The tension between draft-ietf-dnsop-multi-alg-rules and
draft-jabley-dnsop-local-signing-algorithm-policy got me thinking.

What draft-ietf-dnsop-multi-alg-rules tries to do is to allow different
signers for a single zone to sign with different algorithms. In the future we
could have a situation where for a single zone, some signers sign with ECDSA
and others sign with a PQC algorithm (or ECDSA plus PQC).

What draft-jabley-dnsop-local-signing-algorithm-policy proposes is that if
a zone seems to be signed with a PQC algorithm (especially if there a DS record
with a PQC algorithm) then legacy (quantum-unsafe) algorithms are to be
ignored. Combining there two drafts is likely to result in validation errors.

My idea is to introduce an new DNSKEY bit, which I call the Zone Signing
Algorithm (ZSA) bit that signals that the entire zone is signed using the
algorithm of the key that has the bit set to true (and not necessarily that
specific key).

Then any zone that would like to get downgrade protection would set the
bit on a PQC key. And draft-jabley-dnsop-local-signing-algorithm-policy would
only apply to zones that have the ZSA bit set on a PQC DNSKEY record.

This would allow draft-ietf-dnsop-multi-alg-rules to gradually move to PQC
without risk of validation errors.

It would also support two more use cases. The first is one proposed by
Johan Stenstam: a PQC KSK and a quickly rolled ECDSA ZSK. Currently this
fails on validators that implement
draft-jabley-dnsop-local-signing-algorithm-policy. By keeping the ZSA bit
clear this would continue to work.

Another use case goes in the opposite direction: hybrid schemes. A zone
signed with both PQC and ECDSA could set the ZSA bit on both a PQC and
an ECDSA key. That allows paranoid validators to insist in verifying both
the PQC and the ECDSA signatures. The advantage is that the effective
strength is the strongest of the two algorithms.

Note that all of this validation behavior is optional with just the requirement
that more strict forms of validation are only applied to zones that have the
right ZSA bits set.

Signers would be enouraged to set the ZSA bits to benefit from more strict
validation. Signer that cannot guarantee that the entire zone is signed
with a particular algorithm are free to keep the ZSA bits clear. This comes
at the cost of an increased risk of downgrade attacks. But while we
transition to PQC that is probably worth it.

Downsides: this will fail if a single zone is signed with two different PQC
algorithms. In that case, the ZSA bit cannot be set. At the moment signing a
zone with two different PQC algorithms doesn't seem likely. Who knows what
the future will bring.

Note that downgrade attacks are an issue for zones that are signed with both
classical and PQC algorithms. We can assume that in the future when PQC
is common in validators, zones will stop signing with classical algorithms
and the risk of downgrade attacks goes away.
_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to