Detecting Android Inter-App Data Leakage via Compositional Concolic Walking

While many research efforts have been around auditing individual android apps, the security issues related to the interaction among multiple apps are less studied. Due to the hidden nature of Inter-App communications, few existing security tools are able to detect such related vulnerable behaviors. This paper proposes to perform overall security auditing using dynamic analysis techniques. We focus on data leakage as it is one of the most common vulnerabilities for Android applications. We present an app auditing system AppWalker, which uses concolic execution on a set of apps. We use static Inter-App taint analysis to guide the dynamic auditing procedure, so that we can target at potential Inter-App data leakage. To mitigate the exponential blow-up when auditing various combinations of apps, we introduce a novel technique called compositional concolic walking. In the end of the auditing, the event and data inputs created during concolic walking are fed to the app set. By dynamically checking the triggered data-leaking behavior, we are then able to confirm the existence of Inter-App data leakage. AppWalker takes into account both infra- and inter-app communications, and is the first research work on dynamic audit of inter-app vulnerabilities in a path-sensitive way to our knowledge. Experimental results reveal that our method can effectively detect real-world Inter-App data leakage.