CVE-2021-29523: Denial of Service in TensorFlow-GPU

Background

TensorFlow-GPU is an open-source platform for machine learning that provides end-to-end capabilities. It is widely used for developing and deploying machine learning models. TensorFlow-GPU allows users to efficiently train and run models on GPUs, enabling faster computation and improved performance. However, a vulnerability has been identified in TensorFlow-GPU that can lead to a denial of service (DoS) attack.

Vulnerability Detail

The vulnerability (CVE-2021-29523) in TensorFlow-GPU is triggered by a CHECK-fail in the tf.raw_ops.AddManySparseToTensorsMap function. The implementation of this function takes the values specified in the sparse_shape parameter as dimensions for the output shape. However, the TensorShape constructor used in this implementation contains a legacy code that triggers a CHECK operation when the InitDims function returns a non-OK status. This vulnerability can be exploited by an attacker to cause a DoS by providing malicious input that triggers the CHECK-fail.

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

Workaround

As a workaround, users can apply input validation and sanitization techniques to prevent the exploitation of this vulnerability. By carefully validating and sanitizing the input provided to the AddManySparseToTensorsMap function, users can mitigate the risk of a DoS attack. Here is an example of input validation code:

import tensorflow as tf

def add_sparse_to_tensors_map(sparse_indices, sparse_values, sparse_shape):
    # Perform input validation
    if not isinstance(sparse_indices, tf.Tensor):
        raise ValueError('sparse_indices must be a TensorFlow Tensor')
    if not isinstance(sparse_values, tf.Tensor):
        raise ValueError('sparse_values must be a TensorFlow Tensor')
    if not isinstance(sparse_shape, tf.Tensor):
        raise ValueError('sparse_shape must be a TensorFlow Tensor')

    # Call the patched function
    tf.raw_ops.AddManySparseToTensorsMap(sparse_indices, sparse_values, sparse_shape)

By validating the input types and ensuring they are TensorFlow Tensors, the risk of triggering the vulnerability can be reduced.

Conclusion

The vulnerability (CVE-2021-29523) in TensorFlow-GPU can lead to a denial of service (DoS) attack. It is crucial for users to upgrade their TensorFlow-GPU installations to the latest patched version (2.5.0) or the backported patched versions (2.4.2, 2.3.3, 2.2.3, and 2.1.4) to mitigate the risk. Additionally, applying input validation and sanitization techniques can help reduce the likelihood of exploitation. For further support and assistance, contact DevNack.com.

See every vulnerability recorded for tensorflow-gpu →

FAQ