TensorFlow vulnerable to `CHECK` failure in `TensorListReserve` via missing validation
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
In
core/kernels/list_kernels.cc's TensorListReserve
,num_elements
is assumed to be a tensor of size 1. When anum_elements
of more than 1 element is provided, thentf.raw_ops.TensorListReserve
fails theCHECK_EQ
inCheckIsAlignedAndSingleElement
.Patches
We have patched the issue in GitHub commit b5f6fbfba76576202b72119897561e3bd4f179c7.
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 Kang Hong Jin from Singapore Management University.
References