I don't understand the reluctance to grapple with this problem.

Consider:

   - Example.com has a typical mail store server used to send and receive
   employee's personal messages.
   - I receive a message from [email protected], and it passes with SPF PASS
   and strict alignment..   The example.com domain is used for both the
   DMARC policy lookup and the SPF policy lookup.
   - I reply to [email protected] and receive an automatic reply about her
   upcoming vacation.    The auto-reply uses the same outbound path as all
   other messages from Example.com.

If we allow SPF on null sender, the auto-reply is evaluated using a
different SPF policy, based on the HELO name, although the justification
for doing so is vague.

   - If the HELO SPF policy exists and the HELO domain is within
   Example.com, and the domain DMARC policy specifies relaxed alignment, then
   I have DMARC PASS.
   - If the domain DMARC policy specifies strict alignment, I have DMARC
   FAIL.
   - If the server domain is a hosting service, I have DMARC FAIL.
   - If the HELO policy does not exist, I have DMARC FAIL.

Upshot of the above:

   - For many organizations, auto-reply messages require DKIM signatures to
   obtain DMARC PASS.
   - For all organizations, strict alignment is not allowed unless all
   auto-reply messages have DKIM signatures.
   - Because auto-reply messages are expected to use the same
   infrastructure as originating messages, requiring DKIM for auto-reply is
   obliquely requiring DKIM for originating messages.

The flip side:

What if an attacker wants to impersonate the server of another organization?
I am doubtful that this is a useful attack strategy for a bad actor, but
assuming that the intent exists, is SPF on HELO an effective defense
against this attack?

   - If the attacker uses an actual server name, such as "
   realserver.example.com", and the server has an SPF policy, and the
   policy ends with -ALL, then the attack will be detected with SPF FAIL,
   which may be actionable.
   - But the attacker can choose where he is going to attack, and the
   existence of an SPF record is easily checked.   So the attacker can simply
   use a made-up name, such as "fakeserver.example.com", to ensure that the
   the SPF on HELO lookup will return NONE/NXDOMAIN.  That result is a weak
   one.  Even NONE is enough to prevent DMARC alignment, but it is not
   particularly actionable on its own.

So I see SPF on HELO as an insufficient authentication method for DMARC,
because it provides insufficient coverage.
I also see it as an ineffective defense tool when considered separately
from DMARC.

If we want to require DKIM for auto-replies, then let's be consistent and
require it for originated messages.    We could say,

"Domains MUST provide DKIM signatures before publishing p=reject.  SPF
alignment is provided as an evaluation factor when processing messages for
domains that do not specify p=reject.'

But if we want to stick with SPF alignment for everyone, we need to deal
with the auto-reply problem.

Doug



On Sun, Jun 9, 2024 at 1:32 PM Richard Clayton <[email protected]>
wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> In message <CAH48Zfwke97+H66f3+MRqGkmha=MhA7tbzoSzRj2ELsw93ZK-
> [email protected]>, Douglas Foster <[email protected]>
> writes
>
> >    I would be happy to have you or anyone else explain to me
> >    (a) What data indicates that a non-trivial number of servers have
> >    SPF policies on their host name, and
> >     (b) How the answer to that lookup provides information useful to
> >    an evaluation decision.
>
> I looked at the data from $DAYJOB$ for last Wednesday UTC (a typical day
> I believe)
>
> For email with a null "MAIL FROM" (which includes a certain amount of
> spam as well as delivery status reports ...)
>
> I see
>
>         dkim fail, spf fail:   11%
>         dkim fail, spf pass:    6%
>         dkim pass, spf fail:   14%
>         dkim pass, spf pass:   69%
>
> $DAYJOB$ is coy about absolute numbers but the smallest category there
> is more than 10 million and less than 100 million messages.
>
> When I look at the spf=pass results I see more than 650,000 unique EHLO
> strings. I think that qualifies as non-trivial.
>
> Evaluation decisions at $DAYJOB$ are very complex indeed; but that 11%
> will not be an issue RSN (because of "no auth no entry" policies)
>
> YMMV ... but I would be interested in learning what sort of volume you
> are drawing conclusions from.
>
> - --
> richard                                                   Richard Clayton
>
> Those who would give up essential Liberty, to purchase a little temporary
> Safety, deserve neither Liberty nor Safety. Benjamin Franklin 11 Nov 1755
>
> -----BEGIN PGP SIGNATURE-----
> Version: PGPsdk version 1.7.1
>
> iQA/AwUBZmXmrd2nQQHFxEViEQJi/QCgqBl0/6ll/WSu+KKt7qNzE8LPJBAAoM6W
> 4xmGfD6kE8ikBoD87vv1dvgq
> =wD4z
> -----END PGP SIGNATURE-----
>
> _______________________________________________
> dmarc mailing list -- [email protected]
> To unsubscribe send an email to [email protected]
>
_______________________________________________
dmarc mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to