Abstract: Vulnerability root cause localization is an important stage of software debugging,and spectrum-based fault root cause localization method is a hot issue in software automation debugging research,but the effectiveness of the positioning depends to a large extent on the quality of the test cases.Test inputs of different types of software are poorly generalized,and randomly gene-rated test inputs lead to large errors in analysis results due to overfitting or too many confounding items,resulting in limited application scenarios of such techniques at present.In this paper,we propose a phased exploration method Dgenerate based on crash paths to address the test case generation problem and implement the prototype tool Dloc.First,we use binary staking to insert staking path information in the basic block during the input stage of program execution,and then classify the original test inputs into common and guided types based on this information.Then,we use the dynamic energy scheduling algorithm to explore crash-related paths to generate high-quality test cases.Finally,the test cases are executed in the original program and execution information is traced to effectively locate the root cause of program fault through statistical analysis.In this paper,15 real CVE vulnerabilities in six different types of software are selected for experiments,and the results show that test cases generated by Dloc can improve location efficiency by 75% on average compared to previous techniques,and Dloc can output the code fragments related to the root causes of defect in the top five positions with an accuracy of 87%,which verifies the feasibility and practicality of the method system in this paper.

Key words: Root cause location, Test case, Program spectrum, Statistical analysis, Directed greybox fuzzing