[Gllug] Weird networking problem with 2.6 kernel

[adam at thebowery.co.uk]

On Sat, Jun 19, 2004 at 11:58:39AM +0100, john at sinodun.org.uk wrote:
> 
> box has been physically disconnected from it!  The packets carry the new
> box's MAC address and yet are not emanating from the new box.  Anyone any
> idea what could be doing this?  Some other piece of hardware reflecting
> the packets back onto the LAN?

I had a similar problem with a network using Cisco Catalyst 5500 switches a few
years back, it all came down to a broken spanning tree implementation on the
switches, which if a corrupt arp packet got onto the network it would cycle
forever (or until the switches were powercycled). This problem was also made
worse by a Redhat 6.0 (ish) box that had a 2.2 series kernel that would
sometimes flip the ip addresses on its 2 network cards and it would then refuse 
to talk to anything until it flipped the ip addresses back. Of course this 
would cause arp to break as the box would send incorrect responses to any arp 
request and would result in a broadcast storm of arp requests (at one point the
flashy lights on the front of the Cisco switches said that they were passing 
date at about 50% of their total capacity which was 50Gbps, which was no mean 
feat as at the time we were only using 10 meg ethernet).  

So what I am trying to say is look out for spanning tree protocol or broken
vlans on your network equipment although quite why the machine only does this
with kernel 2.6 is unknown but perhaps your 2.6 kernel has some kind of
spanning tree or vlan tagging built in (or some other networking option) which 
the 2.4 series kernel doesn't and is causing the problem?

Adam
-- 
jabberid = quinophex at jabber.earth.li
AFFS || http://www.affs.org.uk/ || Not a filesystem
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug