Dataset of Embedded Systems with Multiple Firmware Versions

This repository provides structured datasets of embedded and industrial systems across multiple firmware versions.
Each dataset includes automated scripts for downloading, unpacking, and preparing firmware samples for controlled experiments such as semantic diffing, retrieval evaluation, and supply-chain security analysis.

---

Datasets Overview

1. OpenWRT Dataset (openwrt_data/)

The OpenWRT dataset offers a consistent, open-source baseline for analyzing software evolution and update behavior across firmware versions.

Contents

• build_openwrt.sh – Automates downloading and unpacking multiple OpenWRT firmware images.
• tags.txt – Lists component identifiers used for semantic retrieval and evaluation.

Use cases

• Measuring function-level change granularity
• Evaluating differential triage methods
• Validating semantic embeddings across versions

---

2. WAGO PLC Dataset (wago_data/)

This dataset contains real-world firmware samples from WAGO PFC200 programmable logic controllers (PLCs).
It supports controlled binary replacement for generating labeled clean/backdoor variants, enabling reproducible semantic diffing experiments.

Contents

• setup_wago_data.sh – End-to-end automation script that:

  1. Downloads firmware versions 03.10.10 and 03.10.08 from WAGO’s public GitHub releases.
  2. Extracts filesystem contents with binwalk.
  3. Locates and removes the original usr/sbin/dropbear binary.
  4. Inserts controlled replacement binaries:
     • dropbear86-backdoor → backdoor variant
     • dropbear86-clean / dropbear83-clean → clean variants
  5. Supports choosing between stripped (firmware-like) and symbol-rich binaries via CLI flag.
  6. Produces three labeled datasets:
     • 03.10.10-backdoor
     • 03.10.10-clean
     • 03.10.08-clean
  7. Moves final datasets into the experiment_samples/ directory.

• dropbear_samples/
  dropbear_samples/
  ├─ stripped/
  │ ├─ dropbear83-clean.stripped
  │ ├─ dropbear86-clean.stripped
  │ └─ dropbear86-backdoor.stripped
  └─ symbols/
  ├─ dropbear83-clean
  ├─ dropbear86-clean
  └─ dropbear86-backdoor

  • stripped/ contains firmware-realistic binaries with symbols removed.
  • symbols/ contains binaries compiled with symbols for analysis and reverse engineering.

• experiment_samples/ – Contains the final processed datasets after running the setup script.

Purpose This dataset is used for evaluating:

• semantic diffing frameworks (e.g., DRIFT),
• function-level retrieval robustness,
• and the impact of controlled binary modification on analysis pipelines.

---

Dependencies

Required Tools

• Python 3
• binwalk (firmware unpacking)
• lzop (decompression helper)
• rsync (optional, faster copying)

Install on Ubuntu/Debian

sudo apt-get update
sudo apt-get install binwalk lzop squashfs-tools rsync
pip install pandas requests

---

Usage

OpenWRT Dataset

cd datasets/openwrt_data
./build_openwrt.sh

This will automatically download and unpack multiple OpenWRT versions for analysis.

---

WAGO Dataset

Generate stripped firmware variants (default):

cd datasets/wago_data
./setup_wago_data.sh

Explicitly use stripped binaries:

./setup_wago_data.sh --stripped

Use symbol-rich binaries:

./setup_wago_data.sh --symbols

Dry-run mode (no changes, just print actions):

./setup_wago_data.sh --dry-run --symbols

After completion, experiment_samples/ will contain:

03.10.10-backdoor/
03.10.10-clean/
03.10.08-clean/

Each contains a fully unpacked firmware root filesystem with:

usr/sbin/dropbear

replaced by the selected clean or backdoor sample.

---

Example Mapping

WAGO Firmware Version	Dropbear Sample (stripped)	Dropbear Sample (symbols)
03.10.08 clean	dropbear83-clean.stripped	dropbear83-clean
03.10.10 clean	dropbear86-clean.stripped	dropbear86-clean
03.10.10 backdoor	dropbear86-backdoor.stripped	dropbear86-backdoor

---

Source Attribution

• OpenWRT Firmware: https://downloads.openwrt.org/
• WAGO PLC Firmware: https://github.com/WAGO/pfc-firmware
• Original Multi-Firmware Dataset: https://github.com/WUSTL-CSPL/Firmware-Dataset

---