Thus spake Nick Mathewson (nickm@freehaven.net):
Title: Improved circuit-creation key exchange Author: Nick Mathewson
Summary:
This is an attempt to translate the proposed circuit handshake from "Anonymity and one-way authentication in key-exchange protocols" by Goldberg, Stebila, and Ustaoglu, into a Tor proposal format.
It assumes that proposal 200 is implemented, to provide an extended CREATE cell format that can indicate what type of handshake is in use.
Protocol:
Take a router with identity key digest ID.
As setup, the router generates a secret key b, and a public onion key B with b, B = KEYGEN(). The router publishes B in its server descriptor.
To send a create cell, the client generates a keypair x,X = KEYGEN(), and sends a CREATE cell with contents:
NODEID: ID -- H_LENGTH bytes KEYID: KEYID(B) -- H_LENGTH bytes CLIENT_PK: X -- G_LENGTH bytes
I mentioned this on the ntor ticket (#7202), but it's probably worth repeating here in case anyone has any suggestions or ideas:
I think we really should consider a proof-of-work field on the client's CREATE cell, so we have some form of response available in the event of circuit-based CPU DoSes against Tor relays.
The first thing that came to my mind was a hash of the entire cell contents plus timestamp and a nonce field. The nonce would be repeatedly altered until the hash prefix matched a value specified by the consensus.
Then, if a relay receives a cell with either a stale timestamp or an invalid hash+hash prefix, that cell can be discarded. If the relay's onionskin queue is full, it can do some additional accounting such as ensuring that the hash values it sees are unique within the valid timestamp window. Or perhaps it only needs to inspect this portion of the cell at all if the onionskin queue is full.
As I see it, there are two serious hurdles to this approach:
1. The timesource of the timestamp needs to not be fingerprintable, yet still current enough to allow us to feasibly track a list of these hashes if we want. The client's consensus timestamp is the most readily available source of such a timestamp, but it would have to be fuzzed to account for the fact that only 1/4 of all clients will posses a consensus for that hour.
Do we have other timesources? I seem to recall seeing some tickets about dirport timesources fly by... Did we ever figure out a way forward for those?
2. If we're not careful, the hash prefix consensus parameter would cause most clients to be unable to build circuits until they download the new consensus instructing them to do the proof of work. Relays would have to wait several consensus periods before enforcing the hash prefix.
Or, alternatively, are there other proof-of-work mechanisms we could use instead?