Just over one year ago (November 2015), I released href="https://github.com/ChrisTruncer/WMIOps">WMIOps, a PowerShell
script that enables a user to carry out different actions via Windows
Management Instrumentation (WMI) on the local machine or a remote
machine. WMIOps can:
As I continued to develop WMIOps and use it during href="https://www.fireeye.com/services/red-team-operations.html">Mandiant
Red Team Operations, I realized that it has some of the same
capabilities that are in Remote Access Tools (RATs). WMIOps’s
capabilities were in a state of disparate functions, but if I wove
what existed along with new functionality, I could create a RAT. After
months of development and internal testing, I’m happy to publicly
release WMImplant.
WMImplant leverages WMI for the command and control channel, the
means for executing actions (gathering data, issuing commands, etc.)
on the targeted system, and data storage. It is designed to run both
interactively and non-interactively. When using WMImplant
interactively, it’s designed to have a menu of commands reminiscent of
Meterpreter, as shown in Figure 1.

Figure 1: WMImplant main menu
After spending some time developing WMImplant, I ran into issues
storing data on systems that used href="https://technet.microsoft.com/en-us/itpro/windows/keep-secure/introduction-to-device-guard-virtualization-based-security-and-code-integrity-policies">Device
Guard, a Microsoft security feature added in Windows 10 and Server
2016. Even though this feature and these operating systems are not
widely deployed today, I wanted WMImplant to support these systems
since I expect Device Guard protected systems to become more common,
especially at security-conscious organizations. Device Guard helps
protect systems by employing (among other capabilities not detailed here):
On a Device Guard protected system, attackers cannot run custom
executables, and the available PowerShell cmdlets are severely
restricted. For example, simple functionality such as base64 encoding
a string is not permitted within Constrained Language mode, as shown
in Figure 2.

Figure 2: PowerShell Constrained Language mode
blocking base64 encoding
At first, I designed WMImplant to use the Windows Registry for data
storage, as described in href="https://www.fireeye.com/content/dam/fireeye-www/global/en/current-threats/pdfs/wp-windows-management-instrumentation.pdf">Matt
Graeber’s WMI research. However, after discussing using the
Windows Registry for data storage with Matt Dunwoody (a Mandiant
coworker), he suggested, “Why not also use WMI itself for storage?”
This conversation led me to research using WMI for data storage. I
found a proof-of-concept for creating custom WMI properties (Figure 3)
in FireEye’s report on href="https://www.fireeye.com/content/dam/fireeye-www/global/en/current-threats/pdfs/wp-windows-management-instrumentation.pdf">WMI
Offense, Defense, and Forensics.

Figure 3: Sample code from FireEye report on WMI
Offense, Defense, and Forensics
However, after testing this code on a Device Guard protected system,
I discovered that this wasn’t permitted within Constrained Language
mode, as shown in Figure 4.

Figure 4: Constrained Language mode blocking WMI
property creation
After some additional research, I found that within Constrained
Language mode, users are able to create custom WMI classes. But, as
evidenced by Figure 4, WMI property creation is not allowed, so this
wouldn’t work for data storage. Therefore, my next thought was to
store data in an existing WMI property. In order to leverage an
existing WMI property, a few conditions would need to be present:
I modified an existing PowerShell script to enumerate all WMI
classes, find the properties of each class, check if each property is
a string type, and determine if it is writable (the script is href="https://gist.github.com/ChrisTruncer/f3fe3f04b9fdd1310507363f8bdad8be">available here).
The script identified a list of candidate WMI properties, but for
one reason or another, modifications to those that I initially tested
resulted in “general failures”. Then, I came across a class I have not
previously used: Win32_OSRecoveryConfiguration. This class has
a property named “DebugFilePath”, which is the file path where Windows
will place a memory dump after a computer failure, as shown in Figure 5.

Figure 5: Win32_OSRecoveryConfiguration’s
DebugFilePath property
The DebugFilePath appears to only accept a file path, and the
property should be limited to the length of a valid Windows paths (260
characters by default or 32k with LongPathsEnabled). In testing,
however, I discovered that I could write an arbitrary string to the
DebugFilePath property within Constrained Language mode without
adversely affecting the targeted system. The final test was to
determine how much data could be placed in the DebugFilePath property.

Figure 6: Data storage test within DebugFilePath
Figure 6 shows that the DebugFilePath property can store over 57
megabytes of data. This satisfied the data storage requirement for
WMImplant, and future testing showed that the DebugFilePath property
could store more than 250 megabytes of data. Additionally, using the
DebugFilePath WMI property for data storage provides the side-benefits
that it is easily retrievable and modifiable remotely.
This discovery shaped WMImplant’s command and control communications
methodology. For commands issued by WMImplant that require data
storage, the communication process is as follows:
This methodology for command and control communications minimizes
the amount of time that the WMI property is modified from its original state.
I’ve developed WMImplant for both interactive and non-interactive
use. Users also have the ability to change the user account that is
authenticating to the targeted machine. As shown in Figure 7, users
can issue the “change_user” command, provide the username and password
to use, and then all future commands through WMImplant will
authenticate with the provided credentials.

Figure 7: Changing the current user context
within WMImplant
The easiest way to use WMImplant is interactively; however, that
isn’t always possible. RATs such as Meterpreter or Cobalt Strike’s
Beacon allow users to load and execute PowerShell scripts, but both of
those tools require non-interactive use. That is, the tools accept a
command to run, execute it, and return the results. They do not allow
the user to interact with the command while running, however.
WMImplant includes a built-in command-line generating feature
specifically for this use case. To generate a command-line, start
WMImplant and specify the “gen_cli” command.
After issuing the “gen_cli” command, the user will be presented with
the normal WMImplant menu and asked for the command to be run.
WMImplant will then ask for any required information for the command
specified. Once the user has provided everything that’s required,
WMImplant will display the command-line command to run in a
non-interactive manner, as shown in Figure 8.

Figure 8: “gen_cli” output
At this point, the user can load WMImplant within the RAT of choice,
and copy and paste the command to run WMImplant non-interactively.
Another of WMImplant’s capabilities is the ability to run a
PowerShell script on a remote machine and receive script output. This
is performed through a multi-step process:
This multi-step process is demonstrated in Figure 9.

Figure 9: Remote PowerShell execution
While I’ve only talked about a limited number of WMImplant’s
features, others include:
I hope that WMImplant can help others as it has helped us on
multiple assessments. If you notice any bugs, please let me know and
I’ll be happy to get a fix pushed!
WMImplant can be downloaded here.
I want to state that I wouldn’t have been inspired to work on this
without the href="https://www.fireeye.com/content/dam/fireeye-www/global/en/current-threats/pdfs/wp-windows-management-instrumentation.pdf">previous
work of Matt Graeber, Willi Ballenthin, and Claudiu Teodorescu.
Their work gave me a lot of great ideas that I was able to build upon
when developing WMImplant.