Proving Memory Safety of the ANI Windows Image Parser using Compositional Exhaustive Testing

Abstract. We report in this paper how we proved memory safety of a complex Windows image parser written in low-level C in only three months of work and using only three core techniques, namely (1) sym-bolic execution at the x86 binary level, (2) exhaustive program path enu-meration and testing, and (3) user-guided program decomposition and summarization. We also used a new tool, named MicroX, for executing code fragments in isolation using a custom virtual machine designed for testing purposes. As a result of this work, we are able to prove, for the first time, that a Windows image parser is memory safe, i.e., free of any buffer-overflow security vulnerabilities, modulo the soundness of our tools and several additional assumptions regarding bounding input-dependent loops, fixing a few buffer-overflow bugs, and excluding some code parts that are not memory safe by design. In the process, we also discovered and fixed several limitations in our tools, and narrowed the gap between systematic testing and verification.