Research On Code Execution Vulnerability Detection Technology For Binary Programs

In today’s era,the network security situation has become more severe,so how to maintain network security has become an urgent problem in the security field.As a core component of many network applications and services,binary programs are responsible for various critical tasks such as processing and storing sensitive data,managing user accounts and permissions,and ensuring data security and privacy.If there are vulnerabilities in binary programs,attackers can exploit these vulnerabilities to attack systems and networks,resulting in serious consequences such as data leakage,system crashes,and network interruptions.Therefore,ensuring the security of binary programs is critical to network security.Vulnerability detection is one of the important means to maintain network security.It can find and fix vulnerabilities in binary programs in time to improve network security and stability.However,the current vulnerability detection based on deep learning is mainly at the granularity of files and functions,and rarely considers the granularity of execution paths;the black-box nature of deep learning technology makes it more difficult to explain the cause of vulnerabilities.Vulnerability detection based on symbolic execution technology is usually assisted by fuzzing,but path explosion and complex solution problems still exist.To solve these problems,this paper uses static taint analysis technology combined with deep learning technology and symbolic execution technology to detect binary program vulnerabilities.The main research results are as follows:1.Aiming at the problem of numerous internal execution paths and complex structures of binary programs,we propose a binary program code execution vulnerability detection method based on deep learning technology,which detects vulnerabilities at the granularity of binary program execution paths.We designed and implemented the BiLSTM+Attention network model and the BERT+RNN network model for detecting vulnerabilities in program execution paths in binary programs.Through experimental verification,the accuracy of our network model has been significantly improved compared with other methods,among which the BERT+RNN network model has the highest accuracy,while the BiLSTM+Attention network model has both high accuracy and low loss rate.2.In binary program analysis,static taint analysis can efficiently analyze programs,but the accuracy rate is low;symbolic execution has high reliability,but has inherent defects such as path explosion and complex solutions.To solve these problems,this paper combines the two to perform efficient and accurate binary program vulnerability detection.To improve efficiency,the static taint analysis method is used to analyze the execution path and find the path that may have loopholes.To improve accuracy,these paths are tested using symbolic execution exploration to confirm whether they are vulnerable paths.