there is a bug in the tcp/ip stack in linux 2.2.*. I tried to track it
down,
but i never get a chance to analyze the bug, because all servers, that
show it are production servers (so i have to reboot it ).
My new policy is now to reboot linux every week. Sigh, where is
the stable things that 2.2.* promises ?
* Environment (different network cards tulip(100MBIT),
eexpress(100MBIT), ne2000)
- a webserver with virtual domains (redhat 5.2)
- a server with netatalk (without virtual domains) redhat5.2
- "normal" linux server (redhat 5.2)
Bug spotted in 2.1.132, 2.2.3, 2.2.6, 2.2.9
No messages, no oops, server runs fine when the network stack goes
crazy
(and you can do a shutdown -r now )
Symptoms: Difficult to describe (Hanging/No Network
Connections/ Connection dropped)
It looks like the server becomes selective which connection
(tcp/udp) it accepts
computer A can connect to server
computer B cannot (same eth0 device)
Example 1: Some network connection from outside to the
webserver worked, but the internal network could not
connect to computer before reboot
* all devices where up
* no firewall
(It was very strange, i tried different providers and some worked
and some others did not (traceroute showed that i could reach
the network of our provider (which reported me our unwilling
server - reboot - and everything worked )))
Example 2: Hanging Network connection
You can make a
> telnet buggyhost 80
GET /
and it returns the webpage but never does a
"Close Connection" (normal case)
so telnet hangs forever( ok 10 minutes before i killed it)
Example 3: Appletalk clients loose connection to server,
this actually shows some messages from appletalk, but
says "connection to client lost" . Clients work ok for
2 Years now.
Method to trigger the bug:
* uptime > 20 Days
* connect to the internet
* and like yesterday setup a new virtual device
ifconfig eth0:22 ... up
and after 2 hours the server gets crazy
Regards
Dietmar Kling
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/