Nous nous intéressons à la mise en œuvre des politiques de confidentialité et d'intégrité des données dans le contexte des applications Web côté client. Étant donné que la plupart des applications Web est développée en JavaScript, on propose des mécanismes statiques, dynamiques et hybrides pour sécuriser le flux d'information en Core JavaScript - un fragment de JavaScript qui retient ses caractéristiques fondamentales. Nous étudions en particulier: une sémantique à dispositif de contrôle afin de garantir dynamiquement le respect des politiques de sécurité en Core JavaScript aussi bien qu'un compilateur qui instrumente un programme avec le dispositif de contrôle proposé, un système de types qui vérifie statiquement si un programme respecte une politique de sécurité donnée, un système de types hybride qui combine des techniques d'analyse statique à des techniques d'analyse dynamique afin d'accepter des programmes surs que sa version purement statique est obligée de rejeter. La plupart des programmes JavaScript s'exécute dans un navigateur Web dans le contexte d'une page Web. Ces programmes interagissent avec la page dans laquelle ils sont inclus parmi des APIs externes fournies par le navigateur. Souvent, l'exécution d'une API externe dépasse le périmètre de l'interprète du langage. Ainsi, une analyse réaliste des programmes JavaScript côté client doit considérer l'invocation potentielle des APIs externes. Pour cela, on présente une méthodologie générale qui permet d'étendre des dispositifs de contrôle de sécurité afin qu'ils prennent en compte l'invocation potentielle des APIs externes et on applique cette méthodologie à un fragment important de l'API DOM Core Level 1. / In this thesis, we address the issue of enforcing confidentiality and integrity policies in the context of client-side Web applications. Since most Web applications are developed in the JavaScript programming language, we study static, dynamic, and hybrid enforcement mechanisms for securing information flow in Core JavaScript --- a fragment of JavaScript that retains its defining features. Specifically, we propose: a monitored semantics for dynamically enforcing secure information flow in Core JavaScript as well as a source-to-source transformation that inlines the proposed monitor, a type system that statically checks whether or not a program abides by a given information flow policy, and a hybrid type system that combines static and dynamic analyses in order to accept more secure programs than its fully static counterpart. Most JavaScript programs are designed to be executed in a browser in the context of a Web page. These programs often interact with the Web page in which they are included via a large number of external APIs provided by the browser. The execution of these APIs usually takes place outside the perimeter of the language. Hence, any realistic analysis of client-side JavaScript must take into account possible interactions with external APIs. To this end, we present a general methodology for extending security monitors to take into account the possible invocation of arbitrary APIs and we apply this methodology to a representative fragment of the DOM Core Level 1 API that captures DOM-specific information flows.

Data-flow (DF) testing was introduced more than thirty years ago aiming at verifying a program by extensively exploring its structure. It requires tests that traverse paths in which the assignment of a value to a variable (a definition) and its subsequent reference (a use) is verified. This relationship is called definition-use association (dua). While control-flow (CF) testing tools have being able to tackle systems composed of large and long running programs, DF testing tools have failed to do so. This situation is in part due to the costs associated with tracking duas at run-time. Recently, an algorithm, called Bitwise Algorithm (BA), which uses bit vectors and bitwise operations for tracking intra-procedural duas at run-time, was proposed. This research presents the implementation of BA for programs compiled into Java bytecodes. Previous DF approaches were able to deal with small to medium size programs with high penalties in terms of execution and memory. Our experimental results show that by using BA we are able to tackle large systems with more than 250 KLOCs and 300K required duas. Furthermore, for several programs the execution penalty was comparable with that imposed by a popular CF testing tool. / Teste de fluxo de dados (TFD) foi introduzido há mais de trinta anos com o objetivo de criar uma avaliação mais abrangente da estrutura dos programas. TFD exige testes que percorrem caminhos nos quais a atribuição de valor a uma variável (definição) e a subsequente referência a esse valor (uso) são verificados. Essa relação é denominada associação definição-uso. Enquanto as ferramentas de teste de fluxo de controle são capazes de lidar com sistemas compostos de programas grandes e que executam durante bastante tempo, as ferramentas de TFD não têm obtido o mesmo sucesso. Esta situação é, em parte, devida aos custos associados ao rastreamento de associações definição-uso em tempo de execução. Recentemente, foi proposto um algoritmo --- chamado \\textit (BA) --- que usa vetores de bits e operações bit a bit para monitorar associações definição-uso em tempo de execução. Esta pesquisa apresenta a implementação de BA para programas compilados em Java. Abordagens anteriores são capazes de lidar com programas pequenos e de médio porte com altas penalidades em termos de execução e memória. Os resultados experimentais mostram que, usando BA, é possível utilizar TFD para verificar sistemas com mais de 250 mil linhas de código e 300 mil associações definição-uso. Além disso, para vários programas, a penalidade de execução imposta por BA é comparável àquela imposta por uma popular ferramenta de teste de fluxo de controle.

Data-flow (DF) testing was introduced more than thirty years ago aiming at verifying a program by extensively exploring its structure. It requires tests that traverse paths in which the assignment of a value to a variable (a definition) and its subsequent reference (a use) is verified. This relationship is called definition-use association (dua). While control-flow (CF) testing tools have being able to tackle systems composed of large and long running programs, DF testing tools have failed to do so. This situation is in part due to the costs associated with tracking duas at run-time. Recently, an algorithm, called Bitwise Algorithm (BA), which uses bit vectors and bitwise operations for tracking intra-procedural duas at run-time, was proposed. This research presents the implementation of BA for programs compiled into Java bytecodes. Previous DF approaches were able to deal with small to medium size programs with high penalties in terms of execution and memory. Our experimental results show that by using BA we are able to tackle large systems with more than 250 KLOCs and 300K required duas. Furthermore, for several programs the execution penalty was comparable with that imposed by a popular CF testing tool. / Teste de fluxo de dados (TFD) foi introduzido há mais de trinta anos com o objetivo de criar uma avaliação mais abrangente da estrutura dos programas. TFD exige testes que percorrem caminhos nos quais a atribuição de valor a uma variável (definição) e a subsequente referência a esse valor (uso) são verificados. Essa relação é denominada associação definição-uso. Enquanto as ferramentas de teste de fluxo de controle são capazes de lidar com sistemas compostos de programas grandes e que executam durante bastante tempo, as ferramentas de TFD não têm obtido o mesmo sucesso. Esta situação é, em parte, devida aos custos associados ao rastreamento de associações definição-uso em tempo de execução. Recentemente, foi proposto um algoritmo --- chamado \\textit (BA) --- que usa vetores de bits e operações bit a bit para monitorar associações definição-uso em tempo de execução. Esta pesquisa apresenta a implementação de BA para programas compilados em Java. Abordagens anteriores são capazes de lidar com programas pequenos e de médio porte com altas penalidades em termos de execução e memória. Os resultados experimentais mostram que, usando BA, é possível utilizar TFD para verificar sistemas com mais de 250 mil linhas de código e 300 mil associações definição-uso. Além disso, para vários programas, a penalidade de execução imposta por BA é comparável àquela imposta por uma popular ferramenta de teste de fluxo de controle.

The thesis deals with a design and implementation of the dynamic analyser of parallel applications on the .NET Framework platform. The problematic of synchronization in parallel applications, the instrumentation of such an applications, testing of parallel applications and a specifics of these problems for C\# language and for the platform .NET Framework are discussed in the theoretical part. Selected algorithms for detection of deadlocks (the algorithm of Goodlock) and data-race errors (the algorithm of FastTrack and AtomRace) are described in detail in this part as well. Requirements for the dynamic analyser and the system design is made in the following part of this thesis. The thesis also contains a description of the implementation of the proposed solution, a description of the entire testing of the implemented tool. Last but not least, the thesis describes the sample of using dynamic analysers in a particular application environment.