To Unpack or Not to Unpack: Living with Packers to Enable Dynamic Analysis of Android Apps

Abstract

Android apps have become a valuable target for app modifiers and imitators due to its popularity and being trusted with highly sensitive data. Packers, on the other hand, protect apps from tampering with various anti-analysis techniques embedded in the app. Meanwhile, packers also conceal certain behavior potentially against the interest of the users, aside from being abused by malware for stealth. Security practitioners typically try to capture undesired behavior at runtime with hooking (e.g., Frida) or debugging techniques, which are heavily affected by packers. Unpackers have been the community's continuous effort to address this, but due to the emerging commercial packers, our study shows that none of the unpackers remain effective, and they are unfit for this purpose as unpacked apps can no longer run. We first perform a large-scale prevalence analysis of Android packers with a real-world dataset of 12,341 apps, the first of its kind, to find out what percentage of Android apps are actually packed and to what extent dynamic analysis is hindered. We then propose Purifire, an evasion engine to bypass packers' anti-analysis techniques and enable dynamic analysis on packed apps without unpacking them. Purifire is based on eBPF, a low-level kernel feature, which provides observability and invisibility to userspace apps to enforce defined evasion rules while staying low-profile. Our evaluation shows that Purifire is able to bypass packers' anti-analysis checks and more importantly, for previous research works suffering from packers, we observe a significant improvement (e.g., a much higher number of detected items such as device fingerprints).