A dataset generator for next generation system call host intrusion detection systems

• Main Authors: Pendleton, Marcus, Shouhuai Xu
• Format: Conference Proceeding
• Language: English
• Subjects: Complexity theory; Computer security; Context; Dataset; Generators; Hidden Markov models; Instruction sets; Intrusion detection; Linux; Operating Systems; Security; System Calls
• ISSN: 2155-7586
• DOI: 10.1109/MILCOM.2017.8170835

Over the years, system calls (syscalls) have become an increasingly popular data source for host intrusion detection systems (HIDS). This is partly due to their strong security semantic implications. As syscalls conform to a program's control-flow graph, a deviation in a syscall sequence may imply a deviation in a program's control-flow graph. This is useful for detecting the control-flow hijacking class of attacks. Additionally, malware must utilize syscalls in order to provide any utility to the attacker, with the exception of some denial-of-service attacks. Because all syscalls are observable from the kernel, this makes evasion difficult for attackers under syscall HIDS. Given their suitability for HIDS, many approaches based on syscalls have been proposed. However, the syscall datasets available are not always the most suitable for these and emerging techniques in analytics, as they may need additional structural or contextual information about syscalls in their decision engine. Furthermore, this flatness of previous datasets often pigeonholes solutions into those which are limited by that data view. It is also burdensome on the researcher to generate his own custom dataset. In this work, we propose an extensible syscall dataset generator which includes structural and limited contextual information regarding syscalls, yet allows for researchers to easily add their own features to more quickly develop and evaluate their systems. Our dataset generator can aid researchers in widening the solution space for syscall HIDS.