README.md

afl fuzzing demos

afl or american fuzzy lop has been getting a lot of buzz recently. So let's have a go at fuzzing things with afl and python-afl

Installing Fuzzing Tools

• create a working directory and python virtual env
• afl
  • http://lcamtuf.coredump.cx/afl/
  • Download and unpack
  • follow instructions in docs/INSTALL
• python-afl
  • pip install cython
  • pip install python-afl

Approach

Usage of afl (without python-afl) (docs):

• afl-fuzz -i testcase_dir -o findings_dir /path/to/program [...params...]

Note: upon first run, you may be advised to take certain actions to ensure afl's performance is reasonable

Usage of python-afl (docs):

• Add this code (ideally, after all other modules are already imported) to the target program:

    import afl
    afl.start()

• Use py-afl-fuzz instead of afl-fuzz:

  `$ py-afl-fuzz [options] -- /path/to/fuzzed/python/script [...]`compound
  

Fuzzing Python's JSON library

assuming a layout:

python-afl/
└── py-afl-fuzz
afl-fuzzing-demos/
├── json
│   ├── afl_findings/
│   ├── afl_testcases/
│   ├── demo.py
│   └── README.md
└── README.md

• cd afl-fuzzing-demos
• ../python-afl/py-afl-fuzz -i json/afl_testcases/ -o json/afl_findings/ -- python json/demo.py

I'm using some very basic testcases to get the fuzzer started, e.g. {"a":true} & {"1":2}, this should start it off with basic json syntax and it'll mutate from there.

Note: afl makes constant reads & writes to the file system, so if you're running on an SSD it's a good idea to replace the output directory with an in-memory RAM disk folder. E.g. on debian based systems, the shared memory directory /run/shm/. But obviously remember, the results won't survive a reboot, so copying the results out like this might be best:

mkdir -p /run/shm/json/afl_findings

../python-afl/py-afl-fuzz \
    -i json/afl_testcases/ \
    -o /run/shm/json/afl_findings \
    -- python json/demo.py; cp -r /run/shm/json/afl_findings json/

When you're finished, Ctrl-C and see what you come up with in json/afl_findings. You'll start to see output after a few seconds, but it's usual to leave a fuzzer running for a few days or more.

Console after all night fuzzing

bidict

fuzzing by script

mkdir -p /run/shm/bidict/afl_findings

../python-afl/py-afl-fuzz \
    -i bidict/afl_testcases/ \
    -o /run/shm/bidict/afl_findings \
    -- python bidict/fuzz_script.py; cp -r /run/shm/bidict/afl_findings bidict/

try running outputs against script:

for f in bidict/afl_findings/hangs/id*; do cat $f | python bidict/fuzz_script.py; done