SRv6 BGP (advertising SID) support

[ipspace]

Trying to untangle the spaghetti mess of #3263. There are two ways we can use BGP for SRv6:

1. To implement global IPv4-over-SRv6 (IPv4-as-a-service) or IPv6-over-SRv6 (BGP-free IPv6 core). That's part of the existing SRv6 module functionality; it's just that it was only implemented on SR/OS. When I find some willpower, I'll take the good parts of FRR: Add SRv6 bgp support #3263 (it contains way too much something-over-something-else chaff) to have this implemented on FRRouting.
2. To carry SRv6 SIDs.

As we have (1) solved (at least the attributes are there, not sure anyone ever tested it), this discussion focuses on (2).

I have to check, but it looks to me like carrying SRv6 SIDs over BGP IPv6 AF is a no-brainer (it's just a prefix), so I'm wondering if there's anything to do on the intermediate nodes. The edge routers obviously have to advertise their SID prefixes.

OTOH, if someone wants to implement multi-AS SRv6, they would have to establish EBGP sessions with SRv6 SID next hops (direct or multihop) with next-hop-unchanged. Plus, we shouldn't forget about IBGP-over-EBGP with local-as crap for truly masochistic people or those looking for the absolute job security 😜.

I'm not sure how well supported that is, and thus whether it makes sense to start thinking about implementing it in netlab.

Any thoughts @jbemmel @nmodena @DanPartelly (would love to hear from Ammiraglio Tofonoto, but he doesn't have a GitHub account 🥲)

[jbemmel]

Any thoughts @jbemmel @nmodena @DanPartelly (would love to hear from Ammiraglio Tofonoto, but he doesn't have a GitHub account 🥲)

My suggestion would be we come up with a minimal topology that we want to make working on FRR, then implement what's needed for that. #3263 wasn't intended to be integrated as-is, it just contains hints as to what needs to be implemented.

I would suggest single-AS IBGP to start, with your choice of IGP - before we do MRC over SRv6

[DanPartelly]

I think this is a good plan.