SaTC: CORE: Medium: Provably Secure, Usable, and Performant Enclaves in Multicore Processors

在云计算和当今其他流行的计算机使用模式中，一台物理计算机由互不信任的用户贡献的不同应用程序共享。共享计算机的资源必须在用户之间安全地划分；错误允许一个用户了解另一个用户的秘密或影响另一个应用程序的执行。解决应用程序之间相互攻击的挑战的一个很有前途的方法是enclaves，最广为人知的是英特尔自2015年以来推出的SGX架构，但SGX允许在应用程序执行时泄露机密。该项目开发具有强大保护的飞地，重点关注可用性、性能和保证。首先，enclave在生产中仍然很少使用，因此该项目研究简化硬件资源共享的新接口。其次，为了激励社区采用，开发的enclave包含硬件优化，以免影响应用程序的运行时性能。第三，该项目开发了硬件安全性的数学证明技术。针对enclave的软件API设计中的计划方法包括有限的I/O通道接口（例如，强制执行固定的或相对可预测的调度）和跨enclave分配资源（如缓存线）的新的灵活方法。新的enclave-to-enclave通信原语是通过修改内存管理硬件实现的。通过Coq定理证明者对证明的机器检查，这些机制被证明是安全的，符合数学严格的最高标准。该技术是为了证明特定的enclave系统循环-精确地模拟由基本网络连接的独立计算机组（每个enclave一个），并通过将证明义务不明显地分解为不同组件（例如，处理器与内存系统）的局部安全条件来模块化地证明此类结果。这项研究产生了可用的飞地系统，调查人员将其作为亚马逊F1云fpga系统的图像提供给公众，从而便于进行安全评估或扩展工作。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

In cloud computing and other popular modes of computer usage today, one physical computer is shared by different applications contributed by users who do not trust each other. The resources of the shared computer must be divided securely across the users; mistakes allow one user to learn another's secrets or influence another applications’ execution. A promising approach to address the challenge of protecting applications from one another is enclaves, best known through Intel's SGX architecture available since 2015, but SGX allows secrets to be leaked while applications execute. This project develops enclaves with strong protections, focusing on usability, performance, and assurance. First, enclaves have still been used relatively rarely in production, and the project therefore studies new interfaces that streamline the sharing of hardware resources. Second, to motivate adoption by the community, the developed enclaves encompass hardware optimizations so as not to affect runtime performance of applications. Third, the project develops techniques for mathematical proofs of hardware security. Planned approaches in software API design for enclaves include limited I/O channel interfaces (e.g., enforcing fixed or otherwise relatively predictable schedules) and new flexible means of allocating resources like cache lines across enclaves. New enclave-to-enclave communication primitives are realized through modifications to memory-management hardware. These mechanisms are being proved secure to the highest standards of mathematical rigor, through machine-checking of proofs with the Coq theorem prover. The technique is to prove that specific enclave systems cycle-accurately simulate groups of separate computers (one per enclave) connected by a basic network, and to prove such results modularly, via nonobvious decomposition of proof obligations into localized security conditions for different components (e.g., processor vs. memory system). This research produces usable enclave systems that the investigators make available to the public as images for Amazon's F1 cloud-FPGA system, facilitating easy bring-up for security evaluation or extending the work.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.