CVE-2021-29522: TensorFlow Conv3DBackprop Operations Division by Zero Denial of Service

Background

TensorFlow is an end-to-end open source platform for machine learning. It provides a comprehensive ecosystem of tools, libraries, and resources for building and deploying machine learning models. TensorFlow includes various operations for convolutional neural networks, including the tf.raw_ops.Conv3DBackprop* operations. These operations are responsible for computing the gradients of the convolutional filters in a 3D convolutional layer. However, a vulnerability has been discovered in these operations that can lead to a denial of service attack.

Vulnerability Detail

The vulnerability (CVE-2021-29522) arises from a failure to validate the input tensors in the tf.raw_ops.Conv3DBackprop* operations. Specifically, the implementation does not check if the input tensors are empty, resulting in a division by zero error. This vulnerability can be exploited by an attacker who controls the input sizes, allowing them to trigger a denial of service attack by causing a division by zero error. The impact of this vulnerability can be severe, as it can lead to the unavailability of the TensorFlow system and any applications relying on it.

pip install --upgrade "tensorflow-cpu==2.21.0"

Workaround

As a temporary workaround, users can manually validate the input tensors before calling the tf.raw_ops.Conv3DBackprop* operations. This can be done by checking if the tensors are empty and avoiding the operation if they are. Here is an example of how to perform this validation in Python:

import tensorflow as tf

# Validate input tensors
if input_tensor1.size == 0 or input_tensor2.size == 0:
    print('Input tensors are empty. Skipping Conv3DBackprop operation.')
    # Handle empty tensors case
else:
    # Perform Conv3DBackprop operation
    output_tensor = tf.raw_ops.Conv3DBackprop*(input_tensor1, input_tensor2)
    # Continue with further processing

Note that this workaround should only be used until the patched version of TensorFlow is available and installed.

Conclusion

The vulnerability (CVE-2021-29522) in the tf.raw_ops.Conv3DBackprop* operations of TensorFlow can lead to a denial of service attack by triggering a division by zero error. Users are advised to update their TensorFlow installations to the latest patched version as soon as it becomes available. In the meantime, a temporary workaround is available by manually validating the input tensors before calling the operations. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow-cpu →

FAQ