Jan Engelhardt wrote:

On Wednesday 2009-07-01 03:15, Herbert Xu wrote:
  

quoted

On Tue, Jun 30, 2009 at 10:16:35PM +0200, Jan Engelhardt wrote:
    

quoted

It makes sense absolutely. Consider:

* packet enters bridge
* NF_HOOK(PF_INET6, NF_INET_PRE_ROUTING, ...) is called by nr_netfilter.c
* (connection tracking entry is set up)
* let bridging decision be "local delivery"
      

No, my question is does it ever make sense to use conntrack as
part of bridge netfilter.  That is, do you ever want to test it
in your rules that are run as part of bridge netfilter.
    

There is the possibility that some users have -m conntrack in their
mangle table in the PREROUTING chain. However, I am pretty sure that
if there are such users, they do it because of the layer-3/4/5/7 part
and not care about bridge so much.

  

quoted

conntrack is inherently a security hole when used as part of
bridging, because it ignores the Ethernet header so two unrelated
connections can be tracked as one.
    

On secondary thought, one could also argue that because conntrack 
ignores the interface, two unrelated connections happening to be routed 
through the same machine(*) are tracked as one, too.
  

I had a similar problem while trying to implement virtual routers using 
different
routing tables and something like 'veth' to connect them.

My solution was to add a 'mark' field to the netdevice and allow 
user-space to set
the mark on the device.  This mark was included as part of the 
connection identifier.

The mark is set before the pkt hits the bridge code on ingress, so a pkt 
entering from
eth1 can get a different connection hash from a pkt entering eth2, even 
if all other
data in the packet is the same.

I'll dig it out of my monster patch if something like this is deemed 
useful upstream.

Thanks,
Ben

-- 
Ben Greear [off-list ref] 
Candela Technologies Inc  http://www.candelatech.com