Skip to content

Latest commit

 

History

History
79 lines (61 loc) · 2.51 KB

README.md

File metadata and controls

79 lines (61 loc) · 2.51 KB
Raw

Auto-Generating Fuzzers and Corpus from Unittests

We decided to use the unittests (which show how to test each of the parsers, and even provides some interesting input cases) to auto-generate both the fuzzers (the cpp files that contain the fuzzing glue) and the corpus.

We provide a script (converter.py) that parses unittests in order to auto-generate both the fuzzers (in the libfuzzer's case this means the LLVMFuzzerTestOneInput() function), and the corpus used as fuzzing seed. The idea is to allow the user to signal the parts of the unittest that are useful for fuzzing, including code that needs to be fuzzed (in order to generate fuzzers), and the unittest buffers (which can be used for the corpus).

  • In order to signal the code that needs to be fuzzed, the unittest writer can use the following syntax:
$ cat test/h265_aud_parser_unittest.cc
...
  // fuzzer::conv: begin
  auto aud = H265AudParser::ParseAud(buffer, arraysize(buffer));
  // fuzzer::conv: end
...

This tells the autogen script to put all the code between the "fuzzer::conv: begin" and "fuzzer::conv: end" tags into a fuzzer function (libfuzzer's LLVMFuzzerTestOneInput() function). The script does a couple of other fixes, including copying the include files, fixing the parameter names, and doing some minor namespace adjustments. The latter should be formalizable if there is interest in reusing the infra for another project.

  • In order to signal the buffer(s) that can be used as seed, the unittest writer can use the following syntax:
$ cat test/h265_vps_parser_unittest.cc
...
  // fuzzer::conv: data
  const uint8_t buffer[] = {
      0x0c, 0x01, 0xff, 0xff, 0x01, 0x60, 0x00, 0x00,
      0x03, 0x00, 0xb0, 0x00, 0x00, 0x03, 0x00, 0x00,
      0x03, 0x00, 0x5d, 0xac, 0x59, 0x00
  };
...

The script will look for C++-defined buffers, extract all the hexadecimal constants, and put them into a binary file.

Operation

(1) In order to re-generate a fuzzing files, use the converter script:

$ ./converter.py ../test/h265_vps_parser_unittest.cc ./

The script will parse the input file (../test/h265_vps_parser_unittest.cc) and generate a similarly-named fuzzer (h265_vps_parser_fuzzer.cc) in the output directory (./). It will also generate some cases in the per-fuzzer corpus directory (corpus/h265_vps_parser_fuzzer/unittest.*.bin).

(2) In order to run the fuzzers, use:

$ cd fuzz
$ make fuzz
...

This will run each of the generated fuzzers a fixed number of times, and produce new cases in the per-fuzzer corpus directory.