TWC: Small: Enhancing the Security and Performance of GPU Access in Web Browsers

现代个人计算机已经接受了越来越强大的图形处理单元（GPU），即实现高性能图形的硬件组件。控制当今计算机中这些GPU编程的软件（即，图形栈）被设计成由从可信赖的开发者处获得并由用户直接安装的应用程序使用。然而，Web应用程序（即，在web浏览器内运行的应用程序）越来越流行，并且WebGL是为web应用程序提供基于GPU的图形的最近的工业努力。然而，WebGL已经引起了严重的安全问题，因为恶意Web应用程序已经使用WebGL来完成对用户系统的攻击。不幸的是，基于Web浏览器的保护技术并没有解决安全问题，并显着恶化了WebGL的性能。本项目旨在开发在不降低性能的前提下，提高网络应用程序中基于GPU的图形的安全性的技术。在第一个推力中，研究人员正在使用GPU虚拟化技术来沙箱化Web应用程序对GPU的访问，为共享GPU提供硬件级隔离。在这种设计中，每个Web应用程序都使用专用的虚拟GPU（或vGPU），使Web应用程序能够实现高图形性能，同时将其与系统的其他部分隔离开来。该项目正在开发一个新的操作系统合成器，以控制和调度vGPU之间的共享资源。不幸的是，基于vGPU的架构无法阻止所有可能的基于GPU的安全攻击，包括某些形式的信息泄漏、拒绝服务和内存完整性攻击。在第二个方面，研究人员正在研究这些剩余的攻击媒介，并开发补充解决方案。

Modern personal computers have embraced increasingly powerful Graphics Processing Units (GPUs), hardware components that enable high performance graphics. The software that controls the programming of these GPUs in today's computers (i.e., the graphics stack) was designed to be used by applications acquired from trustworthy developers and installed directly by the user. However, web applications (i.e., applications running inside a web browser) are gaining in popularity and WebGL is a recent industry effort to provide GPU-based graphics for web applications. However, WebGL has posed serious security concerns as malicious web applications have used WebGL to accomplish attacks on user's systems. Unfortunately, web browser-based protection techniques have not solved the security concerns and have significantly worsened WebGL performance. This project is developing techniques to improve the security of GPU-based graphics for web applications, without worsening performance.The project is proceeding in two research thrusts. In the first thrust, the researchers are using GPU virtualization technology to sandbox a web application's access to the GPU, providing hardware-level isolation for sharing a GPU. In this design, each web application uses a dedicated virtual GPU (or vGPU) allowing the web application to achieve high graphics performance, while isolating it from the rest of the system. The project is developing a new operating system compositor to control and schedule the shared resources between the vGPUs. Unfortunately, the vGPU-based architecture cannot prevent all possible GPU-based security attacks, including some forms of information leakage, denials-of-service, and memory integrity attacks. In the second thrust, the researchers are studying these remaining attack vectors and developing complementary solutions.