Hardware-assisted Security

点击翻译按钮获取中文摘要

Exploiting memory errors is a popular tactic used by remote adversaries attempting to break into computer systems. Many recent high-profile attacks for ransomware (eg, WannaCry) and disruption of daily life (eg, NotPetya) exploited memory errors and caused billions of dollars in damages. Memory errors are also routinely exploited in targeted attacks against individuals, as in the recent WhatsApp hack, which exploited a common type of memory error (buffer overflows) altering the normal program flow in order to let the attacker control the phone just by making a WhatsApp call to the phone's owner. Software-only solutions against such "run-time attacks" are easier to deploy but incur high costs if they are to ensure sufficient security. Hardware-based solutions are more difficult to deploy. Recently, major processor vendors have started to roll out hardware-based defenses (e.g., ARM Pointer Authentication (PA)), against specific classes of run-time attacks. Leveraging such defenses to design broadly applicable solutions for run-time protection is attractive because these solutions can provide effective protections without incurring high deployment costs. However, doing so involves two major challenges: (1) understanding the limitations inherent in the current designs of these defenses (e.g., ARM PA is susceptible to "reuse attacks" where an authenticated pointer taken from one location in the code is reused in another), and (2) building on these insights to design effective run-time protection mechanisms that are simultaneously more efficient and more secure than the state-of-the art. Approach: My long term objective is to design effective hardware-assisted approaches to protect software. To this end, I seek to understand how to use emerging commercial off-the-shelf (COTS) hardware-security defenses to achieve effective run-time protection with a low-barrier for real-world deployment. To address the challenges above, I aim for three short- to medium-term objectives O1-O3: (O1) to systematically analyze these COTS defenses to understand their limitations and to develop effective techniques for run-time protection by using these defenses in new ways (individually or in conjunction), (O2) develop a rigorous framework to compare techniques, and (O3) use the experience gained to identify and realize a set of optimal hardware-security building blocks ("primitives") on RISC-V, an open-source hardware platform. Anticipated outcomes and benefits: The program will result in: software artifacts that allow developers to easily afford run-time protection for their software on commodity hardware, an open-source hardware implementation of selected primitives that can be used for technology transfer or further research, and  train highly-qualified personnel with deep understanding of the intricacies of run-time attacks and equipped to develop effective defenses. If successful, this program will significantly strengthen security for software in the real world.