P-DATA: A Task-Level Parallel Framework for Dependency-Aware Value Flow Taint Analysis

Rulin Xu,Xiaoguang Mao,Luohui Chen

DOI: https://doi.org/10.1109/apsec60848.2023.00035

2023-01-01

Abstract:Taint analysis of value flows, as a static analysis technique, has gained widespread application in the fields of software security and vulnerability mining. However, when dealing with complex programs, it still faces challenges in terms of precision and performance. This research proposes P-DATA, a parallel framework implementing dependency-aware taint analysis. P-DATA employs modeling to capture data and control dependencies, reducing false positives over tools like Clang Static Analyzer and SVF. To accelerate the analysis, P-DATA leverages a task-level parallel framework introducing Preemption of Computational Resources (PCR) and Asynchronous Taint Source Registration, lead to impressive scalability and efficiency. Evaluations demonstrate P-DATA's ability to significantly expedite taint analysis for large programs using multi-core resources, achieving over 25X speedup on 32 cores. P-DATA makes notable contributions by boosting precision, efficiency and scalability of security-critical program analysis through advanced dependency modeling and paral-lelization techniques. It provides an extensible high-performance framework benefiting static analysis advancement.