TraceDroid: A Robust Network Traffic Analysis Framework for Privacy Leakage in Android Apps

Abstract

Network traffic analysis is an appealing approach for the security auditing of mobile apps. Prior research employs various techniques (e.g., Man-in-the-Middle, TCPDUMP) to capture network traffic from apps and further recognize security/privacy risks inside. However, these techniques suffer from limitations such as traffic mixing, proxy evasion, and SSL pinning. Possible solutions are to modify and customize the Android system. However, existing studies are mainly based on Android OS 6/7. Contemporary apps generally cannot work properly on these archaic Android OS, which has become a stumbling block for further traffic analysis research. To address the above problems, we propose a new network traffic analysis framework-TraceDroid. We first leverage the dynamic hooking technique to hook the critical functions for sending network requests, and then save the request data along with code execution traces. Besides, TraceDroid proposes an unsupervised way to identify third-party libraries (TPLs) inside apps for facilitating the liability analysis between apps and TPLs. Utilizing TraceDroid, we conduct a large-scale experiment on 9,771 real-world apps to make an empirical study of the status quo of privacy leakage. Our findings show that TPLs account for 44.45% of privacy leakage in contemporary apps, and files transmitted from user devices contain much more detailed privacy data than network requests. We bring to light the over-data harvest and cross-library data harvest issues in apps. Furthermore, we unveil the relationship between TPLs and their visiting domains that previous research has never discussed.