srTLS: Secure TLS Handshake on Corrupted Machines

TLS 1.3 is widely used to realize secure communication over the Internet. Existing security analyses of TLS 1.3 primarily focus on its handshake protocol which is indeed an authenticated key exchange (AKE) protocol, and implicitly neglect the so-called subversion attacks (e.g., breaking TLS via Dual EC) in the real world. Reverse firewall (RF) is a prevalent approach to defend against subversion attack. To the best of our knowledge, the only two subversion-resilient AKE protocols with RFs are proposed by Dodis et al. (CRYPTO’16) and Bossuat et al. (ESORICS’20). The security of both protocols is proved under game-based model which is insufficient for the concurrent execution of multiple TLS instances in practice. In this paper, we propose $\mathsf {srTLS}$, a variant of the TLS 1.3 full one round-trip time (1-RTT) handshake protocol with RFs under the universally composable (UC) model. In particular, we first present the ideal functionality of unilateral AKE $\mathcal {F}_{\mathsf {uaKE}}$. Then, we use RFs with outer transparency to circumvent the difficulty in sanitizing the messages of handshake protocol, and prove that $\mathsf {srTLS}$ UC-realizes $\mathcal {F}_{\mathsf {uaKE}}$ in the presence of subversion attacks. Finally, we integrate $\mathsf {srTLS}$ and existing subversion-resilient AKE protocols into TLS 1.3. The evaluation result demonstrates that $\mathsf {srTLS}$ achieves at least a 44.86% efficiency improvement over other subversion-resilient AKE protocols.