Link Search Menu Expand Document

Reproducing OSS-Fuzz issues

You’ve been CCed on an OSS-Fuzz issue (examples). Now what? Before attempting to fix the bug, you should be able to reliably reproduce it.


Fuzz target bugs

Every issue has a reproducer file (also know as a “testcase” file) attached. Download it. This file contains the bytes that were fed to the fuzz target.

Note: If the issue is not public, you will need to login using a Google account (why?) that the bug report CCs.

If you have already integrated the fuzz target with your build and test system, all you have to do is run this command:

$ ./fuzz_target_binary <testcase_path>

For timeout bugs, add the -timeout=25 argument. For OOM bugs, add the -rss_limit_mb=2560 argument. Read more on how timeouts and OOMs are handled.

Depending on the nature of the bug, the fuzz target binary needs to be built with the appropriate sanitizer (for example, if it’s a buffer overflow, build with AddressSanitizer).

If you’re not sure how to build the fuzzer using the project’s build system, you can also use Docker commands to replicate the exact build steps used by OSS-Fuzz, then feed the reproducer input to the fuzz target (how?, why?).

Building using Docker

Pull the latest Docker images

Docker images get regularly updated with a newer version of build tools, build configurations, scripts, and other changes. In some cases, a particular issue can be reproduced only with a fresh image being used. Pull the latest images by running the following command:

$ python infra/helper.py pull_images

Build the image and the fuzzers

Run the following commands:

$ python infra/helper.py build_image $PROJECT_NAME
$ python infra/helper.py build_fuzzers --sanitizer <address/memory/undefined> \
    --architecture <x86_64/i386> $PROJECT_NAME

The sanitizer used in the report is the value in the Sanitizer column. It’s one of the following:

  • address for AddressSanitizer.
  • memory for MemorySanitizer.
  • undefined for UndefinedBehaviorSanitizer.

Note: The architecture argument is only necessary if you want to specify i386 configuration.

Reproducing bugs

After you build an image and a fuzzer, you can reproduce a bug by running the following command:

$ python infra/helper.py reproduce $PROJECT_NAME <fuzz_target_name> <testcase_path>

For example, to build the libxml2 project with UndefinedBehaviorSanitizer (undefined) instrumentation and reproduce a crash testcase for a fuzzer named libxml2_xml_read_memory_fuzzer, you would run:

$ python infra/helper.py build_image libxml2
$ python infra/helper.py build_fuzzers --sanitizer undefined libxml2
$ python infra/helper.py reproduce libxml2 libxml2_xml_read_memory_fuzzer ~/Downloads/testcase

Reproduce using local source checkout

You can also mount local sources into the running container by using these commands:

$ python infra/helper.py build_fuzzers \
    --sanitizer <address/memory/undefined> $PROJECT_NAME <source_path>
$ python infra/helper.py reproduce $PROJECT_NAME <fuzz_target_name> <testcase_path>

Once you reproduce the bug, you can do the following:

  • Fix issue: Write a patch to fix the issue in your local checkout, then use the previous command to verify the fix (i.e. no crash occurred). Use gdb if needed.
  • Submit fix: Submit the fix in the project’s repository. ClusterFuzz will automatically pick up the changes, recheck the testcase, and close the issue (in < 1 day).
  • Improve fuzzing support: Consider improving your integration with OSS-Fuzz.

Reproducing build failures

Our infrastructure runs some sanity tests to make sure that your build was correctly configured, even if it succeeded. To reproduce these locally, run these commands:

$ python infra/helper.py build_image $PROJECT_NAME
$ python infra/helper.py build_fuzzers --sanitizer <address/memory/undefined> \
    --engine <libfuzzer/afl/honggfuzz> --architecture <x86_64/i386> $PROJECT_NAME
$ python infra/helper.py check_build  --sanitizer <address/memory/undefined> \
    --engine <libfuzzer/afl/honggfuzz> --architecture <x86_64/i386> $PROJECT_NAME \
    <fuzz_target_name>

Note: Unless you have a reason to think the build is an i386 build, the build is probably an x86_64 build and the architecture argument can be omitted.

If you need to reproduce a coverage build failure, follow the Code Coverage page to build your project and generate a code coverage report.