From mboxrd@z Thu Jan  1 00:00:00 1970
To: 9fans@cse.psu.edu
Subject: Re: [9fans] drawterm sigsegv on linux 2.6.18
From: "Russ Cox" <rsc@swtch.com>
Date: Tue, 29 May 2007 08:26:02 -0400
In-Reply-To: <1180366010.706662.238930@q69g2000hsb.googlegroups.com>
MIME-Version: 1.0
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit
Message-Id: <20070529122316.C7AD51E8C3B@holo.morphisms.net>
Topicbox-Message-UUID: 74527ac0-ead2-11e9-9d60-3106f5b1d025

> When I do a "du -a /" on a plan9 cpu server via drawterm,  drawterm
> gets a sigsegv after a while.  Gdb  reports that the sigsegv arrived
> when tas.c:9 (inline assembler code to do test and set) is called.  I
> cannot see anything wrong with the tas code though.

The problem isn't in tas but its caller -- note that tas
is being passed a null pointer.  Thanks for the gdb 
stack trace, which was very helpful.

The fix this change to kern/devfs-posix.c (a similar
change applies to kern/devfs-win32.c too):

 		cname = addelem(cname, name[i]);
 		wq->qid[i] = nc->qid;
 	}
-	nc->name = nil;
-	cnameclose(cname);
+	nc->name = cname;
 	if(i != nname){
 		cclose(nc);
 		wq->clone = nil;

The bug is only triggered when traversing paths 
more than 16 levels deep on the local machine.

Russ