stalled connections on recent -current

[Robert Watson]

On Wed, 23 Jul 2008, Mike Makonnen wrote:

> I'm wondering if anyone has seen stalled connections to cvsup servers 
> recently. I've tried updating my mirror from several different cvsup 
> servers, but the connection always stalls sometime in the middle of the 
> update process and eventually fails with the following message in my cvsup 
> log file: TreeList failed: Network write failure: Connection timed out
>
> This is what netstat -f inet looks like during the stall:
>
> mtm at rogue ~% netstat -f inet
> Active Internet connections
> Proto Recv-Q Send-Q  Local Address          Foreign Address       (state)
> tcp4       0  32544 192.168.0.98.32829     sanmateo.ecn.pur.cvsup ESTABLISHED
> tcp4       0      0 192.168.0.98.22775     xxx.xxx.xxx.xxx .ssh ESTABLISHED
>
> This started happening after I upgraded the system (which was a couple of 
> months old) to one from around July 16th. Unfortunately I don't have the old 
> kernel to confirm that this is indeed a kernel bug.
>
> Any suggestions as to how to debug this?

I think I'd always start with tcpdump and see if it's at all obvious what is 
going on.  In particular, it would be good to see if any ACKs are coming from 
sanmateo.ecn.pur.whatever:cvsup, and if so, what their window advertisements 
look like.  It also doesn't hurt to confirm the state of local 
processes/threads attached to the socket, perhaps using procstat -k.

Robert N M Watson
Computer Laboratory
University of Cambridge