CVE-2021-37647: TensorFlow GPU - Null Pointer Dereference in SparseTensorSliceDataset

Background

TensorFlow is an end-to-end open source platform for machine learning. It provides a comprehensive ecosystem for building and deploying machine learning models. The TensorFlow GPU package, available on PyPI, is specifically designed to leverage GPU acceleration for faster training and inference. This vulnerability affects versions 2.3.0 to 2.3.3, 2.4.0 to 2.4.2, and 2.5.0 to 2.5.0 of the TensorFlow GPU package.

Vulnerability Detail

The vulnerability in TensorFlow GPU arises when the tf.raw_ops.SparseTensorSliceDataset function is called without providing valid arguments for a sparse tensor. Specifically, if the indices argument is empty while the values argument is not, the implementation fails to handle this scenario correctly. As a result, a null pointer dereference occurs, leading to a crash or potential code execution. This vulnerability can be exploited by an attacker who can supply malicious input to trigger the vulnerable code path.

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

Workaround

If immediate package update is not possible, a temporary workaround is to avoid calling the tf.raw_ops.SparseTensorSliceDataset function with an empty indices argument while providing a non-empty values argument. Developers can add a check in their code to ensure that both indices and values are either both empty or both non-empty before calling the function. Here's an example of a possible workaround:

import tensorflow as tf

def create_sparse_tensor(indices, values):
    if indices and values:
        dataset = tf.raw_ops.SparseTensorSliceDataset(indices, values)
        # Rest of the code
    else:
        # Handle the case when either indices or values is empty
        # Error handling or alternative logic

Conclusion

The vulnerability in TensorFlow GPU (CVE-2021-37647) allows for a null pointer dereference when certain arguments are not properly validated in the tf.raw_ops.SparseTensorSliceDataset function. This can be exploited by an attacker to cause a crash or potentially execute arbitrary code. To mitigate this vulnerability, users should update their TensorFlow GPU package to version 2.3.4, 2.4.3, 2.5.1, or later. Alternatively, a temporary workaround can be implemented by ensuring that both indices and values are either both empty or both non-empty before calling the vulnerable function. For further assistance or support in vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow-gpu →

FAQ