>"normal" consists of a number of more or less independent steps: > * Drop certain malformed or invalid frames. > * Check VLAN. > * Check for other inadmissible frames. > * Learn source MAC. > * Decide on output interface(s). > * Output to each output interface. Yes, that sounds good too.Does a flow rule as shown below make sense ?
table=20, actions=normal,no_learn_inport:4,no_learn_inport:5 (If the packet hits table 20, do normal actions but no SRC MAC learning on OF ports 4 and 5). >Can you explain why you'd want this to be apply per interface instead of >to an entire bond? My focus was on service chaining virtual appliances.Since bonding really applies to physical links, I really was'ntgiving it much thought. And you are right, no reason we can't apply this to an entire bond. Thanks,Farhad. On Tuesday, April 19, 2016 9:57 AM, Ben Pfaff <b...@ovn.org> wrote: On Tue, Apr 19, 2016 at 04:27:19PM +0000, Farhad Sunavala wrote: > >> Proposal: > >> We introduce the interface keyword "learn= no". > >> E.g. ovs-vsctl set interface foo learn=no > >> > >> This will instruct br-int NOT to do MAC learning on packets received on > >> interface foo. > > >This seems less flexible than adding some kind of option to "normal", or > >breaking "normal" into sub-actions. Why do it this way? > > Can you elaborate more on your thoughts about breaking "normal" into > sub-actions ? "normal" consists of a number of more or less independent steps: * Drop certain malformed or invalid frames. * Check VLAN. * Check for other inadmissible frames. * Learn source MAC. * Decide on output interface(s). * Output to each output interface. These, or some of these, could be made into separate actions that could be applied independently. > >> This cannot be applied to a bonded port with 2 or more interfaces. > >Why? > > Just to make it simple for now. These are virtual appliances and > unlikely tosupport bonding anyway. Issues such as what happens if the > bonded port has 3 interfaces and only oneof the interfaces has > learning disabled. Do we disable learning for the entire port? Can you explain why you'd want this to be apply per interface instead of to an entire bond?
_______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss