Re: [dhcwg] WGLC: draft-ietf-dhc-dhcpv4-over-ipv6

[Simon Perreault <[e-mail]>]

Le 2012-10-18 11:31, Ted Lemon a écrit :
> On Oct 18, 2012, at 10:14 AM, Simon Perreault <[e-mail]> wrote:
>> I favour advancing, but there's a technical issue: the draft does not specify how unicast DHCP requests are handled. Specifically in the case of an LCRA, if nothing is done the LCRA will only get broadcast packets while unicast packets will be sent directly to the DHCP server over IPv4, bypassing the LCRA. Unless the LCRA is somehow in the unicast path and is inspecting all packets, which is something you want to avoid for scalability and good architecture reasons.
>
> Renewals would automatically go down the tunnel, so unless the tunnel doesn't provide a route to the DHCP server, I don't think this is a problem.   Am I missing something?

Yes you're missing something. ;)

The message would indeed go down the tunnel to the server. But remember 
that the CPE only has a partial IPv4 address on that tunnel interface. 
Messages sent from the server to the CPE's IPv4 address at port 68 will 
reach the CPE whose port range includes 68, not necessarily the CPE that 
sent the request. The result is that messages from client to server will 
work, but the reply will be lost.

Simon
-- 
DTN made easy, lean, and smart --> http://postellation.viagenie.ca
NAT64/DNS64 open-source        --> http://ecdysis.viagenie.ca
STUN/TURN server               --> http://numb.viagenie.ca