CVShield

Abstract

The emerging Connected Vehicle (CV) technology enables vehicles to wirelessly exchange safety and mobility information (e.g., location and speed) with traffic infrastructure and other vehicles. Existing CV applications heavily rely on sensor inputs (e.g., GPS). However, previous work has shown that the attacker can cause severe congestion or increased safety risks by compromising vehicles and broadcasting falsified sensor data. Thus, it is highly desirable to ensure the integrity of sensor data. In this paper, to prevent compromised vehicles from sending falsified sensor data, we propose a system CVShield, which utilizes the recent advances in hardware-assisted security (e.g., ARM TrustZone). CVShield can ensure the integrity of the sensor data from their reading to their transmission at the vehicle side. In general, we relocate all codes that are related to sensor data reading, processing, encapsulation, and transmission from the rich execution environment (REE) into the trusted execution environment (TEE). However, manually extracting code sections is laborious and error-prone. Also, we should minimize the size of the trusted computing base (TCB) in TEE to reduce the attack surface. To achieve these goals, we propose to leverage program slicing to automatically extract code sections and eliminating irrelevant codes in large codebases. Our initial results demonstrate that CVShield can support GPS data reading, and our optimization can eliminate the time overhead introduced by context switches of TrustZone.