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

Re: AH (without ESP) on a secure gateway

	 [this may be a repeat. Emacs crashed.]
	 
	 -----BEGIN PGP SIGNED MESSAGE-----
	 
	 
	 >>>>> "Steven" == Steven Bellovin <smb@research.att.com> writes:
	     Steven> mode.  To do otherwise is inviting trouble.  In fact, I
	     Steven> had thought that was what was done -- no other possibility
	     Steven> had occurred to me.
	 
	   Are you suggesting we need specific wording in the drafts?

If people are trying to do AH at the firewall, without an intervening
IP header -- yes, we should prohibit that.
	 
	     Steven> There's a second issue that has come up here -- how does
	     Steven> one know which the right firewall is?  This is one of the
	 
	   I favour a firewall discovery system. ICMP Admin denied messages can
	 provide the right info. I am not certain if this is workable, since I
	 haven't been able to prototype it all. (And no longer have the funding
	 to even try.)

We need some mechanism -- but I'm not convinced I know the right answer.
Certainly, firewall discovery is part of it, but we also need some sort
of authentication mechanism, so that you know you've heard from a legitimate
firewall.
	 
	     Steven> points I raised at the last IETF meeting; in my opinion,
	     Steven> it's very closely related to the naming issue and the
	     Steven> certificate issue, and we haven't really tackled either of
	   
	   I am retrieving your slides. [how do you get the AT&T logo inside of
	 slidex? Cool.] The problem I see is that there may be different
	 firewalls involved. Both firewalls in parallel and firewalls in
	 series.
	   While this doesn't sound very likely very soon with most current
	 application layer firewalls, Checkpoint has announced state-sharing
	 facilities. It could be a different firewall that is involved each
	 time! Should the encryption state be shared too? Maybe. Maybe not. 
	   Firewalls in series are more interesting. I expect to see this. I am
	 seeing this.

[I use latex for my slides, and once upon a time, someone wrote a Metafont
definition for the Death Star logo.  So to latex, it's just another
character.]

We certainly will see many combinations of firewalls.  That's part of
the fun...
	   
	   The best system I can imagine is one where an end node is provided
	 with a certificate, signed by its intended destination stating that
	 "firewall X is a legitimate firewall for me". The local node will also
	 need to be able to recognize a certificate from a local authority
	 saying "firewall Y is a legitimate firewall to get to 0.0.0.0/0"

Something like that, though wildcard DNS-like records might be needed
as well.
	  
	    (local node)-------- Y --------- X ------- (end node)
	 
	   The SPKI groups' proposal has the notion of cache certificate that
	 could reduce a series of statements into single self-signed certificat
	e.

All of these are reasonable ideas.  Now we need concrete proposals and
implementations....