Efficient and Precise Information Flow Control for Machine Code through Demand-Driven Secure Multi-Execution

Dynamic Information Flow Control (IFC) systems, like No-Sensitive-Upgrade or Permissive-Upgrade, can guarantee Termination-Insensitive Non-Interference, but reject valid programs due to their inability to track implicit flows. More advanced multi-execution based approaches, like Shadow Execution and Secure Multi-Execution, are precise and guarantee Termination-Sensitive Non-Interference, but require additional resources or, in the case of Faceted Evaluation, deep changes to the execution semantics. In this paper, we propose a novel efficient and precise Information Flow Control system for machine code through Demand-Driven Secure Multi-Execution. Our key idea is to use lightweight single-execution monitoring as long as the execution is secretless and fork multiple copies on-demand when necessary. We present the first Secure Multi-Execution implementation for legacy code in Unix-based environments and show that our demand-driven optimization drastically reduces the run-time overhead for cat and sha256sum. Our results indicate that further acceleration is possible through improved static analyses, making multi-execution based IFC systems applicable to machine code.