Viktor Dukhovni wrote in <[email protected]>: |On Mon, Oct 19, 2026 at 04:49:37PM +0100, Richard Clayton wrote: |>>1. What is signed. Section 3.4 of -07 says the algorithm "computes \ |>>a hash over |>>all the Message-Instance and DKIM2-Signature fields as described \ |>>in Section 12.8 |>>using SHA-512 [FIPS-180-4] as the hash-alg. |> |> I think it's pretty clear we should use either the LAMPs composite |> (where the hash is fed into ed25519 and the ML-DSA-44 gets fed the raw |> data) | |I all this discussion I see no substantive motivation for introducing |composite signature algorithms. What realistic threat are they really |intended to address? | | - If there's not a CrQC in the next decade, classical algorithms | are quite sufficient for now. | | - If there is a CrQC in that timeframe, the classical component of | the composite algorithm is needless baggage. | | - If a pure PQ algorithm is selected and the choice later turns out | broken, the same goes for its composites. Selectors for broken | algorithms can be easily removed from the DNS by the sender, | and a classical algorithm used instead if there is as yet no | known CrQC (cost-effectively available to the attackers, ...). | | If CrQCs are readily available to the attackers, and fallback | to a classical algorithm (with sufficiently frequent key rotation) | is unexpectedly not an option, then a new PQ algorithm is needed, | and composites don't help. | |Also DKIM signatures do not need to meet a high bar of cryptographic |robustness. The DNS lookups to obtain said keys are typically not |protected, and signature forgery just needs to be noticeably more |expensive than various non-cryptographic attacks (look-alike domains, |fresh free email accounts at the major too big to block providers, ...). | |So the simplest choice is just ML-DSA-44 (pure) and one or two classical |alternatives, say either or both of RSA (with a 2048–3072 bit modulus) |and ed25519.
The problem with that is that the public key is so large that DNS over UDP is no longer possible *per se*. Let alone with DNSSEC, let alone with DTLS, totally let alone DNSSEC+DTLS. And that in times when OpenSSL introduced DTLSv1.3, which is top! DNS over DTLS becomes achievable, super-modern! Other algorithms (will) exist (standardized) that fit all the bills, in a not too distant future --- anyway soon enough before any thinkable (afaik) introduction time of CrQCs. |Dragging composite signatures into DKIM makes no sense. It does for my/the alternative DKIM/ACDC approach. (Though not as "composite" but as a "stack" of two different algorithms.) It starts immediately with Ed25519 (but not as via RFC 8463), but flags (Q/q) that no post-quantum part really exists for now. It defines a flag second (0x71717171) after which post-quantum is mandatory. The algorithm will (likely) come from NIST 2027 consideration. At that PQ flag second the Ed25519 could become obsolete, that is not yet defined, but i would tend to leave it in, because it is a short signature and as such could serve for a good cache database key, broken or not. (Otherwise one likely would need to checksum the super large PQ signature first, before using this as a cache key.) (And it doesn't matter whether CrQCs can break the private key of Ed25519, as the algorithm result still would require both sub-parts to succeed, and creating a complete duplicate over both algorithms will likely not work out, for the forseeable future.) That is the key for me. There is no real hurry to introduce PQ for any DKIM. Once it *could* start to make sense there will be new algorithms which fit the bill for DKIM much better, at least in my opinion, and despite the large signatures. What is necessary is to have a clean update path. And to create a proper one of that sort, an algorithm stack does make sense, from the logical point of view. Btw, off-topic for this one here, but to keep the noise level low. My own DKIM/ACDC -- i need some more days for the next, and hopefully final version (at least as an alternative living apart and "in the void") -- i have introduced a new "ehe=" tag to cover all the header fields which are signed. (These come from a database of headers to be signed, represented as a bitset, plus free-form headers listed by the "h=" tag, and do not include any signature header field etc.) Ie i finally replaced DKIM section 3.7 on how hashes have to be created. The signature (DKIX-Sig: here) is no longer covered by any checksum, but only by the signature algorithm. --steffen | |Der Kragenbaer, The moon bear, |der holt sich munter he cheerfully and one by one |einen nach dem anderen runter wa.ks himself off |(By Robert Gernhardt) _______________________________________________ Ietf-dkim mailing list -- [email protected] To unsubscribe send an email to [email protected]
