> On Aug 30, 2016, at 8:40 AM, Ben Pfaff <b...@ovn.org> wrote: > > On Mon, Aug 22, 2016 at 04:31:40PM -0700, Jarno Rajahalme wrote: >> Refactor handle_packet_out() to prepare for bundle support for packet >> outs in a later patch. >> >> Two new callbacks are introduced in ofproto-provider class: >> ->packet_xlate() and ->packet_execute(). ->packet_xlate() translates >> the packet using the flow and actions provided by the caller, but >> defers all OpenFlow-visible side-effects (stats, learn actions, actual >> packet output, etc.) to be explicitly executed with the >> ->packet_execute() call. >> >> Signed-off-by: Jarno Rajahalme <ja...@ovn.org> > > This adds a couple of comments of the form > > /* OUTPUT to a real port can not refer to versionable lookups (flow tables, > * groups), so we do not need to worry about the version to use here. */ > > but I'm not sure that I believe it. In particular, > compose_table_xlate(), which has this comment, outputs to OFPP_TABLE, > which isn't a real port at all but instead does a full flow table > traversal. >
Thanks for spotting this! > This line in packet_execute() needs an extra space: > execute.actions =opo->odp_actions.data; > OK. > I'm surprised to see ofproto.c including ofproto-dpif-xlate.h. That > seems like a layer violation. I think this was due to the plate cache only. The learn action entries in late cache use the struct ofproto_flow_mod, which the ofproto.c can operate on. Apart from bundle revert these could be handled from ofproto-dpif.c with calls to ofproto.c functions from there. To make the revert work I’d need to add a new ofproto-provider class member function and have it iterate through the late cache and then call back to ofproto.c to revert learned flows. Also, making the late cache a module of it’s own would remove the inclusion of ofproto-dpif-xlate.h. Would this alleviate your concern with layer violation, or should I strive to keep the xlate cache opaque to ofproto.c? Jarno _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
