`CHECK`-fails due to attempting to build a reference tensor in Tensorflow
tensorflow
Affected versions
| Product | Affected | Fixed in |
|---|---|---|
| tensorflow PyPI |
|
|
| tensorflow-cpu PyPI |
|
tensorflow
| Product | Affected | Fixed in |
|---|---|---|
| tensorflow PyPI |
|
|
| tensorflow-cpu PyPI |
|
≥ 2.6.0 · < 2.6.3≥ 2.7.0 · < 2.7.1
|
| tensorflow-gpu PyPI |
|
|
Tensorflow is an Open Source Machine Learning Framework. A malicious user can cause a denial of service by altering a `SavedModel` such that Grappler optimizer would attempt to build a tensor using a reference `dtype`. This would result in a crash due to a `CHECK`-fail in the `Tensor` constructor as reference types are not allowed. The fix will be included in TensorFlow 2.8.0. We will also cherrypick this commit on TensorFlow 2.7.1, TensorFlow 2.6.3, and TensorFlow 2.5.3, as these are also affected and still in supported range.
From the CVE record · GitHub_M
A malicious user can alter a SavedModel so the Grappler optimizer attempts to build a tensor with a reference dtype, causing a crash. The crash occurs when the altered SavedModel is processed; the input does not specify network access or authentication requirements.
Check installations of the PyPI packages tensorflow, tensorflow-cpu, and tensorflow-gpu below 2.5.3, from 2.6.0 to below 2.6.3, or from 2.7.0 to below 2.7.1.