Re: RFC3484 destination address selection rule 2 is buggy

Le Friday 14 March 2008 00:27:26 Pekka Savola, vous avez écrit :
> This issue was first reported about 5 years ago by Alain Durand et al and
> yet there is no fix yet (and no mention in the default address selection
> problem statement), see section 2 of:
> http://tools.ietf.org/html/draft-ietf-v6ops-v6onbydefault-03
>
> The main problem is destination address selection rule 2 which requires
> that source and destination address scopes must match (which in the case of
> v4 private and global addresses is not a very useful comparison given the
> prevalence of NAT).

Indeed. And this was (inconclusively) discussed at the mike during the last 
v6ops meeting. I had also asked about this a few months ago. Nobody seemed to 
care (winter vacation?):
http://www.ops.ietf.org/lists/v6ops/v6ops.2007/msg01150.html

> Maybe we need a more systematic approach to deal with RFC3484 issues (as
> in, a numbered list of all the problems noted) instead of doing a nice new
> features to have PPT slideshow every IETF meeting.

I think we need to "simplify" RFC3484 section 3.2 through removing the IPv4 
site-local scope there: we'd be left with only global scope (public addresses 
+ RFC1918) and link-local scope (169.254.0.0/16).

I suspect some implementors (at least Windows) already ignore §3.2 for the 
sake of reliability. I know Linux does implement ¼3.2 to the letter of the 
RFC unfortunately. And I have seen people _remove_ AAAA from their server's 
DNS records because of this issue, combined with deficient 6to4 relays.

Another problem involves incomplete implementation of RFC3484: some stacks 
apply RFC3484 for IPv6, in connect() and sendto() socket APIs, but fail to do 
RFC3484 in getaddrinfo(), and simply assume IPv6 is first, and IPv4 second. I 
suspect this applies to OSX and BSDs, and also "embedded" C run-times.

These two issues are different, though they effectively have the same result: 
favor IPv6 over IPv4, when IPv4 should be favored over IPv6 (IMHO).

-- 
Rémi Denis-Courmont
--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------