TDroid: Exposing App Switching Attacks in Android with Control Flow Specialization

Abstract

The Android multitasking mechanism can be plagued with app switching attacks, in which a malicious app replaces the legitimate top activity of the focused app with one of its own, thus mounting, e.g., phishing and denial-of-service attacks. Existing market-level defenses are still ineffective, as static analysis is fundamentally unable to reason about the intention of an app and dynamic analysis has low coverage. We introduce TDroid, a new market-level approach to detecting app switching attacks. The challenge lies in how to handle a plethora of input-dependent branch predicates (forming an exponential number of paths) that control the execution of the code responsible for launching such attacks. TDroid tackles this challenge by combining static and dynamic analysis to analyze an app without producing any false positives. In its static analysis, TDroid transforms the app into runnable slices containing potentially app switching attacks, one slice per attack. In its dynamic analysis, TDroid executes these slices on an Android phone or emulator to expose their malicious GUIs. The novelty lies in the use of a new trigger-oriented slicing technique in producing runnable slices so that certain input-dependent branch predicates are specialized to execute always some fixed branches. Evaluated with a large set of malware apps, TDroid is shown to outperform the state of the art, by detecting substantially more app switching attacks, in a few minutes per app, on average.