tcp connections dropped in 2.6.7

tcp connections dropped in 2.6.7

[David N. Welton]

[ Please CC replies to me.  Thanks! If there is a more specific list, I 
will go there. ]

Hello,

This one has really got me stumped, and I would like to debug it if 
possible (although my boss would really like the machine to 'just 
work').  I think it's in the kernel because it affects all userspace 
programs equally.

The problem:

When logged into the machine via ssh, every so often the connection goes 
  down (not frozen, it goes away completely) and attempts to log back in 
fail.  Seemingly, networking comes back up ok if, from the server (the 
one with the problem) I generate some outgoing traffic - even a single 
icmp packet.  Pings to the machine in question normally run around 0.12 
ms from my machine, but slow down to 2 or 3 ms during the problem.

On a suggestion from someone, I tried turning off the windowing bit 
which has apparently caused some problems lately, but that doesn't help 
(I'm not going through a router in any case).

There are no messages at all in logfiles.  tcpdump doesn't show anything 
that looks obvious (to me at least... could be me not seeing something).

Seeing as how outgoing traffic seems to kick the connection back into 
shape, as a crutch I tried leaving a ping running, but it went down just 
the same.

As a network card, I'm using a realtek.  Initially I observed the 
problem with a forcedeth card, which we substituted thinking that might 
be the problem.

I don't seem to be able to recreate the problem.  It "just happens" 
every now and then.

I'll put a dmesg up at http://dedasys.com/dmesg.txt

Thankyou for your time,
-- 
David N. Welton
davidw@eidetix.com

Re: tcp connections dropped in 2.6.7

[ViCToRy]

[-- Attachment #1: Type: text/plain, Size: 280 bytes --]

On Tue, Jul 13, 2004 at 06:45:42PM +0200, David N. Welton wrote:
> I'll put a dmesg up at http://dedasys.com/dmesg.txt

Why eth1 goes to promiscuous mode? it's showed on last line of dmesg.

-- 
Victor Sanahuja (ViCToRy) - Registered Linux User #198934 - http://piscue.com

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: tcp connections dropped in 2.6.7

[David N. Welton]

[ Please CC replies to me.  Thanks! ]

ViCToRy wrote:
> On Tue, Jul 13, 2004 at 06:45:42PM +0200, David N. Welton wrote:
> 
>>I'll put a dmesg up at http://dedasys.com/dmesg.txt
> 
> 
> Why eth1 goes to promiscuous mode? it's showed on last line of dmesg.

Because I tried tcpdump'ing ...

Server doesn't show packets arriving at all

Client shows things like this:

19:12:18.875723 IP coosbay.38062 > 192.168.0.240.ssh: S 
675525364:675525364(0) win 5840 <mss 1460,sackOK,timestamp 548888862 
0,nop,wscale 0>

19:12:18.878553 IP 192.168.0.240.ssh > coosbay.38062: R 0:0(0) ack 
675525365 win 0

Or

19:13:07.190788 IP coosbay.38063 > 192.168.0.240.www: S 
725915674:725915674(0) win 5840 <mss 1460,sackOK,timestamp 548937184 
0,nop,wscale 0>
19:13:07.196125 IP 192.168.0.240.www > coosbay.38063: R 0:0(0) ack 
725915675 win 0

yes, I tried it without the wscale... same thing.

I should also mention: 2.4.26 seems to have the same problem.

Thankyou,
-- 
David N. Welton
davidw@eidetix.com

Re: tcp connections dropped in 2.6.7

[David N. Welton]

[ Please CC replies to me.  Thanks! ]

Checked the cabling and switch and that wasn't it either.

This seems related to what I was seeing:

https://qa.mandrakesoft.com/show_bug.cgi?id=1546

I recompiled the realtek nic driver to show me debugging information,
and in fact, it wasn't even showing interrupts on pings during the 'outage'.

If anyone has ideas on how to track down the buggy software, I'm willing
to try them out, but I don't have the know-how to chase this down on my
own.  Maybe it's a known problem and already being worked on...

Thankyou,
-- 
David N. Welton
davidw@eidetix.com