Backdoor detection systems for embedded devices

A system is said to contain a backdoor when it intentionally includes a means to trigger the execution of functionality that serves to subvert its expected security. Unfortunately, such constructs are pervasive in software and systems today, particularly in the firmware of commodity embedded systems and “Internet of Things” devices. The work presented in this thesis concerns itself with the problem of detecting backdoor-like constructs, specifically those present in embedded device firmware, which, as we show, presents additional challenges in devising detection methodologies. The term “backdoor”, while used throughout the academic literature, by industry, and in the media, lacks a rigorous definition, which exacerbates the challenges in their detection. To this end, we present such a definition, as well as a framework, which serves as a basis for their discovery, devising new detection techniques and evaluating the current state-of-the-art. Further, we present two backdoor detection methodologies, as well as corresponding tools which implement those approaches. Both of these methods serve to automate many of the currently manual aspects of backdoor identification and discovery. And, in both cases, we demonstrate that our approaches are capable of analysing device firmware at scale and can be used to discover previously undocumented real-world backdoors.