Squill: Testing DBMS with Correctness Feedback and Accurate Instantiation

Abstract

Database Management Systems (DBMSs) are the core of management information systems. Thus, detecting security bugs or vulnerabilities of DBMSs is an essential task. In recent years, grey-box fuzzing has been adopted to detect DBMS bugs for its high effectiveness. However, the seed scheduling strategy of existing fuzzing techniques does not consider the seeds’ correctness, which is inefficient in finding vulnerabilities in DBMSs. Moreover, current tools cannot correctly generate SQL statements with nested structures, which limits their effectiveness. This paper proposes a fuzzing solution named Squill to address these challenges. First, we propose correctness-guided mutation to utilize the correctness of seeds as feedback to guide fuzzing. Second, Squill embeds semantics-aware instantiation to correctly fill semantics to SQL statements with nested structures by collecting the context information of AST nodes. We implemented Squill based on Squirrel and evaluated it on three popular DBMSs: MySQL, MariaDB, and OceanBase. In our experiment, Squill explored 29% more paths and found 3.4× more bugs than the existing tool. In total, Squill detected 30 bugs in MySQL, 27 in MariaDB, and 6 in OceanBase. Overall, 19 of the bugs are fixed with 9 CVEs assigned. The results show that Squill outperforms the previous fuzzer in terms of both code coverage and bug discovery.

Keywords

• coverage-based grey-box fuzzing
• database testing
• vulnerability