TensorFlow vulnerable to `CHECK` failure in `AvgPoolOp`
Moderate severity
GitHub Reviewed
Published
Sep 15, 2022
in
tensorflow/tensorflow
•
Updated Jan 28, 2023
Description
Published by the National Vulnerability Database
Sep 16, 2022
Published to the GitHub Advisory Database
Sep 16, 2022
Reviewed
Sep 16, 2022
Last updated
Jan 28, 2023
Impact
The
AvgPoolOp
function takes an argumentksize
that must be positive but is not checked. A negativeksize
can trigger aCHECK
failure and crash the program.Patches
We have patched the issue in GitHub commit 3a6ac52664c6c095aa2b114e742b0aa17fdce78f.
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 Jingyi Shi.
References