Portal de Periódicos da CAPES

Mutation Methods for Structured Input to Enhance Path Coverage of Fuzzers

2024; Springer Science+Business Media; Linguagem: Inglês

10.1007/978-981-99-8024-6_20

1611-3349

Yonggon Park, Youngjoo Ko, Jong Kim,

VLSI and Analog Circuit Testing

Existing mutation methods used in coverage-based grey-box fuzzing (CGF), such as those employed by AFL and AFL++, can lead to biased testing for structured inputs. While fuzzing, certain input sections of structured input may receive fewer mutations, resulting in less testing of the code that handles those sections, which leads to lower path coverage in those code parts. In this paper, we propose two mutation methods for the structured input to address the unbalanced problem and improve path coverage. The first method, Uniform Mutation, involves conducting additional mutations in input sections that trigger less testing, thereby achieving a more balanced path coverage across the target program. However, this method requires prior knowledge of the input format, which reduces its usability when the format of the target program changes. To overcome the limitation, we propose the second method, Format-agnostic Mutation, which automatically partitions the input into sections based on coverage feedback. This method redistributes the number of mutations and resizes the sections to improve path coverage without knowing the input format. We evaluate the effectiveness of these methods using two real-world programs (Xpdf and libxml2) and compare them with AFL. The experimental results demonstrate that Uniform and Format-agnostic mutations (weight and resizing) outperform AFL regarding path coverage exploration.