CVE-2021-37641: TensorFlow CPU - Out-of-bounds Read in RaggedGather

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 allows users to build and train neural networks, perform data preprocessing and augmentation, and deploy models in various environments. The TensorFlow CPU package, available on PyPI, is a CPU-only version of TensorFlow that can be used on systems without GPU support.

Vulnerability Detail

The vulnerability identified as CVE-2021-37641 affects the TensorFlow CPU package. It arises from a flaw in the implementation of the tf.raw_ops.RaggedGather function. When the arguments provided to this function do not determine a valid ragged tensor, it can trigger a read from outside the bounds of heap allocated buffers. The implementation fails to check that the tensor has a rank of at least 1 (i.e., it is not a scalar) before reading the first dimension of its shape. Additionally, the implementation does not verify that the list of tensors given by params_nested_splits is not empty. This vulnerability can potentially lead to information disclosure or a crash of the TensorFlow application.

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

Workaround

As a workaround, users can avoid using the tf.raw_ops.RaggedGather function with arguments that may result in an invalid ragged tensor. It is recommended to carefully validate the inputs and ensure that the tensor has a rank of at least 1 and the list of tensors provided is not empty. Here is an example of how to perform the necessary checks:

import tensorflow as tf

def safe_ragged_gather(params, indices, params_nested_splits):
    if tf.rank(params) < 1:
        raise ValueError('Invalid params tensor: rank must be at least 1')
    if not params_nested_splits:
        raise ValueError('Invalid params_nested_splits: list must not be empty')
    return tf.raw_ops.RaggedGather(params=params, indices=indices, params_nested_splits=params_nested_splits)

Conclusion

In conclusion, the TensorFlow CPU package is affected by a vulnerability (CVE-2021-37641) that can result in an out-of-bounds read when using the tf.raw_ops.RaggedGather function with certain inputs. This vulnerability has been addressed in TensorFlow 2.6.0 and will also be patched in earlier versions. Users are advised to update to the patched versions or apply the provided workaround to mitigate the risk. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow-cpu →

FAQ