Milkomeda: Safeguarding the Mobile GPU Interface Using WebGL Security Checks
Title | Milkomeda: Safeguarding the Mobile GPU Interface Using WebGL Security Checks |
Publication Type | Conference Paper |
Year of Publication | 2018 |
Authors | Yao, Zhihao, Mirzamohammadi, Saeed, Amiri Sani, Ardalan, Payer, Mathias |
Conference Name | Proceedings of the 2018 ACM SIGSAC Conference on Computer and Communications Security |
Publisher | ACM |
ISBN Number | 978-1-4503-5693-0 |
Keywords | Human Behavior, human factors, mobile graphics security, pubcrawl, resilience, Resiliency, Scalability, Security Audits, WebGL security |
Abstract | GPU-accelerated graphics is commonly used in mobile applications. Unfortunately, the graphics interface exposes a large amount of potentially vulnerable kernel code (i.e., the GPU device driver) to untrusted applications. This broad attack surface has resulted in numerous reported vulnerabilities that are exploitable from unprivileged mobile apps. We observe that web browsers have faced and addressed the exact same problem in WebGL, a framework used by web apps for graphics acceleration. Web browser vendors have developed and deployed a plethora of security checks for the WebGL interface. We introduce Milkomeda, a system solution for automatically repurposing WebGL security checks to safeguard the mobile graphics interface. We show that these checks can be used with minimal modifications (which we have automated using a tool called CheckGen), significantly reducing the engineering effort. Moreover, we demonstrate an in-process shield space for deploying these checks for mobile applications. Compared to the multi-process architecture used by web browsers to protect the integrity of the security checks, our solution improves the graphics performance by eliminating the need for Inter-Process Communication and shared memory data transfer, while providing integrity guarantees for the evaluation of security checks. Our evaluation shows that Milkomeda achieves close-to-native GPU performance at reasonably increased CPU utilization. |
URL | https://dl.acm.org/citation.cfm?doid=3243734.3243772 |
DOI | 10.1145/3243734.3243772 |
Citation Key | yao_milkomeda:_2018 |