-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
In message <CAAFsWK0q5NJCH_jmF-yYvhsfUqe1p1+SXZWa4Ru_sOTQv-
[email protected]>, Wei Chuang <[email protected]>
writes
>* ORing of multiple signatures - to support buggy implementations with
>greater security risk when there is an algorithm vulnerability.
What are these "buggy implementations" of which you speak .. yes I
mentioned $DAYJOB$ having code that didn't work -- but this wasn't put
into production in such a way as to disrupt email flows.
Designing and specifying a protocol on the basis that some people will
deploy code that has not been rigorously tested and carefully monitored
on rollout is an impossible task. Fortunately, in the real world code is
tested and monitored (and helpful people provide validation suites), so
we should do the right thing from an engineering point of view and
forget about designing for world full of systems that don't really work.
>I would also argue the current -06 spec draft is [.....]
you go on to say full of inconsistencies and I entirely agree (and it is
more than charitable to say they might have been intended).
I am about to shop version -07. Most of the text regarding signatures
and their validation is the pretty much the same as in -06.
Claude tells me that there are still 11 inconsistencies in the text
(despite me doing a bit of tidying up). I have started on fixing the
items on this list because I think the WG made an incorrect decision
when it first considered this issue; and I don't want to spend too much
time documenting what is, I believe, the wrong path.
I think the correct approach is:
* For two signatures with the same algorithm, consider the signatures
valid if both pass or one passes and the other gives a TEMPERROR.
* For multiple algorithms insist that ALL signatures pass.
* Use MLDSA-44 rather than a hybrid (because then you can sign with
another algorithm if you care).
* Allow all variants of "-sha256" and "-sha512" for tidiness.
* Continue with the MUST constraint on verifiers that they have to
implement all algorithms and hashes
* Assume that in the event one of our algorithms is so broken it cannot
be used that the fuss will be sufficient that we can issue a new RFC
and there will be considerable motivation for attention to be paid to
what it says. We have future-proofed the spec (the ABNF will cope)
and since its futile to pretend we can future-proof implementations
we should not waste any energy trying.
This is simple to understand, avoids the hybrid issue and we can change
our mind about whether MLDSA-44 is the right choice all the way up to
"AUTH-48" (but not a day beyond).
We might also like to revisit the statement about "2048" as it applies
to RSA keys now, whilst people are still reading our documents.
> * AND/OR under the control of some sender specified flag- supports
> ORing during bring up and ANDing when security is needed though
> w/greater complexity.
the flag approach has legs, but would have to be combined with a hard
and fast rule of "OR" for the same algorithm otherwise key rotation
would be impossible without (effectively) dropping down to one algorithm
- --
richard @ highwayman . com "Nothing seems the same
Still you never see the change from day to day
And no-one notices the customs slip away"
-----BEGIN PGP SIGNATURE-----
Version: PGPsdk version 1.7.1
iQA/AwUBatVwo2HfC/FfW545EQIrYgCeIHTDKWUjC5jnQkc6i+0UggK++DUAn0uK
rBcGIrJN3T/4SDuzRuXi4hyU
=/GCs
-----END PGP SIGNATURE-----
_______________________________________________
Ietf-dkim mailing list -- [email protected]
To unsubscribe send an email to [email protected]