I/O Interaction Analysis of Binary Code

Abstract

The increasing use of closed-source software in everyday life leads to a need for privacy verification of binary code. Unfortunately, analyses of machine code behavior are often imprecise, do not scale, or require expert interaction. We present a novel and fast algorithm that is capable of soundly approximating I/O behavior of binary code. The key idea of our algorithm is that overall I/O behavior can be approximated by analyzing input and output operations. Since most function parameters are defined close to their use, our backward symbolic execution algorithm can quickly recover most meaningful parameters. We show the applicability and performance of our approach by analyzing the coreutils binaries.