Fuzzing Rust code with american fuzzy lop (AFL)

Screen recording of AFL running on Rust code. The code under test is examples/hello.rs in this repository.

Fuzz testing is a software testing technique used to find security and stability issues by providing pseudo-random data as input to the software. American fuzzy lop is a popular, effective, and modern fuzz testing tool. This library, afl.rs, allows one to run AFL on code written in the Rust programming language.

• Nightly build of Rust from any time after January 24, 2016 (this issue prevented compatibility with previous builds of Rust)
• C++ compiler that supports C++11
• afl.rs needs to compile against a version of LLVM that matches rustc's. The easy solution (if you can wait on a slow build) is to build rustc from source and put it in your PATH. Then afl.rs's build script will find llvm-config automatically. Otherwise, the environment variable LLVM_CONFIG should hold the path to llvm-config when you build afl.rs.

Because of these relatively strict requirements, there is a Vagrantfile provided that assists in bootstraping an afl.rs compatible environment. View the README in the vagrant/ directory for more information.

First, add this project as a Cargo dependency:

[dependencies]
afl = "0.1"
afl-plugin = "0.1"

Then you can add afl instrumentation to one or more crates:

#![feature(plugin)]
#![plugin(afl_plugin)]

You will also need a test executable that exercises the instrumented functions, in a deterministic way based on input from stdin. This executable should link the afl run-time library:

extern crate afl;

This will produce a binary that you can pass to afl-fuzz in the usual manner. afl instrumentation adds some run-time overhead, so it's a good candidate for conditional compilation, perhaps through a Cargo feature:

# You may need to add `optional = true` to the above dependencies.
[features]
afl = ["afl-plugin", "afl"]

// Active only with `cargo [...] --feature afl`
#![cfg_attr(feature = "afl", feature(plugin))]
#![cfg_attr(feature = "afl", plugin(afl_plugin))]

C++ code will be compiled by default with g++, though one can specify a different C++ compiler by setting the CXX environment variable to point to a different compiler binary.

To look for logic errors in safe Rust code, use the no-landing-pads rustc flag upon compilation of the AFL entrypoint. This causes the fuzzer to treat any Rust panic as a crash. Examples of usage:

• rustc -Z no-landing-pads
• cargo rustc -- -Z no-landing-pads

If your program has a slow set-up phase that does not depend on the input data, you can set AFL_DEFER_FORKSRV=1 for a substantial speed-up, provided that you insert a call to afl::init() after the set-up and before any dependence on input. There are various other caveats, described in the section "Bonus feature: deferred instrumentation" in llvm_mode/README.llvm distributed with afl. See also examples/deferred-init.rs in this repository.

See the afl documentation for other configuration variables. Some of these are set at compile time in config.h. For the most part they only affect afl-fuzz itself, and will work fine with this library. However, if you change SHM_ENV_VAR, MAP_SIZE, or FORKSRV_FD, you should update this library's src/config.h to match.

• brotli-rs: #2, #3, #4, #5, #6, #7, #8, #9, #10, #11, #12
• flac: #3
• httparse: #9
• image: #414, #473, #474, #477
• mp4parse-rust: #2, #4, #5, #6
• rustc: #24275, #24276
• rust-url: #108
• regex: #84
• rust-asn1: #32
• rustc-serialize: #109, #110
• serde: #75, #77, #82
• tar-rs: #23
• xml-rs: #93
• Logic errors in tendril and its html5ever integration

These bugs aren't nearly as serious as the memory-safety issues afl has discovered in C and C++ projects. That's because Rust is memory-safe by default, but also because not many people have tried afl.rs yet! Over time we will update this section with the most interesting bugs, whether they're logic errors or memory-safety problems arising from unsafe code. Pull requests are welcome!