Lez-fuzzing

Coverage-guided fuzzing and adversarial testing infrastructure for the Logos Execution Zone (LEZ) protocol.

---

Repository Layout

lez-fuzzing/
├── Cargo.toml              # Workspace root (members: fuzz_props)
├── Justfile                # Turn-key entry-points
├── rust-toolchain.toml     # Pins Rust nightly (required by cargo-fuzz)
├── .gitignore
├── fuzz_props/             # Shared invariant framework + input generators
│   ├── Cargo.toml
│   └── src/
│       ├── lib.rs
│       ├── invariants.rs   # ProtocolInvariant trait + concrete invariants
│       └── generators.rs   # Arbitrary / proptest strategies
├── fuzz/                   # cargo-fuzz crate (own [workspace] sentinel)
│   ├── Cargo.toml
│   ├── fuzz_targets/
│   │   ├── fuzz_transaction_decoding.rs
│   │   ├── fuzz_stateless_verification.rs
│   │   ├── fuzz_state_transition.rs
│   │   └── fuzz_block_verification.rs
│   └── corpus/             # Curated seed inputs (one dir per target)
├── .github/
│   └── workflows/
│       └── fuzz.yml        # CI: smoke-fuzz · regression · proptest · perf
└── docs/
    └── fuzzing.md          # Full developer guide

The LEZ codebase is consumed as a sibling directory — clone logos-execution-zone next to this repository:

parent/
├── lez-fuzzing/            ← this repo
└── logos-execution-zone/   ← LEZ codebase (path deps resolve via ../)

---

Quick Start

Prerequisites

rustup install nightly
rustup component add llvm-tools-preview --toolchain nightly
cargo install cargo-fuzz
# Optional but recommended:
cargo install just

Why nightly? cargo-fuzz passes -Zsanitizer=address and -Zinstrument-coverage (unstable flags) to rustc, and depends on the llvm-tools-preview nightly component for coverage reporting. The rust-toolchain.toml pins the whole repository to nightly so you never need an explicit +nightly flag.

Setup

# Clone both repositories side by side
git clone <LEZ_REPO_URL> logos-execution-zone
git clone <LEZ_FUZZING_REPO_URL> lez-fuzzing
cd lez-fuzzing

Run the fuzz targets

# All targets for 30 s each (RISC0_DEV_MODE=1 is set automatically)
just fuzz

# Specific duration
just fuzz 120

# Single target
RISC0_DEV_MODE=1 cargo fuzz run fuzz_state_transition -- -max_total_time=120

# Corpus regression (replay saved corpus, no mutations)
just fuzz-regression

# Property-based tests only (no libFuzzer)
just fuzz-props

ZK-proof cost: RISC0_DEV_MODE=1 is exported at the top of the Justfile and must be set in every fuzz run to stub out ZK proof generation. Without it each execution takes seconds instead of microseconds.

---

Fuzz Targets

Target	Protocol layer	Entry point
fuzz_transaction_decoding	Borsh decoding of all tx/block types	fuzz/fuzz_targets/fuzz_transaction_decoding.rs
fuzz_stateless_verification	transaction_stateless_check() idempotency	fuzz/fuzz_targets/fuzz_stateless_verification.rs
fuzz_state_transition	V03State transition + state-isolation invariant	fuzz/fuzz_targets/fuzz_state_transition.rs
fuzz_block_verification	Block hash integrity	fuzz/fuzz_targets/fuzz_block_verification.rs

---

Corpus Management

# Minimise all corpora (removes dominated inputs, keeps coverage-equivalent set)
just corpus-cmin

# Minimise a single target's corpus
just corpus-cmin-target fuzz_state_transition

---

Crash / Failure Workflow

# Minimise a crash artifact
just fuzz-tmin fuzz_state_transition fuzz/artifacts/fuzz_state_transition/crash-abc123

# Print the bytes as a Rust literal (for a regression #[test])
cargo fuzz fmt fuzz_state_transition fuzz/artifacts/fuzz_state_transition/crash-abc123

# Promote the minimised input to the corpus so CI catches regressions
cp fuzz/artifacts/fuzz_state_transition/crash-abc123-minimised \
   fuzz/corpus/fuzz_state_transition/regression_001

---

Housekeeping

just clean            # Remove Cargo build artefacts (target/ and fuzz/target/)
just clean-artifacts  # Remove fuzz/artifacts/  (crash/timeout inputs)
just clean-coverage   # Remove fuzz/coverage/   (LLVM coverage reports)
just clean-all        # All of the above

---

CI

GitHub Actions runs four jobs on every push/PR and nightly:

Job	What it does
smoke-fuzz (matrix)	Builds + runs each target for 60 s
regression (matrix)	Replays the saved corpus (-runs=0)
proptest	cargo test -p fuzz_props --release
perf-baseline (nightly only)	Measures exec/sec per target, uploads perf_baseline.txt

---

Documentation

Full developer guide — how to add new targets, interpret crashes, update the LEZ sibling clone, and tune performance — is in docs/fuzzing.md.

---

License

Licensed under the MIT License.