Re: [dmarc-ietf] Ticket #1 - SPF alignment

[Douglas Foster <[e-mail]>]

You missed my main point.   I will restate it, and then address your threat
question.

If there are N ways to establish authentication, then a recipient only
needs to choose one of them, but a sender who wants to be validated by all
recipients must implement all of them.   One purpose of the specification
is to constrain that complexity.

So the convenient solution would be to say that SPF HELO must be ignored,
except for null sender, because doing so reduces the number of variables
that must be considered by senders.   It is also probably consistent with
the logic of many current implementations.   I just do not believe that it
is justified to tell recipients that they MUST NOT evaluate SPF HELO when
MAILFROM is present.

An alternative would be to say that SPF HELO and SPF MAILFROM should both
be evaluated.   In the long run, this is arguably the best theoretical
solution, but the fear of short-term pain from false positives probably
makes this unattractive.

So, I am back to the only solution that seems justifiable:    Senders
SHOULD comply with both MAILFROM and HELO policies because SOME recipients
may evaluate both criteria, but many will only evaluate MAILFROM because
the incremental theoretical benefit seems less than the incremental cost.

- - -

To your point about threat vectors:  DMARC alignment is based on nested
layers of trust, and the trust is broken if you skip layers.  The server
operator provides a mail server environment to a mail domain, and the mail
domain provides delivery services for an author.

SPF proves that the mail domain (indicated by the MAILFROM address) really
uses the services of the server domain, and therefore the server (indicated
by IP address, HELO, and possibly ReverseDNS) is authorized to send
messages on behalf of the mail domain.

DMARC proves that the source server for the message was authorized to speak
on behalf of the author (indicated by the From address).

You suggest that if SPF HELO passes, and aligns with FROM, then we have
established that the server is authorized to speak for the author.   This
is true, but it does not establish that the mail domain in the MAILFROM
address has that right.

When the MAILFROM domain is different from the HELO domain, this implies
that that MAILFROM domain is a client of the server domain.   The server
domain has the right to emit messages for the client, but the client domain
does not have the right to email messages claiming to be from the server
domain.

- - -

It can be asserted that the server domain should be able to use its
firewall to prevent unauthorized servers from emitting mail, and should be
able to use its outbound gateway to prevent client domains from emitting
messages from the server domain, other clients, or any unrelated domain.
These commonly implemented measures could eliminate the need for checking
SPF HELO.   Why should others check what the sender should be able to do
for themselves?
Are we prepared to make this argument?   If so, then the existing wording
stands, but needs a solid justification.

DF

On Thu, Feb 11, 2021 at 8:50 AM Scott Kitterman <[e-mail]>
wrote:

> What problem are you purporting to solve?
>
> By problem, I mean a case were a bad actor can get a DMARC pass result if
> SPF
> HELO results are allowed to be used that they couldn't already get with a
> mail
> from result.
>
> I don't think such a case exists which is why I think this entire line of
> argument is a waste of time.
>
> Scott K
>
> On Thursday, February 11, 2021 6:35:49 AM EST Douglas Foster wrote:
> > Applying SPF to DMARC could become out of scope, if we choose to remove
> SPF
> > from DMARC and make it dependent only on DKIM.   Until then, we need to
> > have a shared understanding of how SPF is applied.  This question asks
> > whether that shared understanding exists.
> >
> > SPF involves two tests, which can be used together.   This WG can insist
> > that for DMARC purposes, only one can be used:
> >
> >     "When the sender is not null, DMARC-evaluation only considers the SPF
> > evaluation of the MAILFROM Address.   SPF evaluation of HELO MUST NOT be
> > considered for DMARC purposes."
> >
> > This wording seems implied by the current language, and by those who want
> > to leave it untouched.  Implication is different from specification, so
> our
> > document should make this explicit.   Unfortunately, an explicit MUST NOT
> > requirement is hard to justify.   When two domains are involved, and both
> > domains have published policy information, what justification exists for
> > ignoring some of the available security-related information?
> >
> > If we back away from MUST NOT, then we have to consider that some
> > recipients MAY evaluate SPF HELO and SPF MAILFROM together, just as the
> SPF
> > RFC expected them to be used, and as outlined in one of my examples.
> If
> > some recipients MAY evaluate HELO, then senders SHOULD take care to
> ensure
> > that HELO will generate a PASS.   Our language becomes something like
> this:
> >
> >     "When the sender is not null, DMARC-evaluation always uses the SPF
> > evaluation of the MAILFROM Address.   Some recipients may evaluate SPF
> HELO
> > as well.   To maximize recipient trust, senders SHOULD publish an SPF
> > policy which ensures that both MAILFROM and HELO will produce SPF PASS
> > results."
> >
> > DF
> >
> > On Wed, Feb 10, 2021 at 6:29 PM Dave Crocker <[e-mail]> wrote:
> > > On 2/10/2021 3:24 PM, Douglas Foster wrote:
> > > > Huh?  Are you asserting that SPF MAILFROM and SPF HELO are
> > > > interchangeable?   They are not, but they can work together.
> > >
> > > Perhaps I misread, but I thought I saw that this really is out of scope
> > > for this working group.
> > >
> > >
> > > d/
> > >
> > > --
> > > Dave Crocker
> > > [e-mail]
> > > 408.329.0791
> > >
> > > Volunteer, Silicon Valley Chapter
> > > American Red Cross
> > > [e-mail]
>
>
>
>
> _______________________________________________
> dmarc mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/dmarc
>