Re: [MMUSIC] [rtcweb] Draft new: draft-wang-mmusic-encrypted-ice-candidates

["Martin Thomson" <[e-mail]>]

On Wed, Nov 6, 2019, at 06:15, Qingsi Wang wrote:
> On the key management, we think this should be deferred to 
> implementations, given different options to solve this problem. Chrome 
> enterprise policy is currently an example for such a solution in 
> managed environments, and it allows admins to push configuration 
> settings to managed endpoints.

If there is an assumption that key provisioning works this way then I can see how you reached this conclusion.  I think that you need to be very careful to describe how that constrains the design in the draft though.  Right now, there is a big hole that screams "key management problems".  I wasn't asking you to define the key management system, just to articulate what properties this design assumes and maybe offer some guidance for implementation.  If this is as simple as you describe, that's great.

I don't think that it is that simple - this is a combination of policy and context that is hard to get right.  If I take that managed device home, why would I hide my address?  Or, why would I share my address with my colleagues?

Finally, you are talking about a secret that many people will know.  For instance, I can easily imagine people posting the Foo company corporate shared key on forums to aid in debugging.

 > As for errors from key rotation, MAC verification will cause such 
> candidates to be discarded, and clients will fall back to mDNS, STUN, 
> or TURN as appropriate. If we feel like that we need to solve this 
> problem more thoroughly, some implementation guidelines can also be 
> included regarding maintaining a brief key history and using trial 
> decryption as needed. We'd appreciate suggestions on further improving 
> these guidelines.

For the very constrained configuration you describe, it might be appropriate to rely on trial decryption, but that assumes that all deployments will be able to manage with a very simple scheme.  That's rarely appropriate in the general case.  I'd suggest that some sort of key identifier is probably necessary.