CVE-2021-37654: TensorFlow GPU - Out-of-Bounds Read Vulnerability

Sponsored Habilume Habit tracker with a companion that grows with you Complete habits, earn Lume Points, and watch your Lume grow as you stay consistent. App Store Google Play

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 widely used in various industries and research domains for training and inference of deep learning models.

Vulnerability Detail

The vulnerability (CVE-2021-37654) in TensorFlow GPU allows an attacker to trigger a crash or read data from outside the bounds of heap allocated data. The issue arises from a lack of validation in the tf.raw_ops.ResourceGather API, specifically in the batch_dims parameter. By supplying a batch_dims value greater than or equal to the rank of the input tensor, an attacker can cause the program to read data beyond the allocated buffer, leading to potential information disclosure or a crash. This vulnerability affects debug builds of TensorFlow GPU and release builds with the tf.raw_ops.ResourceGather API.

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

Workaround

As a workaround, users can avoid using the tf.raw_ops.ResourceGather API with a batch_dims value greater than or equal to the rank of the input tensor. It is recommended to review the codebase and ensure that the batch_dims parameter is properly validated before invoking the API. Here is an example of how to validate the batch_dims parameter:

import tensorflow as tf

def resource_gather_with_validation(tensor, indices, batch_dims):
    if batch_dims >= tf.rank(tensor):
        raise ValueError('Invalid batch_dims value')
    return tf.raw_ops.ResourceGather(tensor=tensor, indices=indices, batch_dims=batch_dims)

By implementing this validation check, the vulnerability can be mitigated.

Conclusion

The vulnerability (CVE-2021-37654) in TensorFlow GPU exposes a potential risk of crashing the program or reading data from outside the bounds of heap allocated data. The TensorFlow development team has addressed this issue and will release patched versions to mitigate the vulnerability. Users are advised to update their TensorFlow GPU installations to the patched versions once they are available. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow-gpu →

FAQ