[secdir] Secdir review of draft-ietf-dime-extended-naptr-06

[Charlie Kaufman <[e-mail]>]

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.

This document makes some straightforward extensions to the DNS records that support the Diameter protocol (RFC 3588). The goal is to improve performance over the previous approach where in some cases multiple servers would have to be contacted to determine which one(s) supported some particular application. As the document notes, there are no security issues beyond those that apply to the base protocol (RFC 3588).

So I found nothing objectionable in this document.

Unfortunately, the story doesn't end there. To understand this spec, I had to read the Diameter document. The approach that Diameter takes to configuration using DNS appears to leave a gaping security hole that should be examined by whoever is reviewing RFC 3588bis (which is also in final stages of approval). The problem is that TLS (or IPsec) is used to secure the Diameter connections themselves, and shared keys or certificates are used to authenticate the peer entities. But the DNS records appear to be used to learn the names of the peer entities which are subsequently authenticated. That means if someone could update DNS they could indicate the Diameter server for some domain was in fact some rogue server. So long as that rogue server could be authenticated, it would be trusted to speak for the domain in the context of Diameter.

I suspect common practice is to have Diameter server names be manually configured, in which case there is no vulnerability. But without looking at this in detail it would appear that the more flexible autoconfiguration via DNS is not secure. If so (I could easily be missing something), there are several ways it could be fixed. DNSSEC would be the obvious solution, but not necessarily the best. DNSSEC is still not widely deployed, and this protocol is unlikely to act as a forcing function. Better would be to put some sort of extension in the certificates of Diameter certificates that indicates the domain(s) over which the server has jurisdiction in this context. Yet another option would be some naming convention for finding the name of Diameter servers for a particular domain rather than looking them up in DNS.

In any case, someone should have a look.

                --Charlie