CVE-2022-35969: TensorFlow Denial of Service Vulnerability in Conv2DBackpropInput

Background

TensorFlow is an open-source machine learning library widely used for various applications, including deep learning and neural networks. It provides a flexible ecosystem for building and deploying machine learning models. However, a vulnerability has been identified in the Conv2DBackpropInput operation, which can lead to a denial of service if exploited.

Vulnerability Detail

The vulnerability arises from the Conv2DBackpropInput implementation, which requires the input_sizes parameter to be a 4-dimensional tensor. If this requirement is not met, it results in a CHECK failure. An attacker can exploit this flaw by providing invalid input sizes, potentially leading to a denial of service. The vulnerability can be triggered with a specific input that causes the system to fail, impacting the availability of the service.

pip install --upgrade "tensorflow==2.21.0"

Workaround

As a workaround, ensure that the input_sizes parameter passed to the Conv2DBackpropInput operation is always a 4-dimensional tensor. Here is an example of how to correctly define input_sizes:

import tensorflow as tf

strides = [1, 1, 1, 1] padding = 'SAME' use_cudnn_on_gpu = True explicit_paddings = [] data_format = 'NHWC' dilations = [1, 1, 1, 1] input_sizes = tf.constant([1, 64, 64, 2], shape=[4], dtype=tf.int32) # Correctly defined 4D tensor filter = tf.constant(0.159749106, shape=[3, 3, 2, 2], dtype=tf.float32) out_backprop = tf.constant(0, shape=[], dtype=tf.float32) tf.raw_ops.Conv2DBackpropInput(input_sizes=input_sizes, filter=filter, out_backprop=out_backprop, strides=strides, padding=padding, use_cudnn_on_gpu=use_cudnn_on_gpu, explicit_paddings=explicit_paddings, data_format=data_format, dilations=dilations)

Conclusion

The CVE-2022-35969 vulnerability in TensorFlow's Conv2DBackpropInput operation poses a risk of denial of service. Users are advised to update to the latest patched versions of TensorFlow and ensure proper input validation to mitigate this risk.

See every vulnerability recorded for tensorflow →

FAQ