CVE-2021-29539: TensorFlow GPU ImmutableConst Segfault

Background

TensorFlow is an end-to-end open source platform for machine learning. It provides a comprehensive set of tools, libraries, and resources for developing and deploying machine learning models. TensorFlow allows users to build and train neural networks, perform data preprocessing, and deploy models in various environments. The TensorFlow GPU package, specifically the tensorflow-gpu PyPI package, is a version of TensorFlow optimized for GPU acceleration. It enables users to leverage the power of GPUs for faster training and inference of machine learning models.

Vulnerability Detail

The vulnerability identified as CVE-2021-29539 affects the TensorFlow GPU package. It arises when calling the tf.raw_ops.ImmutableConst function with a dtype parameter set to tf.resource or tf.variant. This triggers a segmentation fault (segfault) in the implementation code, as it assumes that the tensor contents are pure scalars. The issue has been patched in commit 4f663d4b8f0bec1b48da6fa091a7d29609980fa4 and will be included in TensorFlow 2.5.0. The vulnerability can have severe consequences, including crashes or unexpected behavior in applications utilizing TensorFlow GPU. It is crucial to address this vulnerability to ensure the stability and reliability of machine learning workflows.

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

Workaround

If updating to the latest version of tensorflow-gpu is not immediately feasible, a workaround can be implemented by filtering the dtype argument when using tf.raw_ops.ImmutableConst. By ensuring that only valid scalar types are used, the segfault can be avoided. Here is an example of how to apply the workaround:

def safe_immutable_const(value, dtype):
    valid_dtypes = [tf.float32, tf.float64, tf.int32, tf.int64, tf.bool, tf.string]
    if dtype in valid_dtypes:
        return tf.raw_ops.ImmutableConst(value=value, dtype=dtype)
    else:
        raise ValueError('Invalid dtype')

Conclusion

The CVE-2021-29539 vulnerability in TensorFlow GPU can lead to crashes or unexpected behavior in machine learning applications. It is crucial to update the tensorflow-gpu package to version 2.5.0 or later to fix the vulnerability. Alternatively, a workaround can be implemented by filtering the dtype argument when using tf.raw_ops.ImmutableConst. For further support and assistance in vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow-gpu →

FAQ