IMHO, this is a valid point that the spec should clarify (one way or the other).
Slightly different question: is there a common place to list issues with the current spec? I know we were using trac for the 1.1 development, but now that it's released, is that still being used? - Rob . On Wed, Jun 8, 2011 at 4:22 PM, kk yap <[email protected]> wrote: > Hi, > > Just wanted to bring this up to archive the issue for consideration > later and as a cautionary note to those dealing with ARP using > OpenFlow. The specific issue pertains to clarity and consistency of > the OpenFlow specs. > > In OpenFlow (up till v1.1), we allow IP addresses in ARP to be match > using ofp_match's nw_src and nw_dst. This is very useful for working > with ARP in an actual network. However, the corresponding rewrite > action (i.e., set_nw_src and set_nw_dst) does not work with ARP, since > ARP is not an IP packet with the addresses located in the payload. > This creates an inconsistency from the view of the programmer: > > E.g., I wanted to match on a certain IP address and rewrite it to > another. Naively, I would send a flow_mod with > (if nw_dst == A, then rewrite_nw_dst to B). > The flow will match fine, but no rewrite action would be done. > However, the same semantics will work perfectly for all IP packets. > This inconsistency extends to the Ethernet addresses in the ARP > packets too, because the hardware addresses in the ARP payload will > not be updated. > > While "non-detrimental", this behavior might catch people by surprise, > so I hope those who work on the next iteration of the spec can look > into how the situation can be improved (if applicable). > > FYI. > > Regards > KK > _______________________________________________ > openflow-discuss mailing list > [email protected] > https://mailman.stanford.edu/mailman/listinfo/openflow-discuss > _______________________________________________ openflow-discuss mailing list [email protected] https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
