Fix use after free when advanced indexing tensors with tensors #4559
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
In `THPTensor_(_convertToTensorIndexers)`, a `vector<THPIndexTensor>` is
created by constructing `THPTensor`s from sequences/tensors/etc. Each
`THPIndexTensor` is then freed with the following:
```
for (auto& idx : indexers) {
THIndexTensor_(free)(LIBRARY_STATE idx->cdata);
Py_DECREF(idx);
}
```
This is a problem because `Py_DECREF(idx)` will turn `idx->ob_refcnt` to 0 since this function
created the relevant `THPIndexTensor`s and owns them, causing `THPTensor_(dealloc)` to be
called. `THPTensor_(dealloc)` already has a line that calls
`THIndexTensor_(free)(LIBRARY_STATE idx->cdata)`.
So `THIndexTensor_(free)(LIBRARY_STATE idx->cdata)` gets called twice on the same
`cdata`. After the first call frees `cdata`, the second attempts to access flags/members of `cdata` to
determine if it should free it.
soumith
approved these changes
Jan 9, 2018
apaszke
reviewed
Jan 9, 2018
| // Clean up Indexers | ||
| for (auto& idx : indexers) { | ||
| THIndexTensor_(free)(LIBRARY_STATE idx->cdata); | ||
| Py_DECREF(idx); |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
soumith
pushed a commit
that referenced
this pull request
Feb 7, 2018
In `THPTensor_(_convertToTensorIndexers)`, a `vector<THPIndexTensor>` is
created by constructing `THPTensor`s from sequences/tensors/etc. Each
`THPIndexTensor` is then freed with the following:
```
for (auto& idx : indexers) {
THIndexTensor_(free)(LIBRARY_STATE idx->cdata);
Py_DECREF(idx);
}
```
This is a problem because `Py_DECREF(idx)` will turn `idx->ob_refcnt` to 0 since this function
created the relevant `THPIndexTensor`s and owns them, causing `THPTensor_(dealloc)` to be
called. `THPTensor_(dealloc)` already has a line that calls
`THIndexTensor_(free)(LIBRARY_STATE idx->cdata)`.
So `THIndexTensor_(free)(LIBRARY_STATE idx->cdata)` gets called twice on the same
`cdata`. After the first call frees `cdata`, the second attempts to access flags/members of `cdata` to
determine if it should free it.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When indexing tensors with tensors, like the following code, a use after free occurs. This doesn't actually impact anything in practice (nothing crashes or errors out) but is theoretically undefined behavior.
Advanced indexing on Variables doesn't exhibit this behavior because it's a different codepath.
Sample code:
Explanation
In
THPTensor_(_convertToTensorIndexers), avector<THPIndexTensor>is created by constructingTHPTensors from sequences/tensors/etc. EachTHPIndexTensoris then freed with the following:This is a problem because
Py_DECREF(idx)will turnidx->ob_refcntto 0 since this function created the relevantTHPIndexTensors and owns them, causingTHPTensor_(dealloc)to be called.THPTensor_(dealloc)already has a line that callsTHIndexTensor_(free)(LIBRARY_STATE idx->cdata).So
THIndexTensor_(free)(LIBRARY_STATE idx->cdata)gets called twice on the samecdata. After the first call freescdata, the second attempts to access flags/members ofcdatato determine if it should free it.