CVE-2022-35971: TensorFlow CHECK Fail in FakeQuantWithMinMaxVars Leading to Denial of Service

Background

TensorFlow is an open-source machine learning library developed by Google. It provides a comprehensive ecosystem for building and deploying machine learning models. TensorFlow supports various tasks, including deep learning and neural network training, and is widely used in both research and production environments. However, it has been found to have a vulnerability in the FakeQuantWithMinMaxVars operation, which can lead to a denial of service if improperly used.

Vulnerability Detail

The vulnerability arises when the FakeQuantWithMinMaxVars operation is provided with min or max tensors that have a nonzero rank. This situation triggers a CHECK fail, which can be exploited to cause a denial of service. An attacker can craft inputs that lead to this failure, effectively crashing the application or service utilizing TensorFlow. This vulnerability is particularly concerning for applications that rely on TensorFlow for real-time processing or critical tasks.

pip install --upgrade "tensorflow==2.21.0"

Workaround

As a workaround, ensure that the min and max tensors provided to the FakeQuantWithMinMaxVars operation have zero rank. Here is an example of how to correctly use the operation:

import tensorflow as tf

num_bits = 8
narrow_range = False
inputs = tf.constant(0, shape=[], dtype=tf.float32)  # Zero rank tensor
min = tf.constant(0, shape=[], dtype=tf.float32)      # Zero rank tensor
max = tf.constant(0, shape=[], dtype=tf.float32)      # Zero rank tensor
tf.raw_ops.FakeQuantWithMinMaxVars(inputs=inputs, min=min, max=max, num_bits=num_bits, narrow_range=narrow_range)

Conclusion

In summary, CVE-2022-35971 is a significant vulnerability in TensorFlow that can lead to denial of service if the FakeQuantWithMinMaxVars operation is misused. It is crucial to update to the latest patched versions to mitigate this risk and ensure the stability of applications using TensorFlow.

See every vulnerability recorded for tensorflow →

FAQ