Memory access pattern randomization to defend against a controlled-channel attack

Abstract

New hardware features such as Intel SGX provide trusted execution environments for user processes called enclaves running on top of an untrusted OS. However, recent studies showed that such protected enclaves are vulnerable to controlled-channel attack, which uses page faults to get memory access patterns. Controlled-channel attack infers an enclave’s control flow or sensitive data from the observed memory access patterns. Several defense techniques including software-based and hardware-based approaches are proposed, but a state-of-the-art study proposed a new way of getting memory access pattern of an enclave without incurring page faults. This paper randomizes the memory access pattern of an enclave through fine-grained ASLR and annotated data layout randomization. Even a potentially malicious OS can observe memory access patterns of an enclave, it is hard to infer valuable information from them. We demonstrated that our defense method protects AES from controlled-channel attack with little performance overhead.