| Popis: |
We would all benefit if software were slimmer, thinner, and generally only did what we needed and nothing more. To this end, our research team has been exploring methods for removing unused and undesirable features from compiled programs. Our primary goal is to improve software security by removing rarely used features in order to decrease a pro- gram's attack surface. We describe two different approaches for "thinning" binary images of compiled programs. The first approach removes specific program features using dynamic tracing as a guide. This approach is safer than many alterna- tives, but is limited to removing code which is reachable in a trace when an undesirable feature is enabled. The second ap- proach uses a genetic algorithm (GA) to mutate a program until a suitable variant is found. Our GA-based approach can potentially remove any code that is not strictly required for proper execution, but may break program semantics in unpredictable ways. We show results of these approaches on a simple program and real-world software and explore some of the implications for software security. |
