trezor-firmware/crypto/fuzzer

Fuzz Testing trezor-crypto

Selected functions can be fuzzed via specific libFuzzer harnesses for increased test coverage and issue detection.

Note: the following commands are relative to the trezor-crypto main directory.

Build

A modern C compiler with built-in libFuzzer support is required. The build process will use clang by default. Set the CC= environment variable if you want to use a special compiler variant.

make clean
FUZZER=1 make fuzzer

Sanitizers

Recommended: ASAN / UBSAN / MSAN flags for error detection can be specified via the special SANFLAGS.

Examples:

• SANFLAGS="-fsanitize=address,undefined"
• SANFLAGS="-fsanitize=memory -fsanitize-memory-track-origins"

Optimizations

Override OPTFLAGS to test the library at different optimization levels or simplify the debugging of detected issues.

Examples:

• OPTFLAGS="-O0 -ggdb3"
• OPTFLAGS="-O3 -march=native"

Other Flags

To be determined:

• -DNDEBUG
• -DUSE_BIP39_CACHE=0 -DUSE_BIP32_CACHE=0

Operation

See the libFuzzer documentation for valid options and usage. Detailed fuzzer usage and relevant considerations are out of scope of this document.

Warning: fuzzing is resource-intensive and can have a negative impact on your system stability.

Basic fuzzer call:

./fuzzer/fuzzer

Here is a more sophisticated multithreading example with a persistent input corpus and other optimizations:

mkdir fuzzer/fuzzer_corpus
./fuzzer/fuzzer -max_len=2048 -use_value_profile=1 -jobs=16 -timeout=1 -reload=5 -print_pcs=1 -print_funcs=42  fuzzer/fuzzer_corpus

Hint: for more permanent setups, consider invoking the fuzzer from outside of the source directory to avoid cluttering it with logfiles and crash inputs. Similarly, it is recommended to store the fuzzer corpus in another location.

Automated Fuzzer Dictionary Generation

Dictionaries are a useful mechanism to augment the capabilities of the fuzzer. Specify them via the -dict= flag.

Collect Interesting Strings From Unit Tests

cd fuzzer
./extract_fuzzer_dictionary.sh fuzzer_crypto_tests_strings_dictionary1.txt

The resulting file can be used as a fuzzer dictionary.

Evaluate Source Coverage

1. build the fuzzer binary with CFLAGS="-fprofile-instr-generate -fcoverage-mapping"
2. run with suitable -runs= or -max_total_time= limits
3. convert the recorded data llvm-profdata merge -output=default.profdata -instr default.profraw
4. render the data llvm-cov show fuzzer/fuzzer -instr-profile=default.profdata -format=html -output-dir=coverage-report
5. analyze report at coverage-report/index.html
6. (optional) remove artifacts with rm default.profraw default.profdata && rm -r coverage-report