🕵️ NetGuard – Hybrid Python/C NIDS & Observability Engine

A full end-to-end real-time Network Intrusion Detection & Prevention System (NIDS/NIPS).
Combines a Multi-threaded Python (Scapy) capture and analysis engine with a dual-layer DPI pipeline — an Aho-Corasick automaton for keyword signatures and a Native C Extension (bounds-checked, called live via ctypes) for injection-pattern signatures — plus Sliding-Window anomaly detection, OS Firewall Active Defense, and a fully code-managed monitoring stack (Dashboard as Code) on Docker (Grafana + Loki + Promtail).

---

🔎 Overview & Architecture

NetGuard provides a complete solution for monitoring, analyzing, and responding to network security events across OSI layers 3, 4, and 7. The architecture is built on a continuous Producer-Consumer pipeline that separates low-level packet capture, real-time threat analysis, and structured observability shipping.

Architecture Highlight: NetGuard runs a dual-layer DPI pipeline on non-encrypted traffic: an Aho-Corasick automaton for Layer-7 keyword signatures, and a native C Extension (ctypes) for injection-pattern detection. Encrypted payloads (e.g., port 443) are automatically bypassed to optimize CPU usage and eliminate false positives.

flowchart LR
%% Traffic Input
NIC[📡 NIC] -->|Raw Packets| SnifferThread[🐍 Sniffer Thread<br>Scapy store=0]
%% Core Engine
subgraph Engine [Python NIDS Core Engine]
SnifferThread -->|Non-blocking Put| Queue[📦 Queue<br>maxsize=20000]
Queue -->|Get Packet| Worker[⚙️ Worker Pool]
subgraph Detection [Detection & Defense]
Worker -->|L3/L4 Window| Anomaly[🛡️ DoS / Scan Detector]
Worker -->|L7 Payload| DPI[⚡🐍 Aho-Corasick + Native C DPI]
Anomaly -->|Threshold Breach| Firewall[🧱 OS Firewall<br>netsh / iptables]
Anomaly & DPI -->|Update| State[🔒 State & Blacklist]
end
GC[🧹 GC Thread] -->|Clean Every 30s| State
end
%% Logging & Observability
Worker -->|JSON Log| LogFile[📄 logs/network_security.json]
LogFile --> Promtail[🔄 Promtail Container]
Promtail -->|Push| Loki[🗄️ Loki DB]
Loki --> Grafana[📊 Grafana Dashboard]

Loading

---

📊 Dashboard Preview

Live view of the NetGuard Security Overview Grafana dashboard, provisioned automatically from code.

Live JSON log stream with DNS query events, alongside real-time Security Events Distribution and Threat Timeline panels.

Active Defense in action — a DoS/SYN Flood attack triggers automatic IP isolation, alongside a DNS Tunneling detection alert (Shannon entropy-based).

Full event-type breakdown (Port Scans, Stealth Scans, DPI Alerts, DoS Attacks) with Top Suspicious Source IPs and Total Security Alerts panels.

---

⚡ Performance & Resilience Analysis

• Memory Backpressure & Drop Policy — The engine utilizes a bounded Queue(maxsize=20000) combined with store=0 in Scapy to ensure zero in-memory packet buffering by the sniffer thread. Under high-throughput conditions, excess packets are dropped safely rather than causing Out-Of-Memory (OOM) fatal crashes.
• Concurrency & C-Level Unlocking — Low-level packet capture executes within native socket primitives (C-level libpcap/WinPcap), releasing Python's Global Interpreter Lock (GIL) and allowing the background worker thread and garbage collector thread to execute processing tasks concurrently.
• Thread Safety via a Shared Lock — Multi-threaded access to volatile state structures (syn_history, port_history, blacklist) is protected by a single shared threading.Lock to guarantee atomic read/write state transitions without data races.
• Active Defense & OS Firewall Integration — Automatically triggers dynamic OS firewall mitigation rules (netsh advfirewall on Windows, iptables on Linux) upon identifying DoS/SYN Flood attacks. Commands execute asynchronously in detached daemon threads to keep processing queues zero-latency.
• Deterministic Resource Cleanup (Garbage Collector) — Dormant IP records and expired blacklist entries are purged every 30 seconds by a background garbage collection thread in bounded $O(N)$ time, ensuring steady memory utilization under sustained traffic.

---

🔬 DPI Engines: Live Detection vs. Throughput Benchmark

NetGuard runs two DPI engines together in the live pipeline (main.py), each covering a different signature class, plus a separate standalone script for measuring the C engine's raw throughput in isolation:

	Live: Aho-Corasick (_check_dpi_keywords)	Live: Native C (_check_dpi_native)	Standalone: benchmark_dpi.py
Technology	pyahocorasick automaton, pure-Python substring fallback if unavailable	Native C via ctypes, loaded once in NetworkGuardian.__init__	Same libdpi shared library, loaded independently
Where it runs	Inside NetworkGuardian._check_dpi, on every captured packet	Inside NetworkGuardian._check_dpi, on every captured packet	Standalone script only — not imported by main.py
Signature set	admin, password, etc/passwd, select * from	' OR '1'='1, UNION SELECT, <script>, ../../, etc/passwd, cmd.exe, ; whoami	Same as the live C engine (fixed in dpi.c)
Purpose	Real-time keyword/credential-leak alerting	Real-time injection-pattern alerting (SQLi, XSS, path traversal, cmd injection)	Measuring native-C vs. pure-Python throughput on synthetic payloads
Required?	Recommended (pip install pyahocorasick); falls back to pure Python if missing	Optional — compile with make -C c_src; NetGuard runs on Aho-Corasick alone if the shared library isn't present	N/A — dev/benchmarking tool only

In short: the ~8x acceleration number below is a controlled, isolated measurement of the C engine's throughput on synthetic payloads — it describes the engine's capability, not a claim about end-to-end NIDS throughput (which also includes Scapy capture, queueing, and the Aho-Corasick pass). Build libdpi (make -C c_src) before running main.py to get the C-accelerated injection-signature detection live; without it, NetGuard still runs correctly on Aho-Corasick alone.

---

🏎️ DPI Native C Engine Performance

• Batch Processing & Bounds-Checked Safety — By eliminating Python FFI execution overhead and passing contiguous memory blocks directly to native C primitives, payload scanning avoids GIL bottlenecks. The engine uses bounds-checked memchr/memcmp scanning to prevent Out-Of-Bounds reads and Null-Byte truncation issues on raw binary network traffic.
• Micro-Benchmark Results (100,000 Packets Evaluation):

Engine Implementation	Execution Time	Throughput	Acceleration
🐍 Python Pure	0.3134 sec	319,038 Packets/sec	Baseline (1.0x)
⚡ Native C Extension (Safe)	0.0395 sec	2,532,165 Packets/sec	7.94x Faster

Run Benchmark Locally:

# Linux / macOS
make -C c_src
# Windows (GCC) — requires MinGW/MSYS2 installed
gcc -shared -O3 -march=native -o libdpi.dll c_src/dpi.c
python benchmark_dpi.py

---

📂 Project Structure

NetGuard/
├── assets/ # Static Documentation Assets (Dashboard Screenshots)
│ ├── dashboard_overview.png
│ ├── dashboard_threat_detection.png
│ └── dashboard_analytics.png
├── c_src/ # Low-Level Native C Extensions
│ ├── dpi.c # Native C DPI Engine (Batch Engine)
│ └── Makefile # C Compilation Setup
├── grafana/
│ ├── dashboards/ # Standard JSON Dashboard (Git Version-Controlled)
│ │ └── dashboard-NetGuard Security Overview.json
│ └── provisioning/ # Grafana Automated Provisioning Configs
│ ├── dashboards/
│ │ └── dashboards.yml
│ └── datasources/
│ └── datasources.yml
├── logs/ # Runtime Log Directory (Ignored by Git, not tracked)
├── .env.example
├── .gitignore
├── LICENSE
├── README.md
├── benchmark_dpi.py # Native C vs Python DPI Micro-Benchmark
├── docker-compose.yml
├── main.py # NIDS Core Engine (Thread-Safe & GC Refactored)
├── promtail-config.yml
├── requirements.txt
└── test_attack.py # Traffic Simulator

---

🚀 Core Features

Domain	Feature	Status	Description	Performance Indicator
📡 Network	Real-time L3-L7 Sniffing	✅	Real-time capture and analysis of IP, TCP, UDP, and DNS traffic while preventing memory overflow (store=0).	Bounded memory capture (no packet buffering), O(1) enqueue
🛡️ Cyber Security	Sliding-Window & Stealth Detection	✅	Detects DoS (SYN Flood), standard port scans, and Stealth Scans (NULL, FIN, XMAS) via moving time windows.	O(1) queue operations
🧬 DNS Security	DNS Tunneling Detection	✅	Shannon Entropy calculation & query length evaluation to catch exfiltration over DNS.	O(N) entropy check
⚡ Active Defense	Dynamic IP Isolation & OS Firewall	✅	Active mitigation mechanism that dynamically injects OS firewall rules (netsh / iptables) to block malicious hosts upon threshold breach.	Non-blocking async execution, O(1) blacklist check
🔍 DPI Engine	Deep Packet Inspection	✅	Dual-layer L7 payload scanning: an Aho-Corasick automaton for keyword signatures, plus a Native C Extension (loaded live via ctypes) for bounds-checked injection-pattern signatures. Falls back to Aho-Corasick alone if the C library isn't compiled.	O(N+M) string matching; native C path measured at ~8x throughput vs. pure Python in isolation
⚙️ Architecture	Producer-Consumer & Thread-Safety	✅	Bounded Queue, threading.Lock primitives, and a dedicated background Garbage Collector thread to prevent memory leaks.	Bounded queue, backpressure-safe
📊 Observability & IaC	Dashboard as Code (Grafana + Loki)	✅	A single all-in-one "NetGuard Security Overview" dashboard in standard JSON format, automatically loaded on container startup via Provisioning files.	Instant provisioning on boot
📝 Logging	Structured JSON Dual-Stream	✅	Colorized console output alongside structured JSON log writes (logs/network_security.json), tailored for collection by Promtail.	Low-overhead async writes
🧪 Testing	Traffic Attack Simulator	✅	Simulation script (test_attack.py) that generates synthetic attack traffic to validate detection mechanisms.	Configurable synthetic load

---

🛠️ Technologies & Architectural Highlights

• Python & Scapy — Raw-socket-level packet capture, protocol parsing, and deep payload-level inspection (DPI).
• Aho-Corasick DPI (live) — Multi-pattern automaton (pyahocorasick) used inside main.py for real-time keyword/signature matching against packet payloads, with a pure-Python substring fallback when the library is unavailable.
• Native C Extension (ctypes, live) — Batch-safe, memory-bounds-checked C DPI engine loaded via ctypes in NetworkGuardian.__init__ and called on every packet's payload in _check_dpi_native alongside the Aho-Corasick pass, delivering ~8x (7.94x measured) throughput over pure Python for its signature set. The same shared library is also loaded independently by benchmark_dpi.py for isolated throughput measurement. Requires make -C c_src first; NetGuard degrades gracefully to Aho-Corasick-only if the compiled library isn't present.
• Producer-Consumer Architecture — Full separation between packet capture and analysis via queue.Queue(maxsize=20000), preventing packet loss under load.
• Thread-Safety & Active Defense — Whitelist/Blacklist state management guarded by threading.Lock to prevent data races, integrated with background dynamic OS Firewall rule injection (netsh / iptables).
• Background Garbage Collector — A dedicated background thread that cleans up stale data structures (Sliding Window History & Blacklist) from memory every 30 seconds, synchronously and thread-safely, ensuring zero memory leaks from dormant IP addresses.
• Promtail & Grafana Loki — Shipping of structured JSON logs from the local logs directory and indexing them in Loki.
• Dashboards as Code (IaC) — The "NetGuard Security Overview" dashboard is version-controlled in Git under grafana/dashboards/, automatically loaded into Grafana on container startup.
• Docker Compose Stack — One-click deployment of the entire observability infrastructure.

---

🧬 Algorithmic Detection: Shannon Entropy for DNS Tunneling

NetGuard identifies covert communications and data exfiltration over DNS by analyzing the randomness (entropy) of domain query strings.

Standard domain names exhibit predictable natural-language patterns, whereas encrypted/encoded data streams (e.g., Base64/Hex DNS Tunneling) produce significantly higher entropy scores.

• Mathematical Model: Calculates Shannon Entropy $H(X)$ over the unique byte/character distribution of each query string: $$H(X) = -\sum_{i=1}^{n} P(x_i) \log_2 P(x_i)$$
• Detection Threshold: Queries exceeding the configured threshold ($H(X) > 4.3$) alongside anomalous string lengths trigger an immediate DNS Tunneling Alert and log entry.

---

📋 Prerequisites

• Git — Required to clone the repository.
• Docker & Docker Compose — For running Loki, Promtail, and Grafana.
• Python 3.10+ — Required for running the NIDS engine and test suite.
• GCC / Make — Required to compile the native C DPI engine shared object (libdpi.so on Linux, libdpi.dylib on macOS, libdpi.dll on Windows). Recommended before running main.py, since it's now loaded live for injection-signature detection; also used to run benchmark_dpi.py. NetGuard still runs correctly on Aho-Corasick alone if you skip this.
• pyahocorasick — Optional but recommended for the live DPI engine's full performance; a pure-Python fallback is used automatically if it's not installed.
• Administrator / Root Privileges — Required to capture raw socket traffic via Scapy and inject OS Firewall blocking rules (or use the least-privilege setcap option below on Linux).
• Npcap (Windows only) — Required for Scapy to capture raw packets on Windows network adapters.

---

📝 JSON Log Structure (Structured Logging)

{
"timestamp": "2026-08-06T10:30:15.123456+03:00",
"level": "WARNING",
"message": "[PORT SCAN DETECTED] Host 10.0.0.4 scanned 18 unique ports",
"logger": "NetworkGuardian",
"src_ip": "10.0.0.4",
"event_type": "PORT_SCAN",
"details": "18 ports scanned"
}

---

⚙️ Installation & Quick Start

# 1. Clone the repository
git clone https://github.com/RazEini/NetGuard.git
cd NetGuard
# 2. Environment Setup
cp .env.example .env # Set your Grafana password in .env# 3. Start Observability Stack (Grafana, Loki, Promtail)# Grafana will automatically provision all dashboards from grafana/dashboards/
docker compose up -d
# 4. Setup Python Environment
python -m venv .venv
.\.venv\Scripts\activate # On Windowssource .venv/bin/activate # On Linux/Mac
pip install -r requirements.txt
# 5. Compile the Native C DPI Engine (recommended)# main.py auto-detects and loads this at startup for live injection-signature# detection; if you skip this step NetGuard still runs fine on Aho-Corasick alone.
make -C c_src # Linux / macOS
gcc -shared -O3 -march=native -o libdpi.dll c_src/dpi.c # Windows — requires MinGW/MSYS2 installed# Optional: run the standalone throughput benchmark (same shared library, isolated measurement)
python benchmark_dpi.py # Verify ~8x memory-safe speedup

6a. Run NIDS Engine — Linux / macOS

# Principle of Least Privilege - grant raw socket capability without full sudo:
sudo setcap cap_net_raw,cap_net_admin=eip $(readlink -f .venv/bin/python)
.venv/bin/python main.py
# Or run directly with root:
sudo .venv/bin/python main.py

6b. Run NIDS Engine — Windows

# Run PowerShell / CMD as Administrator:
python main.py

7. Run Attack Simulator (in a separate terminal)

# Automatically targets local active IP:
python test_attack.py
# Or target a specific IP address explicitly:
python test_attack.py <TARGET_IP>

📊 Accessing Grafana: Open your browser to http://localhost:3000 (username: admin, password set in .env). All dashboards will already be loaded and ready to use!

---

📄 License

This project is distributed under the MIT license – free to use and modify for educational and research purposes.

---

👨‍💻 Raz Eini (2026)