Re: [radext] BoF request for IETF 115

[Stefan Winter <[e-mail]>]

Hi,


> TBH, we should just deprecate PSKs. Perhaps allow the use of a PSK, 
> but mark it as NOT RECOMMENDED. And require that all implementations 
> MUST support client certificates. That won't solve the problem, but it 
> will at least embarrass insecure implementations. 

 From what I see in operations, the biggest problem people have with certificates seems to be that they have an expiry date you need to attend to.

Configure a client->server with a shared secret or PSK: it will keep running forever

Configure a client->server with an X.509 certificate: the connection will "totally unexpectedly and unpredictably" break when it reaches its Not-After. And the person that requested the original certificate x years ago has long left, and the knowledge about renewals is gone.

If PSKs are too low-entropy then one of the more recent public/private keypair constructs could be an alternative: RFC7250, "Using Raw Public Keys in TLS and DTLS".

This reduces "certificates" to the pure public/private keypair, with no expiry, no names in need of vetting, and able to meet all the expectations on entropy etc. It also integrates nicely into TLS, so no custom code beyond TLS itself is needed on client nor server.

Greetings,

Stefan Winter

-- 
This email may contain information for limited distribution only, please treat accordingly.

Fondation Restena, Stefan WINTER
Chief Technology Officer
2, avenue de l'Université
L-4365 Esch-sur-Alzette