Security analyses for detecting deserialisation vulnerabilities : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Computer Science at Massey University, Palmerston North, New Zealand

Abstract

An important task in software security is to identify potential vulnerabilities. Attackers exploit security vulnerabilities in systems to obtain confidential information, to breach system integrity, and to make systems unavailable to legitimate users. In recent years, particularly 2012, there has been a rise in reported Java vulnerabilities. One type of vulnerability involves (de)serialisation, a commonly used feature to store objects or data structures to an external format and restore them. In 2015, a deserialisation vulnerability was reported involving Apache Commons Collections, a popular Java library, which affected numerous Java applications. Another major deserialisation-related vulnerability that affected 55\% of Android devices was reported in 2015. Both of these vulnerabilities allowed arbitrary code execution on vulnerable systems by malicious users, a serious risk, and this came as a call for the Java community to issue patches to fix serialisation related vulnerabilities in both the Java Development Kit and libraries.

Despite attention to coding guidelines and defensive strategies, deserialisation remains a risky feature and a potential weakness in object-oriented applications. In fact, deserialisation related vulnerabilities (both denial-of-service and remote code execution) continue to be reported for Java applications. Further, deserialisation is a case of parsing where external data is parsed from their external representation to a program's internal data structures and hence, potentially similar vulnerabilities can be present in parsers for file formats and serialisation languages.

The problem is, given a software package, to detect either injection or denial-of-service vulnerabilities and propose strategies to prevent attacks that exploit them. The research reported in this thesis casts detecting deserialisation related vulnerabilities as a program analysis task. The goal is to automatically discover this class of vulnerabilities using program analysis techniques, and to experimentally evaluate the efficiency and effectiveness of the proposed methods on real-world software. We use multiple techniques to detect reachability to sensitive methods and taint analysis to detect if untrusted user-input can result in security violations.

Challenges in using program analysis for detecting deserialisation vulnerabilities include addressing soundness issues in analysing dynamic features in Java (e.g., native code). Another hurdle is that available techniques mostly target the analysis of applications rather than library code.

In this thesis, we develop techniques to address soundness issues related to analysing Java code that uses serialisation, and we adapt dynamic techniques such as fuzzing to address precision issues in the results of our analysis. We also use the results from our analysis to study libraries in other languages, and check if they are vulnerable to deserialisation-type attacks. We then provide a discussion on mitigation measures for engineers to protect their software against such vulnerabilities.

In our experiments, we show that we can find unreported vulnerabilities in Java code; and how these vulnerabilities are also present in widely-used serialisers for popular languages such as JavaScript, PHP and Rust. In our study, we discovered previously unknown denial-of-service security bugs in applications/libraries that parse external data formats such as YAML, PDF and SVG.