Biblio

The rapid rise of cyber-crime activities and the growing number of devices threatened by them place software security issues in the spotlight. As around 90% of all attacks exploit known types of security issues, finding vulnerable components and applying existing mitigation techniques is a viable practical approach for fighting against cyber-crime. In this paper, we investigate how the state-of-the-art machine learning techniques, including a popular deep learning algorithm, perform in predicting functions with possible security vulnerabilities in JavaScript programs. We applied 8 machine learning algorithms to build prediction models using a new dataset constructed for this research from the vulnerability information in public databases of the Node Security Project and the Snyk platform, and code fixing patches from GitHub. We used static source code metrics as predictors and an extensive grid-search algorithm to find the best performing models. We also examined the effect of various re-sampling strategies to handle the imbalanced nature of the dataset. The best performing algorithm was KNN, which created a model for the prediction of vulnerable functions with an F-measure of 0.76 (0.91 precision and 0.66 recall). Moreover, deep learning, tree and forest based classifiers, and SVM were competitive with F-measures over 0.70. Although the F-measures did not vary significantly with the re-sampling strategies, the distribution of precision and recall did change. No re-sampling seemed to produce models preferring high precision, while re-sampling strategies balanced the IR measures.

Nowadays, most vendors apply the same open source code to their products, which is dangerous. In addition, when manufacturers release patches, they generally hide the exact location of the vulnerabilities. So, identifying vulnerabilities in binaries is crucial. However, just searching source program has a lower identifying accuracy of vulnerability, which requires operators further to differentiate searched results. Under this context, we propose VMPBL to enhance identifying the accuracy of vulnerability with the help of patch files. VMPBL, compared with other proposed schemes, uses patched functions according to its vulnerable functions in patch file to further distinguish results. We establish a prototype of VMPBL, which can effectively identify vulnerable function types and get rid of safe functions from results. Firstly, we get the potential vulnerable-patched functions by binary comparison technique based on K-Trace algorithm. Then we combine the functions with vulnerability and patch knowledge database to classify these function pairs and identify the possible vulnerable functions and the vulnerability types. Finally, we test some programs containing real-world CWE vulnerabilities, and one of the experimental results about CWE415 shows that the results returned from only searching source program are about twice as much as the results from VMPBL. We can see that using VMPBL can significantly reduce the false positive rate of discovering vulnerabilities compared with analyzing source files alone.