Mitigating Side-Channel Leaks in Next-Generation Cloud Systems

点击翻译按钮获取中文摘要

Background, literature, and recent progress Online services, such as social media, education, financial, healthcare, and e-commerce services, are increasingly hosted in public clouds. However, a growing concern among the service providers is that their sensitive data hosted in the cloud (e.g., cryptographic keys, ML algorithms, medical and financial records) can be revealed to adversaries via side channels. Side channels arise in shared physical resources (e.g., CPUs, memory, caches, network). Upcoming trends in cloud, serverless computing and hardware disaggregation, further increase fine-grained resource sharing among applications of untrusting tenants, leading to more side-channel leaks. Side-channel leaks can be broadly categorized into three classes: (i) leaks via network, (ii) leaks via access patterns of server resources, such as memory or caches, and (iii) leaks due to speculative execution in CPUs. Existing mitigation solutions for side-channel leaks either do not eliminate all leaks or are too inefficient for cloud workloads, and apply to only a limited class of applications. My recent work provides foundations for eliminating side-channel leaks via network and memory access patterns by design however, challenges to scale and generalize the solutions remain. Objective, Methodology and HQP My goal is to build secure, efficient, and general-purpose solutions to mitigate all classes of side-channel leaks by design in the next-generation cloud systems, specifically in the context of emerging trends of serverless computing and hardware disaggregation. For each class of leaks, I will first build tools to identify correlations between the application secrets and the side-channel observations and compute strategies to break the correlations. Next, I will build sound and complete mitigation systems that implement the strategies, and will empirically evaluate their security properties and performance characteristics. Over the five-year term of this Discovery Grant, I will train 2 PhD, 3 MSc, and 5 undergraduate students. These HQP will receive training in systems and security research, and in various technologies in operating systems, networking, programming languages and compiler tools, and CPU architecture. They will be able to provide valuable expertise on handling data privacy challenges in the Canadian software industry. Impact The proposed research will develop principles, methods, and tools to mitigate side-channel leaks by design, while still allowing efficient resource sharing in the next-generation cloud systems. The tools will enable cloud providers to offer a trustworthy environment for their tenants, while tenants will be able to guarantee stronger privacy for the sensitive user data they process. Furthermore, the research will provide principles for designing side-channel mitigation systems for other resource-sharing environments, such as mobile and IoT devices.