On 1/28/15 12:09 PM, Ben Pfaff wrote:
My big comment on this series is that I don't understand how this is
going to interact with OpenFlow, especially in a backward-compatibility
sense.
This is certainly a concern. The intention was not to break any legacy
behavior, but only to add the push and pop capability as specified in OF
1.1 and above.
Here's an example: in OpenFlow 1.0, one can drop all double-tagged VLANs
by matching on packets that have a Ethertype of 0x88aa or 0x8100, since
the second VLAN header shows up as the Ethertype. After this patch
series, I guess that that will no longer be true.
Have you thought about these issues?
I hadn't thought of this case specifically. I thought that legacy 1.0
controllers would detecting either single or double Vlans by matching on
DL_TYPE only. I also thought that the existing VLAN tests would be
pretty good coverage for regression testing.
The intention of this patch is not to break any legacy behavior.
Instead, only to distinguish among single tagged and double tagged Vlans
in user space and put enough information in the flow key so the matches,
pushes and pops could be done quickly in the fast path as specified by
OF 1.1+.
Do you think that additional tests should be added to V7 for the case
you cited and to make sure that similar cases of legacy use aren't
broken? Would adding tests be sufficient to verify legacy behavior that
may not now be covered by the existing VLAN tests?
Thank you very much for the thoughtful comments.
--Tom
--
Thomas F. Herbert
Network Implementation Engineer
Entry Point LLC
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
