sci-apps-fuzz

AFL++ fuzz harnesses + utils for scientific applications

Prerequisites

Installed on the host machine:

• bash
• docker
• docker buildx

You can use Docker's remote contexts feature to delegate fuzzing workloads to other machines.

Usage

1. Make a directory to store fuzzer data. This directory will grow large, so place it on an appropriate filesystem.

   Set the variable

   export SHARED_FUZZ_FS=/path/to/my/directory

   Note that when using a remote Docker context, this path is a remote path.

2. Use the convenience script to build and run the container with Docker:

   ./run_with_docker.sh $APPLICATION_NAME $CONTEXT $CPUS

   where $APPLICATION_NAME is the name of a target in src/targets/, $CONTEXT is the name of the Docker context you wish to use (default, if you did not configure a remote context), and $CPUS is the set or range of CPUs to which to limit the container (eg. 0, 3,4, 16-31, etc.).

3. Inside the container, use just to invoke the recipes in the justfile. Remember to keep CPU, memory, and disk usage under control. You can safely background the watcher interface with ^Z, and resume with fg.

   For more details, see the appendix of our SC'26 workshop paper.