[openstack-dev] [neutron] Enable arp_responder without l2pop
thomas.morin at orange.com
Wed Feb 22 12:49:22 UTC 2017
Tue Feb 21 2017 22:47:46 GMT-0500 (EST), Anil Venkata:
> Currently arp_resonder can enabled only if l2pop is enabled.
> Can we have arp_responder feature enabled without l2pop(i.e Remove the
> dependency between arp_responder and l2_pop)?
I agree that it would be useful.
networking-bgpvpn ovs/bagpipe driver is relying on arp_responder, and
hence currently draws this dependency on l2pop (not an issue I find, but
still an artefact rather than a design decision).
> Also setup arp_responder on OVS integration bridge(and not on br-tun)?
While relevant, I think this is not possible until br-int allows to
match the network a packet belongs to (the ovsdb port tags don't let you
do that until the packet leaves br-int with a NORMAL action).
Ajo has told me yesterday that the OVS firewall driver uses registers
precisely to do that. Making this generic (and not specific to the OVS
firewall driver) would be a prerequisite before you can add ARP
responder rules in br-int.
I think this question (of where to put the ARP responder rules) also
relates to https://review.openstack.org/#/c/320439/ .
More information about the OpenStack-dev