search

tlswg / tls13-spec

TLS 1.3 Specification
565 stars 157 forks source link

Tracking by replaying a flight of 0-RTT data #1334

Closed emanjon closed 6 months ago

emanjon commented 8 months ago

Information that the replay attack described in Section 8 can be used for server tracking is missing.

ekr commented 8 months ago

@emanjon I'm not sure if this is correct. Suppose instead that we set of servers with some distributed state system so that they behave as one even though they are physically distinct

martinthomson commented 8 months ago

Yeah, I'm similarly unsure.

The basic idea of the attack relies on the server recognizing 0-RTT in a way that is observable, but not having anti-replay protections in place such that replay is possible. Generally speaking, those two things are aligned.

A server can recognize that this is a 0-RTT attempt - like NSS could recognize 0-RTT attempts from other NSS server implementations - but if it recognizes the attempt to the point that it is observable beyond that, then it should also share the anti-replay state. Having decryption keys, but not anti-replay state is a pretty big error.

emanjon commented 8 months ago

I was mostly thinking of a single IoT device that is moving around together with a person. IoT devices using TLS are often TLS servers.

ekr commented 8 months ago

I feel like I'm still missing something, because this device will have the same certificate (or I guess maybe PSK?), right? So when you connect to it, won't you be able to tell that it's the same just from that?

I had assumed that the idea here was that you wanted to distinguish two devices which had the same identity.

ekr commented 6 months ago

@emanjon: following up here. I'm planning to submit a clean version for pubreq at end of month and will close this then unless I hear from you by 2/29.

emanjon commented 6 months ago

Sorry that I forgot to follow up on this even even Sean poked me.

feel like I'm still missing something, because this device will have the same certificate (or I guess maybe PSK?), right? So >when you connect to it, won't you be able to tell that it's the same just from that?

I had assumed that the idea here was that you wanted to distinguish two devices which had the same identity.

Yes, that is correct. The attack already exists. An active attacker can identify a sever by sending a new CH (in the cert case) or by replaying a CH in the PSK case. These seems like simpler attacks achieving the same thing. I think we can close this and not do any changes.