Memory corruption in Tensorflow
tensorflow
Affected versions
| Product | Affected | Fixed in |
|---|---|---|
| tensorflow PyPI |
|
|
| tensorflow-cpu PyPI |
|
tensorflow
| Product | Affected | Fixed in |
|---|---|---|
| tensorflow PyPI |
|
|
| tensorflow-cpu PyPI |
|
≥ 2.3.0 · < 2.3.1< 2.2.1≥ 2.3.0rc0 · < 2.3.1
|
| tensorflow-gpu PyPI |
|
|
In Tensorflow before versions 2.2.1 and 2.3.1, the implementation of `dlpack.to_dlpack` can be made to use uninitialized memory resulting in further memory corruption. This is because the pybind11 glue code assumes that the argument is a tensor. However, there is nothing stopping users from passing in a Python object instead of a tensor. The uninitialized memory address is due to a `reinterpret_cast` Since the `PyObject` is a Python object, not a TensorFlow Tensor, the cast to `EagerTensor` fails. The issue is patched in commit 22e07fb204386768e5bcbea563641ea11f96ceb8 and is released in TensorFlow versions 2.2.1, or 2.3.1.
From the CVE record · GitHub_M
TensorFlow's dlpack.to_dlpack can use uninitialized memory when passed a Python object instead of a tensor, causing further memory corruption. Exploitation requires the ability to pass such an object to the function; the input does not specify network access or authentication requirements.
Engineers using PyPI tensorflow, tensorflow-cpu, or tensorflow-gpu versions before 2.2.1 or versions from 2.3.0rc0 before 2.3.1 should check their installations.