Linux eth stops working ?!

[Pedro d'Alte]

I have a Linux Server connected to a SpeedStream 5660 Router. This Linux Server is a proxy/firewall server (squid+iptables).
The problem is that after some hours or a day of work, the Linux server stops accessing web services like HTTP, SMTP or FTP (but DNS queries and ICMP work!). Only rebooting the server everything starts working again (tried stopping iptables and squid, restarting the network, changing the mac address,...). I already changed to another server, different cables, different OS version, and the same happens on that different server. There are some Windows machines also directly connected to the ADSL Router which never stop working. I already installed several scenarios like this one on other places without a single problem...
Does anyone have a clue of what's happening?
Oh, I have also looked on the system logs, but no problem is reported.

[Steven E. Protter]

When this happens, collect some data.

service squid status
service iptables status

It may be a software situation of some kind, you'll want to proceed after checking this information.

Also when it happens, check this:

telnet www.yahoo.com 80

See if you talk to yahoo or not.


Then.

traceroute yahoo.com

See where it stops.

It could be the router is losing its RIP settings. To help, data is needed.

SEP

[Pedro d'Alte]

The squid and iptables status is ok when it occurs. The logfiles also do not report nothing wrong except a message on cache.log saying it is closing the client connection due to lifetime timeout.
I already tryed telnet some_site on port 80 and also to a mailserver on port 25, and the connection fails (timeout), although nslookups and pings work fine. At the same time if I go to a Windows machine using the same router as the default gateway, everything works fine.

Pedro.

[Steven E. Protter]

Additional steps:

vi /etc/services

look for conflicts on ports 21,25 and 80

service sendmail status

Check the mail log, maybe there is an event there.

When it happens, ifconfig, check for anomolies.

I still think traceroute will be extemely useful.

When it happens:

sendmail -v -d8 -d38 someone@yahoo.com

Type some text

.


You'll get diagnostic trace. You should see dns resolution and then get an error.

Try replacing or rebooting the router next.

If answers are helpful, please consider assigning points.

SEP

[Pedro d'Alte]

Thnx for the reply, again.
I uninstalled sendmail a log time ago. I tested smtp by telneting to port 25 of a mail server, without success.
Rebooting the router was one of the first tests I've done, but the only thing that makes it all work again is rebooting the server.
I havent got any port conflict.

[Olivier Drouin]

I had similar problem when running squid at full debug...
I had to reset to a normal debug level, rm -rf the cache and recreate it with squid -z.

just my 0.02 cents.

[Steven E. Protter]

I don't believe you have a port conflict.

If you remove sendmail and then telnet to port 25 the result you got is expected. The connection will be refused if the sendmail daemon is not running.

How is work going on our other suggestions? Any progress?

SEP

[Balaji N]

tell me one thing. what distro are u running. By any chance do u have some kind of apm configured.

read on a mailing list out here abt a RHL box installed as a workstation but running some services and eth0 used to get stopped. fixed by disabling APM. or some power management service. unable to recall / locate the exact message
-balaji

[Stuart Browne]

When you say you tried it on a 'another server', was this an identically spec'd server, or a completely different server?

[Stuart Browne]

When you say you tried it on a 'another server', was this an identically spec'd server, or a completely different server?

[Pedro d'Alte]

Hi. Thanks for the replies. I'll try to answer you all.

I believe squid has nothing to do with it. If I stop the squid daemon I still haven't got any access to web services on the Linux box itself.

Concerning the sendmail posting, when I said I tried telneting port 25, I didn't mean I was doing a telnet on port 25 of my server. So, forget sendmail. What I cannot do is telnet mail servers on the Internet on port 25.

The suggestion of trying to telnet port 80 of a webserver was also tried, and it also doesn't work. The only services that seem to work when it happens is DNS queries and ICMP.

I'm running RedHat 9.0.

The other server where I had the same problem was a completely different machine. The hardware was all different, including the ethernet interface cards, and it was running RedHat 8.0.

Yes, apm was running. I just stopped it to see if it has any effect.

[Alexander Chuzhoy]

Well.. Try to change the NIC also?

[Pedro d'Alte]

As I said before, I had the same problem with another server with different NICs.

[Paulo A G Fessel]

Have you tried to flash the SpeedStream? What's the version of the firmware? In the web interface of my SpeedStream, it shows

2.3.0(7) Mar 4 2002 13:44:50s

Also, have you looked at ARP tables both in your servers and routers? In the SpeedStream this is shown with "show arp".

HTH
Paulo Fessel

[Paulo A G Fessel]

Oh, and BTW I'm running SpeedStream connected to a Mandrake Linux PPC machine which happens to be my firewall. I have not had no problems to date.

[]'s
Paulo Fessel

[Pedro d'Alte]

Yes, I checked the router's firmware version, which is the same as yours 2.3.0(7), and I've confirmed that this is the last version available.

I've looked on the arp entries on both machine and router, and while I was having the problem I deleted those entries... nothing happened, the problem remained. Once I even changed the MAC address of the server's external network card to see if it had any effect, but it didn't.

[Paulo A G Fessel]

Well, this is my last straw...

Duplex configuration mismatch?

I've seen things like these when one side of the connection duplexing configuration is set to "autodetect" and the other has fixed duplex configuration, particularly when the fixed duplex configuration side is set to "full".

In the SpeedStream:

Command-> show ethcfg

--- Ethernet Configuration
Ethernet Mode - Half

To change:

Command-> set ethcfg ?
----------------------------------------------------------------------------
usage: set ethcfg
----------------------------------------------------------------------------
Allows configuration of the data transmission direction of the Ethernet interface. The interface can be set to 'half' (one-way transmission at a time) or 'full' duplex(transmission in both directions at once). The default setting is 'half'.

In Linux, you shoud set this per the configuration instructions of your network card. More details are available at http://www.scyld.com/network_index.html.

HTH
Paulo Fessel