Re: 9.0-RC1 panic in tcp_input: negative winow.

From: Pawel Jakub Dawidek <pjd_at_FreeBSD.org>
Date: Sun, 23 Oct 2011 08:10:38 +0200
On Sun, Oct 23, 2011 at 12:35:15PM +1100, Lawrence Stewart wrote:
> On 10/22/11 19:49, Pawel Jakub Dawidek wrote:
> > The panic message says:
> >
> > 	panic: tcp_input negative window: tp 0xfffffe007763e000 rcv_nxt 3718269252 rcv_adv 3718268291
> >
> > I only have picture of the backtrace:
> >
> > 	http://people.freebsd.org/~pjd/misc/panic_negative_window.jpg
> >
> 
> ewww that is not good. Can you give us any more information about the 
> machine and what it's doing? Is it terminating TCP connections from the 
> internet at large or only local LAN (i.e. is there likely to be packet 
> loss happening)? Are you doing TSO or LRO? Do you have any non-default 
> tuning in place?

It is my local file server. It is doing NFS and AFP over LAN and also
downloads files from the internet. It is triggered after few hours.
I changed the KASSERT() into printf() and added printing 'win' variable
and this is what got logged during the night:

05:16:24 tcp_input negative window: tp 0xfffffe0026772b70 rcv_nxt 1107827269 rcv_adv 1107826256 win=242
05:16:29 tcp_input negative window: tp 0xfffffe0026772b70 rcv_nxt 1107833451 rcv_adv 1107832977 win=880
05:16:41 tcp_input negative window: tp 0xfffffe0026772b70 rcv_nxt 1107849563 rcv_adv 1107848860 win=639
05:20:02 tcp_input negative window: tp 0xfffffe0026772b70 rcv_nxt 1108108230 rcv_adv 1108107331 win=567
05:24:30 tcp_input negative window: tp 0xfffffe0026772b70 rcv_nxt 1108433302 rcv_adv 1108432272 win=974
05:24:46 tcp_input negative window: tp 0xfffffe0026772b70 rcv_nxt 1108450385 rcv_adv 1108450060 win=751
05:26:44 tcp_input negative window: tp 0xfffffe0026772b70 rcv_nxt 1108574818 rcv_adv 1108573851 win=71
05:28:03 tcp_input negative window: tp 0xfffffe0026772b70 rcv_nxt 1108654103 rcv_adv 1108653166 win=0
05:28:43 tcp_input negative window: tp 0xfffffe0026772b70 rcv_nxt 1108692396 rcv_adv 1108691451 win=0
05:30:06 tcp_input negative window: tp 0xfffffe0026772b70 rcv_nxt 1108781258 rcv_adv 1108780372 win=235
05:35:05 tcp_input negative window: tp 0xfffffe0026772b70 rcv_nxt 1109067578 rcv_adv 1109067335 win=663
05:37:03 tcp_input negative window: tp 0xfffffe0026772b70 rcv_nxt 1109180403 rcv_adv 1109179411 win=0
05:41:03 tcp_input negative window: tp 0xfffffe0026772b70 rcv_nxt 1109428265 rcv_adv 1109427375 win=170

And the systems seems to be fine.

I'm happy to test patches, but one round would take 24h.

My suggestion would be that if we won't be able to fix it before 9.0,
we should turn this assertion off, as the system seems to be able to
recover.

-- 
Pawel Jakub Dawidek                       http://www.wheelsystems.com
FreeBSD committer                         http://www.FreeBSD.org
Am I Evil? Yes, I Am!                     http://yomoli.com

Received on Sun Oct 23 2011 - 04:11:27 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:19 UTC