[ietf-dkim] Issue #1524: Signature semantics

[Douglas Otis]

On Dec 11, 2007, at 11:31 PM, Jim Fenton wrote:

> SSP does require one additional semantic over that of DKIM-base:  in  
> addition to taking responsibility for the message, those domains  
> that publish SSP records other than "unknown" must assert that, when  
> the address in the From: header field is really their domain, that  
> this is actually true.  Whether this assertion extends to the local- 
> part of the address is the subject of another issue (1399) so let's  
> discuss that part separately.

Jim,

What can be safely deduced when a signature is not "on-behalf-of" a  
particular header's email-address?

The "all" assertion offers no assurance a signature is "on-behalf-of"  
of any header.  The "all" assertion does not require the "originator  
signature" (soon "author's signature").  The "all" simply indicates  
_all_ messages from the domain are signed by _some_ domain.  Such an  
assertion makes no assurance that identities related to email- 
addresses contained within signed messages have been authenticated.   
Such an assertion of authentication of identities related to email- 
addresses goes well beyond DKIM WG's charter!

> In other words, those publishing SSP records must make sure that  
> they don't sign spoofed messages claiming to come from their own  
> domain.  Aside from that, no assertion is made regarding the  
> content.  As I have frequently said, I would not want a DKIM  
> signature to authorize a transfer from my bank account.

Your bank _MUST NOT_ assume identities associated with email-addresses  
contained within DKIM signed messages have been authenticated!  Use of  
DKIM may mean when there is a problem, those causing problems lose  
access.  Review DKIM WG's charter regarding strong external assurances  
of identities related to email-addresses.

> This new semantic only applies to those who deploy (publish) SSP  
> records and therefore does not extend to DKIM-base (and therefore  
> does not require a modification to 4871).  This semantic is not  
> currently described in the SSP draft, and needs to be in my opinion.

I do not agree.  A DKIM signature ONLY means the message was handled  
by the domain adding the signature.  Not even "strict" (should)  
require a domain's signature be "on-behalf-of" the From header.  A  
"strict" assertion likely needs an exception for restricted keys where  
the i= parameter must be "on-behalf-of" the From header.  (Per user  
keys will likely prove to be a mistake.)  When a domain makes an SSP  
"strict" assertion, a message with a From email-address containing  
this domain should only require signature by this domain (or a parent  
of) regardless of which or whether any "on-behalf-of" header can be  
determined.

How domain's making "strict" assertions avoid the spoofing of email- 
addresses within their domain is an internal matter for that domain.   
While such a problem is a concern, SSP does not dictate how this is  
resolved.

-Doug