Open-source differential fuzzing of Ethereum 2.0 Phase 0 implementations. Maintained by Sigma Prime for the Ethereum Foundation.
A differential fuzzer of Eth2.0 implementations using libfuzzer. By default, fuzzing progresses indefinitely unless an implementation panics or differing output is identified.
This is a continuation of Guido Vranken's eth2.0-fuzzing.
This project and its inner workings are subject to change.
A note on terminology: "client" and "implementation" are used interchangeably here to mean a specific Eth2 implementation.
Operational Fuzz Targets:
(and their relevant spec function)
All currently use the "mainnet" config: https://github.com/ethereum/eth2.0-specs/blob/v0.9.1/configs/mainnet.yaml
See corpora repository for explanation of input structure.
See corpora for examples and explanation of structure.
$ git clone --depth 1 https://github.com/sigp/beacon-fuzz.git $ git clone --depth 1 https://github.com/sigp/beacon-fuzz-corpora.git $ cd beacon-fuzz $ sudo ./runfuzzer.sh block_header ../beacon-fuzz-corpora/0-9-1/mainnet/block_header/ ../beacon-fuzz-corpora/0-9-1/mainnet/beaconstate
$ git clone --depth 1 https://github.com/sigp/beacon-fuzz.git $ cd beacon-fuzz $ sudo docker build . -t beacon_fuzz $ sudo docker run -it beacon_fuzz bash # git clone --depth 1 https://github.com/sigp/beacon-fuzz-corpora.git # export ETH2_FUZZER_STATE_CORPUS_PATH="/eth2/beacon-fuzz-corpora/0-9-1/mainnet/beaconstate" # /eth2/fuzzers/attestation/fuzzer /eth2/beacon-fuzz-corpora/0-9-1/mainnet/attestation
jobs=N to run N simultaneous jobs (by default in
help=1 for more arguments (see also libfuzzer docs)
- Add more implementations
- Add more fuzz targets
- Improved onboarding, ease of adding new targets and implementations
- Improved coverage measurements and visibility
- Deploy on dedicated production fuzzing infrastructure
- Structure-aware fuzzing mutations
- Mutate input
The following implementations will be added to the various fuzzing targets:
$ pre-commit install
(see also .pre-commit-config.yaml)
If build fails, comment the
RUN /eth2/build.sh in
Dockerfile, and run it manually from within the container.
Can adjust Makefiles as needed.
It is generally fine to run
build.sh multiple times, and previously built components will be ignored.
makeing fuzzers directly
This is quicker than re-running
./build.sh and is useful when troubleshooting specific build issues.
build.sh once, a file
/eth2/exported_env.sh will be created.
Sourcing this will ensure you have all the environment variables required by the Makefiles.
NOTE: the fuzzer's Makefiles do not currently identify changes to dependent file outside of the
So, for example, a change to a dependent file in
./files/lib will require a
make clean (or equivalent) for the modifications to be visible.
The following make variable flags are exposed (and are enabled by setting them to anything other than ''):
make BFUZZ_NO_DISABLE_BLS=1 all
BFUZZ_NO_DISABLE_BLS- don't disable bls verification
Adding new implementations for a target
harness.pys for a succinct, readable harness implementation example without much boilerplate.
For state transition functions, each client should expect to receive a correctly-encoded SSZ container containing a
and an input object relevant for the transition. (As described in https://github.com/sigp/beacon-fuzz-corpora/blob/master/0-8-3/README.md,
state_id has been replaced with a corresponding
Please panic/abort if SSZ decoding fails, as this indicates an error in preprocessing or the SSZ libraries.
e.g. even though a client implementing the Attestation fuzz target can expect to receive any arbitrary
it should be in the form of a validly-encoded SSZ blob.
Currently clients will only receive known, valid
ETH2_FUZZER_STATE_CORPUS_PATH) so the actual fuzzing/mutation is performed with the other part of the input.
This is because clients generally maintain their own
BeaconStates, so don't expect to receive arbitrary states from untrusted sources.
(It is also highly unlikely that current mutation will ever produce a valid
There are 3 types of results/outputs that a client is expected to return to the fuzzer:
- A bytestring/bytearray/blob (usually a SSZ-encoding of the
- An error result (usually a
- To be returned when the operation failed but the client is in a consistent state (e.g. supplied Attestation data does not refer to an appropriate epoch).
- The c++ module returns this as a
- This is necessary to differentiate from the few cases where an empty bytestring is a valid and successful result e.g. shuffling an empty list.
- To occur when a client is in an inconsistent state and indicates a bug is present.
TODO disabling BLS verification?
Adding a new fuzzing target
- Python editable installs in Venvs aren't detected.
MIT - see LICENSE