CVE-2021-29575: TensorFlow ReverseSequence 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 is widely used in various industries and research domains for tasks such as image classification, natural language processing, and reinforcement learning. The vulnerability described in this report affects the tf.raw_ops.ReverseSequence implementation in TensorFlow.

Vulnerability Detail

The implementation of tf.raw_ops.ReverseSequence in TensorFlow allows for a stack overflow and/or denial of service (DoS) attack. The vulnerability arises from the failure to validate the seq_dim and batch_dim arguments, which can result in negative values or invalid inputs. Negative values for seq_dim can lead to a stack overflow, while invalid values for batch_dim can cause a CHECK failure. The impact of this vulnerability depends on the version of Eigen code used to implement the operation. The issue has been assigned the CVE identifier CVE-2021-29575 and has been addressed in TensorFlow 2.5.0. Fixes have also been backported to TensorFlow 2.4.2, TensorFlow 2.3.3, TensorFlow 2.2.3, and TensorFlow 2.1.4, as these versions are still within the supported range.

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

Workaround

As a workaround, users can avoid using the tf.raw_ops.ReverseSequence operation or modify their code to validate the seq_dim and batch_dim arguments before calling the function. Here is an example of how to perform the validation check in Python:

import tensorflow as tf

def reverse_sequence_safe(data, seq_lengths, seq_dim, batch_dim):
    if seq_dim < 0 or seq_dim >= data.shape.ndims:
        raise ValueError('Invalid seq_dim')
    if batch_dim < 0 or batch_dim >= data.shape.ndims:
        raise ValueError('Invalid batch_dim')
    return tf.raw_ops.ReverseSequence(data, seq_lengths, seq_dim, batch_dim)

Conclusion

The vulnerability in TensorFlow's tf.raw_ops.ReverseSequence operation can result in a stack overflow or denial of service (DoS) attack. It is crucial to update to TensorFlow 2.5.0 or apply the appropriate patches for the affected versions. Users can also implement the provided workaround to validate the input arguments before calling the function. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow-cpu →

FAQ