[openstack-dev] networking-sfc: unable to use SFC (ovs driver) with multiple networks

Banszel, MartinX martinx.banszel at intel.com
Thu Jun 16 11:48:39 UTC 2016


I'd need some help with using the SFC implementation in openstack.

I use liberty version of devstack + liberty branch of networking-sfc.

It's not clear to me if the SFC instance and it's networks should be separated
from the remaining virtual network topology or if it should be connected to it.

E.g. consider the following topology, where SFC and its networks net2 and net3
(one for ingress port, one for egress port) are connected to the tenants
networks. I know that all three instances can share one network but a use case I
am trying to implement requires that every instance has it's separated network
and there is a different network for ingress and egress port of the SF.

 +-------+                     +---------+                 +-------+
 | VMSRC |                     |  VMSFC  |                 | VMDST |
 +---+---+                     +--+---+--+                 +---+---+
     | p1 (             p2|   |p3                      |p4 (
     |                            |   |                        |
-----+--------+------- net1       |   |                  --+---+--------- net4
              |                   |   |                    |
              |          ---+-----+---)---- net2           |
              |          ---)--+------+---- net3           |
              |             |  |                           |
              |          +--+--+--+                        |
              +----------+ ROUTER +------------------------+

All networks are connected to a single router ROUTER. I created a flow
classifier that matches all traffic going from VMSRC to VMDST
(--logical-source-port p1 --source-ip-prefix=
--destination-ip-prefix=, port pair p2,p3, a port pair group
containing this port pair and a port chain containing this port pair group and
flow classifier.

If I try to ping from VMSRC the address, it is correctly steered through
the VMSFC (where just the ip_forwarding is set to 1) and forwarded back through
the p3 port to the ROUTER.  The router finds out that there are packets with
source address coming from port where is should not (the router expects
those packets from the net1 interface), they don't pass the reverse path filter
and the router drops them.

It works when I set the rp_filter off via sysctl command in the router namespace
on the controller. But I don't want to do this -- I expect the sfc to work
without such changes.

Is such topology supported? What should the topology look like?

I have noticed, that when I disconnect the net2 and net3 from the ROUTER, and
add new routers ROUTER2 and ROUTER3 to the net2 and net3 networks respectivelly
and don't connect them anyhow to the ROUTER nor the rest of the topology, the
OVS is able to send the traffic to the p2 port on the ingress side. However, on
the egress side the packet is routed to the ROUTER3 which drops it as it doesn't
have any route for it.

Thanks for any hints!

Best regards
Martin Banszel
Intel Research and Development Ireland Limited
Registered in Ireland
Registered Office: Collinstown Industrial Park, Leixlip, County Kildare
Registered Number: 308263

This e-mail and any attachments may contain confidential material for the sole
use of the intended recipient(s). Any review or distribution by others is
strictly prohibited. If you are not the intended recipient, please contact the
sender and delete all copies.

More information about the OpenStack-dev mailing list