Re: Understanding IPv6-PD over PPPoE

[Bjørn Mork <bjorn mork no>]

Steve Hill <steve opendium com> writes:
> On 15/06/2021 18:19, Bjørn Mork wrote:
>
> > Yes, that is buggy.  I wonder... I did hit a similar issue many many
> > years ago while testing IPv6 over PPPoE (which we didn't end up doing in
> > the end).
>
> I'm not entirely sure how the default route is assigned - on ethernet
> it would be assigned on receipt of the router advertisement, but is
> that still the case with PPP, or is the default route negotiated by
> IPv6CP?

The only defined options are Interface-Identifier and
IPv6-Compression-Protocol, ref
https://www.iana.org/assignments/ppp-numbers/ppp-numbers.xhtml#ppp-numbers-28

The interface-id is required to form a valid link-local address since
there is no EUI-48 or EUI-64

Address and routing is otherwise identical to any other Ipv6 interface.
I.e. SLAAC, DHCPv6, and/or static.

> > Maybe you are hitting something similar with your ISP?  The typical
> > symptom would be that DHCPv6 worked once again after the lease expires.
> > If this is the issue, then you can probably assign a static interface-id
> > on your end to work around it.
>
> I've got 2 internet connections on the same ISP.
>
> One of them was set up years ago and the ISP was trialling IPv6 at the
> time, so I requested an IPv6 prefix.  This one is working ok (the
> DHCPv6 advertise contains my prefix).  I have no idea whether the ISP
> ever rolled out IPv6 to everyone by default, but this works for me
> since I enrolled on the trial years ago.
>
> The other connection is new.  The ISP sends router advertisements and
> we get a default route, but the DHCPv6 advertise doesn't contain a
> prefix or IP.  I think it's a reasonable assumption that the ISP never
> enabled IPv6 for everyone by default and the result is that when you
> send a DHCPv6 solicit you don't get a prefix on this connection.

Yes, that sounds likely. They shouldn't be sending RAs either then. But
clients blindly trying IPv6 over PPP is probably not a common problem.
I believe most clients disable IPV6CP by default. 

> > I think it would be too risky to try to be smart here.
> > How can the system know whthere there will be an address available
> > later?  Should it cache all the routing?  What if the system has an
> > address which isn't routable on the interface with the default route?
> > Etc.
> > Installing the routes you receive is the only sane thing to do,
> > IMHO.
>
> But it results in a configuration that causes real-world brokenness
> (long timeouts waiting for connections that can never happen).

Whether you want to trust an RA or not is a local policy
decision. Forcing a new default policy is sure to break existing
configurations.

If the default route from a specific router or interface doesn't work
for you, then ignore it.

> The correct thing would probably be for the kernel to treat all global
> addresses as unroutable if the machine only has link-local addresses, 
> irrespective of the existence of a default route.  Is there a reason
> why you would ever want to generate traffic with a link-local source
> address and a global destination address?

You can reach any directly connected system by their global address
using a link-local source.  There is no way to know that a global
address is more than one hop away without trying.


Bjørn