Implementation and Evaluation of a Static Backwards Data Flow Analysis in FlowDroid

In the last decade, many static taint analysis tools based on IFDS have been proposed. Nearly all of them only have a single analysis direction, with most opting for a forward-directed analysis and some for a backward-directed analysis.

In this thesis, we investigate the effect of the analysis direction on the runtime and memory consumption. We extended FlowDroid to feature a backward-directed analysis, which is equally precise as FlowDroid's forward analysis. We evaluated both implementations on DroidBench and 200 real-world apps. The results show that there is a favorable analysis direction for many apps and the right choice can reduce the runtime dramatically.