A novel architecture to virtualise a hardware-bound trusted platform module

Abstract

Security and trust are particularly relevant in modern softwarised infrastructures, such as cloud environments, as applications are deployed on platforms owned by third parties, are publicly accessible on the Internet and can share the hardware with other tenants. Traditionally, operating systems and applications have leveraged hardware tamper-proof chips, such as the Trusted Platform Modules (TPMs) to implement security workflows, such as remote attestation, and to protect sensitive data against software attacks. This approach does not easily translate to the cloud environment, wherein the isolation provided by the hypervisor makes it impractical to leverage the hardware root of trust in the virtual domains. Moreover, the scalability needs of the cloud often collide with the scarce hardware resources and inherent limitations of TPMs. For this reason, existing implementations of virtual TPMs (vTPMs) are based on TPM emulators. Although more flexible and scalable, this approach is less secure. In fact, each vTPM is vulnerable to software attacks both at the virtualised and hypervisor levels. In this work, we propose a novel design for vTPMs that provides a binding to an underlying physical TPM; the new design, akin to a virtualisation extension for TPMs, extends the latest TPM 2.0 specification. We minimise the number of required additions to the TPM data structures and commands so that they do not require a new, non-backwards compatible version of the specification. Moreover, we support migration of vTPMs among TPM-equipped hosts, as this is considered a key feature in a highly virtualised environment. Finally, we propose a flexible approach to vTPM object creation that protects vTPM secrets either in hardware or software, depending on the required level of assurance.