dynmx Prototype

dynmx (spoken dynamics) is a signature-based detection approach for behavioural malware features based on Windows API call sequences. In a simplified way, you can think of dynmx as a sort of YARA for API call traces (so called function logs) originating from malware sandboxes. Hence, the data basis for the detection approach are not the malware samples themselves which are analyzed statically but data that is generated during a dynamic analysis of the malware sample in a malware sandbox. Currently, dynmx supports function logs of the following malware sandboxes:

• VMRay (function log, text-based and XML format)
• CAPEv2 (report.json file)
• Cuckoo (report.json file)

The detection approach is described in detail in the master thesis Signature-Based Detection of Behavioural Malware Features with Windows API Calls. This project is the prototype implementation of this approach and was developed in the course of the master thesis. The signatures are manually defined by malware analysts in the dynmx signature DSL and can be detected in function logs with the help of this tool. Features and syntax of the dynmx signature DSL can also be found in the master thesis. Furthermore, you can find sample dynmx signatures in the repository dynmx-signatures. In addition to detecting malware features based on API calls, dynmx can extract OS resources that are used by the malware (a so called Access Activity Model). These resources are extracted by examining the API calls and reconstructing operations on OS resources. Currently, OS resources of the categories filesystem, registry and network are considered in the model.

Example

In the following section, examples are shown for the detection of malware features and for the extraction of resources.

Detection

For this example, we choose the malware sample with the SHA-256 hash sum c0832b1008aa0fc828654f9762e37bda019080cbdd92bd2453a05cfb3b79abb3. According to MalwareBazaar, the sample belongs to the malware family Amadey. There is a public VMRay analysis report of this sample available which also provides the function log traced by VMRay. This function log will be our data basis which we will use for the detection.

If we would like to know if the malware sample uses an injection technique called Process Hollowing, we can try to detect the following dynmx signature in the function log.

dynmx_signature:
  meta:
    name: process_hollow
    title: Process Hollowing
    description: Detection of Process hollowing malware feature
  detection:
    proc_hollow:
      # Create legit process in suspended mode
      - api_call: ["CreateProcess[AW]", "CreateProcessInternal[AW]"]
        with:
          - argument: "dwCreationFlags"
            operation: "flag is set"
            value: 0x4
          - return_value: "return"
            operation: "is not"
            value: 0
        store:
          - name: "hProcess"
            as: "proc_handle"
          - name: "hThread"
            as: "thread_handle"
      # Injection of malicious code into memory of previously created process
      - variant:
        - path:
          # Allocate memory with read, write, execute permission
          - api_call: ["VirtualAllocEx", "VirtualAlloc", "(Nt|Zw)AllocateVirtualMemory"]
            with:
              - argument: ["hProcess", "ProcessHandle"]
                operation: "is"
                value: "$(proc_handle)"
              - argument: ["flProtect", "Protect"]
                operation: "is"
                value: 0x40
          - api_call: ["WriteProcessMemory"]
            with:
              - argument: "hProcess"
                operation: "is"
                value: "$(proc_handle)"
          - api_call: ["SetThreadContext", "(Nt|Zw)SetContextThread"]
            with:
              - argument: "hThread"
                operation: "is"
                value: "$(thread_handle)"
        - path:
          # Map memory section with read, write, execute permission
          - api_call: "(Nt|Zw)MapViewOfSection"
            with:
              - argument: "ProcessHandle"
                operation: "is"
                value: "$(proc_handle)"
              - argument: "AccessProtection"
                operation: "is"
                value: 0x40
      # Resume thread to run injected malicious code
      - api_call: ["ResumeThread", "(Nt|Zw)ResumeThread"]
        with:
          - argument: ["hThread", "ThreadHandle"]
            operation: "is"
            value: "$(thread_handle)"
  condition: proc_hollow as sequence

Based on the signature, we can find some DSL features that make dynmx powerful:

• Definition of API call sequences with alternative paths
• Matching of API call function names with regular expressions
• Matching of argument and return values with several operators
• Storage of variables, e.g. in order to track handles in the API call sequence
• Definition of a detection condition with boolean operators (AND, OR, NOT)

If we run dynmx with the signature shown above against the function of the sample c0832b1008aa0fc828654f9762e37bda019080cbdd92bd2453a05cfb3b79abb3, we get the following output indicating that the signature was detected.

$ python3 dynmx.py detect -i 601941f00b194587c9e57c5fabaf1ef11596179bea007df9bdcdaa10f162cac9.json -s process_hollow.yml


    |
  __|         _  _    _  _  _
 /  |  |   | / |/ |  / |/ |/ |  /\/
 \_/|_/ \_/|/  |  |_/  |  |  |_/ /\_/
          /|
          \|
            
 Ver. 0.5 (PoC), by 0x534a


[+] Parsing 1 function log(s)
[+] Loaded 1 dynmx signature(s)
[+] Starting detection process with 1 worker(s). This probably takes some time...

[+] Result
process_hollow	c0832b1008aa0fc828654f9762e37bda019080cbdd92bd2453a05cfb3b79abb3.txt

We can get into more detail by setting the output format to detail. Now, we can see the exact API call sequence that was detected in the function log. Furthermore, we can see that the signature was detected in the process 51f0.exe.

$ python3 dynmx.py -f detail detect -i 601941f00b194587c9e57c5fabaf1ef11596179bea007df9bdcdaa10f162cac9.json -s process_hollow.yml


    |
  __|         _  _    _  _  _
 /  |  |   | / |/ |  / |/ |/ |  /\/
 \_/|_/ \_/|/  |  |_/  |  |  |_/ /\_/
          /|
          \|
            
 Ver. 0.5 (PoC), by 0x534a


[+] Parsing 1 function log(s)
[+] Loaded 1 dynmx signature(s)
[+] Starting detection process with 1 worker(s). This probably takes some time...