The following reply was made to PR kern/34293; it has been noted by GNATS.

From: Michael van Elst <mlelstv@serpens.de>
To: gnats-bugs@netbsd.org
Cc: 
Subject: Re: kern/34293 vnd deadlocks on I/O buffers
Date: Sun, 27 Aug 2006 19:45:47 +0200

 Using a separate pool for nested buffers helps a little, but
 the same condition also prevents normal buffers from being
 allocated leading to the situation similar to the one described
 in kern/10731 and kern/20296.
 
 The following patch works against the root cause, which is the
 exhaustion of buffers.
 
 Index: vnd.c
 ===================================================================
 RCS file: /cvsroot/src/sys/dev/vnd.c,v
 retrieving revision 1.148
 diff -u -r1.148 vnd.c
 --- vnd.c	21 Jul 2006 16:48:48 -0000	1.148
 +++ vnd.c	27 Aug 2006 17:42:54 -0000
 @@ -507,6 +507,9 @@
  	if (vnddebug & VDB_FOLLOW)
  		printf("vndstrategy(%p): unit %d\n", bp, unit);
  #endif
 +	while (vnd->sc_maxactive > 0 && vnd->sc_active >= vnd->sc_maxactive) {
 +		tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
 +	}
  	BUFQ_PUT(vnd->sc_tab, bp);
  	wakeup(&vnd->sc_tab);
  	splx(s);
 
 
 -- 
                                 Michael van Elst
 Internet: mlelstv@serpens.de
                                 "A potential Snark may lurk in every tree."