Obfuscation-Resilient Code Recognition in Android Apps

Many Android developers take advantage of third-party libraries and code snippets from public sources to add functionality to apps. Besides making development more productive, external code can also be harmful, introduce vulnerabilities, or raise critical privacy issues that threaten the security of sensitive user data and amplify an app's attack surface. Reliably recognizing such code fragments in Android applications is challenging due to the widespread use of obfuscation techniques and a variety of ways, how developers can express semantically similar program statements.

We propose a code recognition technique that is resilient against common code transformations and that excels in identifying code fragments and libraries in Android applications. Our method relies on obfuscation-resilient features from the Abstract Syntax Tree of methods and uses them in combination with invariant attributes from method signatures to derive well-characterizing fingerprints. To identify similar code, we elaborate an effective scoring metric that reliably compares fingerprints at method, class, and package level. We investigate how well our solution tackles obfuscated, shrunken, and optimized code by applying our technique to real-world applications. We thoroughly evaluate our solution and demonstrate its practical ability to fingerprint and recognize code with high precision and recall.