1 # Fuzzing 2 3 Each fuzzing target can be built with multiple engines. 4 Zstd provides a fuzz corpus for each target that can be downloaded with 5 the command: 6 7 ``` 8 make corpora 9 ``` 10 11 It will download each corpus into `./corpora/TARGET`. 12 13 ## fuzz.py 14 15 `fuzz.py` is a helper script for building and running fuzzers. 16 Run `./fuzz.py -h` for the commands and run `./fuzz.py COMMAND -h` for 17 command specific help. 18 19 ### Generating Data 20 21 `fuzz.py` provides a utility to generate seed data for each fuzzer. 22 23 ``` 24 make -C ../tests decodecorpus 25 ./fuzz.py gen TARGET 26 ``` 27 28 By default it outputs 100 samples, each at most 8KB into `corpora/TARGET-seed`, 29 but that can be configured with the `--number`, `--max-size-log` and `--seed` 30 flags. 31 32 ### Build 33 It respects the usual build environment variables `CC`, `CFLAGS`, etc. 34 The environment variables can be overridden with the corresponding flags 35 `--cc`, `--cflags`, etc. 36 The specific fuzzing engine is selected with `LIB_FUZZING_ENGINE` or 37 `--lib-fuzzing-engine`, the default is `libregression.a`. 38 Alternatively, you can use Clang's built in fuzzing engine with 39 `--enable-fuzzer`. 40 It has flags that can easily set up sanitizers `--enable-{a,ub,m}san`, and 41 coverage instrumentation `--enable-coverage`. 42 It sets sane defaults which can be overridden with flags `--debug`, 43 `--enable-ubsan-pointer-overflow`, etc. 44 Run `./fuzz.py build -h` for help. 45 46 ### Running Fuzzers 47 48 `./fuzz.py` can run `libfuzzer`, `afl`, and `regression` tests. 49 See the help of the relevant command for options. 50 Flags not parsed by `fuzz.py` are passed to the fuzzing engine. 51 The command used to run the fuzzer is printed for debugging. 52 53 Here's a helpful command to fuzz each target across all cores, 54 stopping only if a bug is found: 55 ``` 56 for target in $(./fuzz.py list); do 57 ./fuzz.py libfuzzer $target -jobs=10 -workers=10 -max_total_time=1000 || break; 58 done 59 ``` 60 Alternatively, you can fuzz all targets in parallel, using one core per target: 61 ``` 62 python3 ./fuzz.py list | xargs -P$(python3 ./fuzz.py list | wc -l) -I__ sh -c "python3 ./fuzz.py libfuzzer __ 2>&1 | tee __.log" 63 ``` 64 Either way, to double-check that no crashes were found, run `ls corpora/*crash`. 65 If any crashes were found, you can use the hashes to reproduce them. 66 67 ## LibFuzzer 68 69 ``` 70 # Build the fuzz targets 71 ./fuzz.py build all --enable-fuzzer --enable-asan --enable-ubsan --cc clang --cxx clang++ 72 # OR equivalently 73 CC=clang CXX=clang++ ./fuzz.py build all --enable-fuzzer --enable-asan --enable-ubsan 74 # Run the fuzzer 75 ./fuzz.py libfuzzer TARGET <libfuzzer args like -jobs=4> 76 ``` 77 78 where `TARGET` could be `simple_decompress`, `stream_round_trip`, etc. 79 80 ### MSAN 81 82 Fuzzing with `libFuzzer` and `MSAN` is as easy as: 83 84 ``` 85 CC=clang CXX=clang++ ./fuzz.py build all --enable-fuzzer --enable-msan 86 ./fuzz.py libfuzzer TARGET <libfuzzer args> 87 ``` 88 89 `fuzz.py` respects the environment variables / flags `MSAN_EXTRA_CPPFLAGS`, 90 `MSAN_EXTRA_CFLAGS`, `MSAN_EXTRA_CXXFLAGS`, `MSAN_EXTRA_LDFLAGS` to easily pass 91 the extra parameters only for MSAN. 92 93 ## AFL 94 95 The default `LIB_FUZZING_ENGINE` is `libregression.a`, which produces a binary 96 that AFL can use. 97 98 ``` 99 # Build the fuzz targets 100 CC=afl-clang CXX=afl-clang++ ./fuzz.py build all --enable-asan --enable-ubsan 101 # Run the fuzzer without a memory limit because of ASAN 102 ./fuzz.py afl TARGET -m none 103 ``` 104 105 ## Regression Testing 106 107 The regression test supports the `all` target to run all the fuzzers in one 108 command. 109 110 ``` 111 CC=clang CXX=clang++ ./fuzz.py build all --enable-asan --enable-ubsan 112 ./fuzz.py regression all 113 CC=clang CXX=clang++ ./fuzz.py build all --enable-msan 114 ./fuzz.py regression all 115 ``` 116 117 ## Fuzzing a custom sequence producer plugin 118 Sequence producer plugin authors can use the zstd fuzzers to stress-test their code. 119 See the documentation in `fuzz_third_party_seq_prod.h` for details. 120 121 ## Adding a new fuzzer 122 There are several steps involved in adding a new fuzzer harness. 123 124 ### Build your harness 125 1. Create a new your fuzzer harness `tests/fuzz/your_harness.c`. 126 127 2. Add your harness to the Makefile 128 129 2.1 Follow [this example](https://github.com/facebook/zstd/blob/e124e39301381de8f323436a3e4c46539747ba24/tests/fuzz/Makefile#L216) if your fuzzer requires both compression and decompression symbols (prefix `rt_`). If your fuzzer only requires decompression symbols, follow [this example](https://github.com/facebook/zstd/blob/6a0052a409e2604bd40354b76b86272b712edd7d/tests/fuzz/Makefile#L194) (prefix `d_`). 130 131 2.2 Add your target to [`FUZZ_TARGETS`](https://github.com/facebook/zstd/blob/6a0052a409e2604bd40354b76b86272b712edd7d/tests/fuzz/Makefile#L108). 132 133 3. Add your harness to [`fuzz.py`](https://github.com/facebook/zstd/blob/6a0052a409e2604bd40354b76b86272b712edd7d/tests/fuzz/fuzz.py#L48). 134 135 ### Generate seed data 136 Follow the instructions above to generate seed data: 137 ``` 138 make -C ../tests decodecorpus 139 ./fuzz.py gen your_harness 140 ``` 141 142 ### Run the harness 143 Follow the instructions above to run your harness and fix any crashes: 144 ``` 145 ./fuzz.py build your_harness --enable-fuzzer --enable-asan --enable-ubsan --cc clang --cxx clang++ 146 ./fuzz.py libfuzzer your_harness 147 ``` 148 149 ### Minimize and zip the corpus 150 After running the fuzzer for a while, you will have a large corpus at `tests/fuzz/corpora/your_harness*`. 151 This corpus must be minimized and zipped before uploading to GitHub for regression testing: 152 ``` 153 ./fuzz.py minimize your_harness 154 ./fuzz.py zip your_harness 155 ``` 156 157 ### Upload the zip file to GitHub 158 The previous step should produce a `.zip` file containing the corpus for your new harness. 159 This corpus must be uploaded to GitHub here: https://github.com/facebook/zstd/releases/tag/fuzz-corpora 160 161 162