Mandiant
recently responded to an incident at a critical infrastructure
organization where an attacker deployed malware designed to manipulate
industrial safety systems. The targeted systems provided emergency
shutdown capability for industrial processes. We assess with moderate
confidence that the attacker was developing the capability to cause
physical damage and inadvertently shutdown operations. This malware,
which we call TRITON, is an attack framework built to interact with
Triconex Safety Instrumented System (SIS) controllers. We have not
attributed the incident to a threat actor, though we believe the
activity is consistent with a nation state preparing for an attack.
TRITON is one of a limited number of publicly identified malicious
software families targeted at href="/content/fireeye-www/en_US/solutions/industrial-systems-and-critical-infrastructure-security.html">industrial
control systems (ICS). It follows href="/content/fireeye-www/en_US/company/press-releases/2014/fireeye-reveals-rise-in-advanced-threat-activities-by-iranian-linked-ajax-security-team-in-post-stuxnet-era.html">Stuxnet
which was used against Iran in 2010 and Industroyer which we believe
was deployed by Sandworm Team against Ukraine in 2016. TRITON is
consistent with these attacks, in that it could prevent safety
mechanisms from executing their intended function, resulting in a
physical consequence.
|
|
|
TRITON | width="189"> | Main |
library.zip | Custom |
Table 1: Description of TRITON Malware
The attacker gained remote access to an SIS engineering workstation
and deployed the TRITON attack framework to reprogram the SIS
controllers. During the incident, some SIS controllers entered a
failed safe state, which automatically shutdown the industrial process
and prompted the asset owner to initiate an investigation. The
investigation found that the SIS controllers initiated a safe shutdown
when application code between redundant processing units failed a
validation check -- resulting in an MP diagnostic failure message.
We assess with moderate confidence that the attacker inadvertently
shutdown operations while developing the ability to cause physical
damage for the following reasons:
FireEye has not
connected this activity to any actor we currently track; however, we
assess with moderate confidence that the actor is sponsored by a
nation state. The targeting of critical infrastructure as well as the
attacker’s persistence, lack of any clear monetary goal and the
technical resources necessary to create the attack framework suggest a
well-resourced nation state actor. Specifically, the following facts
support this assessment:
The attacker targeted the SIS suggesting an interest in causing a
high-impact attack with physical consequences. This is an attack
objective not typically seen from cyber-crime groups.
The attacker deployed TRITON shortly after gaining access to the SIS
system, indicating that they had pre-built and tested the tool which
would require access to hardware and software that is not widely
available. TRITON is also designed to communicate using the
proprietary TriStation protocol which is not publicly documented
suggesting the adversary independently reverse engineered this protocol.
The targeting of critical infrastructure to disrupt, degrade, or
destroy systems is consistent with numerous attack and reconnaissance
activities carried out globally by Russian, Iranian, North Korean,
U.S., and Israeli nation state actors. Intrusions of this nature do
not necessarily indicate an immediate intent to disrupt targeted
systems, and may be preparation for a contingency.
Figure 1: ICS Reference Architecture
Modern industrial process control and automation systems rely on a
variety of sophisticated control systems and safety functions. These
systems and functions are often referred to as href="/content/fireeye-www/en_US/solutions/industrial-systems-and-critical-infrastructure-security.html">Industrial
Control Systems (ICS) or Operational Technology (OT).
A Distributed Control System (DCS) provides human operators with the
ability to remotely monitor and control an industrial process. It is a
computerized control system consisting of computers, software
applications and controllers. An Engineering Workstation is a computer
used for configuration, maintenance and diagnostics of the control
system applications and other control system equipment.
A SIS is an autonomous control system that independently monitors
the status of the process under control. If the process exceeds the
parameters that define a hazardous state, the SIS attempts to bring
the process back into a safe state or automatically performs a safe
shutdown of the process. If the SIS and DCS controls fail, the final
line of defense is the design of the industrial facility, which
includes mechanical protections on equipment (e.g. rupture discs),
physical alarms, emergency response procedures and other mechanisms to
mitigate dangerous situations.
Asset owners employ varied approaches to interface their plant's DCS
with the SIS. The traditional approach relies on the principles of
segregation for both communication infrastructures and control
strategies. For at least the past decade, there has been a trend
towards integrating DCS and SIS designs for various reasons including
lower cost, ease of use, and benefits achieved from exchanging
information between the DCS and SIS. We believe TRITON acutely
demonstrates the risk associated with integrated designs that allow
bi-directional communication between DCS and SIS network hosts.
Figure 2: Temporal Relationship Between
Cyber Security and Safety
The attack lifecycle for disruptive attacks against ICS is similar
to other types of cyber attacks, with a few key distinctions. First,
the attacker’s mission is to disrupt an operational process rather
than steal data. Second, the attacker must have performed OT
reconnaissance and have sufficient specialized engineering knowledge
to understand the industrial process being controlled and successfully
manipulate it.
Figure 2 represents the relationship between cyber security and
safety controls in a process control environment. Even if cyber
security measures fail, safety controls are designed to prevent
physical damage. To maximize physical impact, a cyber attacker would
also need to bypass safety controls.
The SIS threat model below highlights some of the options available
to an attacker who has successfully compromised an SIS.
Attack Option 1: Use the SIS to shutdown the process
Attack Option 2: Reprogram the SIS to allow an unsafe state
Attack Option 3: Reprogram the SIS to allow an unsafe state – while
using the DCS to create an unsafe state or hazard
We assess with moderate confidence that the attacker’s long-term
objective was to develop the capability to cause a physical
consequence. We base this on the fact that the attacker initially
obtained a reliable foothold on the DCS and could have developed the
capability to manipulate the process or shutdown the plant, but
instead proceeded to compromise the SIS system. Compromising both the
DCS and SIS system would enable the attacker to develop and carry out
an attack that causes the maximum amount of damage allowed by the
physical and mechanical safeguards in place.
Once on the SIS network, the attacker used their pre-built TRITON
attack framework to interact with the SIS controllers using the
TriStation protocol. The attacker could have caused a process shutdown
by issuing a halt command or intentionally uploading flawed code to
the SIS controller to cause it to fail. Instead, the attacker made
several attempts over a period of time to develop and deliver
functioning control logic for the SIS controllers in this target
environment. While these attempts appear to have failed due one of the
attack scripts’ conditional checks, the attacker persisted with their
efforts. This suggests the attacker was intent on causing a specific
outcome beyond a process shutdown.
Of note, on several occasions, we have observed evidence of long
term intrusions into ICS which were not ultimately used to disrupt or
disable operations. For instance, Russian operators, such as Sandworm
Team, have compromised Western ICS over a multi-year period without
causing a disruption.
The TRITON attack tool was built with a number of features,
including the ability to read and write programs, read and write
individual functions and query the state of the SIS controller.
However, only some of these capabilities were leveraged in the
trilog.exe sample (e.g. the attacker did not leverage all of TRITON’s
extensive reconnaissance capabilities).
The TRITON malware contained the capability to communicate with
Triconex SIS controllers (e.g. send specific commands such as halt
or read its memory content) and remotely reprogram them with an
attacker-defined payload. The TRITON sample Mandiant analyzed added an
attacker-provided program to the execution table of the Triconex
controller. This sample left legitimate programs in place, expecting
the controller to continue operating without a fault or exception. If
the controller failed, TRITON would attempt to return it to a running
state. If the controller did not recover within a defined time window,
this sample would overwrite the malicious program with invalid data to
cover its tracks.
Asset owners who wish to defend against the capabilities
demonstrated in the incident, should consider the following controls:
Figure 3: Triconex Key Switch (source)
Figure 4: TRITON Architecture and Attack Scenario
TRITON was deployed on an SIS engineering workstation running the
Microsoft Windows operating system. The malware was named to
masquerade as the legitimate Triconex Trilog application. This
application is used for reviewing logs and is a part of the TriStation
application suite. The malware was delivered as a Py2EXE compiled
python script dependent on a zip file containing standard Python
libraries, open source libraries, as well as the attacker-developed
Triconex attack framework for interacting with the Triconex
controllers. Along with the executable, two binary files,
inject.bin (malicious function code) and imain.bin
(malicious control logic), were deployed as the controller’s
payload. These file names were hard coded in the Py2EXE compiled
python script.
Trilog.exe took one option from the command line, which was a single
IP address of the target Triconex device. It did not leverage the
underlying TRITON library’s capability for Triconex device discovery,
instead an instance of trilog.exe had to be invoked separately for
each target controller in the environment. Once invoked, trilog.exe
checked the status of the controller, then read the configuration
information exposed by the TriStation protocol. If the controller was
in a running state, trilog.exe encoded the two payload files
inject.bin and imain.bin and passed them to the
communication libraries to be appended to the controller’s program
memory and execution table.
After payload files were inserted into memory on the Triconex
controller, the script initiated a countdown, periodically checking
the status of the controller. If an error was detected, the
communication library’s method SafeAppendProgramMod attempted
to reset the controller to the previous state using a TriStation
protocol command. If this failed, trilog.exe attempted to write a
small ‘dummy’ program to memory. We assess that this was an
anti-forensics technique to hide the presence of the attacker code on
the Triconex controller.
Working with the asset owner, Mandiant ran trilog.exe in a lab
environment with a valid Triconex controller and discovered a
conditional check in the malware that prevented the payload binary
from persisting in the environment. Mandiant confirmed that, after
correcting patching the attack script to remove this check, the
payload binary would persist in controller memory, and the controller
would continue to run.
TRITON implements the TriStation protocol, which is the protocol
used by the legitimate TriStation application, to configure controllers.
TsHi is the high-level interface created by the
malware’s authors that allows the threat actor’s operators to
implement attack scripts using the TRITON framework. It exposes
functions for both reconnaissance and attack. The functions generally
accept binary data from the user, and handle the code ‘signing’ and
check sums prior to passing the data to lower level libraries for
serialization on to the network.
TsBase, another attacker-written module, contains the
functions called by TsHi, which translate the attacker’s
intended action to the appropriate TriStation protocol function code.
For certain functions, it also packs and pads the data in to the
appropriate format.
TsLow is an additional attacker module that implements the
TriStation UDP wire protocol. The TsBase library primarily
depends on the ts_exec method. This method takes the function
code and expected response code, and serializes the commands payload
over UDP. It checks the response from the controller against the
expected value and returns a result data structure indicating success
or a False object representing failure.
TsLow also exposes the connect method used to check
connectivity to the target controller. If invoked with no targets, it
runs the device discovery function detect_ip. This leverages a
"ping" message over the TriStation protocol using IP
broadcast to find controllers that are reachable via a router from
where the script is invoked.
|
|
trilog.exe | MD5: 6c39c3f4a08d3d78f2eb973a94bd7718 SHA-256: e8542c07b2af63ee7e72ce5d97d91036c5da56e2b091aa2afe737b224305d230 |
imain.bin | MD5: |
inject.bin | MD5: |
library.zip | MD5: |
TS_cnames.pyc | MD5: |
TsBase.pyc | MD5: |
TsHi.pyc | MD5: |
TsLow.pyc | MD5: |
sh.pyc | MD5: |
rule TRITON_ICS_FRAMEWORK $py_cnames_01 = $py_tslibs_01 = $py_tsbase_01 = "TsBase.py" $py_tslow_01 = "TsLow.py" nocase ascii $py_crc_01 = "crc.pyc" nocase $py_sh_01 = "sh.pyc" nocase $py_keyword_01 = " condition: |