Stealing Webpages Rendered on Your Browser by Exploiting GPU Vulnerabilities

Abstract

Abstract-Graphics processing units (GPUs) are important components of modern computing devices for not only graphics rendering, but also efficient parallel computations. However, their security problems are ignored despite their importance and popularity. In this paper, we first perform an in-depth security analysis on GPUs to detect security vulnerabilities. We observe that contemporary, widely-used GPUs, both NVIDIA's and AMD's, do not initialize newly allocated GPU memory pages which may contain sensitive user data. By exploiting such vulnerabilities, we propose attack methods for revealing a victim program's data kept in GPU memory both during its execution and right after its termination. We further show the high applicability of the proposed attacks by applying them to the Chromium and Firefox web browsers which use GPUs for accelerating webpage rendering. We detect that both browsers leave rendered webpage textures in GPU memory, so that we can infer which webpages a victim user has visited by analyzing the remaining textures. The accuracy of our advanced inference attack that uses both pixel sequence matching and RGB histogram matching is up to 95.4%.

Veil: Private Browsing Semantics Without Browser-side Assistance

“…Veil does not try to stop information leaks from GPU RAM [31], but GPU RAM is never swapped to persistent storage. Poorly-written or malicious browser extensions that leak sensitive page data [32] are also outside the scope of this paper.…”

Vulnerable GPU Memory Management: Towards Recovering Raw Data from GPU

“…For example, Lee et al [20] found that: although they could not decode or recover original images from GPU memories, they were able to infer the sites visited by victims, by linking a targeted memory trunk with the most similar one in a labeled training set. Nevertheless, this attack is limited by some restrictions.…”

“…Second GPU memories also hold considerable amounts of nonimage data, because modern GPUs are frequently used for nongraphical computations such as encryption and matrix calculation, which makes the problem even more complex and difficult. To address these challenges, we leverage several distinctive features of image data and have improved the prime-probe method proposed in the previous study [20]. More details are elaborated below.…”

“…First, currently, adversaries are able to recover raw images from GPU memory residues and then extract much more sensitive information directly from such raw data (e.g., the email contents displayed in a browser that requested a GPU to render web pages). By contrast, previously, such adversaries can only perform side-channel attacks over nondecoded data and infer less critical information indirectly [20]. Second, adversaries can now attack a large group of vulnerable applications with the GPUacceleration feature, because they are able to recover raw images rendered by GPUs (because of acceleration), whereas previously, their attacks could only infer insensitive information indirectly through statistical analysis of memory residues.…”

“…Some recent studies have revealed several concrete attacks against GPU memory [14,18,20,21,24]. In particular, Lee et al demonstrated that adversaries could infer the websites visited by victims on a multiuser system by using statistical analysis of pixel colors [20].…”