An Analysis of Universal Information Flow Based on Self-Composition

Abstract

We introduce a novel way of proving information flow properties of a program based on its self-composition. Similarly to the universal information flow type system of Hunt and Sands, our analysis explicitly computes the dependencies of variables in the final state on variables in the initial state. Accordingly, the analysis result is independent of specific information flow lattices, and allows to derive information flow w.r.t. any of these. While our analysis runs in polynomial time, we prove that it never loses precision against the type system of Hunt and Sands, and may gain extra precision by taking similarities between different branches of conditionals into account. Also, we indicate how it can be smoothly generalized to an interprocedural analysis.