• tries to unpack password protected attachments using the mail body as a knowledge base for passwords
• can also use a LLM to extract the password from the mail body
• unpacks

  mime type	password protection
  application/vnd.rar	yes
  application/x-7z-compressed	yes
  application/x-bzip	yes
  application/x-bzip2	yes
  application/x-tar	yes
  application/x-iso9660-image	yes
  application/x-rar	yes
  application/x-rar-compressed	yes
  application/zip	yes

• scans each dropped sample and attachment using the provided YARA rules; independent of the file type

1. use the provided docker-compose.yml and spawn the container
2. got into the etc docker-volume of the container, create a rules directory and put your .yar rule files into it
   (Note: a good start for rules may be https://yarahq.github.io/)
3. restart the container and check that it's running
4. Rspamd integration
   • go to your Rspamd configuration files and add the following to your external_services.conf

      malunpacker {
        servers = "172.22.1.1:10055";
        # needs to be set explicitly for Rspamd < 1.9.5
        scan_mime_parts = false;
        type = "icap";
        scheme = "respmod";
        x_client_header = true;
        # mime-part regex matching in content-type or filename
        # block all macros
        max_size = 3145728;
        timeout = 60.0;
        retransmits = 1;
        x_client_header = true; # Add X-Client-IP: $IP header
        x_rcpt_header = true; # Add X-Rcpt-To: $SMTP_RCPT header
        x_from_header = true; # Add X-Mail-From: $SMTP_FROM header
      }
     

   • adapt your VIRUS_FOUND symbol in your composites.conf and add the MALUNPACKER symbol to it's expression f.e.:

     VIRUS_FOUND {
       expression = "( CLAM_VIRUS | MALUNPACKER ) & !WHITELIST";
       score = 2000.0;
     }
     

5. That's it! You can now test malunpacker by sending some archive containeing malware or some ISO.

You may also enable the use of a LLM for credential extraction by specifing the environment variable USE_ML_FOR_CREDS_EXTRACTION=true in your docker-compose file.

The service uses rust-bert for the ML part.