CVE-2022-21727: Integer Overflow Vulnerability in TensorFlow-GPU

Background

TensorFlow-GPU is an open-source machine learning framework that provides a comprehensive ecosystem for developing and deploying machine learning models. It is widely used in various domains for tasks such as image recognition, natural language processing, and recommendation systems. TensorFlow-GPU leverages the power of GPUs to accelerate the training and inference processes, making it suitable for large-scale machine learning tasks. However, a vulnerability has been discovered in the implementation of shape inference for the Dequantize function in TensorFlow-GPU.

Vulnerability Detail

The vulnerability arises from an integer overflow weakness in the Dequantize function's shape inference implementation. The axis argument, which specifies the axis along which to dequantize the input, can be set to a positive value up to the number of dimensions of the input tensor. However, the upper bound is not properly checked, and an attacker can trigger an integer overflow by providing a large positive value for axis. This can lead to unexpected behavior, memory corruption, or even remote code execution. The issue affects versions of TensorFlow-GPU up to 2.5.3, 2.6.3, and 2.7.1.

pip install --upgrade "tensorflow-gpu==2.12.0"

Workaround

As a temporary workaround, users can avoid the vulnerability by explicitly setting the axis argument to a valid value within the range of the input tensor's dimensions. By carefully inspecting the input tensor's shape, users can determine the appropriate value for axis and pass it as an argument to the Dequantize function. Here is an example of how to use the workaround in Python:

import tensorflow as tf

# Assuming `input_tensor` is the input tensor to be dequantized
axis = 0  # Set the `axis` argument to a valid value
output_tensor = tf.dequantize(input_tensor, axis=axis)

By explicitly setting the axis argument, users can ensure that the vulnerability is not triggered.

Conclusion

In conclusion, TensorFlow-GPU versions up to 2.5.3, 2.6.3, and 2.7.1 are vulnerable to an integer overflow weakness in the Dequantize function's shape inference implementation. This vulnerability can be exploited by providing a large positive value for the axis argument, leading to unexpected behavior, memory corruption, or remote code execution. Users are advised to update their TensorFlow-GPU installation to one of the fixed versions (2.8.0, 2.7.1, 2.6.3, or 2.5.3) and always stay up to date with the latest stable releases to ensure the security of their machine learning models. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow-gpu →

FAQ