kern/87094: 5.4 system with SMP and IPFW crashes under load (mbuf underrun)

Robert Watson rwatson at FreeBSD.org
Sat Oct 8 10:50:20 PDT 2005


The following reply was made to PR kern/87094; it has been noted by GNATS.

From: Robert Watson <rwatson at FreeBSD.org>
To: Chris Ulrich <cdu at ucr.edu>
Cc: freebsd-gnats-submit at FreeBSD.org
Subject: Re: kern/87094: 5.4 system with SMP and IPFW crashes under load
 (mbuf	underrun)
Date: Sat, 8 Oct 2005 18:47:42 +0100 (BST)

 On Fri, 7 Oct 2005, Chris Ulrich wrote:
 
 > Run SMP and IPFW and natd on a 5.4 i386 system and run iperf between one 
 > interface and the other (server outside the nat interface).
 >
 > After a minute or so, netstat -m will show something like:
 > 4294266316 mbufs in use
 > 4294737989/25600 mbuf clusters in use (current/max)
 > 3560445 KBytes allocated to network
 > 201 calls to protocol drain routines
 >
 > After another period of time, the system will eventually reboot.  The 
 > behavior is not related to load or firewall rule changes.
 
 FYI, statistics errors in the mbuf allocator in 5.x are probably unrelated 
 to the reboot, and occurs due to races in statistics gathering.  6.x 
 corrects this problem, and I'm considering a merge of the changes to the 
 5.x branch following 6.0-RELEASE and some additional in-field testing.
 
 The reboot is, however, concerning.  Do you get a panic and reboot, or 
 just a reboot?  In particular, if you compile in the kernel debugger, does 
 it drop to the debugger at some point, or reboot?
 
 Could you try a 5-STABLE kernel and see if that works better?  There have 
 been a number of synchronization fixes that have not yet been released as 
 errata, which might affect this.
 
 Thanks,
 
 Robert N M Watson


More information about the freebsd-bugs mailing list