CVE-2021-29521: TensorFlow GPU - Segmentation Fault in SparseCountSparseOutput

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 GPU is a variant of TensorFlow that leverages GPU acceleration for faster computation. It is commonly used in scenarios where large-scale machine learning tasks require high-performance computing capabilities.

Vulnerability Detail

The vulnerability (CVE-2021-29521) in TensorFlow GPU arises when specifying a negative dense shape in the tf.raw_ops.SparseCountSparseOutput function. This leads to a segmentation fault being thrown from the standard library due to broken invariants in the implementation. The issue stems from the assumption made in the implementation that the first element of the dense shape is always positive, which is used to initialize a BatchedMap<T> data structure. If the shape tensor has more than one element, num_batches is the first value in shape. By ensuring that the dense_shape argument is a valid tensor shape with all non-negative elements, this vulnerability can be mitigated. The fix for this issue will be included in TensorFlow 2.5.0, and it has also been backported to TensorFlow 2.4.2 and TensorFlow 2.3.3.

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

Workaround

As a workaround, users can validate the dense_shape argument before passing it to the tf.raw_ops.SparseCountSparseOutput function. This can be done by checking if all elements of the dense_shape tensor are non-negative. If any negative element is found, it should be replaced with a valid non-negative value. Here is an example code snippet in Python:

import tensorflow as tf

def validate_dense_shape(dense_shape):
    return tf.where(dense_shape < 0, 0, dense_shape)

# Usage
input_dense_shape = tf.constant([-1, 10, 20])
valid_dense_shape = validate_dense_shape(input_dense_shape)
output = tf.raw_ops.SparseCountSparseOutput(dense_shape=valid_dense_shape, ...)

Conclusion

In conclusion, the vulnerability (CVE-2021-29521) in TensorFlow GPU can result in a segmentation fault when specifying a negative dense shape. It is crucial to update TensorFlow GPU to the recommended versions or apply the provided patches to mitigate this vulnerability. Additionally, validating the dense_shape argument can serve as a temporary workaround. For further assistance or support in vulnerability remediation, please reach out to DevNack.com.

See every vulnerability recorded for tensorflow-gpu →

FAQ