A Study on the Vulnerabilities of Mobiles Apps associated with Software Modules

Takuya Watanabe, Mitsuaki Akiyama, Fumihiro Kanei, Eitaro Shioji, Yuta Takata, Bo Sun, Yuta Ishii, Toshiki Shibahara, Takeshi Yagi, Tatsuya MoriarXiv 2017 · arXiv

TL;DR

This paper presents a large-scale study that quantifies how software libraries contribute to mobile app vulnerabilities. By developing novel frameworks to detect and classify libraries, the authors found that a significant majority of vulnerabilities in both free and paid apps stem from these third-party components, with surprisingly, more expensive apps exhibiting more vulnerabilities.

What Problem Does This Paper Address?

Mobile applications are increasingly complex, relying heavily on third-party software libraries to accelerate development and enhance functionality. While these libraries offer numerous benefits, they can also introduce security vulnerabilities. Existing research has touched upon this issue, but a systematic, large-scale quantification of the extent to which libraries are the source of these vulnerabilities, and how this varies across different app types (free vs. paid, popular vs. random), has been lacking. This paper aims to fill that gap by providing a comprehensive analysis of library-associated vulnerabilities in mobile apps.

Methods

The authors developed two key frameworks:

  • Droid-L (Library Detector): This system automatically detects and classifies software libraries within an Android application package (APK). It works by de-compiling the APK, extracting packages, computing fingerprints for these packages, and comparing them against a fingerprint database. Libraries are then categorized as "official," "private," or "third-party." A "dead code checker" is also employed to identify and exclude code that is not actively used.
  • Droid-V (Vulnerability Checker): This framework uses multiple vulnerability scanners to detect specific types of vulnerabilities within an app. By combining the outputs of Droid-L and Droid-V, the researchers can pinpoint which specific libraries are responsible for detected vulnerabilities.

The study also employed a unique approach by analyzing both free and paid apps, and by re-analyzing apps six months later to observe changes in vulnerabilities.

Datasets

The study utilized two primary datasets:

  • Fingerprint Database Construction: 2 million free apps and 30,000 paid apps were collected from the official Android marketplace to build the library fingerprint database.
  • Vulnerability Analysis: A dataset of 30,000 paid apps and a large sample of free apps (details on the exact number of free apps for this phase are not explicitly stated but implied to be substantial, potentially from the 2M collected for the DB) were used for the vulnerability analysis. The paid apps were further categorized into "top-K" (most influential) and "randomly sampled" sets, with similar categorizations for free apps. The dataset was also updated six months later to track changes.

Key Insights and Ideas

  • Libraries are the Primary Culprits: The most striking finding is that a vast majority of app vulnerabilities originate from software libraries. This highlights a critical dependency and a significant security risk that is often outside the direct control of the app developer.
  • The Paradox of Paid Apps: Counterintuitively, more expensive and popular paid apps tend to have more vulnerabilities. This is attributed to their increased functionality, which translates to more code and, consequently, more libraries, thus increasing the attack surface.
  • Stale Libraries in Paid Apps: Paid apps are updated less frequently than free apps, meaning vulnerable libraries within them remain unpatched for longer periods. This suggests that the "sell-once-and-that's-it" model for some paid apps can lead to prolonged security risks.
  • Dead Code is a Significant Factor: Approximately half of the detected vulnerabilities were found in dead code. This underscores the importance of dead code detection in vulnerability analysis tools to avoid false positives and focus on actionable issues.

Key Results

  • Vulnerability Origin: Approximately 70% of free apps with vulnerabilities and 50% of paid apps with vulnerabilities were vulnerable due to libraries.
  • Third-Party Dominance: Third-party libraries accounted for roughly 70-80% of all detected libraries, and were also the primary source of vulnerabilities.
  • Paid App Characteristics: Paid apps generally have fewer installs and are updated less frequently than free apps. More expensive paid apps tend to have more libraries and, paradoxically, more vulnerabilities.
  • Dead Code Impact: Around 40-70% of detected vulnerabilities in libraries were found in dead code, depending on the dataset.

Limitations and Future Work

  • Static Analysis Limitations: Both library detection and vulnerability checking rely on static code analysis, which has inherent limitations. It cannot dynamically track code execution, and may miss vulnerabilities introduced by dynamic code loading or reflection.
  • Native Code: The analysis primarily focused on Java-written components, excluding native code (C/C++) which is common in game apps and can be used to hide malicious actions. Investigating vulnerabilities in native code remains a challenge.
  • Future Work: The authors suggest exploring dynamic code analysis for more comprehensive vulnerability assessment, addressing the challenges of scalability and test scenario generation. They also propose further research into app development environments, economic models, and code reuse to understand the broader origins of mobile app vulnerabilities.

Summary updated

BibTeX
@misc{watanabe2017study,
  author = {Takuya Watanabe and Mitsuaki Akiyama and Fumihiro Kanei and Eitaro Shioji and Yuta Takata and Bo Sun and Yuta Ishii and Toshiki Shibahara and Takeshi Yagi and Tatsuya Mori},
  title = {{A Study on the Vulnerabilities of Mobiles Apps associated with Software Modules}},
  year = {2017},
  eprint = {1702.03112},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1702.03112},
}

Loading related content...

Figure