NARRATOR: Secure and Practical State Continuity for Trusted Execution in the Cloud

Public cloud platforms have leveraged Trusted Execution Environment (TEE) technology to provide confidential computing services. However, TEE-protected applications still suffer from rollback or forking attacks, in which their states could be rolled back to a stale version or be forked into multiple versions, resulting in state continuity violations. Existing solutions against these attacks either rely on weak threat models based on centralized trust (e.g., trusted server) or suffer from large performance overheads (e.g., tens of state updates per second). In this paper, we propose Narrator, a secure and practical system, (1) that relies on a blockchain (i.e., decentralized trust) and TEEs, and ( 2 ) that provides high-performance state continuity protection like unlimited and fast state updates for applications in cloud TEEs. The intuition behind our design is simple. Our design uses the blockchain to initialize a distributed system of TEEs, laying down the decentralized trust base with a small interaction overhead, while the distributed system provides performant state continuity protection. Our distributed system adopts a customized version of the consistent broadcast protocol and leverages advanced techniques to make state updates processed with one round trip delay on average. We build a proof-of-concept of Narrator on Intel SGX (i.e., a representative design of TEEs) and do extensive experiments to evaluate its performance. Our evaluation results show that in a LAN environment with 5 nodes, Narrator can support about 6k state updates per second, meanwhile keeping the latency as low as 3 -8ms. The throughput is 30× larger than that in ROTE and 70× larger than using a TPM counter. CCS CONCEPTS • Security and privacy → Security in hardware.