Improving systems software security through program analysis and instrumentation

Security and reliability bugs are prevalent in systems software. Systems code is often written in low-level languages like C/C++, which offer many benefits but also delegate memory management and type safety to programmers. This invites bugs that cause crashes or can be exploited by attackers to take control of the program. This thesis presents techniques to detect and fix security and reliability issues in systems software without burdening the software developers. First, we present code-pointer integrity (CPI), a technique that combines static analysis with compile-time instrumentation to guarantee the integrity of all code pointers in a program and thereby prevent all control-flow hijack attacks. We also present code-pointer separation (CPS), a relaxation of CPI with better performance properties. CPI and CPS offer substantially better security-to-overhead ratios than the state of the art in control flow hijack defense mechanisms, they are practical (we protect a complete FreeBSD system and over 100 packages like apache and postgresql), effective (prevent all attacks in the RIPE benchmark), and efficient: on SPEC CPU2006, CPS averages 1.2% overhead for C and 1.9% for C/C++, while CPIâ s overhead is 2.9% for C and 8.4% for C/C++. Second, we present DDT, a tool for testing closed-source device drivers to automatically find bugs like memory errors or race conditions. DDT showcases a combination of a form of program analysis called selective symbolic execution with virtualization to thoroughly exercise tested drivers and produce detailed, executable traces for every path that leads to a failure. We applied DDT to several closed-source Microsoft-certified Windows device drivers and discovered 14 serious new bugs that can cause crashes or compromise security of the entire system. Third, we present a technique for increasing the scalability of symbolic execution by merging states obtained on different execution paths. State merging reduces the number of states to analyze, but the merged states can be more complex and harder to analyze than their individual components. We introduce query count estimation, a technique to reason about the analysis time of merged states and decide which states to merge in order to achieve optimal net performance of symbolic execution. We also introduce dynamic state merging, a technique for merging states that interacts favorably with search strategies employed by practical bug finding tools, such as DDT and KLEE. Experiments on the 96 GNU Coreutils show that our approach consistently achieves several orders of magnitude speedup over previously published results.