Re: [PATCH 5/5] xfs: handle CIl transaction commit failures correctly

[Dave Chinner <david@fromorbit.com>]

On Mon, Jan 24, 2011 at 04:12:49AM -0500, Christoph Hellwig wrote:
> > diff --git a/fs/xfs/xfs_trans.c b/fs/xfs/xfs_trans.c
> > index 50753d3..504a804 100644
> > --- a/fs/xfs/xfs_trans.c
> > +++ b/fs/xfs/xfs_trans.c
> > @@ -1754,15 +1754,26 @@ xfs_trans_commit_cil(
> >  	 */
> >  	log_vector = xfs_trans_alloc_log_vecs(tp);
> >  	if (!log_vector)
> > -		return ENOMEM;
> > +		goto out_enomem;
> >  
> >  	error = xfs_log_commit_cil(mp, tp, log_vector, commit_lsn, flags);
> > -	if (error)
> > -		return error;
> > +	if (error) {
> > +		/*
> > +		 * We will only get an error if no modifications have been
> > +		 * made to the items in the transaction. Hence treat it
> > +		  the same as a memory allocation failure.
> > +		 */
> > +		goto out_enomem;
> > +	}
> >  
> >  	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
> >  	xfs_trans_free(tp);
> >  	return 0;
> > +
> > +out_enomem:
> > +	/* caller cleans up transaction */
> > +	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
> > +	return ENOMEM;
> 
> _xfs_trans_commit already restores the process flags for an ENOMEM
> return, so the failure from xfs_trans_alloc_log_vecs is already
> handled correctly.  If we want to handle the EIO return from
> xfs_log_commit_cil the same way it just needs to be turned into an
> ENOMEM.  The big questions is if there's any point in having the
> shutdown check in xfs_trans_commit_cil - we already do one just before
> applying the trans deltas in _xfs_trans_commit, which is handled
> correctly and should be sufficient.

True. Removing the shutdown check makes xfs_log_commit_cil() a
function that can have a void return. That's probably a better
solution, so I'll modify it that way.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

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