LogicMEM: Automatic Profile Generation for Binary-Only Memory Forensics via Logic Inference

Zhenxiao Qi (UC Riverside), Yu Qu (UC Riverside), Heng Yin (UC Riverside)

Memory forensic tools rely on the knowledge of kernel symbols and kernel object layouts to retrieve digital evidence and artifacts from memory dumps. This knowledge is called profile. Existing solutions for profile generation are either inconvenient or inaccurate. In this paper, we propose a logic inference approach to automatically generating a profile directly from a memory dump. It leverages the invariants existing in kernel data structures across all kernel versions and configurations to precisely locate forensics-required fields in kernel objects. We have implemented a prototype named LOGICMEM and evaluated it on memory dumps collected from mainstream Linux distributions, customized Linux kernels with random configurations, and operating systems designed for Android smartphones and embedded devices. The evaluation results show that the proposed logic inference approach is well-suited for locating forensics-required fields and achieves 100% precision and recall for mainstream Linux distributions and 100% precision and 95% recall for customized kernels with random configurations. Moreover, we show that false negatives can be eliminated with improved logic rules. We also demonstrate that LOGICMEM can generate profiles when it is otherwise difficult (if not impossible) for existing approaches, and support memory forensics tasks such as rootkit detection.