Avoid double-free in content branch of NettyHttpRequest.buildBody #6712
Conversation
This file contains 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
|
How is this issue manifesting itself? Are you seeing warnings etc? |
|
You get this stack trace in logs: basically, |
|
Is this related to #6705? |
|
I found it when debugging #6705 but it's not directly related |
|
I don't understand the CI failures. I can't reproduce them locally, and the github actions logs aren't helpful. Is there a way to see more details, maybe a gradle build scan? |
yawkat
force-pushed
the
mixed-refcnt
branch
2 times, most recently
from
74352ee
to
5fb78d8
Compare
When concatenating `receivedContent` data into a CompositeByteBuf, we call `addComponent` with the content buffers of the individual `receivedContent` items. `addComponent` takes "ownership" of the buffers, so we need to retain them once so that freeing the composite buffer does not free the `receivedContent` items.
graemerocher
approved these changes
Jan 12, 2022
jameskleeh
reviewed
Feb 7, 2022
| @@ -304,7 +304,8 @@ protected Object buildBody() { | |||
| for (ByteBufHolder holder : receivedContent) { | |||
| ByteBuf content = holder.content(); | |||
| if (content != null) { | |||
| byteBufs.addComponent(true, content); | |||
| // need to retain content, because for addComponent "ownership of buffer is transferred to this CompositeByteBuf." | |||
| byteBufs.addComponent(true, content.retain()); | |||
There was a problem hiding this comment.
The content is already retained when the addContent method is called. Can you explain why it needs to be retained again and how retaining the content twice wouldn't lead to a memory leak without an additional release?
There was a problem hiding this comment.
The retain in addContent is on the attribute. It does not affect the buffer itself.
The release of the attribute in receivedContent is done in NettyHttpRequest.release
There was a problem hiding this comment.
How can you be sure that addContent is always called with an attribute? It can be called even if the request is not multipart
There was a problem hiding this comment.
There was a problem hiding this comment.
The other cases are probably HttpData, which is handled by the receivedData branch?
But it should be safe anyway. When a holder is added, it is retained, and it is released when the request is released. For this new retain, the corresponding release is when the composite buffer is released by the caller. So there is a corresponding release for each retain.
yawkat
added a commit
that referenced
this pull request
Feb 9, 2022
This is an attempt at a fix for the flaky FuzzyInputSpec. Can't reproduce that locally, but this code is obviously buggy, so this might help. `body` is a memoized supplier, but `NettyHttpRequest.release` checks whether it's a reference counted object. The supplier is never reference counted, but the object it supplies might be. This patch stores the unwrapped value as a separate field so that it can be released properly. The test for #6712 still passes, so this doesn't seem to regress the issue that the patch that caused this leak fixed.
jameskleeh
added a commit
that referenced
this pull request
Feb 10, 2022
* Release unwrapped body in NettyHttpRequest This is an attempt at a fix for the flaky FuzzyInputSpec. Can't reproduce that locally, but this code is obviously buggy, so this might help. `body` is a memoized supplier, but `NettyHttpRequest.release` checks whether it's a reference counted object. The supplier is never reference counted, but the object it supplies might be. This patch stores the unwrapped value as a separate field so that it can be released properly. The test for #6712 still passes, so this doesn't seem to regress the issue that the patch that caused this leak fixed. * fix setBody * Release buffers in tests Co-authored-by: jameskleeh <james.kleeh@gmail.com>
alvarosanchez
pushed a commit
that referenced
this pull request
Feb 17, 2022
* Release unwrapped body in NettyHttpRequest This is an attempt at a fix for the flaky FuzzyInputSpec. Can't reproduce that locally, but this code is obviously buggy, so this might help. `body` is a memoized supplier, but `NettyHttpRequest.release` checks whether it's a reference counted object. The supplier is never reference counted, but the object it supplies might be. This patch stores the unwrapped value as a separate field so that it can be released properly. The test for #6712 still passes, so this doesn't seem to regress the issue that the patch that caused this leak fixed. * fix setBody * Release buffers in tests Co-authored-by: jameskleeh <james.kleeh@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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 concatenating
receivedContentdata into a CompositeByteBuf, we calladdComponentwith the content buffers of the individualreceivedContentitems.addComponenttakes "ownership" of the buffers, so we need to retain them once so that freeing the composite buffer does not free thereceivedContentitems.