Runqing Yang (Zhejiang University), Shiqing Ma (Rutgers University), Haitao Xu (Arizona State University), Xiangyu Zhang (Purdue University), Yan Chen (Northwestern University)
Existing attack investigation solutions suffer from a few limitations such as inaccuracy (because of the dependence explosion problem), requiring instrumentation, using non-deterministic approaches and providing very low visibility. Such limitations have hindered their widespread and practical deployment. In this paper, we present UIScope, a novel accurate, instrumentation-free and deterministic attack investigation system, which also provides high visibility. The core idea of UIScope is to perform causality analysis on both UI elements/events which represent users' perspective and low level system events which provide detailed information of what happens under the hood, and then correlate system events with UI events to provide high accuracy and visibility. Long running processes are partitioned to individual UI transitions, to which low level system events are attributed, making the results accurate. The produced graphs contain (causally related) UI elements with which users are very familiar, making them easily accessible. We deployed UIScope on 7 machines for a week, and also utilized UIScope to conduct an investigation of 6 real-world attacks. Our evaluation shows that UIScope can precisely identify attack provenance while offering users thorough visibility into the attack context. UIScope only introduces negligible runtime overhead (0.2%) and space overhead (3.05 MB event logs per hour on average).