Abstract
As an essential component of critical infrastructure, Industrial Control Systems (ICS) face increasing cyber threats, exacerbated by the emergence of the Shodan search engine. By identifying and indexing Internet-connected industrial devices, Shodan has become a primary tool for both researchers and potential attackers. This paper explores Shodan's scanning behavior through the deployment of a distributed honeypot network. We propose a hierarchical Deterministic Finite Automata (DFA) and Support Vector Machine (SVM) recognition model to identify Shodan scans based on function codes and traffic features. Our results demonstrate that this model outperforms traditional reverse IP resolution. Furthermore, we provide an in-depth analysis of Shodan's scanning patterns, including frequency, regional preferences, and protocol targeting, and suggest defensive measures to mitigate these threats.
Introduction
Industrial Control Systems (ICS)—the hardware and software that run power plants, water treatment facilities, and factories—were historically isolated from public networks. However, the modern push for connectivity has exposed these systems to the Internet, often without adequate security. The Shodan search engine serves as a powerful tool that indexes these devices, effectively creating a "hit list" for attackers.
Traditional methods for identifying automated scanners, such as reverse DNS lookups, are often incomplete or easily bypassed by stealthy actors. This research aims to unmask Shodan’s scanning logic and develop a robust detection framework to protect critical infrastructure from reconnaissance efforts.
Methodology and Data Collection
To capture real-world scanning data, researchers deployed a distributed honeypot system consisting of six virtual servers located in the US, China, Russia, Singapore, and Brazil. These nodes simulated various Programmable Logic Controllers (PLCs) and supported common industrial protocols, including Modbus, S7comm, IEC 60870-5-104, and BACnet.
Over a three-month period, the study collected 145,720 packets, focusing on 32,522 "interactive" packets where a successful handshake occurred. To classify this traffic, a hierarchical DFA-SVM recognition model was developed:
- DFA Layer: Analyzes the logic of communication by matching sequences of industrial "function codes" against known Shodan scanning signatures.
- SVM Layer: Uses machine learning to identify "Shodan-like" scanners that lack known signatures. The model extracts 13 traffic features, such as packet length and arrival intervals, to distinguish automated probes from background network noise.
Analysis of Shodan Scanning Behavior
The investigation revealed several key characteristics of Shodan’s operation:
- The Reconnaissance Pipeline: A significant surge in general attack traffic was observed immediately after a honeypot was indexed by Shodan, confirming its role as a primary discovery tool for attackers.
- Scanning Cycles: Data suggests Shodan completes a full IPv4 scan approximately every 66 hours. It typically takes between 3.5 to 15 days for a new device to appear in Shodan’s searchable web interface.
- Regional and Protocol Preferences: Scanning intensity is highest in Asia, followed by North America and Europe. Furthermore, Shodan’s scanning logic appears tailored to specific protocols; for instance, scanners targeting S7comm showed higher success rates than those targeting Modbus.
Experimental Results
The proposed hierarchical model demonstrated exceptional performance in identifying scanning activity across various industrial protocols.
As shown in the figure above, the DFA-SVM model consistently achieved higher accuracy than the standalone DFA approach, particularly when dealing with the full interactive dataset ("All Data"). The model reached an average recognition accuracy of 99.3% for known Shodan traffic and 95.6% across the broader dataset. Crucially, it successfully unmasked 16 "Shodan-like" scanners that lacked PTR records, making them invisible to traditional reverse IP resolution methods.
Conclusion and Defensive Measures
While Shodan provides valuable transparency for security researchers, it also lowers the barrier to entry for attacking critical infrastructure. The DFA-SVM model provides a highly accurate means of detecting these reconnaissance efforts in real-time.
To mitigate the risks posed by Shodan, defensive measures such as modifying "banner" information (the identifying text a device sends to scanners) and implementing network-level filtering based on the traffic signatures identified in this study are recommended. Future work will focus on optimizing the model for deployment on low-power, real-time Intrusion Detection Systems (IDS) specifically designed for industrial environments.
Summary updated