[ovs-dev] [RFC flow tunnels 5/8] tunnel: Userspace implementation of tunnel manipulation.

Ethan Jackson ethan at nicira.com
Sun Jan 27 00:50:20 UTC 2013

> tnl_match is only used to match on incoming packets. Therefore it would be
> much clearer if "ip_src" actually meant the "ip source address" and "ip_dst"
> "ip destination address" w.r.t the received packet. See below.

Personally I think it's clearer to keep it consistent with
netdev_tunnel_config, and struct flow_tnl.

> How about the struct ofport_dpif pointing to this? That way we could get rid of
> the ofport_map (and find_ofport()), as the calling sites in ofproto_dpif.c would
> have the ofport_dpif->tnl_port available?

Not a bad idea, this turned out to clean up the code so I went ahead
and folded it in.

> Is this really necessary? I imagine we could want to match on also the other
> tunnel fields on the flow-basis in the future. Also, this seems like a change
> to the current tunneling behavior, so it should not be a problem to let the
> tunnel data be?

I'm not sure it's strictly necessary, but it feels defensive to me.
If we add matches for the outer tunnel header at some point, we can
revisit it.


More information about the dev mailing list