CVE-2021-29580: Undefined behavior and CHECK-fail in FractionalMaxPoolGrad

Background

The TensorFlow package is an open-source machine learning framework developed by Google. It provides a comprehensive set of tools and libraries for building and deploying machine learning models. TensorFlow includes various operations and functions for data manipulation, model training, and inference. One of these operations is tf.raw_ops.FractionalMaxPoolGrad, which is used for fractional max pooling in neural networks. Fractional max pooling is a technique that allows for non-integer pooling sizes, providing more flexibility in pooling operations. However, a vulnerability has been discovered in the implementation of FractionalMaxPoolGrad that can lead to undefined behavior and denial of service attacks.

Vulnerability Detail

The vulnerability in tf.raw_ops.FractionalMaxPoolGrad arises from the lack of validation for empty input and output tensors. If one of the input tensors is empty, the implementation triggers undefined behavior. This can result in unexpected program behavior and potential security risks. Additionally, a CHECK condition in the code becomes false when the input and output tensors are empty, leading to a process abort and denial of service. The vulnerability has a CVSS score of 2.5, indicating a low severity level. It affects versions of TensorFlow up to 2.4.2.

pip install --upgrade "tensorflow==2.21.0"

Workaround

As a workaround, users can ensure that the input and output tensors provided to tf.raw_ops.FractionalMaxPoolGrad are not empty and have the same rank. This can be achieved by validating the tensors before passing them to the function. Here is an example of how to perform the validation:

import tensorflow as tf

orig_input = tf.constant([2, 3], shape=[1, 1, 1, 2], dtype=tf.int64)
orig_output = tf.constant([], dtype=tf.int64)
out_backprop = tf.zeros([2, 3, 6, 6], dtype=tf.int64)
row_pooling_sequence = tf.constant([0], shape=[1], dtype=tf.int64)
col_pooling_sequence = tf.constant([0], shape=[1], dtype=tf.int64)

if orig_input.size > 0 and orig_output.size > 0 and orig_input.ndim == orig_output.ndim:
  tf.raw_ops.FractionalMaxPoolGrad(
    orig_input=orig_input, orig_output=orig_output, out_backprop=out_backprop,
    row_pooling_sequence=row_pooling_sequence,
    col_pooling_sequence=col_pooling_sequence, overlapping=False)
else:
  print('Invalid input or output tensors')

By performing this validation, the undefined behavior and denial of service issues can be avoided.

Conclusion

The vulnerability in tf.raw_ops.FractionalMaxPoolGrad can lead to undefined behavior and denial of service attacks. It has been patched in TensorFlow 2.5.0, and users are advised to update their installations. Backported fixes are also available for TensorFlow 2.4.2, TensorFlow 2.3.3, TensorFlow 2.2.3, and TensorFlow 2.1.4. For further support and information on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow →

FAQ