BGP4+ for IPv6

[Francis Dupont]

 In your previous mail you wrote:

   	let us assume the following diagram:
   
   	A --- B --- C		(---: ethernet segment or whatever)
   
   	A and C are peering with BGP4+.

=> I believe A and C are using iBGP between them and B is an IGP only
router (this scenario is well known, it is the first which crashes
until all iBGP/IGP bugs are fixed, this can take *time* :-).

   	since A and C do not share the same link, they will exchange
   	single address in next hop attribute, which is a global address.
   
=> yes

   	we need to put link-local address into nexthop field in IPv6 routing
   	table (like kernel routing table in BSD).  so:
   	- A will consult IGP routing table, understands that C is behind B,
   	  installs the following route:
   		BGP route from C/prefixlen -> nexthop is B's linklocal
   		(left leg)

=> I agree but the gateway field of a kernel route is not the same
than the next-hop attribute of BGP (perhaps this is why I can't see a
problem there?)

   	- B will consult IGP routing table, understands that A is behind B,
   	  installs the following route:
   		BGP route from A/prefixlen -> nexthop is B's linklocal
   		(right leg)
   
=> B doesn't know BGP, usually in this scenario BGP is redistributed
into the IGP (then B can choice between A and C) and A will see C
(and C will see A) only when B will announce the reachability between
them (the less obvious interaction between IGP and BGP).
 The redistribution uses link-local address (for instance when A will
say "route this destination via me" it uses its own link-local address).
If BGP and IGP conflict then you have two BGP speakers on the same link
and you should be in the case where link-local must be in next-hop
attributes (a common case with the self-next-hop hack).

Thanks

Francis.Dupont@enst-bretagne.fr