Robust and Efficient Covert Channel Communications in Operating Systems: Design, Implementation and Evaluation

Abstract

Covert channel has been studied for years due to its ability to divulge sensitive information in computer systems. Constructing covert communication scenarios is the first step to learn the threat of a channel. There are several challenges in the existing design of covert channel communications: lacking general communicating model description, low transmission accuracy and weak anti-interference ability. In this paper, we explore how to construct robust and efficient covert channel communications in operating systems. Firstly, we design three general covert communicating protocol models: the Basic Protocol (BP), the Two-Channel Transmission Protocol (TCTP) and the Self-Adaptive Protocol (SAP). Then we implement them in Linux operating systems. To simulate real attack scenarios, a toy Trojan program extracting passwords to cooperate with the covert protocols is presented. To identify potential covert channels in Linux kernel, we use Directed Information Flow Graph (DIFG) to analyze the source code and choose last_pid and temporary files channels in our implementation. Finally we evaluate the transmitting rate and accuracy of the three protocols. The results demonstrate that without special protective measures, the TCTP can achieve rather high accuracy and rate (100% and 31bps in our lab). When equipped with some restricting or interfering mechanisms, the SAP can achieve 97% accuracy and 18bps rate. This result reveals that attackers can bypass countermeasures to steal sensitive data from victims by well-designed covert protocols.