CVE-2022-41885: TensorFlow FusedResizeAndPadConv2D Overflow Vulnerability

Background

TensorFlow is an open-source machine learning library developed by Google, widely used for building and training machine learning models. It provides a flexible ecosystem of tools, libraries, and community resources that allows researchers and developers to create and deploy ML-powered applications. The library includes various operations for manipulating tensors, including the FusedResizeAndPadConv2D operation, which is designed to resize and pad convolutional layers efficiently. However, a vulnerability has been identified in this operation that can lead to overflow issues when handling large tensor shapes.

Vulnerability Detail

The vulnerability arises in the FusedResizeAndPadConv2D operation when it is provided with a large tensor shape, leading to an overflow condition. This can result in unexpected behavior, including potential crashes or incorrect computations. The issue is triggered when the input tensor size exceeds certain limits, causing the operation to fail or produce erroneous results. This vulnerability can be exploited by attackers to manipulate the input tensor dimensions, potentially leading to denial of service or other unintended consequences.

pip install --upgrade "tensorflow==2.21.0"

Workaround

As a workaround, users can avoid using excessively large tensor shapes when calling the FusedResizeAndPadConv2D operation. Here is an example of how to call the operation safely with smaller tensor sizes:

import tensorflow as tf

mode = "REFLECT"
strides = [1, 1, 1, 1]
padding = "SAME"
resize_align_corners = False
input = tf.constant(147, shape=[3,3,1,1], dtype=tf.float16)
size = tf.constant([100, 100], shape=[2], dtype=tf.int32)  # Use smaller size
paddings = tf.constant([3,4], shape=[2], dtype=tf.int32)
filter = tf.constant(123, shape=[1,3,4,1], dtype=tf.float16)
tf.raw_ops.FusedResizeAndPadConv2D(input=input, size=size, paddings=paddings, filter=filter, mode=mode, strides=strides, padding=padding, resize_align_corners=resize_align_corners)

Conclusion

The CVE-2022-41885 vulnerability in TensorFlow's FusedResizeAndPadConv2D operation poses a risk when handling large tensor shapes, leading to potential overflow issues. Users are advised to update to the latest patched versions of TensorFlow to mitigate this risk. Additionally, using smaller tensor sizes can serve as a temporary workaround.

See every vulnerability recorded for tensorflow →

FAQ