Abstract
The rapid expansion of the Internet of Things (IoT) has introduced significant security vulnerabilities, including DDoS attacks and embedded malware. Effective vulnerability analysis is hampered by the proprietary nature of diverse manufacturers, making timely firmware patching difficult. Identifying specific firmware versions is a critical prerequisite for securing these devices. We propose a novel identification method that analyzes webpage content accessed via weak passwords. Our approach uses clustering to determine device type and brand, followed by page segmentation to identify specific models and firmware versions. Evaluating 74,307 devices, our method achieved a 95.97% accuracy rate, outperforming existing techniques.
Introduction
The Internet of Things (IoT) landscape is a security minefield. When vulnerabilities are discovered, researchers must identify which devices are at risk. However, obtaining "fine-grained" data—the exact firmware version—is challenging. Existing search engines like Shodan provide only "coarse-grained" information (brand or type). Traditional fine-grained identification often requires reverse engineering firmware binaries, which are frequently encrypted or proprietary. This paper addresses the "granularity gap" by identifying firmware through web management interfaces without needing binary access.
Methodology: A Three-Stage Pipeline
The proposed method shifts focus from the firmware binary to the structured web management interface generated by the device.
1. Coarse-grained Clustering
The process begins by scanning the IPv4 space using Zmap to find active hosts on ports 80 and 8080. For each discovered login page, three primary features are extracted: the Document Object Model (DOM) tree, text content, and HTTP response headers. The DBSCAN algorithm then clusters these devices by brand and type based on structural and content similarity.
2. Weak Password Entry
For each identified cluster, the system attempts to log in using known factory default credentials (e.g., admin/admin). This targeted check leverages the widespread issue of default passwords to gain access to internal management pages.
3. Fine-grained Identification
Once authenticated, the system must locate the firmware version string, which is typically buried in sub-menus. A One-Class SVM identifies the relevant navigation links (e.g., "System Status"). The Block-o-Matic (BoM) page segmentation algorithm then isolates the specific logical block containing the version information, which is extracted via regular expressions.
Experimental Results
The method was evaluated using a dataset of 74,307 active IoT devices. To verify accuracy, a manually labeled validation set of 950 samples across 19 brands was used.
- Accuracy: The system achieved a 95.97% identification accuracy, significantly higher than traditional segmentation methods like VIPS (57.1%) or BCS (70.1%).
- Efficiency: Identification takes an average of 73ms per device, enabling large-scale internet scanning.
- Comparative Performance: In tests against frameworks like PANDORA and MACF using 1,000 TP-Link routers, this method consistently provided more specific version data across various release iterations.
Conclusion and Future Work
By utilizing web interfaces as structural "fingerprints" and leveraging default credentials, this research provides a scalable way to map the IoT firmware landscape. While currently dependent on weak passwords and web-based interfaces, future work will focus on identifying firmware through unauthenticated information leaks and expanding support for industrial control equipment.
Summary updated