Cautiously Optimistic Program Analyses for Secure and Reliable Software

Modern computer systems still have various security and reliability vulnerabilities. Well-known dynamic analyses solutions can mitigate them using runtime monitors that serve as lifeguards. But the additional work in enforcing these security and safety properties incurs exorbitant performance costs, and such tools are rarely used in practice. Our work addresses this problem by constructing a novel technique- Cautiously Optimistic Program Analysis (COPA). COPA is optimistic- it infers likely program invariants from dynamic observations, and assumes them in its static reasoning to precisely identify and elide wasteful runtime monitors. The resulting system is fast, but also ensures soundness by recovering to a conservatively optimized analysis when a likely invariant rarely fails at runtime. COPA is also cautious- by carefully restricting optimizations to only safe elisions, the recovery is greatly simplified. It avoids unbounded rollbacks upon recovery, thereby enabling analysis for live production software. We demonstrate the effectiveness of Cautiously Optimistic Program Analyses in three areas: Information-Flow Tracking (IFT) can help prevent security breaches and information leaks. But they are rarely used in practice due to their high performance overhead (>500% for web/email servers). COPA dramatically reduces this cost by eliding wasteful IFT monitors to make it practical (9% overhead, 4x speedup). Automatic Garbage Collection (GC) in managed languages (e.g. Java) simplifies programming tasks while ensuring memory safety. However, there is no correct GC for weakly-typed languages (e.g. C/C++), and manual memory management is prone to errors that have been exploited in high profile attacks. We develop the first sound GC for C/C++, and use COPA to optimize its performance (16% overhead). Sequential Consistency (SC) provides intuitive semantics to concurrent programs that simplifies reasoning for their correctness. However, ensuring SC behavior on commodity hardware remains expensive. We use COPA to ensure SC for Java at the language-level efficiently, and significantly reduce its cost (from 24% down to 5% on x86). COPA provides a way to realize strong software security, reliability and semantic guarantees at practical costs.PhDComputer Science & EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/170027/1/subarno_1.pd