Cache Attacks and Defenses

Abstract

In the digital age, as our daily lives depend heavily on interconnected computing devices, information security has become a crucial concern. The continuous exchange of data between devices over the Internet exposes our information vulnerable to potential security breaches. Yet, even with measures in place to protect devices, computing equipment inadvertently leaks information through side-channels, which emerge as byproducts of computational activities. One particular source of such side channels is the cache, a vital component of modern processors that enhances computational speed by storing frequently accessed data from random access memory (RAM). Due to their limited capacity, caches often need to be shared among concurrently running applications, resulting in vulnerabilities. Cache side-channel attacks, which exploit such vulnerabilities, have received significant attention due to their ability to stealthily compromise information confidentiality and the challenge in detecting and countering them. Consequently, numerous defense strategies have been proposed to mitigate these attacks. This thesis explores these defense strategies against cache side-channels, assesses their effectiveness, and identifies any potential vulnerabilities that could be used to undermine the effectiveness of these defense strategies. The first contribution of this thesis is a software framework to assess the security of secure cache designs. We show that while most secure caches are protected from eviction-set-based attacks, they are vulnerable to occupancybased attacks, which works just as well as eviction-set-based attacks, and therefore should be taken into account when designing and evaluating secure caches. Our second contribution presents a method that utilizes speculative execution to enable high-resolution attacks on low-resolution timers, a common cache attack countermeasure adopted by web browsers. We demonstrate that our technique not only allows for high-resolution attacks to be performed on low-resolution timers, but is also Turing-complete and is capable of performing robust calculations on cache states. Through this research, we uncover a new attack vector on low-resolution timers. By exposing this vulnerability, we hope to prompt the necessary measures to address the issue and enhance the security of systems in the future. Our third contribution is a survey, paired with experimental assessment of cache side-channel attack detection techniques using hardware performance counters. We show that, despite numerous claims regarding their efficacy, most detection techniques fail to perform proper evaluation of their performance, leaving them vulnerable to more advanced attacks. We identify and outline these shortcomings, and furnish experimental evidence to corroborate our findings. Furthermore, we demonstrate a new attack that is capable of compromising these detection methods. Our aim is to bring attention to these shortcomings and provide insights that can aid in the development of more robust cache side-channel attack detection techniques. This thesis contributes to a deeper comprehension of cache side-channel attacks and their potential effects on information security. Furthermore, it offers valuable insights into the efficacy of existing mitigation approaches and detection methods, while identifying areas for future research and development to better safeguard our computing devices and data from these insidious attacks.