On Thu Jul 05 2007 at 10:07:20 +0100, Andrew Doran wrote:
> > +	simple_lock(&vp->v_interlock);
> > +	vxlock = vp->v_flag & VXLOCK;
> > +	simple_unlock(&vp->v_interlock);
> > +
> >  	/*
> >  	 * ..., but don't cache the device node. Also, if we did a
> > -	 * remove, don't cache the node.
> > +	 * remove, don't cache the node.  Finally, if the node is
> > +	 * already being nuked, don't do this, as we would re-enter
> > +	 * the same path and deadlock.
> >  	 */
> > -	if (vp->v_type == VBLK || vp->v_type == VCHR
> > +	if ((vp->v_type == VBLK || vp->v_type == VCHR
> >  	    || (VTOLAYER(vp)->layer_flags & LAYERFS_REMOVED))
> > +	  && (vxlock == 0))
> >  		vgone(vp);
> >  	return (0);
> >  }
> 
> I think it's better to hold the interlock across the if statement and do
> a vgonel() instead.

Good call.  O coffee, coffee, wherefore art thou, coffee?

-- 
Antti Kantee <pooka@iki.fi>                     Of course he runs NetBSD
http://www.iki.fi/pooka/                          http://www.NetBSD.org/
    "la qualité la plus indispensable du cuisinier est l'exactitude"