probably we can assign distinct pseudo odp port numbers for each userland-only ports like patch ports and make datapath accept them eg. for match. in that way, we can associate a packet in_port corresponding to the port for the recirculating bridge, rather than of the receiving bridge.
YAMAMOTO Takashi > It seems recirculation action should be scoped within the same OF-bridge. > > On Tue, Dec 16, 2014 at 4:43 PM, Alex Wang <al...@nicira.com> wrote: >> Hey Yamamoto, >> >> Justin and Ethan just raised the same question you asked, >> >> Yeah, there is an issue. If say, the vlan is set before recirc action, >> the recirc'ed flow will have a vlan id which could lead to totally >> different handling result, >> >> We will discuss further and address it, >> >> Thanks, >> Alex Wang, >> >> On Tue, Dec 16, 2014 at 4:16 PM, YAMAMOTO Takashi <yamam...@valinux.co.jp> >> wrote: >>> >>> hi, >>> >>> > Hey Yamamoto, >>> > >>> > >>> >> what will happen if the receiving bridge also has a bond port? >>> >> >>> >> YAMAMOTO Takashi >>> >> >>> > >>> > >>> > For each 'balance-tcp' bond, it will have a unique 'recirc_id'. >>> > >>> > If the receiving bridge has a 'balance-tcp' bond with recirc_id=ABCD, >>> > the bridge's internal table (Table 254) will contain flows matching on >>> > this one 'recirc_id' and different 'dp_hash' values. >>> > >>> > If the flow miss does not have 'recirc_id' (i.e. recirc_id == 0), lookup >>> > will >>> > start from Table 0. >>> > >>> > If the flow miss has 'recirc_id' == ABCD, it will match one of the rules >>> in >>> > internal table and get output to the corresponding interface. >>> > >>> > If the flow miss has 'recirc_id' != ABCD, it will match the the rule in >>> bold >>> > below, then lookup from Table 0, continue on the pipeline until reaching >>> > the bridge with the 'recirc_id'. >>> > >>> > priority=2,recirc_id=0,actions=drop >>> > *priority=1,actions=resubmit(,0)* >>> >>> thank you for explanation. >>> >>> i have another question: what ensures that the recirculated packet >>> reaches to the same bridge as before? esp. when the packet has >>> been modified. >>> >>> YAMAMOTO Takashi >>> >>> > >>> > Thanks, >>> > Alex Wang, >>> > _______________________________________________ >>> > dev mailing list >>> > dev@openvswitch.org >>> > http://openvswitch.org/mailman/listinfo/dev >>> >> _______________________________________________ >> dev mailing list >> dev@openvswitch.org >> http://openvswitch.org/mailman/listinfo/dev > _______________________________________________ > dev mailing list > dev@openvswitch.org > http://openvswitch.org/mailman/listinfo/dev _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
