[PATCH 1/2] xfs: fix xfs_mark_inode_dirty during umount

[Dave Chinner]

On Tue, Aug 30, 2011 at 02:39:49AM -0400, Christoph Hellwig wrote:
> On Tue, Aug 30, 2011 at 04:24:16PM +1000, Dave Chinner wrote:
> > >  xfs_mark_inode_dirty_sync(
> > > @@ -82,6 +81,10 @@ xfs_mark_inode_dirty_sync(
> > >  
> > >  	if (!(inode->i_state & (I_WILL_FREE|I_FREEING)))
> > >  		mark_inode_dirty_sync(inode);
> > > +	else {
> > > +		barrier();
> > > +		ip->i_update_core = 1;
> > > +	}
> > >  }
> > 
> > Why the barrier()? Isn't that just a compiler barrier? If you are
> > worried about catching the update vs clearing it in transaction
> > commit, shouldn't that use smp_mb() instead (in both places)?
> 
> It's a blind copy & past from xfs_fs_dirty_inode.  The comments
> there suggests it is for update ordering.

Right, I've always wondered about that, because the corresponding
code talks about requiring strongly ordered memory semantics and
that the compiler does this via SYNCHRONIZE() (barrier).