Subject: Re: TCP hang on netbsd-2
To: None <tech-net@NetBSD.org>
From: Manuel Bouyer <firstname.lastname@example.org>
Date: 06/17/2005 15:18:07
On Sun, May 29, 2005 at 07:55:24PM +0200, Manuel Bouyer wrote:
> I'm getting hung ssh connections from a 2.0_RC2 to a 2.0_STABLE box.
> This ssh has lots of traffic on stdout.
> Here is the state on the 2.0_STABLE box:
> Proto Recv-Q Send-Q Local Address Foreign Address State
> tcp 0 32144 pop.ssh barder.53768 ESTABLISHED
> on the 2.0_RC2 box:
> tcp 0 0 barder.53768 pop.ssh ESTABLISHED
> So stdout in the ssh session is locked, there are data pending in the
> socket on the sender's side but it is never sent to the other end.
This happened again, this time between a 2.0_STABLE box (the sender) and
a 1.6.2 box. Here's the output of netstat on the sender side:
pop:/local/pop1/bouyer/current/src/sys/dev/pci>netstat -f inet | grep X11
tcp 0 100 pop.65491 armandeche.X11 ESTABLISHED
The connection is working fine in the other way (from armandeche to pop),
as I can see packets coming in on pop when I move the mouse in the hung window.
I never noticed this problem when pop was running 2.0_BETA (I was using
the same X11 software with 2.0_BETA).
You can find the output of 'tcpdump host armandeche and port 6000' running
on pop at http://www-asim.lip6.fr/~bouyer/tcp_bug.txt.gz I started it when the
window was already open and ^C'd it once the X11 window was wedged.
I also have the raw tcpdump -s 1500 output if anyone is interested.
It would be nice to have this fixed, because it's a regression.
Manuel Bouyer <email@example.com>
NetBSD: 26 ans d'experience feront toujours la difference