Interactive and targeted runtime verification using a debugger-based architecture

Abstract

Runtime verification of software (RV) often relies on two categories of tools : dynamic heavy-weight tools, which significantly impact performance, and lighter and more efficient but static tools, which require recompiling the binary. In this paper we propose a new framework for building efficient and targeted dynamic RV tools, bridging the gap between those two categories. This framework is separated into two domains : source and binary. On the source level, a modular development environment provides a custom user interface which allows for precise targeting of instrumentation, as well as advanced interactivity. The binary level revolves around a debugger, which controls binary manipulation and library loading. In order to create fully dynamic tools, we added new instrumentation capabilities to the GNU debugger, using trampoline-based probes to inject code in the binary efficiently and interactively. Our framework focuses on accessibility for users via the graphical interface, and for developers by making it easy to adapt existing tools and by relying on popular programming languages such as Python and C++. As a demonstration of our framework capabilities, we provide a significantly faster implementation of conditional breakpoints for GDB, as well as targeted and fully dynamic versions of two state-of-the-art runtime verification tools : Address Sanitizer and Data Watch.