[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: CPE equipments and stateful filters



On Jul 24, 2007, at 11:31, Rémi Denis-Courmont wrote:

To be fair, I was told UPnP does support authentication. Only nobody uses it, very much like almost nobody uses authenticated DHCP.

I have not heard that before, but it doesn't surprise me. I do know that UPnP IGD is typically used without any authentication and that Apple's support for it in mDNSResponder assumes that the unauthenticated mode of operation is all that's necessary.

I am sure James could add some kind of authentication mechanism for ALD, but I doubt it would see much real-life use. Unless people feel it should be usable from the outside, much like you can open your house outside door with the appropriate key.

In fact, the security considerations section in my current draft offers the opinion that IPsec and IKE are the appropriate ways to secure the ALD exchange between listeners and firewalls. I don't feel confident that I have adequately explored the issue, but my hunch is that network layer security mechanisms ought to be used for securing network layer control messaging. It's possible that ALD needs specific signaling in Firewall Advertisements to indicate that listeners must have an IPsec SA to be established if they expect their notifications to be processed. However, I wouldn't expect this feature to be necessary in most residential / small-office CPE deployments.

In any case, I think people advocating stateful firewalls on CPEs should really get familiar with the work done in BEHAVE wg. I am very sorry that the use and abuse of stateful firewalls in IPv4 world has lead to the unusability of just about any IETF transport protocol but UDP (Christian already hinted at this). If we are to require stateful firewalls for unmanaged network, we REALLY REALLY REALLY need to clearly explain how TCP, SCTP, DCCP and any other connection oriented protocol is expected to operate home-to- home. Not doing that would be equivalent to stating that the only valid model for
unmanaged IPv6 is client at home and server in the server farm.

At this point, I'm beginning to believe that we are seeing the outlines of a very serious philosophical debate between two factions, most recently represented on this list by Itojun and Fred respectively, who have foundationally different views about the ethics of network security methodologies. As a Rorty-ian neo- Pragmatist myself, I'm sensitive to the possibility that these two philosophical systems may not, in fact, have any objective principles in common, and that the best we might hope to do is try to get as much intersubjective agreement about our principles as we can muster, and proceed accordingly. I believe there may be more overlap than most participants think.

I'm thinking about how to gather together the points where I think there might be some philosophical agreement. I may have to write a draft. Before I do that, I may have to spew random thoughts onto the mailing list and see what sticks. Sorry about that.

Is there a tradition in IETF of writing philosophy essays as Internet drafts?

By the way, to amend James's slideset, only UDP really works with ICE (or any
form of hole punching), while TCP works to a very-lesser extent, and
everything else does not: DCCP, SCTP, IPsec...

I think I was careful to use "ICE-like" as the construction in my slides. I was trying to hint at the various techniques in currency within the BEHAVE working group products for traversing the stateful filters in IPv4/NAT, e.g. TCP simultaneous open, relay services, etc.


--
james woodyatt <jhw@apple.com>
member of technical staff, communications engineering