CVE-2022-29201: TensorFlow QuantizedConv2D Undefined Behavior Vulnerability

Sponsored Habilume Habit tracker with a companion that grows with you Complete habits, earn Lume Points, and watch your Lume grow as you stay consistent. App Store Google Play

Background

TensorFlow is an open-source machine learning library developed by Google, widely used for building and training machine learning models. It provides a flexible ecosystem for deploying machine learning applications across various platforms. The library includes various operations, including quantized operations that optimize performance on hardware with limited precision. However, a vulnerability in the QuantizedConv2D operation can lead to undefined behavior when input arguments are not properly validated.

Vulnerability Detail

The vulnerability arises from the implementation of the tf.raw_ops.QuantizedConv2D function, which fails to validate input arguments adequately. When empty arguments are passed, the function binds references to nullptr, leading to potential crashes or undefined behavior during execution. This can be exploited by providing invalid input shapes, which may result in unexpected application behavior, making it critical for users to address this issue promptly.

pip install --upgrade "tensorflow==2.21.0"

Workaround

As a workaround, ensure that all input arguments to the QuantizedConv2D operation are properly validated before calling the function. For example, avoid passing empty tensors as shown below:

import tensorflow as tf

input = tf.constant(1, shape=[1, 2, 3, 3], dtype=tf.quint8)
filter = tf.constant(1, shape=[1, 2, 3, 3], dtype=tf.quint8)
min_input = tf.constant(0, shape=[], dtype=tf.float32)
max_input = tf.constant(0, shape=[], dtype=tf.float32)
min_filter = tf.constant(0, shape=[], dtype=tf.float32)
max_filter = tf.constant(0, shape=[], dtype=tf.float32)

# Valid arguments should be provided
# Call the QuantizedConv2D operation with valid inputs
result = tf.raw_ops.QuantizedConv2D(
  input=input,
  filter=filter,
  min_input=min_input,
  max_input=max_input,
  min_filter=min_filter,
  max_filter=max_filter,
  strides=[1, 1, 1, 1],
  padding="SAME")

Conclusion

The CVE-2022-29201 vulnerability in TensorFlow's QuantizedConv2D operation poses a risk of undefined behavior due to inadequate input validation. Users are advised to update to the latest patched versions to mitigate this risk and ensure the stability of their applications.

See every vulnerability recorded for tensorflow →

FAQ