Hi, On 21.08.2025 00:03, Al Iverson wrote:
The idea of per-MIME-part hash makes me nervous; I am not a fan because I feel that it could enable a scenario where somebody will decide whether or not they trust a message based on verification of hashes for only certain parts of the message. In my mind this is about trust verification for the entire message, not just bits of it. I want that trust measurement to be 0 or 100% for the whole message, not, well we only checked mime part X and that's fine, so we'll assume that the whole message must be safe.
Yes, this does feel like a new iteration of the length tag and its problems. But in the case of DKIMv2 it should be possible to differentiate between untrusted modifications and trusted ones. There aren't any modifications that could be done with this new suggestion that couldn't already be done with existing diff algebra. Thus if one trusts body modifications described by DKIMv2's diff algebra then trusting any MIME structure changes should not be significantly different?
Assuming of course that if this suggestion is to be incorporated into DKIMv2, that it doesn't introduce cases where practically equivalent modifications are somehow in practice treated different just because of being represented different. So it's probably reasonable to be extra cautious and forbid any such different equivalent representations - for example removal of a MIME part must be recorded using this proposed approach instead of say, "raw" text deletion diff, or vice versa with any MIME part additions. (Heavily depends on the details of course, but the general point still stands.)
Best Regards, Taavi _______________________________________________ Ietf-dkim mailing list -- [email protected] To unsubscribe send an email to [email protected]
