RELENG-7 tcp connectivity problems with certain clients

[Jake Rizzo]

The patch didn't work, but doing this:

sysctl net.inet.tcp.tso=0
sysctl net.inet.tcp.sack.enable=0

Allows the client to connect again. Since my ifconfig doesn't show that my
nic supports TSO then I presume it's the sack sysctl that's making it work.

The machine's a webserver serving almost exclusively dynamic content - with
sack disabled is it likely to affect performance much?

Thanks,
Jake
On Fri, Mar 7, 2008 at 11:52 PM, Mike Silbersack <silby at silby.com> wrote:

>
> On Fri, 7 Mar 2008, Jake Rizzo wrote:
>
> > Hi,
> >
> > I had two 6.3-STABLE boxen which have been happily running away for
> months
> > on end now without any problems. Last week I upgraded  (via buildworld)
> both
> > boxes to 7.0-STABLE.  Since then I've had reports of some clients being
> > unable to connect via tcp.  I've seen this happen first hand on an
> affected
> > remote machine. Traceroute & ping gets to the machine just fine
> > but connecting to an open tcp port on the machine just times out.  The
> > remote box was a XP machine and so I didn't have the luxury of tcp dump
> on
> > that end, however I did get a chance to run it at the freebsd end:
>
> Try this change; it was too late to be put in 7.0, but it will be merged
> to 7-stable in a few days:
>
>
> http://www.freebsd.org/cgi/cvsweb.cgi/src/sys/netinet/tcp_var.h.diff?r1=1.160;r2=1.161
>
> -Mike
>