Arancino is dynamic protection framework that can be used to defend Intel Pin against anti-instrumentation attacks. Arancino is a research project developed at NECSTLab.

We present the findings of this work in a research paper:

Measuring and Defeating Anti-Instrumentation-Equipped Malware
Mario Polino, Andrea Continella, Stefano D’Alessio, Lorenzo Fontana, Fabio Gritti, Sebastiano Mariani, and Stefano Zanero In Proceedings of the Conference on Detection of Intrusions and Malware and Vulnerability Assessment (DIMVA), July 2017

[PDF]

If you use Arancino in a scientific publication, we would appreciate citations using this Bibtex entry:

@InProceedings{polino_arancino_2017,
  author     = {Polino, Mario and Continella, Andrea and Mariani, Sebastiano and D’Alessio, Stefano and Fontata, Lorenzo and Gritti, Fabio and Zanero, Stefano},
  title      = {{Measuring and Defeating Anti-Instrumentation-Equipped Malware}},
  booktitle  = {{Proceedings of the Conference on Detection of Intrusions and Malware and Vulnerability Assessment (DIMVA)}},
  location   = {{Bonn, Germany}},
  date       = {2017},
}

Dynamic Binary Instrumentation (DBI) Tools are useful in malware analysis. However, DBI frameworks are not completely transparent to the analyzed malware, and, in fact, anti-instrumentation techniques have been developed to detect the instrumentation process. We classified such anti-instrumentation techniques in four categories:

• Code Cache Artifacts. These techniques aim at detecting artifacts that are usually inherent of a DBI cache. For example, the Instruction Pointer is different if a binary is instrumented. In fact, in a DBI Tool the code is usually executed from a different memory region, called code cache, rather than from the main module of the binary.

• Environment Artifacts. The memory layout of an instrumented binary is deeply different respect to the one of a not instrumented one. Searching for DBI artifacts such as strings or particular code patterns in memory can eventually reveal the presence of a DBI tool inside the target process memory. Also, the parent process of an instrumented binary is often the DBI tool itself.

• JIT Compiler Detection. JIT compilers make a lot of noise inside the process in terms of Windows API calls and pages allocation. These artifacts can be leveraged by the instrumented program to detect the presence of a DBI tool.

• Overhead Detection. Instrumentation adds an observable overhead in the execution of the target program. This overhead can be noticed by malware samples by estimating the execution time of a particular set of instructions.

Our approach leverages the complete control that a DBI Tool has on the instrumented binary to hide the artifacts that the DBI tool itself introduces during the instrumentation process. In fact, by instrumenting a binary, we can identify when it tries to leverage such artifacts to evade the analysis. In practice, we designed a set of countermeasures for the anti-instrumentation techniques we described above.

On top of Arancino, we implemented a generic, anti-instrumentation-resilient unpacker.

In the spirit of open science we are happy to release our datasets to the community. You can find our data here.

In order to automate the analysis of malware samples, we also developed a lightweight analysis framework, which we release here.