Web browsers are increasingly designed to be extensible to keep up with the Web's rapid pace of change. This extensibility is typically implemented using script-based extensions. Script extensions have access to sensitive browser APIs and content from untrusted web pages. Unfortunately, this powerful combination creates the threat of privilege escalation attacks that grant web page scripts the full privileges of script extensions and control over the entire browser process. This thesis describes the pitfalls of script-based extensibility based on our study of the Firefox Web browser, and is the first to offer a classification of script-based privilege escalation vulnerabilities. We propose a taint-based system to track the spread of untrusted data in the browser and to detect the characteristic signatures of privilege escalation attacks. We show that this approach is effective by testing our system against exploits in the Firefox bug database and finding that it detects the vast majority of attacks with no false alarms.

Web browsers are increasingly designed to be extensible to keep up with the Web's rapid pace of change. This extensibility is typically implemented using script-based extensions. Script extensions have access to sensitive browser APIs and content from untrusted web pages. Unfortunately, this powerful combination creates the threat of privilege escalation attacks that grant web page scripts the full privileges of script extensions and control over the entire browser process. This thesis describes the pitfalls of script-based extensibility based on our study of the Firefox Web browser, and is the first to offer a classification of script-based privilege escalation vulnerabilities. We propose a taint-based system to track the spread of untrusted data in the browser and to detect the characteristic signatures of privilege escalation attacks. We show that this approach is effective by testing our system against exploits in the Firefox bug database and finding that it detects the vast majority of attacks with no false alarms.

Internet application servers are currently designed to maximize resource efficiency by servicing many thousands of users that may fall within disparate privilege classes. Pooling users into a shared execution context in this way enables adversaries not only to laterally propagate attacks against other clients, but also to use the application server as a "confused deputy" to gain escalated privileges against sensitive backend data. In this work, we present the Single-use Server (SuS) model, which detects and defeats these attacks by separating users into isolated, containerized application servers with tailored backend permissions. In this model, exploited servers no longer have unfettered access to the backend data or other users. We create a prototype implementation of the SuS model for the WordPress content management system and demonstrate our model's ability to neutralize real-world exploits against vulnerable WordPress versions. We find that the SuS model achieves a high level of security while minimizing the amount of code modification required for porting an application server. In our performance evaluation, we find that the CPU and latency overheads of the SuS model are very low, and memory consumption scales linearly. We generalize the SuS model to be applicable to a wide range of application server and backend resource pairs. With our modularized codebase, we port IMAP, a widely-used mail retrieval protocol, to the SuS model and find that doing so requires minimal effort.

Cette thèse propose une approche originale de test de vulnérabilité Web à partir de modèles etdirigée par des patterns de tests, nommée PMVT. Son objectif est d’améliorer la capacité de détectionde quatre types de vulnérabilité majeurs, Cross-Site Scripting, Injections SQL, Cross-Site RequestForgery, et Privilege Escalation. PMVT repose sur l’utilisation d’un modèle comportemental del’application Web, capturant ses aspects fonctionnels, et sur un ensemble de patterns de test devulnérabilité qui adressent un type de vulnérabilité de manière générique, quelque soit le type del’application Web sous test.Par l’adaptation de technologies MBT existantes, nous avons développé une chaîne outillée complèteautomatisant la détection des quatre types de vulnérabilité. Ce prototype a été exprimenté et évaluésur deux applications réelles, actuellement utiliseés par plusieurs dizaines de milliers d’utilisateurs.Les résultats d’expérimentation démontrent la pertinence et de l’efficience de PMVT, notamment enaméliorant de façon significative la capacité de détection de vulnérabilités vis à vis des scannersautomatiques d’applications Web existants. / This thesis proposes an original approach, dubbed PMVT for Pattern-driven and Model-basedVulnerability Testing, which aims to improve the capability for detecting four high-profile vulnerabilitytypes, Cross-Site Scripting, SQL Injections, CSRF and Privilege Escalations, and reduce falsepositives and false negatives verdicts. PMVT relies on the use of a behavioral model of theapplication, capturing its functional aspects, and a set of vulnerability test patterns that addressvulnerabilities in a generic way. By adapting existing MBT technologies, an integrated toolchain that supports PMVT automates thedetection of the four vulnerability types in Web applications. This prototype has been experimentedand evaluated on two real-life Web applications that are currently used by tens of thousandsusers. Experiments have highlighted the effectiveness and efficiency of PMVT and shown astrong improvement of vulnerability detection capabilities w.r.t. available automated Web applicationscanners for these kind of vulnerabilities.