Software Integrity Checking Expressions (ICEs) for Robust Tamper Detection

We introduce software integrity checking expressions (SoftICEs), which are program predicates that can be used in software tamper detection. We present two candidates, probabilistic verification conditions (PVCs) and Fourier-learning approximations (FLAs), which can be computed for certain classes of programs,. We show that these predicates hold for any valid execution of the program, and fail with some probability for any invalid execution (e.g., when the output value of one of the variables is tampered). PVCs work with straight-line integer programs that have operations { ∗, +,− }. We also sketch how we can extend this class to include branches and loops. FLAs can work over programs with arbitrary operations, but have some limitations in terms of efficiency, code size, and ability to handle various classes of functions. We describe a few applications of this technique, such as program integrity checking, program or client identification, and tamper detection. As a generalization of oblivious hashing (OH), our approach resolves several troublesome issues that complicate practical application of OH towards