kern/176667: libalias locks on uninitalized data

[Gleb Smirnoff]

The following reply was made to PR kern/176667; it has been noted by GNATS.

From: Gleb Smirnoff <glebius at FreeBSD.org>
To: Lutz Donnerhacke <lutz at iks-service.de>
Cc: freebsd-gnats-submit at FreeBSD.org
Subject: Re: kern/176667: libalias locks on uninitalized data
Date: Thu, 7 Mar 2013 20:30:26 +0400

 On Tue, Mar 05, 2013 at 03:54:50PM +0000, Lutz Donnerhacke wrote:
 L> 
 L> >Number:         176667
 L> >Category:       kern
 L> >Synopsis:       libalias locks on uninitalized data
 L> >Confidential:   no
 L> >Severity:       non-critical
 L> >Priority:       low
 L> >Responsible:    freebsd-bugs
 L> >State:          open
 L> >Quarter:        
 L> >Keywords:       
 L> >Date-Required:
 L> >Class:          sw-bug
 L> >Submitter-Id:   current-users
 L> >Arrival-Date:   Tue Mar 05 16:00:00 UTC 2013
 L> >Closed-Date:
 L> >Last-Modified:
 L> >Originator:     Lutz Donnerhacke
 L> >Release:        FreeBSD 8.3-RELEASE (GENERIC)
 L> >Organization:
 L> IKS Service GmbH
 L> >Environment:
 L> FreeBSD server7.net.encoline.de 8.3-RELEASE FreeBSD 8.3-RELEASE #0: Mon Apr  9 21:23:18 UTC 2012     root at mason.cse.buffalo.edu:/usr/obj/usr/src/sys/GENERIC  amd64
 L> 
 L> >Description:
 L> While testing terminating a huge number of PPPoX clients the kernel panics while doing in-kernel NAT.
 L> 
 L> #4 0xffffffff808e8775 at calltrap+0x8
 L> #5 0xffffffff80fa0f01 at HouseKeeping+0xa1
 L> #6 0xffffffff80f9e6ab at LibAliasOutLocked+0x3b
 L> 
 L> Please note, that the stack trace is incomplete. There are calls to IncrementalCleanup() and DeleteLink(), which are not reported in the stack trace.
 L> 
 L> The problem seems to come from incorrect locking, so the contents of the libalias database get corrupted.
 L> 
 L> This patch might be not the full solution, but is an obvious fix for an obvious bug.
 L> >How-To-Repeat:
 L> Setting up ipfw nat, add more then 9000 clients using mpd5.6, generate traffic
 L> >Fix:
 L> --- sys/netinet/libalias/alias_db.c.ORIG        2013-03-05 16:49:13.000000000 +0100
 L> +++ sys/netinet/libalias/alias_db.c     2013-03-05 16:50:09.000000000 +0100
 L> @@ -2767,8 +2767,8 @@
 L>         struct ip_fw rule;      /* On-the-fly built rule */
 L>         int fwhole;             /* Where to punch hole */
 L> 
 L> -       LIBALIAS_LOCK_ASSERT(la);
 L>         la = lnk->la;
 L> +       LIBALIAS_LOCK_ASSERT(la);
 L> 
 L>  /* Don't do anything unless we are asked to */
 L>         if (!(la->packetAliasMode & PKT_ALIAS_PUNCH_FW) ||
 L> @@ -2841,8 +2841,8 @@
 L>  {
 L>         struct libalias *la;
 L> 
 L> -       LIBALIAS_LOCK_ASSERT(la);
 L>         la = lnk->la;
 L> +       LIBALIAS_LOCK_ASSERT(la);
 L>         if (lnk->link_type == LINK_TCP) {
 L>                 int fwhole = lnk->data.tcp->fwhole;     /* Where is the firewall
 L>                                                          * hole? */
 
 The code edited isn't correct and the patch is neither.
 
 The fw punching isn't supported when libalias is compiled into kernel.
 
 The LIBALIAS_LOCK_ASSERT(la) on not initialized variable couldn't even
 pass compiler, if only the entire fw punching code was enabled.
 
 So these lines need to be just removed for sanity. Unfortunately this isn't
 related to panic you are hitting.
 
 Do you have cores of that panic?
 
 -- 
 Totus tuus, Glebius.