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

[Alessandro Vesely <[e-mail]>]

Just to clarify:


On Wed 10/Feb/2021 05:19:38 +0100 Scott Kitterman wrote:
> No one has demonstrated that if someone has implemented SPF (RFC 7208) without
> worrying about DMARC that there are any associated problems for DMARC.


I think I did.  OpenDMARC, for example, seems to read a single result, either 
Authentication-Results: or Received-SPF:, assuming that it contains the mfrom 
identity unless empty.  Note that it has an option to disable SPF entirely, 
presumably as a means to tackle non-DMARC oriented SPF filters.

Google apparently works similarly.  Given a valid helo and a neutral mfrom, the 
spf= clause of its (ARC-)Authentication-Results: only reports the latter.  That 
is to say, you need a non-RFC7208 compliant SPF filter to instruct DMARC.


> On Tuesday, February 9, 2021 10:13:37 PM EST Douglas Foster wrote:
>> [...]
>> 
>> My interest is interoperability:    We want recipient requirements and
>> sender compliance measures to align.
>> 
>> RFC 7208 says that recipients MAY want to use SPF HELO and SPF MAILFROM
>> together.  An argument can be made that this is not necessary:   SPF
>> MAILFROM shows that the server is authorized to send messages for the
>> specific domain in the MAILFROM, while SPF HELO says only that the server
>> is authorized to send message for the server domain and an unknowable set
>> of other domains.


What unknowable set of other domains?  If the server has an SPF record, it 
presumably authorizes just its IP address(es).


>> Todd's assertion is that SPF HELO will cause an excessive number of false
>> positives.


I'd let Todd speak for himself, but I never saw that assertion.  Todd said the 
set of messages that would get a different DMARC status in case we linearize 
the spec is immeasurable  —which I believe is true, and a valid basis to carry 
out the linearization without fear of disruption.


>> A second assumption is that no significant recipients are evaluating SPF
>> MAILFROM and SPF HELO together in a way that would be of interest to
>> senders. This may also be true, but I don't think this is something that
>> can be tested.


It makes no sense to require both values to pass simultaneously.

Linearization would be that *any* validated identifier, as long as it's 
aligned, produces a DMARC pass.


Best
Ale
--