Yet another IPv6 Prefix Delegation question

[Scott Shambarger]

On 2012-07-04 12:23, Vladimir-M. Obelic wrote:
> How am I supposed to do that with just 'dhclient -P eth0' here?
> I'm aware only one interface can be used for PD for now, but how do I
> specify which downstream interface gets the prefix?
> I am able to get the prefix, but what is supposed to happen next?
>

I struggled with the same issue, and basically came to the conclusion 
that dhclient will only allow you to process the prefix in your 
dhclient-script (or in my case, dhclient-exit-hooks, called from my 
dhclient-script on Fedora), and doesn't do anything with it internally.  
So I had my script-hook assign an address from the prefix to the LAN 
interface, and ping radvd to advertise it.

The real problem I ran into was when I wanted to get a public address 
for the WAN at the same time (rfc3633 says the PD shouldn't be used on 
the requesting interface).  You _can_ run dhclient -P -N to re-enable 
address assignment, but the mix does weird things in the state machine 
since INIT-REBOOT becomes a REBIND6 when a prefix is requested... 
leading to a address REBIND6 with old_ip == new_ip for the assigned 
address, which dhclient-script seems to think is a noop.

I ended up patching dhc6.c to look for a specially named pseudo 
interface (hard coded as "pd-handler"), and if it finds it, assigns any 
delegated prefix queries to it; since the pseudo interface is an 
independent state machine, that leaves the regular address query to work 
as normal (which is actually how the RFC says it should work).  Of 
course there are issues with needing to set the client-identifier, since 
dhclient doesn't follow rfc4361 and mix the IAID and DUID in creating 
the automatic ones... but that's another matter.

I submitted the patch to dhcp-bugs, but haven't heard any comments on 
it.  I also submitted it downstream at 
https://bugzilla.redhat.com/show_bug.cgi?id=836702 in case anyone wants 
to check it out.

(also had to fix a bug with lease6 not saving the "name" attribute, 
which was causing a restart to get quite confused...)

Thanks,
Scott