Towards a Provably Secure Implementation of TLS 1.3

We report ongoing work towards a verified reference implementation of TLS 1.3 in the F* programming language. Our code supports selected ciphersuites for all versions of TLS from 1.0 to 1.3. It is being developed on http://github.com/mitls as the next version of MITLS, written in F instead of F#. We intend to prove a strong, joint cryptographic security theorem for TLS 1.3 clients and servers, even when they run alongside older versions of the protocol with weaker security. Our verification approach adapts and extends that of MITLS in several ways: (1) we adopt an idiomatic stateful style that promises higher performance than the pure functional style of MITLS; (2) we seek to prove stronger security properties with fewer ad hoc assumptions; (3) we rely on advanced F type inference to substantially reduce both the code base and the typed proof annotations. This paper describes our implementation architecture, our new composite state machine for TLS 1.0–1.3, and our target security theorem. By the time of the TRON workshop, we will present concrete verification results for our implementation. As far as we are aware, these will be the first cryptographic proofs for an implementation of TLS 1.3, and a fortiori for an implementation that is backward compatible with older versions of the protocol. We anticipate that our effort will identify early problems and offer implementation guidelines for other TLS libraries.