LTL Model-Checking for Malware Detection

Nowadays, malware has become a critical security threat. Traditional antiviruses such as signature-based techniques and code emulation become insufficient and easy to get around. Thus, it is important to have efficient and robust malware detectors. In [23,21], CTL model-checking for PushDown Systems (PDSs) was shown to be a robust technique for malware detection. However, the approach of [23,21] lacks precision and runs out of memory in several cases. In this work, we show that several malware specifications could be expressed in a more precise manner using LTL instead of CTL. Moreover, LTL can express malicious behaviors that cannot be expressed in CTL. Thus, since LTL model-checking for PDSs is polynomial in the size of PDSs while CTL model-checking for PDSs is exponential, we propose to use LTL model-checking for PDSs for malware detection. Our approach consists of: (1) Modeling the binary program as a PDS. This allows to track the program’s stack (needed for malware detection). (2) Introducing a new logic (SLTPL) to specify the malicious behaviors. SLTPL is an extension of LTL with variables, quantifiers, and predicates over the stack. (3) Reducing the malware detection problem to SLTPL model-checking for PDSs. We reduce this model checking problem to the emptiness problem of Symbolic Büchi PDSs. We implemented our techniques in a tool, and we applied it to detect several viruses. Our results are encouraging.