote:
> > I'm still trying to get a reliable reproducer, at least exactly with what I have
> > seen a few days ago.
> > 
> > Shyam, could you try to reproduce it with a recent/upstream kernel? That would
> > be great to make sure we have been seen the same issue.
> > 
> > AFAICT, it happens in the following situation:
> > 
> > 1 - Something is written to the filesystem
> > 2 - log checkpoint is done for the previous write
> > 3 - Disk failure
> > 4 - XFS tries to writeback metadata logged in [2]
> > 
> > When [4] happens, I can't trigger xfs_log_force messages all the time, most of
> > time I just get an infinite loop in these messages:
> > 
> > [12694.318109] XFS (dm-0): Failing async write on buffer block
> > 0xffffffffffffffff. Retrying async write.
> >
> > Sometimes I can trigger the xfs_log_force() loop.
> 
> This all smells like the filesystem is getting IO errors but it not
> in a shutdown state. What happens when you run 'xfs_io -x -c
> "shutdown" /mnt/pt' on a filesystem in this state? Can you then
> unmount it?
> 

I'll give it a try today, although, I can't do it while umount command is hung,
since, before the command get stuck, the mount point is removed from the
user namespace, so I have no access to the mountpoint from userspace while the
command is 'running'.


> Cheers,
> 
> Dave.
> -- 
> Dave Chinner
> david@fromorbit.com
> 
> _______________________________________________
> xfs mailing list
> xfs@oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs

-- 
Carlos

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs