route: treat nhid changes as full route replacements

When updating a route targeting a nexthoup group with id changing its
nexthop to a single nexthop, we treat this as an update and instead
add that nexthop to the group instead of replacing the route in the
cache.

So e.g. with

nexthop id 10 fe80::1 dev eth1
nexthop id 11 fe80::2 dev eth2
nexthop id 12 fe80::3 dev eth3
nexthop id 20 grp 10/11

adding a route
dst 2001:db8:1:/64 nhid 20

libnl will store this as 2001:db8:1:/64 via fe80::1, fe80::2

if we now update the route with

dst 2001:db8:1:/64 nhid 12

libnl will see a new route with a single nexthop. Since the existing
route for dst 2001:db8:1:/64 is a multipath route, and the new route has
a single nexthop, libnl assumes this is a new nexthop for the route, and
merges it to 2001:db8:1:/64 via fe80::1, fe80::2, fe80::3 instead of
replacing it.

Fix this by checking the nhid of the routes, and only merge if they
match.

Fixes: 29b7137 ("route cache: Fix handling of ipv6 multipath routes")
Signed-off-by: Jonas Gorski <[email protected]>

#449

Author: KanjiMonster

lib/route/route_obj.c

@@ -528,6 +528,13 @@ static int route_update(struct nl_object *old_obj, struct nl_object *new_obj)
 	if (rtnl_route_get_nnexthops(new_route) != 1)
 		return -NLE_OPNOTSUPP;
 
+	/*
+	 * If the route's nexthop id has changed, the new route replaces the
+	 * old one
+	 */
+	if (rtnl_route_get_nhid(old_route) != rtnl_route_get_nhid(new_route))
+		return -NLE_OPNOTSUPP;
+
 	/*
 	 * Get the only nexthop entry from the new route. For
 	 * IPv6 we always get a route with a 0th NH