[ovs-dev] [PATCH v5] tunneling: Avoid recirculation on datapath by computing the recirculate actions at translate time.

[Zoltán Balogh]

Hi Joe,

> Thanks for taking a look. Andy and I have been throwing some thoughts
> around about this, but I'm not sure we came to a concrete solution yet
> either. My main thought is that I think that the 'flow' needs to be
> modified in the similar way that the 'push_tnl' would work in the
> datapath - updating the IP/UDP fields in a protocol-specific manner.
> Then the shared code between patch ports and this tunneling
> translation needs close attention to check everything is lined up
> correctly, restored after output translation, etc.

I agree, the flow and base_flow need to be updated according the tunnel
header before applying nested clone actions, then they should be restored. 

> Given that master is broken, it would be nice to restore it to a good
> state. The quickest way to do so would be to revert this patch on
> master. Then you could re-propose the patches to achieve this 'direct
> translation of tunneling' logic. That said, if you expect this to be
> fixed shortly then perhaps we could just wait for a fix. The main
> worry I have is that the translation code tends to be pretty
> elaborate/subtle so getting a solid fix in this area may take some
> time (and my regular tester box has already been complaining at me for
> a while now).
> 
> What do you think? I'm happy to give you a bit more time if you think
> that's the best approach.

I think it's ok if the patch is reverted, then a new patch with fix is going
to be sent to the list. Since I'm a co-author, I would like to ask Sugesh 
as well. Sugesh what do you think?

Best regards,
Zoltan