Experiences with PDG-Based IFC

Information flow control systems provide the guarantees that are required in today's security-relevant systems. While the literature has produced a wealth of techniques to ensure a given security policy, there is only a small number of implementations, and even these are mostly restricted to theoretical languages or a subset of an existing language. Previously, we presented the theoretical foundations and algorithms for dependence-graph-based information flow control (IFC). As a complement, this paper presents the implementation and evaluation of our new approach, the first implementation of a dependence-graph based analysis that accepts full Java bytecode. It shows that the security policy can be annotated in a succinct manner; and the evaluation shows that the increased runtime of our analysis—a result of being flow-, context-, and object-sensitive—is mitigated by better analysis results and elevated practicability. Finally, we show that the scalability of our analysis is not limited by the sheer size of either the security lattice or the dependence graph that represents the program.