Detecting Communication Protocol Security Flaws by Formal Fuzz Testing and Machine Learning

摘要

Network-based fuzz testing has become an effective mechanism to ensure the security and reliability of communication protocol systems. However, fuzz testing is still conducted in an ad-hoc manner with considerable manual effort, which is mainly due to the unavailability of protocol model. In this paper we present our on-going work of developing an automated and measurable protocol fuzz testing approach that uses a formally synthesized approximate formal protocol specification to guide the testing process. We adopt the Finite State Machine protocol model and study two formal methods for protocol synthesis: an active black-box checking algorithm that has provable optimality and a passive trace minimization algorithm that is less accurate but much more efficient. We also present our preliminary results of using this method to implementations of the MSN instant messaging protocol: MSN clients Gaim (pidgin) and aMSN. Our testing reveals some serious reliability and security flaws by automatically crashing both of them.