Fuzzers

These are fuzzers designed for use with libFuzzer or afl. They can be used to run on Google's OSS-Fuzz (https://github.com/google/oss-fuzz/).

The convention used here is that the initial values for each parser fuzzer are taken from the $NAME.in directory.

Crash reproducers from OSS-Fuzz are put into $NAME.repro directory for regression testing with top dir 'make check' or 'make check-valgrind'.

The ./configure runs below are for libidn2. To test libicu replace 'libidn2' with 'libicu', to test with libidn replace 'libidn2' by 'libidn'. To test without IDNA libraries replace --enable-runtime=... with --disable-runtime and replace --enable-builtin=... with --disable-builtin.

Running a fuzzer using clang

Use the following commands on top dir:

export CC=clang
export LIB_FUZZING_ENGINE="-lFuzzer -lstdc++"
export UBSAN_OPTIONS=print_stacktrace=1
export ASAN_SYMBOLIZER_PATH=/usr/bin/llvm-symbolizer

# set flags for clang asan
CFLAGS="-O1 -fno-omit-frame-pointer -gline-tables-only -DFUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION -fsanitize=address,leak,nonnull-attribute -fsanitize-address-use-after-scope -fsanitize-coverage=trace-pc-guard,trace-cmp"

# set flags for clang ubsan
CFLAGS="$CFLAGS -fsanitize=bool,array-bounds,float-divide-by-zero,function,integer-divide-by-zero,return,shift,signed-integer-overflow,vla-bound,vptr,undefined,alignment,null,enum,integer,builtin,float-divide-by-zero,function,object-size,returns-nonnull-attribute,unsigned-integer-overflow,unreachable -fsanitize=fuzzer-no-link"

# unsigned-integer-overflow is not UB, so recover when we see it.
CFLAGS="$CFLAGS -fno-sanitize-recover=all -fsanitize-recover=unsigned-integer-overflow"

export CFLAGS

./configure --enable-static --disable-gtk-doc --enable-fuzzing --enable-runtime=libidn2 --enable-builtin=libidn2
make -j$(nproc)
cd fuzz

# run libpsl_fuzzer
./run-clang.sh libpsl_fuzzer

If you see a crash, then a crash corpora is written that can be used for further investigation. E.g.

==2410==ERROR: AddressSanitizer: heap-use-after-free on address 0x602000004e90 at pc 0x00000049cf9c bp 0x7fffb5543f70 sp 0x7ff
fb5543720
...
Test unit written to ./crash-adc83b19e793491b1c6ea0fd8b46cd9f32e592fc

To reproduce the crash:

./libpsl_fuzzer < ./crash-adc83b19e793491b1c6ea0fd8b46cd9f32e592fc

You can also copy/move that file into libpsl_fuzzer.repro/ and re-build the project without fuzzing for a valgrind run, if you like that better. Just a ./configure and a make check-valgrind should reproduce it.

Running a fuzzer using AFL

Use the following commands on top dir:

$ CC=afl-clang-fast ./configure --disable-gtk-doc --enable-runtime=libidn2 --enable-builtin=libidn2
$ make -j$(nproc) clean all
$ cd fuzz
$ ./run-afl.sh libpsl_fuzzer

Fuzz code coverage using the corpus directories *.in/

Code coverage reports currently work best with gcc+lcov+genhtml.

In the top directory:

CC=gcc CFLAGS="-O0 -g" ./configure --disable-gtk-doc --enable-runtime=libidn2 --enable-builtin=libidn2
make fuzz-coverage
xdg-open lcov/index.html

Each fuzzer target has it's own functions to cover, e.g. libpsl_fuzzer covers psl_is_public_suffix.

To work on corpora for better coverage, cd fuzz and use e.g. ./view-coverage.sh libpsl_fuzzer.

Enhancing the testsuite for issues found

Each reproducer file should be dropped into the appropriate *.repro/ directory.

Clang CFI instrumentation

CC=clang CFLAGS="-B/usr/bin/gold -O0 -fsanitize=cfi -flto -fvisibility=default -fno-sanitize-trap=all" ./configure
make clean
make
make check