| 1 | # Oversight fuzz harnesses |
| 2 | |
| 3 | Two libFuzzer-based harnesses for the security-critical parsers: |
| 4 | - `container_parser` - hammers the `.sealed` binary format parser |
| 5 | - `manifest_parser` - hammers the canonical-JSON manifest parser |
| 6 | |
| 7 | ## Setup (one time) |
| 8 | |
| 9 | ```bash |
| 10 | cargo install cargo-fuzz |
| 11 | ``` |
| 12 | |
| 13 | Requires a nightly Rust toolchain for fuzzing (sanitizers, coverage): |
| 14 | ```bash |
| 15 | rustup install nightly |
| 16 | ``` |
| 17 | |
| 18 | ## Run |
| 19 | |
| 20 | ```bash |
| 21 | cd oversight-rust/fuzz |
| 22 | cargo +nightly fuzz run container_parser -- -max_total_time=300 |
| 23 | cargo +nightly fuzz run manifest_parser -- -max_total_time=300 |
| 24 | ``` |
| 25 | |
| 26 | ## What "pass" looks like |
| 27 | |
| 28 | The harness runs until you stop it. "Pass" means: no panics, no hangs, |
| 29 | no OOMs, no memory safety violations (Rust + libFuzzer's AddressSanitizer |
| 30 | catches memory bugs). Any crash input is saved to `fuzz/artifacts/...` for |
| 31 | reproduction. |
| 32 | |
| 33 | Target: run continuously for at least 24 hours before a paid security audit |
| 34 | engagement, per our ROADMAP.md prerequisites. |