Hi, > On 6. Dec 2021, at 16:48, Ondrej Zajicek <[email protected]> wrote: > On Mon, Dec 06, 2021 at 09:18:30AM +0100, Sebastian Hahn wrote: >> I've been trying to remove v4 transit nets from my setup. This seems to >> work well for BGP (both eBGP and iBGP) - I have routes for all external >> v4 addresses with the appropriate next hop set. For my internal routing, >> I'm using OSPF, but it seems OSPF does not allow v6 next hops for v4 >> routes. Checking RFC 5838 seems to imply that this is a general design >> choice for OSPF, but I am not sure if my reading is correct. Also, are >> there any practical drawbacks for this kind of setup that I may have >> missed when choosing it for BGP? > > You are right that RFC 5838 does not allow v6 next hops for v4 routes. > I am not sure why, it could be easily done. Perhaps they just did not > expect universal support for that in forwarding planes in time when > RFC 5838 was written and they did not want to complicate negotiation > of capabilities.
OK, good to know you read it the same way. > But note that RFC 5838 explicitly allows nexthops outside of iface > network prefix. So in principle, you do not need v4 transit nets, > just one v4 host/loopback address per router. I am not sure that BIRD > implementation would work with that setup but if not, that could be > easily fixable. Thanks, I ended up doing it and it works like a charm with bird 2.0.8. All the best Sebastian
