Security for multithreaded programs under cooperative scheduling

Information flow exhibited by multithreaded programs is subtle because the attacker may exploit scheduler properties when deducing secret information from publicly observable outputs. Volpano and Smith have introduced a protect command that prevents the scheduler from observing sensitive timing behavior of protected commands and therefore prevents undesired information flows. While a useful construct, protect is nonstandard and difficult to implement. This paper presents a transformation that eliminates the need for protect under cooperative scheduling. We show that both termination-insensitive and termination-sensitive security can be enforced by variants of the transformation in a language with dynamic thread creation.