CVE-2021-29548: Division by 0 in tensorflow - Division by 0 in QuantizedBatchNormWithGlobalNormalization

Background

TensorFlow is an open-source machine learning framework that provides a comprehensive ecosystem of tools, libraries, and community resources. It is widely used for building and deploying machine learning models. The vulnerability affects the QuantizedBatchNormWithGlobalNormalization function in TensorFlow, which is responsible for performing batch normalization on quantized tensors. Batch normalization is a technique used to improve the training and performance of deep neural networks. The vulnerability allows an attacker to trigger a runtime division by zero error, leading to a denial of service.

Vulnerability Detail

The vulnerability arises due to the lack of proper validation of constraints in the implementation of the QuantizedBatchNormWithGlobalNormalization function. An attacker can exploit this vulnerability by providing specific inputs that result in a division by zero error. This can cause the affected TensorFlow application to crash, resulting in a denial of service. The impact of this vulnerability is limited to the availability of the application and does not directly lead to unauthorized access or data leakage.

pip install --upgrade "tensorflow==2.21.0"

Workaround

As a temporary workaround, users can avoid using the QuantizedBatchNormWithGlobalNormalization function in their TensorFlow applications. By replacing the function with an alternative implementation or using a different approach for batch normalization, users can mitigate the risk of encountering the division by zero error. Here is an example of a workaround using a custom batch normalization implementation:

import tensorflow as tf

def custom_batch_norm(x):
    # Custom batch normalization implementation
    # ...
    return normalized_x

x = tf.constant([], shape=[0, 0, 0, 0], dtype=tf.quint8)
normalized_x = custom_batch_norm(x)

Conclusion

The division by zero vulnerability in the QuantizedBatchNormWithGlobalNormalization function of TensorFlow can be exploited by an attacker to cause a denial of service. The vulnerability has been patched in the latest version of TensorFlow, and users are advised to update their installations once the fix is released. In the meantime, a temporary workaround is available by avoiding the use of the affected function. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow →

FAQ