Semantics for Noninterference with Interaction Trees

Noninterference is the strong information-security property that a program does not leak secrets through publicly-visible behavior. In the presence of effects such as nontermination, state, and exceptions, reasoning about noninterference quickly becomes subtle. We advocate using interaction trees (ITrees) to provide compositional mechanized proofs of noninterference for multi-language, effectful, nonterminating programs, while retaining executability of the semantics. We develop important foundations for security analysis with ITrees: two indistinguishability relations, leading to two standard notions of noninterference with adversaries of different strength, along with metatheory libraries for reasoning about each. We demonstrate the utility of our results using a simple imperative language with embedded assembly, along with a compiler into that assembly language.