Fuzzer for libhevc decoder and encoder

This describes steps to build hevc_dec_fuzzer and hevc_enc_fuzzer binary.

Linux x86/x64

Requirements

• cmake (3.9.1 or above)
• make
• clang (6.0 or above) needs to support -fsanitize=fuzzer, -fsanitize=fuzzer-no-link

Steps to build

Clone libhevc repository

$ git clone https://android.googlesource.com/platform/external/libhevc

Create a directory inside libhevc and change directory

 $ cd libhevc
 $ mkdir build
 $ cd build

Build fuzzer with required sanitizers (-DSANITIZE=fuzzer-no-link is mandatory to enable fuzzers)

 $ cmake .. -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \
   -DCMAKE_BUILD_TYPE=Debug -DSANITIZE=fuzzer-no-link,address
 $ make

Steps to run

Create a directory CORPUS_DIR and copy some elementary hevc files (for hevc_dec_fuzzer) or yuv files (for hevc_enc_fuzzer) to that directory

To run the fuzzers

$ ./hevc_dec_fuzzer CORPUS_DIR
$ ./hevc_enc_fuzzer CORPUS_DIR

Android

Steps to build

Build the fuzzers

  $ mm -j$(nproc) hevc_dec_fuzzer
  $ mm -j$(nproc) hevc_enc_fuzzer

Steps to run

Create a directory CORPUS_DIR and copy some elementary hevc files (for hevc_dec_fuzzer) or yuv files (for hevc_enc_fuzzer) to that folder Push this directory to device

To run hevc_dec_fuzzer on device

  $ adb sync data
  $ adb shell /data/fuzz/hevc_dec_fuzzer CORPUS_DIR

To run hevc_enc_fuzzer on device

  $ adb sync data
  $ adb shell /data/fuzz/arm64/hevc_enc_fuzzer/hevc_enc_fuzzer CORPUS_DIR

To run hevc_dec_fuzzer on host

  $ $ANDROID_HOST_OUT/fuzz/x86_64/hevc_dec_fuzzer/hevc_dec_fuzzer CORPUS_DIR

To run hevc_enc_fuzzer on host

  $ $ANDROID_HOST_OUT/fuzz/x86_64/hevc_enc_fuzzer/hevc_enc_fuzzer CORPUS_DIR

Appendix

libhevc encoder fuzzer

Plugin Design Considerations

The fuzzer plugin for HEVC is designed based on the understanding of the codec and tries to achieve the following:

Maximize code coverage

The configuration parameters are not hardcoded, but instead selected based on incoming data. This ensures more code paths are reached by the fuzzer.

HEVC supports the following parameters:

1. Frame Width (parameter name: i4_width)
2. Frame Height (parameter name: i4_height)
3. Intra max transform tree depth (parameter name: i4_max_tr_tree_depth_I)
4. Inter max transform tree depth (parameter name: i4_max_tr_tree_depth_nI)
5. CU level Qp modulation (parameter name: i4_cu_level_rc)
6. Rate control mode (parameter name: i4_rate_control_mode)
7. Frame level Qp (parameter name: ai4_frame_qp)
8. Encoder quality preset (parameter name: i4_quality_preset)
9. Target Bitrate (parameter name: ai4_tgt_bitrate)
10. Enable entropy sync (parameter name: i4_enable_entropy_sync)
11. Deblocking type (parameter name: i4_deblocking_type)
12. Default scaling matrices (parameter name: i4_use_default_sc_mtx)
13. Max temporal layers (parameter name: i4_max_temporal_layers)
14. Max difference b/w IDR frames (parameter name: i4_max_closed_gop_period)
15. Min difference b/w IDR frames (parameter name: i4_min_closed_gop_period)
16. Max difference b/w I frames (parameter name: i4_max_i_open_gop_period)
17. Max difference b/w CRA frames (parameter name: i4_max_cra_open_gop_period)
18. Automatic insertion of SPS at each CDR (parameter name: i4_sps_at_cdr_enable)
19. Enable VUI output (parameter name: i4_vui_enable)
20. Enable SEI messages (parameter name: i4_sei_enable_flag)
21. Architecture type (parameter name: e_arch_type)
22. Enable force IDR frame test(parameter name: mIsForceIdrEnabled)
23. Enable dynamic bitrate test (parameter name: mIsDynamicBitrateChangeEnabled)
24. Force IDR frame number (parameter name: mForceIdrInterval)
25. Frame number for dynamic bitrate (parameter name: mDynamicBitrateInterval)

This also ensures that the plugin is always deterministic for any given input.

Maximize utilization of input data

The plugin feeds the entire input data to the codec using a loop. If the encode operation was successful, the input is advanced by the frame size. If the encode operation was un-successful, the input is still advanced by frame size so that the fuzzer can proceed to feed the next frame.

This ensures that the plugin tolerates any kind of input (empty, huge, malformed, etc) and doesnt exit() on any input and thereby increasing the chance of identifying vulnerabilities.

References:

• http://llvm.org/docs/LibFuzzer.html
• https://github.com/google/oss-fuzz