Crash caused by corrupted stack in System.Net.Quic tests

[jkotas]

Crash dump: runfo get-helix-payload -j 5b5b3a98-af71-44f7-96d2-3e4df7dcae00 -w System.Net.Http.Functional.Tests -o c:\helix_payload\System.Net.Http.Functional.Tests

Hit by #62590

Tagging subscribers to this area: @dotnet/ncl
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Crash dump: runfo get-helix-payload -j 5b5b3a98-af71-44f7-96d2-3e4df7dcae00 -w System.Net.Http.Functional.Tests -o c:\helix_payload\System.Net.Http.Functional.Tests

Hit by #62590

Author:	jkotas
Assignees:	-
Labels:	area-System.Net.Http, untriaged
Milestone:	-

[jkotas]

The crash is caused by stack corrupted around the call to ConfigurationLoadCredentialDelegate.

There is one obvious interop problem in mismatched definition of CredentialConfig structure. The unmanaged version of this structure has AllowedCipherSuites field that is missing in the managed definition. This problem alone cannot explain the crash, there are likely more similar interop issues, potentially in interop calls done that happened earlier.

Tagging subscribers to this area: @dotnet/ncl
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Crash dump: runfo get-helix-payload -j 5b5b3a98-af71-44f7-96d2-3e4df7dcae00 -w System.Net.Http.Functional.Tests -o c:\helix_payload\System.Net.Http.Functional.Tests

Hit by #62590

Author:	jkotas
Assignees:	-
Labels:	area-System.Net.Http, untriaged, area-System.Net.Quic
Milestone:	-

[wfurt]

thanks for looking into it @jkotas. AFAIK the AllowedCipherSuites was added later AFAIK e.g. after the initial code was written and failed to notice the difference. There was debate that the low-level binding would move to msquic it self but that did not happened (yet?)

cc: @nibanks @ThadHouse

[thhous-msft]

The AllowedCipherSuites field is only read if QUIC_CREDENTIAL_FLAG_SET_ALLOWED_CIPHER_SUITES is passed as one of the flags. And it doesn't seem like that is happening from the .NET side, so I suspect its unrelated.

[jkotas]

Yes, I agree that the crash is unrelated to the mismatched definition of CredentialConfig. It seems to be regression in the runtime introduced by recent change (#62271)