Moin,
in the spirit of doing stupid things, I gave a 'somewhat' test of this
a shot yesterday (using traditional v4v6 NH instead of resolving
neighbors with NDP; But the principle remains the same).
- FRR 10.3 connected to a set of my favorite toaster & other household
appliance IXes, whose RS send me v4 routes with a v4 NH over a v6
session
- bgp parameters allow-martian-nexthops
- bgp neighbor $rs-ipv6 disable-connected-check
- /32 for all v4 for which I could find a matching v6 addr in the
peering LAN injected as v4v6 routes into my IGP
The v4v6 routes are basically the proxy for 'working neighbor discovery
with ARP/"NDP4" '; The allow-martian-nexthops / disable-connected-check
seems to be needed in this case as connected checks for the nexthop
fail as the ingesting router does not have v4 (at all), leading to
routes not becoming valid and hence not being exported in the iBGP.
However, doing that, I can actually forward traffic from a host with a
v4 address in another pop to an IP behind the IX:
My traceroute [v0.95]
v4less02.ring.nlnog.net (45.91.12.160) -> 45.159.140.1 (45.159.140.1)
2025-04-09T06:36:48+0000
Keys: Help Display mode Restart statistics Order of fields
quit
Packets Pings
Host Loss%
Snt Last Avg Best Wrst StDev
1. 192.0.0.8 (192.0.0.8) 0.0%
42 0.4 0.5 0.3 3.9 0.5
2. 192.0.0.8 (192.0.0.8) 0.0%
42 0.6 0.8 0.6 1.1 0.1
3. 192.0.0.8 (192.0.0.8) 0.0%
42 1.2 1.3 0.8 10.0 1.4
4. 192.0.0.8 (192.0.0.8) 0.0%
42 6.6 6.2 5.9 6.6 0.2
5. (waiting for reply)
6. eno6-514.ams2.as212635.net (45.159.140.10) 0.0%
42 14.1 14.5 12.6 34.7 4.0
7. ams1.as212635.net (45.159.140.1) 0.0%
41 13.2 14.3 12.6 38.1 3.9
So, technically, for an IX to support outbound (from the member)
traffic for an RFC8950 only member to dualstack-but-not RFC8950
members, it would suffice to inject those routes from the RS, ideally
with a specific community and only sending them to RFC8950 peers.
NDP4 would still help with the connected/martian issue here, i guess.
With best regards,
Tobias
_______________________________________________
GROW mailing list -- [email protected]
To unsubscribe send an email to [email protected]
