[Bug 55203] New: ws2_32:sock - test_AcceptEx() sometimes gets WSAEWOULDBLOCK and times out on Windows 10

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=55203

Bug ID: 55203
           Summary: ws2_32:sock - test_AcceptEx() sometimes gets
                    WSAEWOULDBLOCK and times out on Windows 10
           Product: Wine
           Version: unspecified
          Hardware: x86-64
                OS: Windows
            Status: NEW
          Severity: normal
          Priority: P2
         Component: winsock
          Assignee: [e-mail]
          Reporter: [e-mail]

ws2_32:sock - test_AcceptEx() sometimes gets WSAEWOULDBLOCK and times out on
Windows 10:

sock.c:8704: Test failed: failed to accept, error 10035
sock.c:8716: Test failed: got 0xc0000023
sock.c:8751: Test failed: got 00000000
sock.c:8765: Test failed: AcceptEx on too small remote address size returned 0
+ errno 10022
sock.c:8785: Test failed: AcceptEx returned 0 + errno 10022
sock.c:8805: Test failed: got error 10056
ws2_32:sock:13d4 done (258) in 120s 488B

See https://test.winehq.org/data/patterns.html#ws2_32:sock

Where:
  10022 == WSAEINVAL
  10035 == WSAEWOULDBLOCK
  10056 == WSAEISCONN
  0xC0000023 == STATUS_BUFFER_TOO_SMALL

This happened 6 times in the past 8 months, all on Windows 10:
* 2023-02-07 win22H2_fgtb-w10pro64-64
* 2023-03-23 win21H1_newtb-w10pro64-zh-CN-64
* 2023-03-31 win2004_newtb-w10pro64v2004-64
* 2023-05-30 win1909_newtb-w1064v1909-64
* 2023-06-23 win21H1_newtb-w10pro64-ja-64
* 2023-07-06 win21H1_newtb-w10pro64-de-64

[Bug 55203] ws2_32:sock - test_AcceptEx() sometimes gets WSAEWOULDBLOCK and times out on Windows 10

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=55203

François Gouget  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |source, testcase

[Bug 55203] ws2_32:sock - test_AcceptEx() sometimes gets WSAEWOULDBLOCK and times out on Windows 10

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=55203

Zeb Figura  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #1 from Zeb Figura  ---
This is very strange; I can't figure out any reasonable reason for this to
fail.

The test queues an asynchronous accept with an incorrect size. On Windows this
results in the accept being successfully queued. When a connection request
comes in the socket will try to fill that async, realize that the size is too
small, and instead return failure from the async, with the connection request
remaining in the queue for the next accept.

What's odd is that that "next" accept—the one we do just to prove that the
connection is still queued—fails to complete synchronously. This doesn't make
any sense. There has to be a connection request queued, otherwise the invalid
accept wouldn't return. Why would the valid accept fail to complete
synchronously?

I thought it might be because the invalid accept doesn't actually wait for a
valid connection to come in, but a quick test shows that's not the case.