Re: idea for work in ovsdb netvirt L3: icmp responder

Tim Rozet <trozet@...>


Let me know if I'm missing some required info.


Tim Rozet
Red Hat SDN Team

----- Original Message -----
From: "Flavio Fernandes" <ffernand@...>
To: "Marcelo Amaral" <marcelo.amaral@...>
Cc: "Swaminathan Vasudevan (PNB Roseville)" <swaminathan.vasudevan@...>, "Tim Rozet" <trozet@...>, "Sam Hague" <shague@...>, "Andre Fredette" <afredett@...>, "ovsdb-dev" <ovsdb-dev@...>
Sent: Wednesday, August 12, 2015 4:26:20 AM
Subject: idea for work in ovsdb netvirt L3: icmp responder

[cc: ovsdb-dev]

Hi Marcelo,

It was really nice meting at the Summit! Sorry for not having a lot of extra time to mingle.

You mentioned you were looking for some work on ovsdb L3 and I thought of a very specific task that would have
a great impact in the L3 work; which may be within the time you have left to work on your internship.

While doing the demo [demo], Tim Rozet pointed out how tenant vms cannot get a response when pinging their default
gateway. As explained in the "Final thoughts” of my blog [1], this is the case because there is no ip stack to
handle any packets destined to the neutron router.

A solution to this would be to extend the arp responder table (20) to have rules that match on a specific icmp query
packets, move things around to turn it into a ping response and send the packet out of the in_port. the arp
handling code [2] can be a guide in how that gets accomplished. A way of prototyping this would be to create the
rule ‘manually’ [3] after doing the demo [demo] and coming up with what it would look like; similar to how Dave Tucker
and I started off when doing L3 [4].

@Tim: did you ever open the bug on this? If not, please do.
@Marcelo: would this be something you would be interested to work on?


— flavio

[3]: sh ovs-ofctl add-flow -OOpenFlow13 br-int “table=20, dl_type=0x800,nw_proto=1,icmp_code=..."

Join to automatically receive all group messages.