Writing a libemu/Unicorn Compatability Layer

In this post we are going to take a quick look at what it takes to
write a libemu compatibility layer for the Unicorn engine. In the
course of this work, we will also import the libemu Win32 environment
to run under Unicorn.

For a bit of background, libemu is a lightweight x86 emulator
written in C by Paul Baecher and Markus Koetter. It was released in
2007 and includes a built-in Win32 environment that allows shellcodes
to resolve API at runtime. The library also provides end users with a
convenient way to receive callbacks when API functions are hit. The
original project supported 5 Windows dlls, 51 hooks and 234 opcodes
all wrapped in a tight 1mb package. Unfortunately it is no longer
being updated.

In late 2015, we saw the Unicorn engine project released by Nguyen
Anh Quynh and Dang Hoang Vu. This project takes the processor
emulators from QEMU and wraps them into an easy to use library.
Unicorn, however, does not provide a Win32 layer.

As an experiment, we were curious to see what it would take to bring
the libemu Win32 environment into Unicorn. This task actually turned
out to be quite simple since it was nicely self contained. In the
process of exploring this it also made sense to write a basic shim
layer to support the libemu API and translate its inner workings over
to Unicorn.

Lets start with the common libemu API:

The API is actually very similar to Unicorn:

The major differences are that Unicorn does everything through an
opaque uc_engine* handle, while libemu uses a series of structs such
as emu, emu_cpu, and emu_memory:

In general, the emu and emu_memory structures are passed directly as
arguments to API wrappers such as emu_cpu_get, emu_memory_get and the
emu_memory_read/write functions. There is one common case of direct
member access to the emu_cpu structure that requires some special
attention. This structure gives the user direct read/write access to
the emulator’s virtual processor and is commonly utilized by user
code. Examples to support include:

The next task was to see if we could mimic the direct access to the
emu_cpu elements as if they were static struct fields. Here we enter
the world of C++ operator overloading.

With these tasks complete, porting existing code from libemu over to
Unicorn should be a pretty straightforward task.

In Figure 1 we see an initial test, we put together that includes
the Win32 environment, shim layer, several API hooks and a hard coded payload.

Figure 1: Initial test of the libemu Win32
environment and hooks running under Unicorn

With this working, the next stage was to try it out against a larger
code base. Here we imported the userhooks.cpp from scdbg, an extension
of the libemu sctest that includes some 250 API hooks. As it turns
out, very few changes were required to get it working.

In Figure 2, we can see the results of testing it against a fairly
complex shellcode that:

  • allocates virtual memory
  • copies code to the new
  • creates a new thread
  • downloads an
  • checks the registry for the presence of Antivirus

Note that while this shellcode would normally do process injection,
scdbg handles it all inline for simplified analysis.

Figure 2: Complex shellcode running with
hooks imported from scdbg

Another large feature to test was the scdbg debug shell. When
testing software in an emulated environment, having interactive debug
tools available is extremely handy.

Figure 3 shows an example of setting a breakpoint, single stepping,
and examining memory of code running in the emulator.

Figure 3: Imported scdbg debug shell
running with Unicorn Engine and libemu shim layer


In this article we took a quick look at the differences between the
libemu and Unicorn emulators API. This allowed us to create a shim
layer to import legacy libemu code and use it with Unicorn largely unchanged.

Once the shim layer was in place, we next imported the libemu Win32
Environment so we could run it under Unicorn.

As a final test we ported several large portions of the scdbg
project, which was originally written to run under libemu. Here our
previous work allowed for the importation of scdbg’s 250+ API hooks
and debug shell to run under Unicorn with only minimal changes.

Overall the entire process went quite smoothly and should provide
benefits for developers of libemu and/or Unicorn. If you would like to
experiment for yourself you can download a copy of our test project here.

*** This is a Security Bloggers Network syndicated blog from Threat Research Blog authored by Threat Research Blog. Read the original post at: http://www.fireeye.com/blog/threat-research/2017/04/libemu-unicorn-compatability-layer.html