IPv6 EBGP

[Svante Bolander]

I have a EBGP-peering connecting two AS with both address-family ipv4 and ipv6 but the prefixes advertized from each AS does get dropped by the other end of the session by local policy:

"MP_REACH NEXT_HOP Martian: n/a 2239"

The BGP session is set up between the IPv4 addresses on the link.

What does this message in "sh ip bgp ipv6 unicast neighbors" mean? Why isn?t the IPv6 prefixes accepted?

[Harold Ritter]

You are exchanging both v4 and v6 over a v4 session, right? The issue is that the v6 updates are received with the ebgp peer v4 address as the NH, which causes the directly connected check to fail.

To solve this issue, you can either set a ebgp v6 session to exchange the v6 prefixes, which will set the BGP NH to the ebgp peer v6 address or you can set "ebgp-multihop 2" on the v4 session, which has the effect of disabling the directly connected check. Both approaches will work.

Hope this helps,

[jackyoung]

Hi Harold, could you please advise if we setup these two sessions, is there any negative impact to the system ? e.g. CPU loading, large size of routing table, memory ? And which is the usual practice and recommended solution ? Many thx.

[Harold Ritter]

The impact of running two sessions is obviously that you will consume more resources on the box. It might not make a difference if you only have a handful of peers.

One more thing to remember. If you are going to use the v4 session to carry v6 prefixes between iBGP peers, you will need to explicitly set the v6 NH using a route-map. Otherwise, the NH will be the v4 of the iBGP peer and it will not work.

As for best practices, in my experience, people are most of the time using a v6 session to carry v6 updates.

Hope this helps,

[jackyoung]

Thanks Harold. It clarified my concern.