panic during work with jailed postgresql8.4

[pluknet]

On 1 April 2010 22:18, Oleg Lomaka <oleg.lomaka at gmail.com> wrote:
> Hello,
>
> I have a kernel panic when connect to postgresql8.4 server installed in one of jails from another jail. It's 100% reproducible.
> Also I have tried to connect from host machine to jailed pg server. That way it works fine without crash.
>
> Server configuration uses geli and zfs. Four disks encrypted using geli. And raidz2 is using ad8.eli, ad10.eli, ad12.eli, ad14.eli providers. All jails located at this raidz2 pool.
>
> Also I use ezjail for jails management. And it uses NFS to mount directories with base system.
>
> atal double fault
> rip = 0xffffffff8063510a
> rsp = 0xffffff80eaec5f50
> rbp = 0xffffff80eaec6040
> cpuid = 1; apic id = 02
> panic: double fault
> cpuid = 1
> Uptime: 7m11s
> Physical memory: 8169 MB
>
> uname -a
> FreeBSD cerberus.regredi.com 8.0-STABLE FreeBSD 8.0-STABLE #7 r206031: Thu Apr  1 13:43:57 EEST 2010     root at cerberus.regredi.com:/usr/obj/usr/src/sys/GENERIC  amd64
>
> Link to dmesg.boot:
> http://docs.google.com/leaf?id=0B-irbkAqk9i7OGY2ZWJiODgtOWJmMy00NDQ1LTliZDctZjU3N2YwNmMxNjZl&hl=en
>
> Link to kernel core backtrace:
> http://docs.google.com/Doc?docid=0AeirbkAqk9i7ZGc5Yzc2ZndfM2M4NzYydmRw&hl=en

Looking at backtrace, I wonder whether tp->t_maxseg changes in
tcp_mtudisc() at all.
You should be able to extract its value on each 2*n frame in that big
recursive call.

-- 
wbr,
pluknet