[ovs-discuss] Openvswitch: Rules matching ports behaving strangely - Rules translation problem?

[Justin Pettit]

Thanks for the report, Andreas.  This does seem odd.  I'll try to  
reproduce it locally.

I can answer this part, though:

> root at loadgen134:~# ovs-dpctl dump-flows br_out
> port0001:vlan65535 mac00:1b:21:10:8c:7e->00:16:3e:76:4f:93 type0800
> proto1 ip192.168.10.1->192.168.10.2 port0->0, packets:125, bytes: 
> 12250,
> used:0.708s, actions:0,2,5,4,3
...
> port0002:vlan65535 mac00:16:3e:76:4f:93->00:1b:21:10:8c:7e type0800
> proto1 ip192.168.10.2->192.168.10.1 port8->0, packets:125, bytes: 
> 12250,
> used:0.708s, actions:0,1,5,4,3
>
> I am confused by the port0->0 and port8->0 in these listings -- my dp
> does not have a port 8?!

The vswitch process caches out "exploded" wildcard entries in the  
datapath for performance.  While these flows should have been limited  
to ports 1, 2, and 4, the flows themselves are not unexpected, since  
you allow traffic between these ports.  You can see that the IP  
protocol type is 1, which is ICMP.  These are just flow related to  
ping traffic, ("port" 8 is an echo request and "port" 0 is an echo  
reply).  This overloading of the transport ports for ICMP is part of  
the OpenFlow specification.  Does this all make sense?

--Justin