Cache-based Side-channel Attacks in Multi-tenant Public Clouds and Their Countermeasures

Yinqian Zhang: Cache-based Side-Channel Attacks in Multi-Tenant Public Clouds and Their Countermeasures (Under the direction of Michael Reiter) Cloud computing is gaining traction due to the business agility, resource scalability and operational efficiency that it enables. However, the murkiness of the security assurances offered by public clouds to their tenants is one of the major impediments to enterprise and government adoption of cloud computing. This dissertation explores one of the major design flaws in modern public clouds, namely insufficient isolation among cloud tenants as evidenced by the cloud’s inability to prevent side-channel attacks between co-located tenants, in both Infrastructure-as-a-Service (IaaS) clouds and Platform-as-a-Service (PaaS) clouds. Specifically, we demonstrate that one virtual machine (VM) can successfully exfiltrate cryptographic private keys from another VM co-located on the same physical machine using a cache-based side-channel attack, which calls into question the established belief that the security isolation provided by modern virtualization technologies remains adequate under the new threat model in multi-tenant public IaaS clouds. We have also demonstrated in commercial PaaS clouds that cache-based side channels can penetrate container-based isolation by extracting sensitive information from the execution paths of the victim applications, thereby subverting their security. Finally, we devise two defensive techniques for the IaaS setting, which can be adopted by cloud tenants immediately on modern cloud platforms without extra help from cloud providers, to address side-channel threats: (1) for tenants requiring a high degree of security and physical isolation, a tool to facilitate cloud auditing of such isolation; and (2) for tenants who use multi-tenant cloud services, an operating-system-level defense to defend against cache-based side-channel threats on their own.