CVE-2021-29526: TensorFlow GPU Division by Zero Vulnerability

Background

TensorFlow is an end-to-end open source platform for machine learning. It provides a comprehensive ecosystem of tools, libraries, and resources for developing and deploying machine learning models. TensorFlow GPU is a specific package that enables GPU acceleration for TensorFlow. It allows users to leverage the power of GPUs to accelerate their machine learning computations, resulting in faster training and inference times.

Vulnerability Detail

The vulnerability (CVE-2021-29526) exists in the tf.raw_ops.Conv2D function of TensorFlow GPU. An attacker can exploit this vulnerability by triggering a division by zero. The vulnerable code can be found in the conv_ops.cc file of the TensorFlow source code. The division operation is performed using a value controlled by the caller, allowing an attacker to manipulate it and cause a division by zero. This vulnerability can lead to a crash or unexpected behavior in applications that utilize the affected function.

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

Workaround

As a temporary workaround, users can avoid the vulnerability by refraining from using the tf.raw_ops.Conv2D function in their code or by implementing input validation to ensure that the divisor value is not zero before calling the function. Here is an example of input validation code:

import tensorflow as tf

def safe_conv2d(input, filter, strides, padding):
    if filter == 0:
        raise ValueError('Divisor cannot be zero.')
    return tf.raw_ops.Conv2D(input, filter, strides, padding)

Conclusion

The TensorFlow GPU Division by Zero vulnerability (CVE-2021-29526) allows an attacker to trigger a division by zero in the tf.raw_ops.Conv2D function. This can result in a crash or unexpected behavior in applications that use TensorFlow GPU. Users are advised to update their TensorFlow GPU package to the latest patched version as soon as it becomes available. In the meantime, a temporary workaround is available by avoiding the vulnerable function or implementing input validation. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow-gpu →

FAQ