On Sun, Oct 25, 2026 at 01:59:18PM -0400, John R Levine wrote:
> > For interoperability, a receiver can treat an unknown algorithm signature
> > as local policy passing i.e., with a SHOULD and Authentication-Result of
> > "policy". If a vulnerability has been announced or for some operator
> > defined reason, that same local policy can set the Authentication-Result to
> > "softfail".
>
> See Viktor's recent messages where he points out that DKIN signatures are
> not long lived so if you think a signing scheme is broken, stop using it.
> That's straightforward and doesn't require inventing new mechanisms that
> operators are unlikely to understand. ("I sign with RSA, ED25519, and ML-44
> using AND on all three because that makes my mail extra secure!")
If downgrade protection is wanted, e.g. for interoperability with both a
classical-only and a pq-only/pq-capable validator with the pq-capable
validator checking at least the pq signature, there needs to be a
downgrade-resistant signal that the sender commits to apply pq signatures.
That signal CANNOT be in the DKIM2 signature itself, because some signers
will be classical-only, and if/when the classical algorithm is broken
attackers can simply delete the pq signature and modify the classical
one to elide any hint that a pq signature should also be present.
Therefore, the signal can only be in the DNS datra for the indicated
selector, which needs to commit to a *set* of signature algorithms.
The validator can then expect all those signature algorithms to be
present, and can validate any desired subset of the ones it supports.
This is a "soft" AND, in that the validator MAY choose to check them all
(and consider the message unsigned if any are invalid), or MAY choose to
check whichever subset it considers strong enough, while ignoring weaker
signatures.
Validators that don't support a given algorithm simply ignore that
signature, and check some subset of the signatures they support.
For this to work the DNS record needs to be modified to carry a
commitment to multiple key algorithms, the simplest approach is
to support multiple TXT records in the response RRset:
selector._domainkey.acme.example. IN TXT \
"v=DKIM2; k=rsa; h=sha256; p=..." "..." ...
selector._domainkey.acme.example IN TXT \
"v=DKIM2; k=mldsa44; p=..." "..." ...
And this represents a commitment to sign with all the listed keys,
of which the validator can choose to check any supported subset.
--
Viktor. 🇺🇦 Слава Україні!
_______________________________________________
Ietf-dkim mailing list -- [email protected]
To unsubscribe send an email to [email protected]