[dhcwg] Re: Relay Source address advice

Ben Patton <[email protected]> Mon, 31 Mar 2025 16:20:01 -0400
Newsgroups gmane.ietf.dhc
Message-ID <CAJBaQUqOEFEjC9SAMRq0f6Rk=Dfyfp4sXbfkUd2cxOEEahHGjA@mail.gmail.com>
Hi Michael,

Watching the DHC WG presentation, the lack of advice for DHCP relay agent
> source addresses.  My understanding is that the relay agent, despite having
> two interfaces, managed to pick the interface which was pointing at the
> clients, which might not actually be routable by the DHCPv6 server.
>
> Did I understand this correctly?
>

Yep, we did see during our test event that one of the relays picked a
client-facing GUA for Relay-forward messages on the Server and/or other
relay-side, which led to some fun routing problems for the server. It's
also worth pointing out that we set the relays up with unicast destinations
for forwarding rather than multicast, so we don't really have any data on
the source address selection behavior for that relay when we set the
destination to ff02::1:2 instead.


> I am still surprised that we do not do better with Relay Agents and PD
> assignments.   My understanding is that current relay agents do deep
> inspection of the DHCP responses to learn what prefix to route to which
> end-point.  Those installed routes, ideally, get installed into the IGP by
> the router in which the relay agent is installed.
>

While we were checking through the latest 8415bis, section 19 had language
about what to set for Peer Address and Link Address, but not the source of
the IP datagram itself as far as we could tell. Other relays that we tested
(besides the one mentioned above) picked their source addresses as we
expected, with a GUA that matched the configured destination address.

~Ben

_______________________________________________
dhcwg mailing list -- [email protected]
To unsubscribe send an email to [email protected]