Implementing and Evaluating Quantum-Safe Fully Encrypted Protocols

The Tor network enables anonymous communication by routing traffic through volunteer-operated relays, but access to the network is often restricted through censorship that targets identifiable protocol patterns. In order to counteract such restrictions, protocols must resist fingerprinting, for example, by obfuscating their traffic to appear like random bytes on the wire. One protocol targeting exactly this property is the obfs4 protocol. In this thesis, we build on the Tor Project’s lyrebird repository implementing obfs4. Later, the obfs4 protocol was extended by the obfuscated key exchange protocol drivel with post-quantum security and stronger obfuscation. We implement drivel and integrate it into the lyrebird repository. As drivel requires ciphertexts to look random, we analyze the Classic-McEliece and HQC key encapsulation mechanisms, defining mappings from ciphertexts to random byte strings. These schemes are then integrated with our implementation, in addition to the previously examined ML-KEM and FrodoKEM schemes. Finally, we empirically evaluate our implementation using benchmarks and local deployments and suggest concrete changes to drivel to reduce identifiable traffic patterns in the handshake and increase performance.