Daniela Lopes (INESC-ID / IST, Universidade de Lisboa), Jin-Dong Dong (Carnegie Mellon University), Pedro Medeiros (INESC-ID / IST, Universidade de Lisboa), Daniel Castro (INESC-ID / IST, Universidade de Lisboa), Diogo Barradas (University of Waterloo), Bernardo Portela (INESC TEC / Universidade do Porto), João Vinagre (INESC TEC / Universidade do Porto), Bernardo Ferreira (LASIGE, Faculdade de Ciências, Universidade de Lisboa), Nicolas Christin (Carnegie Mellon University), Nuno Santos (INESC-ID / IST, Universidade de Lisboa)
Tor is one of the most popular anonymity networks in use today. Its ability to defend against flow correlation attacks is essential for providing strong anonymity guarantees. However, the feasibility of flow correlation attacks against Tor onion services (formerly known as "hidden services") has remained an open challenge. In this paper, we present an effective flow correlation attack that can deanonymize onion service sessions in the Tor network. Our attack is based on a novel distributed technique named Sliding Subset Sum (SUMo), which can be deployed by a group of colluding ISPs worldwide in a federated fashion. These ISPs collect Tor traffic at multiple vantage points in the network, and analyze it through a pipelined architecture based on machine learning classifiers and a novel similarity function based on the classic subset sum decision problem. These classifiers enable SUMo to deanonymize onion service sessions effectively and efficiently. We also analyze possible countermeasures that the Tor community can adopt to hinder the efficacy of these attacks.