Re: Not! the same DNS records for different checks

Hadmut Danisch <[email protected]> Sun, 26 Oct 2003 13:56:03 +0100
Newsgroups gmane.ietf.asrg.rmx
Message-ID <[email protected]>
On Sun, Oct 26, 2003 at 12:41:05AM -0400, Yakov Shafranovich wrote:
> My question to you guys is as follows:
> 
> There are two approachs outlined here:
> 1. HELO checks.
> 2. MAIL FROM checks.
> 
> These different types of checks - can they rely on data stored in the 
> same DNS records - can we use the same DNS record for both checks?


I'm currently writing down what I consider to be a unified proposal, 
and I am elaborating a little bit into security engineering, thus
focussing on the difference between authentication and authorization,
and such things. 


While writing I found a significant and important difference between
HELO and MAIL FROM:


The MAIL FROM verification is solely an authorization check. The
identity is basically the IP address itself (which we get cheap and do
not need to care about), and we ask "Is the entity with this identity
(=ip address) authorized to do this and that?"

In contrast, the HELO check is a completely different story:

- Two identities: IP address and Hostname
- Our question: "Are you really Hostname?"
- So it is only a matter of authentication
- No authorization

Since we want to ask the domain authority whether this particular
machine has really this particular hostname, it is just a 1:1 
relation and authentication. We do not need the RMX/SPF/... overhead
to do so, because our proposals are mainly authorization mechanisms.
To be precise, they are useless here because we don't need 
authorization at HELO.

So the question is: Wouldn't a simple A query to the given hostname
and the verification whether the response meets the peer's IP address 
to the appropriate tool and just cover the authentication question?

If we already agree that the EHLO parameter is a hostname, and the
domain owner has to just confirm that this hostname belongs to this
machine, so what could our proposals do any better than a simple
A record?

Hadmut