Dynamic taint tracking is an important field of study with many

Java-based tools and systems created to implement it, including

Phosphor, a general purpose taint tracking tool designed for

commodity JVMs like Oracle and OpenJDK. Phosphor works by

instrumenting core Java libraries and the entire application bytecode

with operations to accurately propagate taint information. Prior work

strived to reduce the performance overhead of Phosphor by doing

partial instrumentation. The analysis that determined which parts of

the program to instrument was effective but flawed.

This paper aims to improve that analysis and further reduce the

performance overhead by instrumenting less of the program. We use the

Petablox program analysis tool and custom Datalog rules to perform an

information flow analysis that better models Phosphor's behavior,

including calls across native library boundaries. We find that we

obtain a reduction in the amount of a program that needs to be

instrumented by 79.9%.