No, I am not saying that we need to develop a tool to detect the difficult
cases.

You say we need to preach at domain owners to lower defenses on 100% of
their mail because one or more unthinking evaluators may do something
foolish with a small percentage of their mail.

I say that we need to preach at evaluators to not make foolish unthinking
decisions.

I say that DMARC is oversold because "reject" is the wrong assertion,
although we are stuck with it for historical reasons,   All that "p=reject"
can mean is that "I believe all of my in-house mail is dual authenticated
at origin and all of my service provider mail is DKIM signed at origin."
 The domain owner cannot assert anything more than he knows, and this is
all that he can know.   The DMARC specification and the "p=reject"
terminology imply that this is all that the evaluator needs to know.   The
reality is that there are some messages which the evaluator (or at least
his users) consider acceptable and desirable, which will not meet the DMARC
criteria.   The evaluator needs to consider that issue when formulating
local policy.

A large number of message sources can be trusted as impersonation-free
because of the source, either the Mail From address or the server
identity.    This includes email service providers ,like
ConstantContact.com and SendGrid.net, mailing list providers like IETF.org,
outbound filtering services like ProofPoint and Mimecast, and necessarily
trusted organizations like the U.S. Government.   Some of these will
actually violate DMARC, but all of their mail can be trusted as
representing the true author.  The specifics of these trust decisions are a
matter of local policy, but identifying the ones that an evaluator will
trust is a necessary part of any DMARC implementation.  (And providing the
appropriate exception mechanisms is an implied requirement for software
developers.)

Having evaluators understand how to act in their own self interest makes
more sense than  telling domain owners to weaken defenses for all of their
mail and increase vulnerability for all of their recipients.

Doug

On Thu, Aug 11, 2022 at 12:29 AM Murray S. Kucherawy <[email protected]>
wrote:

> On Wed, Aug 10, 2022 at 10:44 AM Douglas Foster <
> [email protected]> wrote:
>
>> "Breaking long-standing practice" is not the fault of the domain owner
>> policy, it is the fault of DMARC being oversold and the DMARC result being
>> applied by the evaluator in a way that undermines the interest of his own
>> recipients.
>>
>
> It's worse than that: It also sabotages normal operation of third
> parties.  RFC 6377 describes the damage we're talking about here, although
> the tool was called ADSP back then.
>
>
>> However, the domain owner has no reliable way of knowing whether
>> conditions 4-5 will ever apply, and applicability will be different for
>> different recipients.   Therefore, the burden falls on the recipient's
>> evaluator to determine whether "p=reject" is caused by condition 6 or by
>> one of the other conditions.   Telling domain owners not to use p=reject is
>> not the solution; the real solution is for evaluators to act wisely, and to
>> review other available evidence carefully. Our document can provide
>> guidance on wise use, starting with a discussion of possible failure modes.
>>
>
> I think you're suggesting that we need a way to identify a failure that's
> caused by, say, MLMs altering messages (your case 4), and handle those
> differently -- perhaps with less severity -- to avoid the collateral damage
> DMARC causes.  But that gives attackers a recipe for creating a message
> that falls in your cases 5 or 6 yet will get less severe treatment than it
> deserves.
>
> Similarly, creating a de-munging strategy presents a cookbook that might
> be able to construct a message that will fail DKIM in a way that something
> later in the processing order will forgive.  Moreover, we would have to be
> sure of being able to describe a very high percentage of all mutations MLMs
> do in a manner that's hard for receivers to reverse incorrectly.  But those
> mutations range from simple (subject tagging) to quite complex (MIME
> wrapping or restructuring).  We've considered both of these approaches
> before, and have never managed to convince ourselves we could achieve an
> acceptable level of success.  There was, so far as I know, not even a
> single experimental implementation of any of the proposals.
>
> We seem to be left with the idea of telling domain owners that "p=reject"
> causes damage at a level that does not justify the protection it provides.
> Domain owners wishing to protect themselves obviously have disagreed with
> that value judgement, but the community for which the IETF speaks, I would
> argue, is larger than that.
>
> -MSK, no hat
>
_______________________________________________
dmarc mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dmarc

Reply via email to