bgpd: fix updating ipv6 global nexthop #17152
Conversation
|
@leonshaw. Here is the fix for the unsent global nexthop that was mentionned here #17073 (comment) Please open an issue or publish a fix for the similar issue with link-local nexthop |
Link-local address is complicated as it affects the number of next-hops. I'd like to have a general discussion on this (#17128) before seeking a fix. |
bgpd/bgp_zebra.c
Outdated
| for (ALL_LIST_ELEMENTS(bgp->peer, node, nnode, peer)) { | ||
| v6_ll_in_nh_global = false; | ||
|
|
||
| if (IN6_IS_ADDR_LINKLOCAL(&peer->nexthop.v6_global)) { |
There was a problem hiding this comment.
If the address is removed and re-added, peer->nexthop.v6_global may have been zeroed in bgp_interface_address_delete(). Can we also handle this case?
There was a problem hiding this comment.
I have also fixed the bgp_interface_address_delete() for global
bgpd/bgp_zebra.c
Outdated
| if (connected->address->family != AF_INET6) | ||
| continue; | ||
| if (!IPV6_ADDR_SAME(&connected->address->u.prefix6, | ||
| &peer->nexthop.v6_global)) |
There was a problem hiding this comment.
Different interfaces may have identical link-local address. How about comparing interface name with peer->nexthop.ifp->name?
There was a problem hiding this comment.
fixed. I now compare nexthop.ifp and ifc->ifp
Move common checks outside of the loop. Signed-off-by: Louis Scalbert <[email protected]>
Reduce bgp_interface_address_add indentation Signed-off-by: Louis Scalbert <[email protected]>
Log new IPv6 global address in bgp_interface_address_add Signed-off-by: Louis Scalbert <[email protected]>
bgpd keeps on advertising IPv6 prefixes with a IPv6 link-local nexthop or a IPv4-mapped IPv6 after a valid IPv6 global appears. For example, at bgpd startup, the IPv6 global is announced by zebra after the link-local. Only the link-local is advertised. Clearing the BGP sessions make the global to to be announced. Update the nexthops with the global IPv6 when available. Do not override any previous global address that is still valid. Signed-off-by: Louis Scalbert <[email protected]>
Add bgp_nexthop_mp_ipv4_6 topotest to test to nexhop value with MP-BGP IPv4 and IPv6 on IPv4 peering. The test has route-reflector, route-server, iBGP and eBGP peers. Signed-off-by: Louis Scalbert <[email protected]>
Move common checks outside of the loop. Signed-off-by: Louis Scalbert <[email protected]>
When the IPv6 global is removed on an interface towards a peer, the IPv6 nexthop global is reset. It should be another global or a link-local. At removal, replace the global by the next global address or the link-local as last resort. Signed-off-by: Louis Scalbert <[email protected]>
Check that we can set back the ipv6 global nexthop. Signed-off-by: Louis Scalbert <[email protected]>
louis-6wind
force-pushed
the
fix-bgp-ipv6-global
branch
from
812faf5
to
3f485c7
Compare
louis-6wind
changed the title
|
Looks good to me. But seems a test case is failing... I'm not sure whether the next-hop should be IPv4-mapped or link-local in the absence of global address. |
bgpd keeps on advertising IPv6 prefixes with a IPv6 link-local nexthop
after a valid IPv6 global appears.
At bgpd startup, the IPv6 global is announced by zebra after the
link-local. Only the link-local is advertised. Clearing the BGP sessions
make the global to to be announced.
Update the nexthops with the global IPv6 when available.