No subject

protect xfs_sync_worker with s_umount semaphore' we have very solid test runs.
Unfortunately, as of commit 8866fc6 'xfs: shutdown xfs_sync_worker before the
log' we have started hitting these crashes again on an occasional basis.  With
this newest patch we have not reproduced these crashes and it has been well by
Mark and on my dev boxes.

> > Index: b/fs/xfs/xfs_mount.c
> > ===================================================================
> > --- a/fs/xfs/xfs_mount.c
> > +++ b/fs/xfs/xfs_mount.c
> > @@ -1490,6 +1490,11 @@ xfs_unmountfs(
> >  
> >  	xfs_qm_unmount(mp);
> >  
> > +	/* flush the worker queues while the log still exists and
> > +	 * before the final sync and unmount record.
> > +	 */
> > +	xfs_syncd_stop(mp);
> 
> xfs_syncd_stop() needs to die rather than being moved from place to
> place every time some problem is seemd.  I outlined what we need to
> do to solve the problems once and for all a couple of months ago:
> 
> http://oss.sgi.com/archives/xfs/2012-06/msg00064.html

Yikes, that patchset you've posted is very nice.  Probably it is appropriate
for the 3.7 merge window.  We also need a fix for 3.5-stable and 3.6.  Do you
have a suggestion for that?  It looks like you've also moved the final log
force to after cancellation of the work queue in patch 6, similar to what Mark
has done... so it would seem that Mark has the right idea.  I think Mark's
patch is appropriate for 3.6 and 3.5-stable inclusion to fix the regression
introduced by me in 8866fc6.

What is your opinion?

Regards,
	Ben