A semantic approach to secure information flow

A classic problem in security is to determine whether a program has secure information flow. Informally, this problem is described as follows: Given a program with variables partitioned into two disjoint sets of “high-security” and “low-security” variables, check whether observations of the low-security variables reveal any information about the initial values of the high-security variables. Although the problem has been studied for several decades, most previous approaches have been syntactic in nature, often using type systems and compiler data flow analysis techniques to analyze program texts. This paper presents a considerably different approach to checking secure information flow, based on a semantic characterization. A semantic approach has several desirable features. Firstly, it gives a more precise characterization of security than that provided by most previous approaches. Secondly, it applies to any programming constructs whose semantics are definable; for instance, the introduction of nondeterminism and exceptions poses no additional problems. Thirdly, it can be used for reasoning about indirect leaking of information through variations in program behavior (e.g., whether or not the program terminates).