[release/6.0] Fix a memory leak in runtime interop stubs when using an array of structs of types that use old-style managed marshalers #93147
Conversation
…n array of structs of types that use old-style managed marshalers
jkoritzinsky
added
Servicing-consider
|
/backport to release/7.0-staging |
|
/backport to release/8.0 |
|
Started backporting to release/7.0-staging: https://github.com/dotnet/runtime/actions/runs/6435796905 |
|
Started backporting to release/8.0: https://github.com/dotnet/runtime/actions/runs/6435797694 |
jeffschwMSFT
added
Servicing-approved
|
/backport to release/7.0-staging |
|
/backport to release/8.0 |
|
Started backporting to release/7.0-staging: https://github.com/dotnet/runtime/actions/runs/6460257377 |
|
Started backporting to release/8.0: https://github.com/dotnet/runtime/actions/runs/6460258950 |
|
Failures are unrelated (also showing up in release/6.0-staging without this change). |
jkoritzinsky
merged commit 4f5047f
into
dotnet:release/6.0-staging
|
@jkoritzinsky thanks for doing this fix. Can I get a drop of this change to test and see if it fixes the leak I am experiencing? |
ghost
locked as resolved and limited conversation to collaborators
Fixes Issue (Issue was reported by a 1st party over email)
main PR #93089
Description
When passing an array of structs that contains an array field marshalled as a SAFEARRAY or "ByValArray" field to a P/Invoke or a COM interop call, the native array and its contents are not released. This PR changes the marshalling logic such that the native array and contents will be released when they should be.
Customer Impact
Customers without this fix that have interop code with this shape will have unmanaged memory leaks.
Regression
This is a regression that was introduced in .NET 5.
Testing
This fix has been manually validated in main and a unit test has been added to the PR in main. This fix has also manually been validated in this branch with the customer and it fixes their scenario.
Risk
The risk for this PR is low. The logic here will only be hit in these two scenarios, and the expected and documented behavior is that we do not leak this memory. It is very unlikely that customers have taken a dependency on this memory being leaked.