README.md

metadata

license: cc-by-4.0
task_categories:
  - other
tags:
  - fuzzing
  - memory-safety
  - security
  - benchmark
  - ARVO
  - OSS-Fuzz
  - C
  - C++
language:
  - en
  - zh
pretty_name: MemVulBench
size_categories:
  - 10GB<n<100GB

MemVulBench v1.0.0

A real-world multi-bug benchmark for C/C++ memory-safety fuzzing: 20 admitted test units, 292 verified unique memory-vulnerability fingerprints (287 expected + 5 audited extra). Each unit is one real upstream project at one unmodified historical commit, driven through a single final entry point (identity or byte-prefix dispatch), with per-fingerprint triggering PoCs, persisted first-fault ASan logs and SHA-256 manifests.

Layout

units/<project>/                            # browsable metadata (3 small files each)
  <project>-<commit12>-v1.0.0.tar           # final aggregate image + PoCs + logs
  manifest.json                             # per-file SHA-256, upstream commit, entry map
  run-config.json                           # docker load / replay commands, ASAN_OPTIONS
fingerprint-index.json                      # 292 fingerprints -> unit, PoC, log, fix commit
manifest.json                               # release manifest (20 units, 292 fingerprints)
checks/hash-verification.json               # 20/20 image sha256 recomputed & matched
README.md                                   # this card

Quickstart

pip install -U "huggingface_hub[hf_xet]"
huggingface-cli download Fisho0/MemVulBench --repo-type dataset \
    --include "units/arrow/*" --local-dir memvulbench

docker load -i memvulbench/units/arrow/arrow-8b09ecc5c690-v1.0.0.tar  # inside the tar
# replay per units/arrow/run-config.json; each PoC's first fault should match
# the corresponding final-entry log in the tar.

Mainland-China users can download without a VPN by prepending:

export HF_ENDPOINT=https://hf-mirror.com

Provenance & license

• Software, catalog and evidence: https://github.com/explorerlxy/MemVulBench (tag memvulbench-v1.0.0; code MIT, data/docs CC BY 4.0).
• Every unit is an unmodified upstream commit recorded in its manifest; the images embed upstream sources under their original licenses.
• All bugs are historically public, upstream-fixed OSS-Fuzz issues (fix commits recorded per fingerprint in fingerprint-index.json).

Citation

@dataset{memvulbench-v1.0.0,
  title  = {MemVulBench: A Real-World Multi-Bug Benchmark for C/C++ Memory-Safety Fuzzing},
  author = {Lu, Xiaoyu and Wei, Qiang and Wang, Yunfeng},
  year   = {2026},
  version= {v1.0.0},
  url    = {https://huggingface.co/datasets/Fisho0/MemVulBench}
}