Reverse-engineering and analysis of access control models in web applications.

摘要

Nowadays, Web applications are ubiquitous and deal with increasingly large amounts of confidential data. In order to protect these data from malicious users, security mechanisms must be put in place. Securing software, however, is an extremely difficult task since a single breach is often sufficient to compromise the security of a system. Therefore, it is not surprising that day after day, we hear about cyberattacks and confidential data leaks in the news. To give the reader an idea, various reports suggest that between 85% and 98% of websites contain at least one serious vulnerability.;In this thesis, we focus on one particular aspect of software security that is access control models. Access control models are critical security components that define the actions a user can and cannot do in a system. Year after year, several security organizations report access control flaws among the most prevalent and critical flaws in Web applications. However, contrary to other types of security flaws such as SQL injection (SQLi) and cross-site scripting (XSS), access control flaws comparatively received little attention from the research community. This research work attempts to reverse this trend.;While application security and access control models are the main underlying themes of this thesis, our research work is also strongly anchored in software engineering. You will observe that our work is always based on real-worldWeb applications and that the approaches we developed are always built in a such way as to minimize the amount of work on that is required from developers. In other words, this thesis is about practical software security.;In the context of this thesis, we tackle the highly challenging problem of investigating unspecified and often undocumented access control models in open source Web applications. Indeed, access control flaws occur when some user is able to perform operations he should not be able to do or access data he should be denied access to. In the absence of security specifications, determining who should have the authorization to perform specific operations or access specific data is not straightforward.;In order to overcome this challenge, we first developed a novel approach, called the Security Pattern Traversal (SPT) analysis, to reverse-engineer access control models from the source code of applications in a fast, precise and scalable manner. Results from SPT analysis give a portrait of the access control model as implemented in an application and serve as a baseline for further analyzes.;For example, real-world Web application, often define several hundred privileges that protect hundreds of different functions and modules. As a consequence, access control models, as reverse-engineered by SPT analysis, can be difficult to interpret from a developer point of view, due to their size. In order to provide better support to developers, we explored how Formal Concept Analysis (FCA) could facilitate comprehension by providing visual support as well as automated reasoning about the extracted access control models. Results indeed revealed how FCA could highlight properties about implemented access control models that are buried deep into the source code of applications, that are invisible to administrators and developers, and that can cause misunderstandings and vulnerabilities.;Through investigation and observation of several Web applications, we also identified recurring and cross-application error-prone patterns in access control models. The second half of this thesis presents the approaches we developed to leverage SPT results to automatically capture these patterns that lead to access control flaws such as forced browsing vulnerabilities, semantic errors and security-discordant clone based errors. Each of these approaches interpret SPT analysis results from different angles to identify different kinds of access control flaws in Web applications.;Forced browsing vulnerabilities occur when security-sensitive resources are not protected against direct access to their URL. Using results from SPT, we showed how we can detect such vulnerabilities in a precise and very fast (up to 890 x faster than state of the art) way.;Semantic errors occur when security-sensitive resources are protected by semantically wrong privileges. To give the reader an idea, in the context of a Web application, protecting access to administrative resources with a privilege that is designed to restrict file uploads is an example of semantic error. To our knowledge, we were the first to tackle this problem and to successfully detect semantic errors in access control models. We achieved such results by interpreting results from SPT in the light of a natural language processing technique called Latent Dirichlet Allocation.;Finally, by investigating SPT results in the light of software clones, we were able to detect yet other novel access control flaws. Simply put, we explored the intuition that code clones, that are blocks of code that are syntactically similar, are expected to perform similar operations in a system and, consequently, be protected by similar privileges. By investigating clones that are protected in different ways, called security-discordant clones, we were able to report several novel access control flaws in the investigated systems.