We are announcing Bazel support for developing and testing fuzz tests, with OSS-Fuzz integration, through the new rules_fuzzing Bazel library.
is an effective, well-known testing technique for finding security and stability bugs in software. But writing and testing fuzz tests can be tedious
. Developers typically need to:
- Implement a fuzz driver function, which exercises the API under test;
- Build the code with the proper instrumentation (such as Address Sanitizer);
- Link it with one of the available fuzzing engine libraries (libFuzzer, AFL++, Honggfuzz, etc.) that provide the core test generation logic;
- Run the fuzz test binary with the right set of flags (e.g., to specify corpora or dictionaries);
- Package the fuzz test and its resources for consumption by fuzzing infrastructures, such as OSS-Fuzz.
Unfortunately, build systems don't traditionally offer any support beyond the core primitives of producing executables, so projects adopting fuzzing often end up reimplementing fuzz test recipes.
We are pleased to announce that projects using Bazel can get advanced fuzzing support through the new rules_fuzzing extension library
. The new fuzzing rules take care of all the boilerplate needed to build and run fuzz tests. Developers simply write the fuzz driver code and define a build target for it (example driver
for RE2). Fuzz tests can be built and run using a number of fuzzing engines provided out-of-the-box, such as libFuzzer
, as well as sanitizers. The rule library also provides the ability to define additional fuzzing engines.
You can integrate the fuzzing library with around 10 LOC
in your Bazel WORKSPACE file. Defining a fuzz test in Bazel is as easy as writing the following in your BUILD file:
name = "my_fuzz_test",
srcs = ["my_fuzz_test.cc"],
deps = [":my_library"],
You can easily test the fuzzer locally by invoking its launcher:$ bazel run --config=asan-libfuzzer //:my_fuzz_test_run
To improve the effectiveness of test case generation, fuzz tests also support seed corpora and dictionaries, through additional rule attributes. They will automatically be validated and included in fuzz test runs. Fuzz tests also serve as regression tests on the seed corpus. For example, you can add previously found and fixed crashes to the corpus and have them tested in your CI workflows:$ bazel test --config=asan-replay //:my_fuzz_test
The fuzzing rules provide built-in support for OSS-Fuzz
, our continuous fuzzing service for open source projects. The OSS-Fuzz support
drastically simplifies writing the build scripts
in project integration by automatically packaging the fuzz test and its dependencies using the expected OSS-Fuzz structure.
The Envoy Proxy project
is one of the early adopters of the fuzzing rules library. As a large, mature C++ codebase, Envoy has maintained its own custom implementation of fuzzing support for its over 50 fuzz targets written so far. By switching to the new Bazel fuzzing rules, Envoy's fuzz targets automatically gained new features, such as local running and testing tools and support for multiple fuzzing engines. At the same time, Envoy simplified its OSS-Fuzz integration scripts
. Moreover, it will automatically gain future functionality (e.g., more effective fuzzing engines, better coverage tracking, improved corpus management) as the Bazel fuzzing rules library evolves.
The Bazel rules for fuzzing
draw from Google's experience providing effective fuzzing tools to our internal developers. We hope the new Bazel support for fuzzing will lower the barrier to fuzzing adoption in open source communities, further increasing the security and reliability of many projects. To learn more about integrating the fuzzing rules into your project, take a look at the Getting Started section
in the documentation.By Stefan Bucur, Software Analysis, Asra Ali, Envoy, and Abhishek Arya, OSS-Fuzz – Google