Jef 20d28e80a5 Initial community commit 1 ヶ月 前
..
all_formats.dict 20d28e80a5 Initial community commit 1 ヶ月 前
build.sh 20d28e80a5 Initial community commit 1 ヶ月 前
fuzz-main.sh 20d28e80a5 Initial community commit 1 ヶ月 前
fuzz-secondary1.sh 20d28e80a5 Initial community commit 1 ヶ月 前
fuzz-secondary2.sh 20d28e80a5 Initial community commit 1 ヶ月 前
fuzz-settings.sh 20d28e80a5 Initial community commit 1 ヶ月 前
fuzz.c 20d28e80a5 Initial community commit 1 ヶ月 前
get-afl.sh 20d28e80a5 Initial community commit 1 ヶ月 前
readme.md 20d28e80a5 Initial community commit 1 ヶ月 前

readme.md

libopenmpt fuzz suite

In this directory, you can find the necessary tools for fuzzing libopenmpt with the American Fuzzy Lop fuzzer (afl++).

Contents:

  • all_formats.dict: A dictionary containing magic bytes from all supported module formats to make the life of the fuzzer a bit easier.
  • fuzz-main.sh: Script to launch the main fuzzing process. If you want to use just one fuzzer instance, run this one.
  • fuzz-secondary[1|2].sh: Scripts to launch the secondary fuzzing process. It is recommended to run at least two fuzzer instances, as the deterministic and random fuzz mode have been found to complement each other really well. The two scripts are set up to use different exploration strategies
  • fuzz-settings.sh: Set up your preferences and afl settings here before the first run.
  • fuzz.c: A tiny C program that is used by the fuzzer to test libopenmpt.
  • get-afl.sh: A simple script to obtain the latest version of afl++. You can also make it download from a specific branch or tag, e.g. GET_AFL_VERSION=stable ./get-afl.sh to download the latest stable but unreleased code.

Prerequisites

  • afl++ - the makefile expects this to be installed in contrib/fuzzing/afl, as it is automatically done by the get-afl.sh install script.
  • Clang with LLVM dev headers (llvm-config needs to be installed). afl also works with gcc, but our makefile has been set up to make use of afl's faster LLVM-LTO mode.

How to use

  • Run get-afl.sh, or manually extract afl to contrib/fuzzing/afl, use make source-only to build. If building fails because llvm-config cannot be found, try prepending LLVM_CONFIG=/usr/bin/llvm-config-12 or similar, and read the afl manual.
  • Build libopenmpt with the build.sh script in this directory.
  • Set up fuzz-settings.sh to your taste. Most importantly, you will have to specify the input directory for first use. The default setup mounts a tmpfs folder for all temporary files. You may change this behaviour if you do not have root privileges.
  • Run fuzz-main.sh for the first (deterministic) instance of afl-fuzz.
  • For a "secondary" instance to run on another core, run fuzz-secondary1.sh and/or fuzz-secondary2.sh.
  • If you want to make use of even more cores, create more copies fuzz-secondary2.sh and adjust "infile03" / "fuzzer03" to "infile04" / "fuzzer04" and so o (they need to be unique). Try variying the fuzzing strategey (the -p parameter) to get results more quickly.