>Number:         30621
>Category:       kern
>Synopsis:       TCP hang on netbsd-2
>Confidential:   no
>Severity:       critical
>Priority:       high
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Tue Jun 28 10:37:00 +0000 2005
>Originator:     Manuel.Bouyer@lip6.fr
>Release:        NetBSD 2.0_STABLE, cvs from Apr 10
>Organization:
LIP6/ASIM http://www-asim.lip6.fr/
>Environment:
System: NetBSD pop.lip6.fr 2.0_STABLE NetBSD 2.0_STABLE (GENERIC.MP) #0: Sun Apr 10 15:46:42 CEST 2005 root@pop.lip6.fr:/local/pop1/bouyer/tmp/i386/obj/local/pop1/bouyer/netbsd-2-0-clean/src/sys/arch/i386/compile/GENERIC.MP i386
Architecture: i386
Machine: i386
>Description:
	TCP connections from a 2.0_RC2 or a 1.6.2 box to 2.0_STABLE sometimes
	hang, with data pending in the send-queue on the 2.0_STABLE box.
	Data in the other way (*to* the 2.0_STABLE system) are still
	transmitted, as shown by tcpdump. I've seen this with ssh connections
	to the 2.0_STABLE system with lots of stdout traffic, and X11
	connections from a sofware running on the 2.0_STABLE system
	displaying on the 1.6.2 box. I've never seen this when this box was
	running a 2.0_BETA kernel.
	Note that the 2.0_STABLE box is much faster than the 2.0_RC2 and 1.6.2
	ones (dual-CPU PIII 1Ghz vs Ultra/1 143Mhz and alpha ev4 233Mhz).

	Here's an example netstat output on the 2.0_STABLE system with 
	a connection in this state:
tcp        0    100  pop.65491              armandeche.X11         ESTABLISHED
	The output of 'tcpdump host armandeche and port 6000' running
	on the 2.0_STABLE system is available at
	http://www-asim.lip6.fr/~bouyer/tcp_bug.txt.gz. It was started when
	the X11 window was already open, and interrupted it once the
	X11 window was wedged. 

>How-To-Repeat:
	ssh from a slow system to a fast one, and start something that will
	print a lot of things on stdout, such as a build.sh -j4
	Start a X11 software with large graphical windows and lots of redraw
	(such as pkgsrc/cad/eagle) on a fast box and display on a slow one.
>Fix:
	unknown.