[openpgp] Certificate discovery over HKP

[Andrew Gallagher <[e-mail]>]

Hi, all.

At the OpenPGP Email Summit last week, we discussed OpenpPGP certificate discovery over HKP, and it was again suggested that we use SRV records instead of a .well-known policy file. I could not immediately recall all of the arguments leading to the previous decision to use a policy file, so I want to follow up now with a summary of the various considerations.

The main concern with policy files remains the barrier to entry for small domain owners, who would have to maintain an `openpgpkey.$DOMAIN` web service to serve as little as a single file. While this would not be difficult for the technically experienced, TLS certificate management is still a significant amount of work compared to setting a SRV record. Setting a CNAME to redirect to a shared service provider would shift the burden of TLS management to the provider, who would have to maintain a certificate management infrastructure for an arbitrary number of domains (although it should be pointed out that KOO already does this).

The main concern with SRV records remains the difficulty of looking up arbitrary DNS records from constrained environments such as browser engines and mobile phone sandboxes. If we moved to this method, we would probably want to specify a method of configuring a known keyserver to proxy discovery lookups, by analogy with recursive DNS resolvers. In addition: a) not all domain registrars support setting SRV records (although it is a one-off cost to change registrars), and b) we can require that policy file lookups are covered by a TLS certificate, but cannot yet require that SRV records be covered by DNSSEC - so a MITM could more easily redirect a client to a malicious discovery service (but consider that a similar argument applies to non-DNSSEC MX records).

Some technical details can be found in the "key discovery" document in the HKP draft repository:

https://gitlab.com/andrewgdotcom/draft-gallagher-openpgp-hkp/-/blob/main/openpgp-key-discovery.md

And the head of the original discovery discussion (from 2023!) is here:

https://mailarchive.ietf.org/arch/msg/openpgp/9Y3y3S4RgJGcv6-QofGA57Kh1f8/

My gut feeling is that the SRV method is technically more “correct” and easier to implement on the server side, but the policy file method is easier to implement on many clients. If we agreed to also specify a proxy discovery method, the balance may shift towards SRV.

Any thoughts?

A