Re: [New Issue] Inconsistent behavior with NCIDs

[Martin Thomson <[e-mail]>]

On Thu, Nov 26, 2020, at 00:32, Kashyap Thimmaraju wrote:
> I agree that the privacy of one end-point relies on the other end-point.
> However, I'm actually refering to cases where multiple CIDs are shared 
> within an
> NCID frame. This is definitely not good for both end-points, as this enables
> linkability of the QUIC connection. How could this be used? Well the 
> attacker
> can simply link the flow after connection migration or by observing it for a
> long time. Does she gain anything else? I can't think of other gains.

There should be no case where an endpoint has concurrent use of connection IDs with the same value.  Whether that is as a result of receiving them at the same time or not.  Implementation should be able to detect that case and the specification encourages the use of CONNECTION_CLOSE if that is detected.  However, we cannot require more thorough checking for the aforementioned reasons.

As I said before, different values are not sufficient to prevent linkability if a peer is determined to cooperate with an attacker, so I see no value in pursuing that.  Same for questions of covert channels.