• Show log

    Commit

  • Hash : 59328ed8
    Author : Patrick Steinhardt
    Date : 2018-07-19T13:29:46

    fuzzers: rename "fuzz" directory to match our style Our layout uses names like "examples" or "tests" which is why the "fuzz" directory doesn't really fit in here. Rename the directory to be called "fuzzers" instead. Furthermore, we rename the fuzzer "fuzz_packfile_raw" to "packfile_raw_fuzzer", which is also in line with the already existing fuzzer at google/oss-fuzz. While at it, rename the "packfile_raw" fuzzer to instead just be called "packfile" fuzzer.

  • README.md

  • Fuzzing

    libgit2 is currently using libFuzzer to perform automated fuzz testing. libFuzzer only works with clang.

    Prerequisites** for building fuzz targets:

    1. All the prerequisites for building libgit2.
    2. A recent version of clang. 6.0 is preferred. pre-build Debian/Ubuntu packages

    Build

    1. Create a build directory beneath the libgit2 source directory, and change into it: mkdir build && cd build
    2. Choose one sanitizers to add. The currently supported sanitizers are address, undefined, and leak/address,leak.
    3. Create the cmake build environment and configure the build with the sanitizer chosen: CC=/usr/bin/clang-6.0 cmake -DBUILD_CLAR=OFF -DBUILD_FUZZERS=ON -DUSE_SANIZER=address -DCMAKE_EXE_LINKER_FLAGS="-fsanitize=fuzzer" -DCMAKE_BUILD_TYPE=RelWithDebInfo ... Note that building the fuzzer targets is incompatible with the tests and examples.
    4. Build libgit2: cmake --build .
    5. Exit the cmake build environment: cd ..

    Run the fuzz targets

    1. ASAN_SYMBOLIZER_PATH=/usr/bin/llvm-symbolize-6.0 LSAN_OPTIONS=allocator_may_return_null=1 ASAN_OPTIONS=allocator_may_return_null=1 ./build/fuzz/fuzz_packfile_raw fuzz/corpora/fuzz_packfile_raw/

    The LSAN_OPTIONS and ASAN_OPTIONS are there to allow malloc(3) to return NULL. The LLVM_PROFILE_FILE is there to override the path where libFuzzer will write the coverage report.

    Get coverage

    In order to get coverage information, you also need to add the -DUSE_COVERAGE=ON flag to cmake, and then run the fuzz target with -runs=0. That will produce a file called default.profraw (this behavior can be overridden by setting the LLVM_PROFILE_FILE="yourfile.profraw" environment variable).

    1. llvm-profdata-6.0 merge -sparse default.profraw -o fuzz_packfile_raw.profdata transforms the data from a sparse representation into a format that can be used by the other tools.
    2. llvm-cov-6.0 report ./build/fuzz/fuzz_packfile_raw -instr-profile=fuzz_packfile_raw.profdata shows a high-level per-file coverage report.
    3. llvm-cov-6.0 show ./build/fuzz/fuzz_packfile_raw -instr-profile=fuzz_packfile_raw.profdata [source file] shows a line-by-line coverage analysis of all the codebase (or a single source file).

    Standalone mode

    In order to ensure that there are no regresions, each fuzzer target can be run in a standalone mode. This can be done by passing -DUSE_STANDALONE_FUZZERS=ON to cmake without setting -DCMAKE_EXE_LINKER_FLAGS. This makes it compatible with gcc. This does not use the fuzzing engine, but just invokes every file in the chosen corpus.

    In order to get full coverage, though, you might want to also enable one of the sanitizers. You might need a recent version of clang to get full support.

    References