Total: 1
The monolithic programming model has been favored for high compatibility and easing the programming for SGX enclaves, i.e., running the secure code with all dependent libraries or even library OSes (LibOSes). Yet, it inevitably bloats the trusted computing base (TCB) and thus deviates from the goal of high security. Introducing fine-grained isolation can effectively mitigate TCB bloating while existing solutions face performance issues. We observe that the off-the-shelf Intel MPK is a perfect match for efficient intra-enclave isolation. Nonetheless, the trust models between MPK and SGX are incompatible by design. We hence propose LIGHTENCLAVE, which embraces non-intrusive extensions on existing SGX hardware to incorporate MPK securely and allows multiple light-enclaves isolated within one enclave. Experiments show that LIGHTENCLAVE incurs up to 4% overhead when separating secret SSL keys for server applications and can significantly improve the performance of Graphene-SGX and Occlum by reducing the communication and runtime overhead, respectively.