Browser attack: Using a website with malicious JavaScript, researchers from the Institute for Applied Information Processing and Communication Technology at Graz University of Technology were able to spy on information about data, keyboard entries and encryption keys on third-party computers in three different attacks via WebGPU.
Modern Websites place ever higher demands on the computing power of computers. Therefore received Web browser In addition to the CPU of a computer, for several years now there has also been access to the computing capacity of the graphics card (Graphics Processing Unit or GPU). Via programming interfaces such as WebGL and the new standard WebGPU can use the scripting language JavaScript the resources of the GPU to use. However, this carries risks.
Lead researchers Browser attack on Crome, Edge & Co
Although WebGPU is currently still under active development, browsers such as Chrome, Chromium, Microsoft Edge and the nightly versions of Firefox already support it. Due to the greater flexibility and modernized design compared to WebGL, the interface will be widely used in the coming years. “Our attacks do not require user interaction with a website and they run in a time frame that allows them to be carried out during normal Internet surfing.
The research team carried out its attacks on several systems in which different graphics cards from the manufacturers NVIDIA and AMD were installed - cards from the GTX 1000 series and the RTX series 2000, 3000 and 4000 from NVIDIA, and cards from the RX 6000 series from AMD for use. For all three types of attacks, the researchers used the access to the computer's cache memory available via WebGPU, which is intended for particularly fast and short-term data access from the CPU and GPU. Through this side channel they accessed meta information that allows conclusions to be drawn about security-relevant information.
Changes in the cache as a whistleblower
The team was able to track changes in the cache by filling it themselves using code in JavaScript via WebGPU and paying attention to when their own data was removed from the cache through input. This allowed for a fairly precise analysis of the keyboard input relatively quickly. Through a finer segmentation of the cache, the researchers were also able to use a second attack to set up their own, secret communication channel, in which filled and unfilled cache segments served as zeros and ones and thus as the basis for binary code.
They used 1024 of these cache segments, achieving transfer speeds of up to 10,9 kilobytes per second, which was fast enough to transfer simple information. Attackers can use this channel to extract data that they could read using other attacks in areas of the computer that are separated from the Internet.
Three successful attack variants
The third browser attack targeted AES encryption, which is used to encrypt documents, connections and servers. Here too, they filled up the cache again, but with their own AES encryption. Through the cache's reaction to this, they were able to identify the places in the system that are responsible for encryption and to access the keys of the attacked system from there. “Our AES attack would probably be a little more complicated under real-time conditions because a lot of encryption runs in parallel on a GPU,” says Roland Czerny from the Institute for Applied Information Processing and Communication Technology at Graz University of Technology. “Nevertheless, we were able to demonstrate that we can also attack algorithms very precisely. We have of course communicated the findings of our work to the browser manufacturers in advance and we hope that they will take this topic into account in the further development of WebGPU.”
At the end of last year, other researchers had already succeeded in a similar attack on GPU acceleration: In a research paper, researchers from the universities of Austin Texas, Carnegie Mellon, Washington and Illinois Urbana-Champaign present a method for carrying out a side channel attack on graphics card GPUs , which can be used to retrieve sensitive information from running programs. This “pixel theft” and their values – artifacts – takes place, for example, via a user’s browser.
More at TUgraz.at