In order to forward IP traffic to a distant subnet, surely the router needs at the very least, to have a MAC address for the IP next hop and an interface to forward the packet on.
I can see where setting the egress interface with policy routing would work where the interface is a point to point interface, or a null interface or a loopback interface, there is clearly a defined next hop MAC address that needs no IP address associated for it to work. I can also see that where an IP address for a next hop is available in the routing table, the router can get a MAC using ARP, inverse ARP, Neighbor discovery etc. by looking for the next hop that it learned form the routing table.
But I can't for the life of me see how the "set interface" command could possibly work if the router does not have a way to resolve a next hop MAC address. I see that set interface is explicitely warned-against in the documentation for fast-switching unless there is a next hop in the route cache, but I see no mention of how it would/should work under CEF or process-switching.
So my question is; is a set interface clause reasonable for non-fast-switched-traffic without there being a next hop IP address in either the CEF cache or the main routing table, or there being a set ip next-hop clause to follow later on in the route-map ?
I have tried testing this on a router by implementing a local policy route-map for pings generated by the router and giving the traffic a kinda-valid next hop interface (one that goes nowhere and is not mentioned in the routing table, but is up). I see that the pings no longer reach their destination (as I had expected) and I see that the pings are policy routed in the debug output, but I don't see any ARP or ping packets leaving the interface I set for them.
Many thanks
Alan
