Learning Fingerprints for a Database Intrusion Detection System

Abstract

There is a growing security concern on the increasing number of databases that are accessible through the Internet. Such databases may contain sensitive information like credit card numbers and personal medical histories. Many e-service providers are reported to be leaking customers’ information through their websites. The hackers exploited poorly coded programs that interface with backend databases using SQL injection techniques. We developed an architectural framework, DIDAFIT (Detecting Intrusions in DAtabases through FIngerprinting Transactions) [1], that can efficiently detect illegitimate database accesses. The system works by matching SQL statements against a known set of legitimate database transaction fingerprints. In this paper, we explore the various issues that arise in the collation, representation and summarization of this potentially huge set of legitimate transaction fingerprints. We describe an algorithm that summarizes the raw transactional SQL queries into compact regular expressions. This representation can be used to match against incoming database transactions efficiently. A set of heuristics is used during the summarization process to ensure that the level of false negatives remains low. This algorithm also takes into consideration incomplete logs and heuristically identifies “high risk” transactions.

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.