Detecting Vulnerabilities in Encrypted Software Code while Ensuring Code Privacy

TL;DR

This paper introduces CoCoA, a novel method for detecting software vulnerabilities on encrypted code, preserving privacy while maintaining analysis accuracy with modest performance overhead.

Contribution

It proposes a new approach combining static analysis and searchable encryption to enable confidential code vulnerability detection, defining a new research field -- Confidential Code Analysis.

Findings

CoCoA achieves similar precision to standard static analysis tools.

The approach incurs an average performance overhead of 42.7%.

Experimental evaluation on PHP applications demonstrates effectiveness.

Abstract

Software vulnerabilities continue to be the main cause of occurrence for cyber attacks. In an attempt to reduce them and improve software quality, software code analysis has emerged as a service offered by companies specialising in software testing. However, this service requires software companies to provide access to their software's code, which raises concerns about code privacy and intellectual property theft. This paper presents a novel approach to Software Quality and Privacy, in which testing companies can perform code analysis tasks on encrypted software code provided by software companies while code privacy is preserved. The approach combines Static Code Analysis and Searchable Symmetric Encryption in order to process the source code and build an encrypted inverted index that represents its data and control flows. The index is then used to discover vulnerabilities by carrying out static analysis tasks in a confidential way. With this approach, this paper also defines a new research field -- Confidential Code Analysis --, from which other types of code analysis tasks and approaches can be derived. We implemented the approach in a new tool called CoCoA and evaluated it experimentally with synthetic and real PHP web applications. The results show that the tool has similar precision as standard (non-confidential) static analysis tools and a modest average performance overhead of 42.7%.