UCLA

Single Sign-on (SSO) protocols play a critical role in operating secure systems, granting access and privileges to authenticated entities. However, logical mistakes in their implementation that violate the standard specification expose them to various security threats that can compromise the confidentiality, integrity, and availability of sensitive user resources. SSO protocols are complex as they involve interactions between multiple parties (e.g., service providers, relying parties, and users) and such interactions result in diverse security requirements for different platforms. Unfortunately, the standard specification is written in hundreds of pages of English documents and the security properties are often not well-defined, leading the protocol developers to make severe mistakes. These mistakes can completely compromise the security of the protocol and they have been exploited many times in recent years to conduct severe attacks, including identity theft and complete account or system takeover. In this dissertation, we aim to identify, analyze, and mitigate security threats and vulnerabilities that impact the SSO-based authorization and authentication protocols. To achieve this goal, we incorporate static analysis, formal verification, and program synthesis techniques to design and implement a toolchain to 1) identify logical security errors in the client and server-side implementations of the SSO protocols, 2) provide safety guarantee with respect to the standard specification, and 3) automatically generate patches for the detected buggy implementations. We further propose specification languages to allow developers to formally express the SSO security requirements that can be checked across large-scale SSO implementations on diverse platforms. We evaluate our proposed approaches on 600 widely used relying party applications and the 25 most popular SSO service providers. Our experiments uncover 40 previously unknown vulnerabilities (i.e., zero-days), which have been acknowledged and fixed by the developers after our responsive disclosure. Furthermore, eight classes of these vulnerabilities lead to new entries in the Common Vulnerabilities and Exposures (CVE) database.