Counteracting Data-Only Malware with Code Pointer Examination

As new code-based defense technologies emerge, attackers move to data-only malware, which is capable of infecting a system without introducing any new code. To manipulate the control flow without code, data-only malware inserts a control data structure into the system, for example in the form of a ROP chain, which enables it to combine existing instructions into a new malicious program. Current systems try to hinder data-only malware by detecting the point in time when the malware starts executing. However, it has been shown that these approaches are not only performance consuming, but can also be subverted. In this work, we introduce a new approach, Code Pointer Examination CPE, which aims to detect data-only malware by identifying and classifying code pointers. Instead of targeting control flow changes, our approach targets the control structure of data-only malware, which mainly consists of pointers to the instruction sequences that the malware reuses. Since the control structure is comparable to the code region of traditional malware, this results in an effective detection approach that is difficult to evade. We implemented a prototype for recent Linux kernels that is capable of identifying and classifying all code pointers within the kernel. As our experiments show, our prototype is able to detect data-only malware in an efficient manner less than 1i¾?% overhead.