SymDroid: Symbolic Execution for Dalvik Bytecode

Abstract

Apps on Google's Android mobile device platform are written in Java, but are compiled to a special bytecode language called Dalvik. In this paper, we introduce SymDroid, a symbolic executor that operates directly on Dalvik bytecode. SymDroid begins by first translating Dalvik into mu-Dalvik, a simpler language that has only 16 instructions, in contrast to Dalvik's more than 200 instructions. We present a formalism for SymDroid's symbolic executor, which can be described with a small number of operational semantics rules; this semantics may be of independent interest. In addition to modeling bytecode instructions, SymDroid also contains models of some key portions of the Android platform, including libraries and the platform's lifecycle control code. We evaluated SymDroid in two ways. First, we ran it on the Android Compatibility Test Suite, and found it passed all tests except ones that used library or system routines we have not yet implemented. On this test suite, SymDroid runs about twice as slow as the Dalvik VM, and about twice as fast as the Java VM. Second, we used SymDroid to discover the (path) conditions under which contacts may be accessed on an Android app, and found it was able to do so successfully. These results suggest that SymDroid, while still a prototype, is a promising first step in enabling direct, precise analysis of Android apps.