Last Call Review of draft-ietf-pcp-anycast-06
review-ietf-pcp-anycast-06-secdir-lc-nir-2015-06-10-00

Hi.

I have reviewed this document as part of the security directorate's ongoing
effort to review all IETF documents being processed by the IESG.  These
comments were written primarily for the benefit of the security area directors.
 Document editors and WG chairs should treat these comments just like any other
last call comments.

TL;DR: Ready (with a question)

The document describes an alternative method for nodes behind a middlebox (such
as NAT device or firewall) to contact the middlebox in order to manage port
allocation. Existing methods (described in RFC 6887 and 7291 respectively) are
to either assume that the default router is the device (suitable for small
networks) or specify the middlebox address in a DHCP option (suitable for
larger networks).

This document proposes a third alternative: use of a well-known anycast
address. Sending a request to that anycast address will ensure delivery to the
closest service address (which may or may not be co-located with the middlebox)
by the routing on the network, supported by either BGP or IGP.

There are two specific concerns about this method (other than the usual anycast
or pcp concerns). The first is that information about the internal network
might leak to a PCP service outside the network. Whereas a failure of a service
whose address is given in DHCP will result in black-holed packets, failure of a
service with an anycast address will cause the packets to be forwarded to some
random PCP server on the Internet. Section 5.1 discusses this and recommends
filtering in perimeter gateways and reduced TTL. I believe this addresses that
threat adequately.

The other specific concern is that a rogue machine would push routes to
advertise itself as a PCP service, hijacking PCP traffic and causing network
outages. Section 5.2 deals with this issue. The section makes the claim that
within the first network segment, the nodes do not use dynamic routing
protocols, so an attack there is equivalent to impersonating the default
router. Outside the first segment, routing protocols are used, and there is a
need for routing security anyway. In both cases an attacker capable of
conducting these attacks can do a lot worse than impersonating a PCP service.

I find this argument almost convincing. What is still bothering me is the
question of whether the more damaging attacks would be discovered immediately,
whereas simply advertising a route to the anycast address can “fly under the
radar” so that the attacker can become the PCP server undetected. I don’t have
a firm attack in mind, just a mild concern.

Yoav