[secdir] Secdir review of draft-ietf-dime-nat-control-07

[Matt Lepinski <[e-mail]>]

[I apologize if you received this message twice]

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 defines a Diameter application whereby a Diameter NAT 
Control Application (DNCA) server (either a AAA server or a Network 
Access Server) can control a DNCA client (either a NAT or NAPT device). 
For example, the DNCA server can determine the number of port bindings 
available to a given host, cause the allocation of particular NAT 
bindings, define the external address pool used for by the NAT device, 
or generate reports used for accounting purposes.

The principle security concern is that an unauthorized (or 
unauthenticated) DNCA server could effect a denial of service of attack 
on any or all hosts using a NAT/NAPT device in several ways. (E.g., an 
unauthorized server could exhaust resources in the NAPT device, set 
maximum bindings to zero for all hosts, provide a set of external 
addresses that are in use elsewhere in the network, etc.) An additional 
concern is that an unauthenticated DNCA client could provide incorrect 
reporting data to the DNCA server to prevent correct accounting within 
the system. Therefore, the NAT control application requires mutual 
authentication, authorization of the DNCA server, and integrity 
protection of the Diameter messages in the DNCA interaction. 
Additionally, the application may require confidentiality depending on 
the deployment scenario and, particularly, how authorization is 
accomplished.

The Security Considerations section of the document claims that security 
considerations are essentially the same as in the Diameter QoS (RFC 
5866). I agree with the authors that the security concerns for Diameter 
NAT Control are very similar to the security concerns for Diameter QoS, 
but I think the additional layer of indirection is not helpful to the 
reader. (In particular, the NAT Control Application has no dependencies 
on the Diameter QoS document. Therefore it is reasonable to believe that 
an implementer of Diameter NAT Control may not be familiar with the 
Diameter QoS document.) I would recommend that the authors add 
additional text explaining why mutual authentication, server 
authorization, and message integrity are required (copying a bit of text 
from RFC 5866 may be appropriate). Then I would recommend that the 
authors cite the Diameter specification (RFC 3588) for a discussion of 
how IPsec or TLS can be used to obtain these security properties. (To 
me, this seems preferable to sending a reader to RFC 5866 for discussion 
of required security properties, which in turn sends the reader to RFC 
3588 for an information on the use of IPsec or TLS to achieve these 
properties.)

Finally, the end of the security considerations section claims that 
"Authorization between DNCA Agent and
    DNCA Manager is beyond the scope of this document". I understand 
that the authors do not want to mandate a particular mechanism for 
making an authorization decision. That being said, providing some 
guidance as to how a DNCA client would determine if a DNCA server were 
authorized to issue NAT control commands. I imagine in practice that the 
way authorization is accomplished is that the NAT/NAPT device stores a 
local authorization policy. (E.g., the device stores a list of server 
identities that authorized, and then once the server has been 
authenticated its identity can be checked against the list). At the very 
least having text analogous to first paragraph of RFC 5866 Section 11 
would be helpful (RFC 5866 says the device making the authorization 
decision needs to have sufficient information to make this decision and 
then provides examples of where this information would come from.)

- Matt Lepinski