Towards a framework for algorithm recognition in binary code.

Algorithm recognition, which is the problem of verifying whether a program implements a given algorithm, is an important topic in program analysis. We propose an approach for algorithm recognition in binary code. For this paper, we have chosen the Dalvik Virtual Machine (DVM) bytecode. Given an algorithm A that is compiled into a DVM method M0, and a DVM program P that includes a series of methods {M1,..., Mn}, the approach is able to identify those blocks Mi from P that essentially implement the algorithm A. The technique we propose first translates binary code into Horn clauses. Then we consider programs as implementing the same algorithm if their Horn clause representations can be reduced to a single common set of Horn clauses by means of a sequence of transformations.