Evaluating Security of Executable Steganography for Digital Software Watermarking

Man-at-the-end (MATE) attacks against software programs are difficult to protect. Adversaries have complete access to the binary program and can run it under both static and dynamic analysis to find and break any software protection mechanisms put in place. Even though full-proof protection is not possible practically or theoretically, the goal of software protection should be to make it more difficult for an adversary to find program secrets by increasing either their monetary cost or time. Protection mechanisms must be easy to integrate into the software development lifecycle, or else they are of little to no use. In this paper, we evaluate the practical security of a watermarking technique known as Weaver, which is intended to support software watermarking based on a new transformation technique called executable steganography. Weaver allows hiding of identification marks directly into a program binary in a way that makes it difficult for an adversary to find and remove. We performed instruction frequency analysis on 106 programs from the GNU coreutils package to understand and define Weaver's limitations and strengths as a watermarking technique. Our evaluation revealed that the initial prototype version of Weaver suffers from limitations in terms of standard benchmarks for steganography evaluation, such as its stealth. We found that this initial prototype of Weaver relied heavily on one type of instruction that does not frequently occur in standard programs, namely the mov instruction with an 8-byte immediate operand. Our instruction frequency analysis revealed a negative impact due to Weaver's over-reliance on this mov instruction.