Is multi factor authentication practical/feasible?

[Michael Slusarz]

> On 07/07/2022 5:24 AM Aki Tuomi <aki.tuomi at open-xchange.com> wrote:
>
> FWIW I think OAuth2 is the modern way to do actually MFA authentication.  There is some progress in Mozilla world (and hopefully other mail clients) to allow OAuth2 to work outside the "big three" circle. Mostly this is *client development issue*, the server-side already mostly supports all the bits you need to roll your own MFA with OAuth2 using off the shelf components. No need to pay microsoft or google.
> 
> Alternate to OAuth2, which works pretty well today, is to use device passwords.

A bit late to the game, but wanted to expand a bit on Aki's comments.

It's good that this topic is being discussed.  We've long felt that email authentication (and, related, client auto-configuration) is one of the biggest weaknesses of email as compared to more "modern" messaging technologies.

However, discussions around this topic tend to get sidetracked as everyone wants to try to design their own technical solutions.  However, all the necessary technologies exist and are standardized.  Token auth is handled by OAuth2.  MFA ,and more generally authentication UI, is handled by OpenID Connect.  At the mail protocol levels, token auth is handled by SASL.  Additionally, SASL supports auto-discovery of authentication providers so there is no need to "hard-code" OAuth2 providers (the unfortunate way that some clients are currently handling OAuth2).  Dovecot supports all of these technologies already, so there's nothing left to do on the server side.  (Side note: client auto-configuration is also already fully supported using existing technologies as well.)

Instead, the issue is chicken/egg - all of this is worthless until clients/providers start implementing this.  That's where the focus of efforts need to be, not in trying to determine which technologies to use.

Admittedly, this not insignificant paradigm shift can be a bit confusing technically.  It's been a long-standing TODO to create some kind of implementation guide to help server/client/auth providers to understand what they need to do to create this new "modern email authentication" ecosystem.  This is a classic example of a situation where necessary standards exist, but the education about these standards are lacking AND there remains open questions about how those standards should interact with each other in real-world scenarios.  Dynamic client registration in OpenID Connect, in particular, is a key component to make this work but is somewhat under documented and lesser known, so it will take community involvement, and likely trial and error, to determine best practices.

TL;DR from a Dovecot perspective: we feel we have all the necessary components needed to enable "modern email auth" in the current product, so we don't see any additional engineering efforts needed in Dovecot.  We instead are focusing our attention in building and supporting a broader community of client authors and authentication providers to push for implementation in order to accomplish the goal.

michael

p.s. As mentioned by Aki, app-specific/device passwords is a perfectly acceptable solution, although a bit of an end-user usability nightmare.  It's a hack to improve security today, but not a proper long-term answer.