dhcpcd-discuss

Hi Robert

On 22/06/2014 0:50, Robert White wrote:
> So the expanded prefix delegation works great for port assignment, but
> then no routing can take place (at least on comcast's networks)
> because there is no globally routable address for the router.

That should not matter becase globally routable addresses can be found 
on the delegated interfaces.
Consider this: you are delegated dead:beef::/60
A reject route is installed for delegated dead:beef::/60 to avoid 
sending unspecified sub prefixes upstream.
Prefix dead:beef:1::/64 is assigned to int1 with an address of 
dead:beef:1::1/64

At this point I, sat on my own network, should be able to ping 
dead:beef:1::/64 directly.
I shouldn't need to go via your routers public facing IP.

> (and using extX for external and intX for internal interfaces and ZZ
> for numbers that don't matter).
>
> This will allocate int1 and int2 some nice addresses, and requests a
> la ping6 will go out ext2 (which is getting a nice default route via
> router advertisement)... but no packets are ever returned.
>
> interface ext0
> ia_pd ZZ/::/60 int1/ZZ/64 int2/ZZ/64
>
> restarting dhcpcd with the following to get a globally routable address 
> for ext0
>
> interface ext0
> ia_na
>
> And then restarting again with the original ia_pd config is curative.
>
> Both the router itself _and_ the clients on the allocated /64(s) are
> affected identically.
>
> NOTE: I figured this out because other web pages referring to other
> configurations (e.g. the one for wide-dhcp and one for some cisco
> client or other) showed examples containing _both_ ia_na and ia_pd
> stanzas for the public interface.

Well, it all sounds bogus to me.
From the one of your client machines, can you provide a traceroute6 to a 
public IPv6 address such as roy.marples.name?

> Being able to request any combination of ia_XX entries, or at least
> ia_na and ia_pd seem necessary.

And not that easy to implement correctly either.
There is a work in progress document to make it easier, but it's not a 
RFC yet.
https://datatracker.ietf.org/doc/draft-ietf-dhc-dhcpv6-stateful-issues/?include_text=1

Once that becomes a RFC, this would be much easier to do in dhcpcd.
But if you read the emails through underneath it, it should not be 
needed because you get workable routable addresses via PD by itself.
So is comcast doing something really odd?

Roy