Combining Confidential Computing and Unikernels for Lightweight Confidential Computing

Zenodo (CERN European Organization for Nuclear Research)

Abstract

While the integrity and confidentiality of data can be guaranteed by confidential virtual machines (CVMs), this is only part of the chain-of-trust when deploying sensitive workloads on untrusted infrastructure. Any vulnerability inside the CVM can potentially undermine its integrity, making verification and hardening of the (full) software stack essential parts of the security of the system. Unikernels—library operating systems that statically link to the application to produce a minimal bootable image—are effective approaches to reduce the Trusted Computing Base (TCB). We present SEV-Unikernel, the first general purpose unikernel compatible with AMD’s CVM, SEV-SNP. SEV-Unikernel is language agnostic and runs in a standard hypervisor without any external dependencies. Our proto­type fully supports the native SEV attestation mechanism, enabling the measurement and attestation of the entire soft­ware stack and the proactive detection of integrity faults. We demonstrate the practicality of SEV-Unikernel on a variety of microbenchmarks and real-world workloads. Our experimental results show that SEV-Unikernel achieves a significant TCB reduction with respect to confidential Linux virtual machines. We observe almost no overhead on com­pute intensive workloads, but implementation weaknesses in the network stack cause a 1.3x to 12x overhead for network-intensive benchmarks.

Authors 0

1. Author list not loaded yet.