Re: immediate v6-related crash

["Steven M. Bellovin" <smb%cs.columbia.edu@localhost>]

On Tue, 1 Apr 2008 09:35:53 +0900
Joerg Sonnenberger <joerg%britannica.bec.de@localhost> wrote:

> On Mon, Mar 31, 2008 at 07:56:09PM +0000, Steven M. Bellovin wrote:
> > On an up-to-date -current kernel (i386, Dell Dimension 8400 with bge
> > card), I get a reproducible panic shortly after going multi-user:
> > "in6_cksum: mbuf too short for IPv6 header".
> 
> Backtrace? E.g. ddb.commandonenter=bt
> 
Several more reboots have yielded more data.

First -- and this one is still tentative -- the crash is triggered by
named running.  I need to test a few more times on that, however.

Second, turning off ddb.onpanic and using the sysctl you suggested did
indeed give a useful backtrace.  The original panic is

in6_cksum
icmp6_input
ip6_input
ip6intr
softint_dispatch
DDB lost fram for Xsoftintr
Xsoftintr

I've omitted the addresses because I think the chain is pretty clear.

However, there's a recursive panic:
uvm_fault(0xc0b27d00, 0, 1) -> 0xe
supervisor trap page fault, code=0
Faulted in DDB

I can't get out of that; it repeats indefinitely...

                --Steve Bellovin, http://www.cs.columbia.edu/~smb