CVE-2021-37657: TensorFlow GPU - Null Pointer Dereference in MatrixDiagV* Operations

Background

TensorFlow is an end-to-end open source platform for machine learning. It provides a comprehensive ecosystem of tools, libraries, and resources for building and deploying machine learning models. The TensorFlow GPU package, available on PyPI, is specifically designed to leverage the power of GPUs for accelerated machine learning computations.

Vulnerability Detail

The vulnerability (CVE-2021-37657) in TensorFlow GPU arises from incomplete validation in the implementation of operations of type tf.raw_ops.MatrixDiagV*. An attacker can exploit this vulnerability by binding a reference to a null pointer in any of these operations. The issue stems from inadequate checks on the number of elements in the tensor k. If k is an empty tensor, code that accesses the first element of the tensor will result in undefined behavior. This vulnerability can lead to crashes, memory corruption, or other unpredictable outcomes.

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

Workaround

As a workaround, users can avoid passing an empty tensor as the input to the affected operations. By ensuring that the tensor k is either a scalar or a non-empty vector, the vulnerability can be mitigated. Here is an example of how to check and handle the tensor k before invoking the MatrixDiagV* operations:

import tensorflow as tf

# Check if k is empty
if tf.size(k) == 0:
    # Handle empty tensor case
    print('k is empty')
else:
    # Invoke the MatrixDiagV* operations
    result = tf.raw_ops.MatrixDiagV*(input_tensor, k)
    print(result)

Conclusion

The vulnerability (CVE-2021-37657) in TensorFlow GPU, which allows for a null pointer dereference in MatrixDiagV* operations, has been addressed by the TensorFlow development team. Users are advised to update their TensorFlow GPU package to one of the patched versions (2.6.0, 2.5.1, 2.4.3, or 2.3.4) to mitigate the vulnerability. A workaround is also available by avoiding the use of empty tensors as input to the affected operations. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow-gpu →

FAQ