ipv6 next-hop link-local

[Gert Doering]

Hi,

On Tue, Feb 15, 2011 at 09:35:33PM +0100, Roger Wiklund wrote:
> My understanding is that all routing protocols that support IPv6 are
> using link-local as next-hop.

This is actually somewhat messed up for BGP.  RFC2545 specifies (if I
remember correctly) that BGP *has* to send an global next-hop, and
optionally can send a LL next-hop - while the receiver is free to 
ignore the LL next-hop.

So Cisco uses the LL next-hop, while Juniper uses the global next-hop...

I think both the RFC and the Cisco implementation are stupid, because
they create a new "weird failure" mode - we've seen this at DECIX, one
of the peers messed up their border ACLs, resulting in ND fail for
link-locals, but ND success for globals.  So BGP came up, but trying
to send packets there resulted in LL ND, which failed, causing black-
holing...  routing protocols shouldn't use "address type A" for 
control plane and "address type B" for next-hop setting.

(Now, for IPv6 IGPs, they use LLs for router-to-router communication
in the first place, so using LL next-hops is somewhat logical)

Gert Doering
        -- NetMaster
-- 
did you enable IPv6 on something today...?

SpaceNet AG                        Vorstand: Sebastian v. Bomhard
Joseph-Dollinger-Bogen 14          Aufsichtsratsvors.: A. Grundner-Culemann
D-80807 Muenchen                   HRB: 136055 (AG Muenchen)
Tel: +49 (89) 32356-444            USt-IdNr.: DE813185279