TruZ-View

Abstract

When OS and hypervisor are compromised, mobile devices currently provide a hardware protected mode called Trusted Execution Environment (TEE) to guarantee the confidentiality and integrity of the User Interface (UI). The present TEE UI solutions adopt a self-contained design model, which provides a fully functional UI stack in the TEE, but they fail to manage one critical design principle of TEE: a small Trusted Computing Base (TCB), which should be more easily verified in comparison to a rich OS. The TCB size of the self-contained model is large as a result of the size of an individual UI stack. To reduce the TCB size of the TEE UI solution, we proposed a novel TEE UI design model called delegation model. To be specific, our design reuses the majority of the rich OS UI stack. Unlike the existing UI solutions protecting 3-dimensional UI processing in the TEE, our design protects the UI solely as a 2-dimensional surface and thus reduces the TCB size. Our system, called TruZ-View, allows application developers to use the rich OS UI development environment to develop TEE UI with consistent UI looks across the TEE and the rich OS. We successfully implemented our design on HiKey board. Moreover, we developed several TEE UI use cases to protect the confidentiality and integrity of UI. We performed a thorough security analysis to prove the security of the delegation UI model. Our real-world application evaluation shows that developers can leverage our TEE UI with few changes to the existing app's UI logic.

Rushmore

“…For denial of service, although our threat model considers it out of scope as described in Section 2, we can still provide a notification mechanism that turns on an LED light to indicate that the secure world is displaying confidential images. Although we do not implement it in Rushmore, previous systems [33,68] do and show that it can be a functional solution.…”

“…In other words, the normal world and the secure world use two separate display channels-the frame buffer (used by the normal world) and the overlay buffer (used by the secure world). Therefore, Rushmore does not need to make a pixel-wise copy of the normal world's content to display it unlike a technique used by previous systems [33,68].…”

“…Although we cannot do apples-to-apples comparisons, our TCB size is comparable to previous systems that have used OP-TEE for their implementation. For example, VButton [34] reports additional 1.3K, TrustTokenF [69] reports additional 4.5K, and TruZ-View [68] reports additional 3.4K.…”

“…The normal world can only access its own memory regions and the buffer memory regions shared between the normal world and the secure world for passing image data and arguments. Confidentiality of User Input: Previous UI protection schemes [2,34,36,68] protect not only UI elements but also user input. However, Rushmore only focuses on securing images and does not provide user input security.…”

“…while the secure world is displaying confidential animated images (e.g., a patient's fMRI animation). One can reduce the impact of this limitation by using a technique that existing approaches have used [33,68], where the secure world displays its content over a static "screenshot" (a pixel-wise copy) of the content displayed by the normal world. However, this technique still does not allow the normal world to actively display anything on the screen since the secure world still has exclusive access to the display hardware.…”