Win32 SD_RECEIVE has the following effects on a TCP or UDP socket:
* further calls to recv() et al. fail with ESHUTDOWN. However, Linux and Mac instead return 0.
* currently pending calls to recv() are not affected. However, Linux and Mac cause them to return 0.
* calls to WSAPoll() are unaffected. However, Linux and Mac instead signal both POLLHUP and POLLIN.
* for TCP connections, further data sent by the peer causes the connection to be reset. Mac and BSD will silently drop data, and Linux simply queues it forever.
In short, SHUT_RD does nothing that we want it to do on Linux or Mac (and probably also BSD, though I am unable to test this).
Most of this impedance mismatch can be worked around in Wine, except for WSAPoll(). Since poll() will always return POLLHUP, we can't safely ask for POLLOUT, which we may need to do. Hence, since we aren't getting any benefit from SHUT_RD, don't use it.
Note that this doesn't actually fix any tests, since we have some rather bizarre logic to defer shutdown anyway (for an asynchronous socket, we don't call shutdown(SHUT_RD) until the peer sends data; this is probably an unintended bug, but "fixing" it ends up causing test failures, which is the main impetus for this patch.)
Signed-off-by: Zebediah Figura z.figura12@gmail.com --- dlls/ws2_32/socket.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/dlls/ws2_32/socket.c b/dlls/ws2_32/socket.c index 889f0327baf..aad92d5d4df 100644 --- a/dlls/ws2_32/socket.c +++ b/dlls/ws2_32/socket.c @@ -2290,7 +2290,7 @@ static NTSTATUS WS2_async_shutdown( void *user, IO_STATUS_BLOCK *iosb, NTSTATUS
switch ( wsa->type ) { - case ASYNC_TYPE_READ: err = shutdown( fd, 0 ); break; + case ASYNC_TYPE_READ: break; case ASYNC_TYPE_WRITE: err = shutdown( fd, 1 ); break; } status = err ? wsaErrStatus() : STATUS_SUCCESS; @@ -5550,7 +5550,7 @@ int WINAPI WS_shutdown(SOCKET s, int how) } else /* non-overlapped mode */ { - if ( shutdown( fd, how ) ) + if (how != SD_RECEIVE && shutdown( fd, SHUT_WR )) { err = wsaErrno(); goto error;