Adrian Hada works as Principal Security Researcher at Ixia, a Keysight Business. Adrian holds a MSc in Information Security from Military Technical Academy in Bucharest. Adrian specializes in gathering Threat Intelligence on the latest Malware, Botnets, and exploitation practices being seen in network attacks. Adrian continuously improves the ATI Research Centers ability to detonate and analyze malware, providing Ixia’s customers with actionable intelligence and protection from the newest evolved threats.

Emulating Packed Linux Malware Using the Unicorn Emulator Framework

Binary packing has been around for a good number of years now. Uses include both benign – such as software and intellectual property protection – as well as malicious – malware authors use packing to avoid detection by common security products. The latter is of utmost interest in the security world.

Although most packed executables reside in the world of Windows malware, the past couple of years have shown more and more Linux threats adopting packing for the same purpose.

While common off-the-shelf packers, such as UPX, allow simple decompression of the packed binary, some malware authors rely on minor modifications or custom code to break the decompression process. Analyzing these samples is, generally, a complicated act of manual labor that requires a skilled analyst to break through the layers of obfuscation or relying on the automated yet resource-intensive sandbox. Analyzing malware samples at scale can’t rely on either of these approaches easily or cheaply.

In this presentation I’ll explain a work-in-progress project of emulating packed Linux binaries up to the point that we’re able to dump the underlying executable. This is achieved using the Unicorn Emulator, a framework that is based on QEMU but is much more customizable and easier to use. Although Unicorn is able to execute sequences of CPU instructions extremely well, the project doesn’t aim to perform full-binary execution leaving steps such as loading the binary, handling system calls, library functions and others to the person using the engine. I will explain the challenges of such an undertaking and offer a couple of examples of how this approach can actually work, as well as what its limitations are.