Re: [openpgp] Transporting certificates in messages [was: Re: Transport public keys inside the message body]

[Andrew Gallagher <[e-mail]>]

On 19 Feb 2024, at 16:03, Daniel Kahn Gillmor <[e-mail]> wrote:
> 
> In draft-ietf-lamps-header-protection, the recommended scheme ("Injected
> Headers") includes many headers on internal MIME parts that do not begin
> with Content-* and the extensive testing that was done for that draft
> revealed no clear problems doing so.

Good to know! In that case, the header doesn’t need to be `Content-*`.

> I'll also note that if the signature itself is part of the headers of a
> MIME part (as opposed to of the message itself), that makes the
> header-protection work more challenging; the signature cannot sign all
> the headers if the signature itself is a header.  so there's a bit of
> fancy footwork that needs to be done to be able to calculate the
> underlying stream to be signed, which should really include all the
> Injected Headers.
> 
> This becomes significantly simpler if the signature in a signed-only
> message is just transported as a header field on the message itself, as
> opposed to on a MIME part.

One thing that did occur to me over the weekend was that many OpenPGP plugins (mailvelope, GnuPGTools, etc.) rely on the MUA to expose an API, rather than manipulating the message directly, and that most such APIs restrict access to rfc822 headers and MIME structure. Any proposed replacement should of course take that into account, and may limit what is practical. See for example the “Content-Disposition: attachment; filename=signature.asc” header that Apple Mail will add to the signature MIME part on this message, which is not required by RFC3156, and does nothing to discourage receiving MUAs from incorrectly treating the signature as an attached data file.

A