Application Data Consistency Checking for Anomaly Based Intrusion Detection

Abstract

Host-based intrusion detection systems may be coarsely divided into two categories. Misuse-based intrusion detection systems, which rely on a database of malicious behavior; and anomaly-based intrusion detection systems which rely on the comparison of the observed behavior of the monitored application with a previously built model of its normal behavior called the reference profile. In this last approach, the reference profile is often built on the basis of the sequence of system calls the application emits during its normal executions. Unfortunately, this approach allows attackers to remain undetected by mimicing the attempted behavior of the application. Furthermore, such intrusion detection systems cannot by nature detect anything but violations of the integrity of the control flow of an application. Although, there exist quite critical attacks which do not disturb the control flow of an application and thus remain undetected. We thus propose a different approach relying on the idea that attacks often break simple constraints on the data manipulated by the program. In this perspective, we first propose to define which data are sensitive to intrusions. Then we intend to extract the constraints applying on these data items, afterwards controlling them to detect intrusions. We finally introduce an implementation of such an approach, and some encouraging results.