NNTailor: A Neural Network-Driven Fuzzer for DataBase Management Systems

DataBase Management Systems (DBMS) are essential software for efficient data storage, management, and analysis, playing a crucial role in modern data-intensive applications. Vulnerabilities in DBMS can significantly threaten data security and application functionality, impacting millions of software systems. While fuzz testing(fuzzing) is a prevalent technique for uncovering DBMS vulnerabilities, there is limited research on employing Neural Network Language Models (NNLMs) for this purpose. This paper introduces NNTailor, a novel approach using NNLMs based on AST code fragments for DBMS fuzzing. A key advantage of this method is its effectiveness in black-box testing scenarios. We evaluated NNTailor on SQLite and PostgreSQL, demonstrating its capability to generate effective SQL test cases.