It helps to know the assumptions that have led John to such incorrect
conclusions.

When moving from a PSL to the DNS, the primary obstacle has always been
missing information.    Therefore, one goal of this project is exactly to
induce people to update their DMARC records, to supply information that is
currently missing.   We are attempting to do what DBOUND tried and failed
to do - insert boundary information into DNS using information published by
registrars and the registered organizations.

Ale has correctly assessed that in the new design, a DMARC record can play
one of four roles.   Evaluators must code for all four of these
possibilities, regardless of how many token values we define.   With four
roles, we have five information states:  explicit indicators for Registrar,
Org, Both, None, and the fallback result of NULL.  Anyone who has ever
tried to interpret incomplete data knows that a null result is different
from a default value, and that explicit data is always preferable to
guessing a default value to use when confronted with null.

Evaluators will be required to cope with a NULL result by assigning those
policy records to one of the four roles.   The assigned role will be
context-sensitive, which is one of the reasons that NULL and NONE are not
equivalent.    Fortunately, we expect that the correct default can be
applied most of the time.   However, the proposal to project five states
into three, using "psd=(y,n,null)", is an unforced error that cannot be
justified.

There is also a huge communication error introduced by assuming that
"psd=n" will be reliably configured to mean "org=y".    There are a
relatively small number of registrars who will be publishing DMARC
information, against millions of organizations that are publishing, or will
publish, DMARC policies.   We need the language optimized to ensure that
organizations publish their flag data correctly.

We also have a nomenclature problem connected to the choice of "psd=value"
as a token.   We have recognized that there are public registrars and
private registrars, and that DMARC must cope with both types.   Private
registrars are not PSOs and we should not bless them with a title that they
do not deserve.   Our document needs to be updated to reflect this
reality.   Throughout the text, we should have "Registrars", and
"Registration Domains", not PSOs and PSDs, except for introductory material
where we explain the difference between the two and its implications for
tree walk.   The registrar role should be indicated with "REG", not with
"PSD".

Doug Foster


On Sat, Feb 26, 2022 at 12:14 PM John R Levine <[email protected]> wrote:

> >> No.  This just adds more useless complexity that is unlikely to get
> implemented.
> >
> > While composing a DMARC record, setting role=org seems more likely than
> psd=n.
>
> For the umpteenth time, the goal here is to be as compatible as possible
> with the way that DMARC works now.  An important part of that is not to
> ask people to change their existing DMARC records because we know that
> most of them won't.
>
> The normal case, like 99.99% of the time, is that the PSD does not publish
> a DMARC record at all.  The org domain has a DMARC record if it sends mail
> or its subdomains use relaxed alignment.  The way Scott and I propose to
> do a tree walk, that will get the same alignment as now with no changes to
> the DMARC record.  That includes millions, maybe tens of millions of
> domains.
>
> A few PSDs publish DMARC records, either because they have a policy about
> their registrants' mail, or because the PSD itself has an MX.  We want
> them to add psd=y.  That includes 52 domains.  (I counted them.)
>
> As an extreme corner case, if you are registered under a PSD that
> publishes a DMARC record  but erroneously doesn't include psd=y, you can
> use psd=n as a kludge to prevent evil sibling alignment.  That currently
> includes about 45 of those 52 domains, but I think we can get it close to
> zero because we have contacts at many of them.
>
> I'm finding it hard to understand the advantage of a scheme that requires
> millions of DMARC records to change rather than one that changes 52.
>
> Regards,
> John Levine, [email protected], Taughannock Networks, Trumansburg NY
> Please consider the environment before reading this e-mail. https://jl.ly
>
> _______________________________________________
> dmarc mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/dmarc
>
_______________________________________________
dmarc mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dmarc

Reply via email to