1# Fuzzer for libgsm codec 2 3## Plugin Design Considerations 4The fuzzer plugin for GSM is designed based on the understanding of the 5codec and tries to achieve the following: 6 7##### Maximize code coverage 8The configuration parameters are not hardcoded, but instead selected based on 9incoming data. This ensures more code paths are reached by the fuzzer. 10 11GSM supports the following options: 121. Verbosity (parameter name: `GSM_VERBOSE`) 132. Fastness (parameter name: `GSM_OPT_FAST`) 143. WAV49 Format (parameter name: `GSM_OPT_WAV49`) 154. LTP Cut speed-up (encoder only) (parameter name: `GSM_OPT_LTP_CUT`) 16 17| Parameter| Valid Values| Configured Value| 18|------------- |-------------| ----- | 19| `GSM_VERBOSE` | 0. `Verbosity disabled` 1. `Verbosity enabled` | Bit 0 (LSB) of 1st byte of data. | 20| `GSM_OPT_FAST` | 0. `Regular algorithm will be used` 1. `Faster version of the algorithm will be used` | Bit 1 (LSB) of 1st byte of data. | 21| `GSM_OPT_LTP_CUT` | 0. `Disable LTP cut-off optimization` 1. `Enable LTP cut-off optimization` | Bit 2 (LSB) of 1st byte of data. | 22| `GSM_OPT_WAV49` | 0. `Disable WAV49 output in the encoder.` 1. `Enable WAV49 format in the encoder.` | Bit 3 (LSB) of 1st byte of data. | 23 24This also ensures that the plugin is always deterministic for any given input. 25 26##### Maximize utilization of input data 27The plugin feeds the entire input data to the codec using a loop. 28###### For the Decoder: 29 * If the size of media file is less than 65 bytes, (65 - size) bytes are padded and then sent to 30 the decoder so that decode happens for all the input files, regardless of their size. 31 GSM decoder consumes alternating frames of 33 and 32 bytes, so 65 bytes of input are needed. 32 * If the decode operation was successful, the input is advanced by the frame size 33 * If the decode operation was un-successful, the input is still advanced by frame size so 34that the fuzzer can proceed to feed the next frame. 35###### For the Encoder: 36 * If the size of media file is less than 320 bytes(160 signed 16-bit words), bytes are padded and then sent to 37 the encoder so that encode happens for all the input files, regardless of their size. 38 GSM encoder consumes 160 signed 16-bit words, so 320 bytes of input are needed. 39 * If the encode operation was successful, the input is advanced by the frame size 40 * If the encode operation was un-successful, the input is still advanced by frame size so 41that the fuzzer can proceed to feed the next frame. 42 43This ensures that the plugin tolerates any kind of input (empty, huge, 44malformed, etc) and doesnt `exit()` on any input and thereby increasing the 45chance of identifying vulnerabilities. 46 47## Build 48 49This describes steps to build gsm_dec_fuzzer and gsm_enc_fuzzer binaries. 50 51### Android 52 53#### Steps to build 54Build the fuzzer 55``` 56 $ mm -j$(nproc) gsm_dec_fuzzer 57 $ mm -j$(nproc) gsm_enc_fuzzer 58``` 59 60#### Steps to run 61Create a directory CORPUS_DIR and copy some gsm files (decoder)/raw files (encoder) to that folder 62Push this directory to device. 63 64To run on device 65``` 66 $ adb sync data 67 $ adb shell /data/fuzz/arm64/gsm_dec_fuzzer/gsm_dec_fuzzer CORPUS_DIR 68 $ adb shell /data/fuzz/arm64/gsm_enc_fuzzer/gsm_enc_fuzzer CORPUS_DIR 69``` 70To run on host 71``` 72 $ $ANDROID_HOST_OUT/fuzz/x86_64/gsm_dec_fuzzer/gsm_dec_fuzzer CORPUS_DIR 73 $ $ANDROID_HOST_OUT/fuzz/x86_64/gsm_enc_fuzzer/gsm_enc_fuzzer CORPUS_DIR 74``` 75 76## References: 77 * http://llvm.org/docs/LibFuzzer.html 78 * https://github.com/google/oss-fuzz 79