Copy initial testcases to testcases-directory.
* If you don't have any, run get_testcases_from_github_issues.py. It will fetch testcases (markdown code blocks) from MRuby Github issues.
Run ./runme.sh to start fuzzing.
The script will build the container, configure required host machine values and launch fuzzer container.
The fuzzer will load initial testcases from the host-machine testcases-directory and will save all output (incl. fuzzer binary) to host-machine /dev/shm directory (ramdisk).
When the fuzzer is running, container shows a status screen (afl-whatsup) which has basic info about the AFL-fuzzers.
New crashes are triaged couple times per minute and saved to crashwalk database. A text-based log of unique crashes can be found in results-directory.
Optionally for additional triaging, run docker exec -it CONTAINER_ID /mruby/bin/triage_online.sh.
This will submit all deduplicated crashes from crashwalk database to online sandbox (https://mruby.science/runs).
If the online sandbox fails execution (NO RESULT / NO MEMORY / NO INSTRUCTIONS), the testcase most likely crashes mruby universally and is not a false positive.
TODO
Add support for locating the commit that introduced a crash (git bisect?).
Automatically create Markdown-reports that contain (modify crashwalk?):
Testcase (base64-encoded).
Crashwalk trace.
Commit that introduced the crash.
???
Add support for deduplicating crashes (search existing issue from Github).