Re: [dmarc-ietf] DMARC2 & SPF Dependency Removal

[Douglas Foster <[e-mail]>]

In general, message recipients lack the expertise needed to distinguish
between legitimate and fraudulent identities.    Most users do not know how
to read message headers or how to make sense of them if shown.   Whenour
automated evaluation systems and administrative quarantine cannot resolve
an ambiguity, users cannot be expected to do better with less information.
Domain-specific tolerance for risk needs to be implemented with
domain-specific policies, but not with end-user discretion.

DMARC has limited scope.  In its purest form, it only blocks impersonations
when the domain has a DMARC policy, the policy specifies enforcement
(quarantine or reject), and the policy is detected correctly (absence of
PSL errors and absence of in-transit changes).   Only a subset of domains
have a DMARC policy, and only a subset of those domains have enforcement.
So only a small percentage of impersonation attacks can be protected by
DMARC.  For that very limited defense to work, domain owners and evaluators
need a high level of confidence in a PASS result.  Eliminating the PSL
increases the confidence in PASS even if it creates some increase in false
failures.   Eliminating SPF will have the same effect, and is desirable for
the same reason.   Our assumption is that DMARC-enforcing domain owners are
highly motivated and will reconfigure as necessary to ensure DMARC PASS
results.

This same logic raises questions about whether relaxed authentication
remains necessary or appropriate.  With SPF, relaxed authentication seems
necessary because changing the MailFrom domain to match a desired From
domain is complex.   For DKIM, I see no important obstacles which prevent
the signature domain  from exactly matching the From domain.

ARC addresses the one problem that domain owners cannot prevent, which is
forwarding with in-transit changes.

Evaluators and recipients have a much larger problem.  They need to protect
themselves from all impersonation attacks, which DMARC does not attempt.
Their defenses must be implemented within the limits of information
available at evaluation time.   If the message has a relevant DKIM
signature but no DMARC policy, the DKIM PASS is still a valid indication
that the message is not impersonated.   Similarly, if the message has no
signature but produces SPF PASS with exact alignment, the SPF PASS result
indicates that the message is probably not impersonated.   Conversely, SPF
FAIL and SPF NONE service to increase the suspicion of an impersonation.

Evaluators use sender authentication to perform a three-way split on
messages:
- Messages from verified and highly trusted senders are allowed to bypass
some or all content filtering.
- Messages from unwanted senders are unconditionally blocked prior to
content filtering.
- Messages from all other senders are accepted if they pass content
filtering.

The more effective your sender authentication, the less you need perfect
content filtering.   Similarly, the more perfect your content filtering,
the less you need optimal sender authentication.

In summary, your instincts are correct, SPF has a long term role in your
defenses, even if it is not part of the DMARC specification.

Doug Foster

On Sat, Jun 17, 2023 at 8:28 AM Jan Dušátko <jan=
[e-mail]> wrote:

> Hi
>
> I would like to know your opinion on the options currently available to
> the system administrator. If it is trying to define a policy that allows
> recipients to authenticate emails, its options are, in my opinion, limited.
> - The issue of SPF and cloud systems mentioned, or including over
> networks with mask /8 or /16 is extremely inappropriate
> - It is not possible to set up the enforcement of DKIM signatures
> because DomainKey and its policies are not related to DKIM and there is
> no similar technology for DKIM (ADSP is not used)
> - If I am the administrator of hundreds or thousands of domains, it is
> in my interest to provide the recipient with information about the
> sender's authentication options. For example, all emails use DKIM, all
> emails use ARC, all emails use SPF ... and if not, you can discard them.
> But I do not have this option with DMARC.
> I understand that this is in the middle of what DMARC2 is supposed to
> address. But could there be a possibility for the DMARC2 record
> administrator to define a policy for domains and provide relevant
> information for recipients? It is the recipient's decision whether to do
> this verification, but it is also the sender's decision to offer this
> information. Unfortunately, at this point, either the signature is
> valid, or wrong, or it is not. If it is not possible to accept the
> e-mail, but the recipient does not have the option of verifying whether
> the sender's domain enforces this signature. It is similar with SPF. And
> DMARC2 seems to me to be a good place for these definitions. But I would
> like to avoid of forced removal of SPF, please let it for administrator
> decision.
>
> Regards
>
> Jan
>
> Dne 16. 6. 2023 v 13:28 Sebastiaan de Vos napsal(a):
> > The need for separate DKIM failure codes to be able to separate
> > between in-transit changes and public key errors is more than just
> > valid and I don't consider SPF worthless in general, but I just find
> > it disturbing how the obviously misplaced confidence in SPF currently
> > weakens the whole DMARC standard.
> >
> > Is it not in our best interest to encourage and motivate in particular
> > the less sophisticated senders to use the higher confidence mechanisms?
> >
> >
> > On 16.06.23 13:02, Douglas Foster wrote:
> >> RFC 7489 takes 8 different authentication mechanisms and lumps them
> >> into a single PASS result:
> >> DKIM or SPF, each with up to four types of alignment:  same domain,
> >> parent->child, child->parent, and sibling->sibling
> >>
> >> These eight mechanisms all provide some level of confidence that the
> >> message is not impersonated, but they do not provide an equal level
> >> of confidence.
> >>
> >> Unsophisticated senders have little incentive to use the higher
> >> confidence mechanisms because any PASS result is still PASS.  The
> >> solution is not to pretend that SPF is worthless, because that is an
> >> overstatement.   The solution is to talk about the differences in
> >> confidence provided by the different authentication methods, and note
> >> that evaluators have reason to distrust some of them.   That distrust
> >> could cause a weakly authenticated message to be distrusted by some
> >> evaluators.
> >>
> >> Similarly, needs multiple types of FAIL.   Since the data indicates
> >> that missing (or invalid) public keys are a significant portion of
> >> all failures, it needs a separate failure code from "normal" failures
> >> which suggest in-transit changes.  The DKIM group needs to define the
> >> result code but this group would need to integrate it into our
> >> aggregate reports.
> >
> >
>
> --
> -- --- ----- -
> Jan Dušátko
>
> Tracker number: [phone]
> e-mail:         [e-mail]
> GPG Signature:  https://keys.dusatko.org/E535B585.asc
> GPG Encrypt:    https://keys.dusatko.org/B76A1587.asc
>
> _______________________________________________
> dmarc mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/dmarc
>