On Wed, Jul 25, 2012 at 06:42:19PM +1200, Joe Stringer wrote: > On 25 July 2012 18:26, Ben Pfaff <b...@nicira.com> wrote: > > On Wed, Jul 25, 2012 at 06:21:22PM +1200, Joe Stringer wrote: > >> Regarding the translation of OF1.1+ instructions to vendor extensions > >> on OF1.0, do we want to move towards using the above syntax for > >> packet-out as well? (in a new patch, presumably) > > > > I'm not sure that there's any need to worry about instructions for > > packet-out, since packet-out takes only actions, not instructions. Do > > you see another problem, or maybe you mean something else? > > Ah, in particular with the write_metadata instruction (which seems > more like an action, the more I look at it..) > > I was using packet-out to test that the write_metadata vendor > extension work was working properly. To guarantee the aforementioned > conditions, I need to either handle the metadata action specifically > in str_to_ofpacts, or have a check at the end of the function to > iterate across the ofpacts[] and verify it's in the correct position, > etc. > > When it comes to parsing a flow_mod with this implementation, we can > now have a (vendor action) write_metadata in the [apply,write]_actions > instructions, as well as the write_metadata instruction - So a check > needs to be added to str_to_inst_ofpacts() to iterate across the > ofpacts[] again and make sure we have just one ofpact_metadata. > > This is quite doable, I just wasn't sure which approach is preferred.
My guess is that a post-parse check is easiest. I think that's OK for now. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
