WebShield: Enabling Various Web Defense Techniques without Client Side Modifications

Today, web attacks are increasing in frequency, severity and sophistication. Existing solutions are either host-based which suffer deployment problems or middlebox approaches that can only accommodate certain security protection mechanisms with limited protection. In this paper, we propose four design principles for general middlebox frameworks of web protection, and apply these principles to design WebShield, which can enable various host-based security mechanisms. In particular, we run all the JavaScript from remote web servers only at shadow browser instances inside the middlebox, and only run our trusted JavaScript rendering agent at client browsers. The trusted rendering agent turns browsers into a thin web terminal by reconstructing the encoded DOM of a webpage. We implement a prototype of WebShield. Evaluation demonstrates that a general JavaScript rendering agent can render webpages precisely and be just slightly slower than direct access. We further demonstrate that our design can work well with interactive web applications such as JavaScript games. WebShield can detect attacks deeply embedded in dynamic HTML pages including the ones in complex Web 2.0 applications, and can also detect both known and unknown vulnerabilities. We further show that WebShield is scalable for deployment.