SyzFix¶
The full lifecycle of 7,000+ fixed Linux kernel bugs — from the syzbot
crash report, through every patch revision and reviewer discussion on
lore.kernel.org, to the commit merged into torvalds/linux.
Counts as of July 2026 — the dataset tracks syzbot continuously and grows with each incremental update.
One record = one bug's whole story¶
Most kernel-bug datasets stop at (crash, final patch). SyzFix keeps the process in between — the part that shows how maintainers actually converge on a fix:
[Sep 16, 2024] syzbot: NULL pointer deref in filemap_read_folio
[Sep 17, 2024] Developer → [PATCH v1]: check S_ISREG before proceeding
[Sep 17, 2024] syzbot: patch confirmed working ✅
[Sep 17, 2024] Developer → [PATCH v2]: also fix multi-device/blob case
[Oct 11, 2024] Chao Yu: Reviewed-by ✅
[Final] Commit 416a8b2c merged into torvalds/linux
Each entry carries the raw crash report (oops / KASAN / BUG), syzkaller and C reproducers, every patch revision with its inline review thread, and the final merged commit — aligned and machine-readable.
How it's built¶
flowchart LR
SYZ["syzbot<br/>bug list, crashes,<br/>reproducers"] --> P[collection<br/>pipeline]
LORE["lore.kernel.org<br/>patch versions,<br/>review threads"] --> P
GIT["git.kernel.org<br/>merged diffs"] --> P
PW["patchwork<br/>series fallback"] --> P
P --> PROC[("processed/<br/>one JSON per bug")]
PROC --> AN["13 heuristic<br/>analyzers"]
PROC --> HF[("HuggingFace<br/>Hub")]
AN --> STATS["cross-layer taxonomy,<br/>fix patterns, difficulty…"]The pipeline is resumable and incremental: a weekly
python -m dataset.update pulls only the bugs fixed since the last run and
pushes the refreshed dataset to HuggingFace.
Data collection →
Start in minutes¶
git clone https://github.com/sysec-uic/syzfix.git && cd syzfix
python3 -m venv venv && source venv/bin/activate
pip install -e .
# pull the pre-built dataset from HuggingFace (~2 GB download)
python -m dataset.restore_processed --repo xiaoguangwang/syzfix-dataset
python -m dataset.view build-index
python -m dataset.view list
- Exploring — browse, search, and inspect individual bugs from the CLI
- Analysis — 13 heuristic analyzers over the full corpus
- Collection — the crawl pipeline, rate limits, incremental updates
- Dataset card — schema, splits, and download options (HF link)
Citation & contact¶
Developed at sysec-uic. If you use SyzFix in your research, please cite the repository until the accompanying paper is available. Questions and issues → GitHub issues.