Messaging Layer Security Infrastructure Prototype

This repository contains the code for a Rust implementation of the Phoenix Protocol, an infrastructure protocol built around the Messaging Layer Security (MLS) group messaging protocol.

The Phoenix Protocol aims to enable functionality commonly required by messaging applications while providing strong security and privacy guarantees. It allows federation between different servers and their connected clients.

For security, the protocol relies mainly on the strong security guarantees provided by the underlying MLS protocol. The authentication service required by the MLS protocol is a simple signature-based PKI.

The documentation including the full specification for the Phoenix Protocol can be found here.

Code structure

The implementation spans client, server and test-specific components. It is split across multiple crates:

• types: Structs, Enums and utilities used by multiple other crates in this repository. This crate allows us to compile client and server components separately.
• backend: Implements both the local and the federation part of the protocol logic on the server side. Inspired by the type-based verification design of the OpenMLS crate (on which both coreclient and backend are built), verification of incoming messages is enforced using Rust's type system. The backend is written to work with arbitrary storage providers. The crate itself provides an in-memory and a Postgres-based storage provider.
• server: The server component that makes the logic implemented in the backend available to clients via a REST API. Beside the REST API, the server also supports web-sockets, which the backend can use to notify clients of new messages. The server crate can be compiled into a binary that can be run to expose the HTTP endpoints. A Dockerfile is available to build a Docker image that contains the server binary.
• coreclient: Implements the protocol logic of the client component. The coreclient stores and manages a user's contacts, conversations, as well as the underlying MLS groups. It provides a high-level API to make use of the protocol in the context of a messaging application. Just like the backend, the coreclient uses a type-based message verification approach. The crate can be used to instantiate and persist (via Sqlite) multiple clients parallely.
• apiclient: A shallow layer that the coreclient calls to interact with the backend via the server using HTTP(s).
• applogic: A thin layer for initial testing of higher-level messaging concepts.
• test_harness: Exclusively used for testing. The test_harness contains a test framework to conduct integration tests and can be compiled into a binary for docker-based testing of the protocol's federation capabilities.

Docker-based federation testing

The Phoenix Protocol allows for communication between users across different servers. To properly test federation-related functionalities, the test_harness provides utilities to spin up multiple servers using Docker. Docker allows us to approximate a real-world networking environments, where servers can discover one-another using DNS and facilitate communication between their respective clients across a network. The server crate contains several federated test scenarios that are run in this way.

Since the tests build fresh Docker images from the code and spin up multiple containers, running them is somewhat slow. To make regular testing more ergonomic, the --include-ignored flag has to be used when running cargo test on the server crate. Note that Docker-based tests are run as part of the CI whenever a pull request is made.

Licensing

All crates in this repository are licensed under the AGPL 3.0. This README file is licensed under CC-BY 4.0.