Jazzer

Fuzz Testing for the JVM

Website | Blog | Twitter

Jazzer is a coverage-guided, in-process fuzzer for the JVM platform developed by Code Intelligence. It is based on libFuzzer and brings many of its instrumentation-powered mutation features to the JVM.

Jazzer currently supports the following platforms:

• Linux x86_64
• macOS 12+ x86_64 & arm64
• Windows x86_64

The following steps assume that JUnit 5.9.0 or higher is set up for your project, for example based on the official junit5-samples.

1. Add a dependency on com.code-intelligence:jazzer-junit:<latest version>. All Jazzer Maven artifacts are signed with this key.
2. Add a new fuzz test to a new or existing test class: a method annotated with @FuzzTest and at least one parameter. Using a single parameter of type FuzzedDataProvider, which provides utility functions to produce commonly used Java values, or byte[] is recommended for optimal performance and reproducibility of findings.
3. Assuming your test class is called com.example.MyFuzzTests, create the inputs directory src/test/resources/com/example/MyFuzzTestsInputs.
4. Run a fuzz test with the environment variable JAZZER_FUZZ set to 1 to let the fuzzer rapidly try new sets of arguments. If the fuzzer finds arguments that make your fuzz test fail or even trigger a security issue, it will store them in the inputs directory. In this mode, only a single fuzz test is executed per test run (see #599 for details).
5. Run the fuzz test without JAZZER_FUZZ set to execute it only on the inputs in the inputs directory. This mode, which behaves just like a traditional unit test, ensures that issues previously found by the fuzzer remain fixed and can also be used to debug the fuzz test on individual inputs.

A simple property-based fuzz test could look like this (excluding imports):

class ParserTests {
   @Test
   void unitTest() {
      assertEquals("foobar", SomeScheme.decode(SomeScheme.encode("foobar")));
   }

   @FuzzTest
   void fuzzTest(FuzzedDataProvider data) {
      String input = data.consumeRemainingAsString();
      assertEquals(input, SomeScheme.decode(SomeScheme.encode(input)));
   }
}

A complete Maven example project can be found in examples/junit.

You can also use GitHub release archives to run a standalone Jazzer binary that starts its own JVM configured for fuzzing:

1. Download and extract the latest release from the GitHub releases page.
2. Add a new class to your project with a public static void fuzzerTestOneInput(FuzzedDataProvider data) method.
3. Compile your fuzz test with jazzer_standalone.jar on the classpath.
4. Run the jazzer binary (jazzer.exe on Windows), specifying the classpath and fuzz test class:

./jazzer --cp=<classpath> --target_class=<fuzz test class>

If you see an error saying that libjvm.so has not been found, make sure that JAVA_HOME points to a JDK.

The examples directory includes both toy and real-world examples of fuzz tests.

Support for Jazzer is available in rules_fuzzing, the official Bazel rules for fuzzing. See the README for instructions on how to use Jazzer in a Java Bazel project.

Code Intelligence and Google have teamed up to bring support for Java, Kotlin, and other JVM-based languages to OSS-Fuzz, Google's project for large-scale fuzzing of open-source software. Read the OSS-Fuzz guide to learn how to set up a Java project.

Note: Open source projects can use Jazzer for free and benefit from the OSS-Fuzz infrastructure, including ClusterFuzzLite and OSS-Fuzz-Gen for automated analysis and continuous integration. There is no risk of accidental license violation as long as Jazzer is used for testing open-source code.

Information on building and testing Jazzer for development can be found in CONTRIBUTING.md

• Common options and workflows
• Advanced techniques

A list of security issues and bugs found by Jazzer is maintained here. If you found something interesting and the information is public, please send a PR to add it to the list.

By using Jazzer, you agree to the Code Intelligence Terms & Conditions.

Jazzer can only be used for code bases that are released under an OSI-approved open source license, or to perform academic research. It cannot be used to fuzz projects for or during automated analysis, continuous integration or continuous delivery, whether as part of normal software engineering processes or otherwise. For these use cases, get in contact with us via sales@code-intelligence.com.

You can find the latest version released under the Apache 2 license here.

The following developers have contributed to Jazzer before its public release:

Sergej Dechand, Christian Hartlage, Fabian Meumertzheim, Sebastian Pöplau, Mohammed Qasem, Simon Resch, Henrik Schnor, Khaled Yakdan

The LLVM-style edge coverage instrumentation for JVM bytecode used by Jazzer relies on JaCoCo. Previously, Jazzer used AFL-style coverage instrumentation as pioneered by kelinci.