CVE-2021-29538: TensorFlow GPU 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 developing and deploying machine learning models. TensorFlow GPU is a specific package that enables GPU acceleration for TensorFlow computations. It allows users to leverage the power of GPUs to accelerate training and inference processes in machine learning models.

Vulnerability Detail

The vulnerability (CVE-2021-29538) exists in the Conv2DBackpropFilter implementation of TensorFlow GPU. The implementation computes a divisor based on user-provided data, specifically the shape of the tensors given as arguments. If all shapes are empty, the divisor 'work_unit_size' becomes 0. However, there is no check for this case before division, resulting in a division by zero runtime exception. This vulnerability can be exploited by an attacker to cause a denial of service (DoS) by triggering the division by zero error. The impact of this vulnerability can disrupt the availability of TensorFlow GPU and any applications relying on it.

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

Workaround

As a temporary workaround, users can avoid triggering the division by zero error by ensuring that the shapes of the tensors provided as arguments to Conv2DBackpropFilter are not all empty. This can be achieved by validating the input data and ensuring that at least one tensor has a non-empty shape. Here is an example code snippet demonstrating the workaround:

import tensorflow as tf

# Validate input tensors
if all(tensor.shape == [] for tensor in input_tensors):
    # Add a dummy tensor with a non-empty shape
    dummy_tensor = tf.constant([1], dtype=tf.float32)
    input_tensors.append(dummy_tensor)

# Perform Conv2DBackpropFilter operation
output_tensor = tf.nn.conv2d_backprop_filter(input_tensors, ...)

Conclusion

The vulnerability (CVE-2021-29538) in TensorFlow GPU's Conv2DBackpropFilter can lead to a division by zero error, potentially causing a denial of service. It is crucial for users to update their TensorFlow GPU installations to the patched versions once they are released. In the meantime, a temporary workaround is available to mitigate the risk. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow-gpu →

FAQ