The MirageOS Blogon building functional operating systems


Introducing transport layer security (TLS) in pure OCaml

By Hannes Mehnert, David Kaloper

We announce a beta release of ocaml-tls, a clean-slate implementation of Transport Layer Security (TLS) in OCaml.

What is TLS?

Transport Layer Security (TLS) is probably the most widely deployed security protocol on the Internet. It provides communication privacy to prevent eavesdropping, tampering, and message forgery. Furthermore, it optionally provides authentication of the involved endpoints. TLS is commonly deployed for securing web services (HTTPS), emails, virtual private networks, and wireless networks.

TLS uses asymmetric cryptography to exchange a symmetric key, and optionally authenticate (using X.509) either or both endpoints. It provides algorithmic agility, which means that the key exchange method, symmetric encryption algorithm, and hash algorithm are negotiated.

TLS in OCaml

Our implementation ocaml-tls is already able to interoperate with existing TLS implementations, and supports several important TLS extensions such as server name indication (RFC4366, enabling virtual hosting) and secure renegotiation (RFC5746).

Our demonstration server runs ocaml-tls and renders exchanged TLS messages in nearly real time by receiving a trace of the TLS session setup. If you encounter any problems, please give us feedback.

ocaml-tls and all dependent libraries are available via OPAM (opam install tls). The source is available under a BSD license. We are primarily working towards completeness of protocol features, such as client authentication, session resumption, elliptic curve and GCM cipher suites, and have not yet optimised for performance.

ocaml-tls depends on the following independent libraries: ocaml-nocrypto implements the cryptographic primitives, ocaml-asn1-combinators provides ASN.1 parsers/unparsers, and ocaml-x509 implements the X509 grammar and certificate validation (RFC5280). ocaml-tls implements TLS (1.0, 1.1 and 1.2; RFC2246, RFC4346, RFC5246).

We invite the community to audit and run our code, and we are particularly interested in discussion of our APIs. Please use the mirage-devel mailing list for discussions.

Please be aware that this release is a beta and is missing external code audits. It is not yet intended for use in any security critical applications.

In our issue tracker we transparently document known attacks against TLS and our mitigations (checked and unchecked). We have not yet implemented mitigations against either the Lucky13 timing attack or traffic analysis (e.g. length-hiding padding).

Trusted code base

Designed to run on Mirage, the trusted code base of ocaml-tls is small. It includes the libraries already mentioned, `ocaml-tls`, `ocaml-asn-combinators`, `ocaml-x509`, and `ocaml-nocrypto` (which uses C implementations of block ciphers and hash algorithms). For arbitrary precision integers needed in asymmetric cryptography, we rely on `zarith`, which wraps `libgmp`. As underlying byte array structure we use `cstruct` (which uses OCaml Bigarray as storage).

We should also mention the OCaml runtime, the OCaml compiler, the operating system on which the source is compiled and the binary is executed, as well as the underlying hardware. Two effectful frontends for the pure TLS core are implemented, dealing with side-effects such as reading and writing from the network: Lwt_unix and Mirage, so applications can run directly as a Xen unikernel.

Why a new TLS implementation?

Update: Thanks to Frama-C guys for pointing out that CVE-2014-1266 and CVE-2014-0224 are not memory safety issues, but logic errors. This article previously stated otherwise.

There are only a few TLS implementations publicly available and most programming languages bind to OpenSSL, an open source implementation written in C. There are valid reasons to interface with an existing TLS library, rather than developing one from scratch, including protocol complexity and compatibility with different TLS versions and implementations. But from our perspective the disadvantage of most existing libraries is that they are written in C, leading to:

  • Memory safety issues, as recently observed by Heartbleed and GnuTLS session identifier memory corruption (CVE-2014-3466) bugs;
  • Control flow complexity (Apple's goto fail, CVE-2014-1266);
  • And difficulty in encoding state machines (OpenSSL change cipher suite attack, CVE-2014-0224).

Our main reasons for ocaml-tls are that OCaml is a modern functional language, which allows concise and declarative descriptions of the complex protocol logic and provides type safety and memory safety to help guard against programming errors. Its functional nature is extensively employed in our code: the core of the protocol is written in purely functional style, without any side effects.

Subsequent blog posts over the coming days will examine in more detail the design and implementation of the four libraries, as well as the security trade-offs and some TLS attacks and our mitigations against them. For now though, we invite you to try out our demonstration server running our stack over HTTPS. We're particularly interested in feedback on our issue tracker about clients that fail to connect, and any queries from anyone reviewing the source code of the constituent libraries.

Posts in this TLS series: