On Tue, Jan 26, 2010 at 09:31:40AM +1100, Dave Chinner wrote:
> On Mon, Jan 25, 2010 at 06:53:08AM -0500, Christoph Hellwig wrote:
> > > diff --git a/fs/xfs/linux-2.6/xfs_sync.c b/fs/xfs/linux-2.6/xfs_sync.c
> > > index 98b8937..ca0cc59 100644
> > > --- a/fs/xfs/linux-2.6/xfs_sync.c
> > > +++ b/fs/xfs/linux-2.6/xfs_sync.c
> > > @@ -270,8 +270,7 @@ xfs_sync_inode_attr(
> > > goto out_unlock;
> > > }
> > >
> > > - error = xfs_iflush(ip, (flags & SYNC_WAIT) ?
> > > - XFS_IFLUSH_SYNC : XFS_IFLUSH_DELWRI);
> > > + error = xfs_iflush(ip, (flags & SYNC_WAIT));
> > No need for the masking here, as xfs_iflush simply ignores SYNC_TRYLOCK.
> > > /* Now we have an inode that needs flushing */
> > > error = xfs_iflush(ip, sync_mode);
> > > + if (!(sync_mode & SYNC_WAIT))
> > > + goto requeue_no_flock;
> > So for the !wait case we entirely ignore the return value? We should
> > at least check for an I/O error here I think.
> I'm not sure we can get an error on a delayed write that we
> need to take action on. We don't actually issue the IO from this
> call, so the only error case is on reading the buffer. If we get
> an error there, what should be do?
> My thinking was that if we can't write back the inode, we can't
> reclaim it, and if the error is from a transient read error (e.g. FC
> path failover) we should be retrying anyway as we do in
I forgot to mention that it will get caught by a sync
reclaim pass when there is a context to return the error to....