[ietf-dkim] Re: ISSUE 1525 -- Restriction to posting by first Author breaks email semantics

[Douglas Otis]

On Jan 15, 2008, at 2:54 PM, Jim Fenton wrote:
>
> The goal of SSP is to determine the practices of the (alleged)  
> author of the message.

SSP asserts the signing policy of the _domain_, and not that of the  
"first author".  SSP assertions are referenced from the domain of the  
"first author" as the entity protected by SSP.  When a domain signs  
with an unrestricted key on-behalf-of entity other than the "first  
author", and happens to include a 'first-author" address that is  
within their domain, it MUST BE assumed to be compliant with the  
domain's signing policy.  In this case, there is no need to check  
against the SSP assertions.  In other words, the Sender or the third  
 From email-address could be signed on-behalf-of and yet be compliant  
when the domains match that of the "first author".

> The practices of the (alleged) agent responsible for the  
> transmission of the message aren't relevant; the agent could be the  
> author's secretary, or for that matter the author's (or authors')  
> attorney or PR firm.

Only practices of the signing domain are relevant.  A corner case is  
with respect to a message signed on-behalf-of the office administrator  
that includes a "first author" within the same domain.  Such a message  
is compliant with an assertion that all messages from the domain are  
signed, as in "dkim=all".

When the entity's address within the Sender's header being signed is  
within a different domain from that of the "first author", the message  
would not be complaint with "dkim=all".  SSP policy is about _domains_  
and not about _entities_.  DKIM's charter excludes signatures being  
about _entities_.  DKIM is about _domains_.

> According to RFC 2822 section 3.6.2, the Sender header field MUST  
> appear whenever the From field contains more than one mailbox  
> specification, and SHOULD appear whenever the message is transmitted  
> by other than the author.  However, even when the From field  
> contains more than one mailbox specification, the Sender field still  
> represents the transmitting agent, not the author.  Use of the  
> Sender field would therefore apply SSP incorrectly.

When the entity's address contained within the Sender header is being  
signed, and the signature is within the same domain as that of the  
"first author", then SSP policies apply to both the Sender and "first  
author".

> We then are left with the dilemma of what to do when there is more  
> than one author.

No.  By definition, SSP policy is determined by the "first author"  
domain.  A requirement should be added for a signature using a g=  
restricted key.  For a restricted key, the signature is valid with  
respect to policy compliance when on-behalf-of the "first author".

> One option would be to look up the practices of all of the authors  
> and combine them.

No. This would result in more transactions and highly dubious results,  
as an indication of a signature being valid means what?  Which domain  
is making the assurance?  Policy assertions should stem from the  
"first author", but signatures should be on-behalf-of the entity  
responsible for their introduction.

> An attacker could then potentially make up messages with many  
> alleged authors as a make-work attack on SSP.

It would be doubtful anyone could offer an understandable way of  
conveying who did what as well.  The recipient should only need to  
look at the _domain_ of the "first author" to understand which  
_domain_ originated the message when this domain's policies are either  
"all" or "strict" and the message is marked as DKIM signed.   DKIM is  
not required to indicate _who_ sent the message.

When a large ISP signs all messages, they may wish to include an  
opaque i= parameter within their signatures and use an SSP of  
"dkim=all".  This changed definition would allow their customers to  
place their desired email-address as "first author" (perhaps even the  
email-address offered by the provider) without the message being  
deemed non-compliant.  The opaque i= parameter thereby protects their  
customer's privacy, while also offering a means for providers to curb  
abuse.

> Instead, by looking at the first From address only, we force an  
> attacker who wants to weaken SSP by inserting an extra address to  
> put the bogus address first, causing the message, most likely, to  
> look like it came from someone else entirely.

Only look to the "first author" for locating the _domain's_ policy.   
Keep in mind that DKIM is not about identifying individuals, as you  
appear to be implying.   Ironically, by not restricting which entity a  
signature can be on-behalf-of, enables domains to clarify who sent the  
message.  As SSP is currently defined, domains might be forced to  
falsely identify who sent the message or make who sent the message  
ambiguous.  Placing unnecessary restrictions on the use of the i=  
parameter is counter productive and is highly problematic.

-Doug