Constraints on Automated Key Management for Routing Protocols Ross Callon IETF 71 March 2008, Philadelphia
AKM for Routing Protocols Link State protocol constraints Bootstrapping the routing protocol Operation over Broadcast Media Don’t take down the network Simplicity and Comprehensibility
Link State Protocol Constraints OSPF & IS-IS work because every router in an area has an identical view of the topology And runs identical route computation Authentication can be used to decide whether to bring up a link Or whether two neighbors exchange IGP traffic Authentication must not effect whether I believe the advertisement from a router across the area Different routers may get different results
Bootstrapping the Routing Protocol If something goes wrong with routing (or with security), there has to be a way to recover If the routing protocol depends upon AKM, then AKM can’t depend upon the routing protocol For OSPF & IS-IS, AKM **must** only operate between directly attached devices, using link layer You can’t depend on IP to an arbitrary address BGP can depend upon the IGP being up But can’t depend on a priori inter-domain routes For BGP, authentication probably only effects the preference of routes (in some sense)
Broadcast Media OSPF / IS-IS / RIP operate over broadcast media (eg, Ethernet) A router on a broadcast LAN uses link layer multicast to send one packet to multiple other routers on the same LAN AKM will need to operate over the LAN And provide a key that one router can use to send a single packet to multiple other routers
Don’t Break the Network The point is to keep the network up Authentication has to be more likely to keep things up, than to take the network down It has to be simple, understandable, resilient to mistakes Some configuration is allowed A router has to know which IGP to run Probably one pre-shared secret is okay also But: Keep it simple
Simplicity, Comprehensibility Many router experts are not security experts (and vice versa) This is not a complete mutual understanding Security is much more likely to be deployed if it is understood Including what it protects against, failure modes, and how to deal with problems.
Summary It has to work It (AKM for RPs) has to bootstrap It has to work over broadcast LANs It has to be simple, foolproof It has to solve a perceived problem Requirements may differ by protocol (OSPF, IS-IS, RSVP, LDP, UDP, TCP for BGP, TCP for not-BGP, …)