On Thu, Jan 19, 2012 at 02:21:07PM +0200, Heinonen, Johanna (NSN - FI/Espoo) wrote: > It seems that the problem is related to LLDP messages that come from > "bad in-port" i.e. a port that is not connected to ovs.
Your controller is trying to send a packet with an in_port of OFPP_CONTROLLER. This is incorrect: the correct in_port is OFPP_NONE. This is why OVS is rejecting it It also looks like your controller is sending Nicira extensions messages with type 8 (NXT_MGMT). This has been obsolete for years, so I don't know why you are trying to use it. > My setup is the > following: > > NOX linux1 > | | > -----------------------192.168.4.0/24 > | > linux3----OVS--------linux2 > > OVS has ports 1(eth0, towards linux3), 2(eth1, towards linux2), 3 and 4 > (gre0 and gre1 towards linux1). In addition to this the machine where > OVS is running has eth2 that is not connected as a port to OVS but has > an IP address (.101) towards the 192.168.4.0/24 subnet. (Gre tunnels do > not work if this port is connected to OVS). Are there any possibilities > to have the gre tunnels and connection to NOX via this same eth2? The > limitation is that NOX and Linux1 have only one interface, and I'd like > to avoid adding a router in between also if possible. You'd need to make sure that the flow traffic recognizes the tunnel traffic and treats it as a special case. I have not implemented rules to do this. Naively, it sounds to me like the problem is similar to the in-band control problem. Maybe it is simpler because GRE packets are so distinct from other traffic. Thanks, Ben. _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
