From: Dmitry Timoshkov dmitry@baikal.ru
When an application has both apartment-threaded and multi-threaded apartments then apartment_get_current_or_mta() may return wrong one, and later the oxid check will fail. Probably it's better try to find correct apartment from the start.
Signed-off-by: Dmitry Timoshkov dmitry@baikal.ru --- dlls/combase/rpc.c | 15 ++------------- 1 file changed, 2 insertions(+), 13 deletions(-)
diff --git a/dlls/combase/rpc.c b/dlls/combase/rpc.c index c51b59de4bf..160c5c579b2 100644 --- a/dlls/combase/rpc.c +++ b/dlls/combase/rpc.c @@ -1344,15 +1344,6 @@ static DWORD WINAPI rpc_sendreceive_thread(LPVOID param) return 0; }
-static inline HRESULT ClientRpcChannelBuffer_IsCorrectApartment(ClientRpcChannelBuffer *This, const struct apartment *apt) -{ - if (!apt) - return S_FALSE; - if (This->oxid != apartment_getoxid(apt)) - return S_FALSE; - return S_OK; -} - static HRESULT WINAPI ClientRpcChannelBuffer_SendReceive(LPRPCCHANNELBUFFER iface, RPCOLEMESSAGE *olemsg, ULONG *pstatus) { ClientRpcChannelBuffer *This = (ClientRpcChannelBuffer *)iface; @@ -1365,17 +1356,15 @@ static HRESULT WINAPI ClientRpcChannelBuffer_SendReceive(LPRPCCHANNELBUFFER ifac ORPC_EXTENT_ARRAY orpc_ext_array; WIRE_ORPC_EXTENT *first_wire_orpc_extent = NULL; HRESULT hrFault = S_OK; - struct apartment *apt = apartment_get_current_or_mta(); + struct apartment *apt = apartment_findfromoxid(This->oxid); struct tlsdata *tlsdata;
TRACE("%p, iMethod %ld\n", olemsg, olemsg->iMethod);
- hr = ClientRpcChannelBuffer_IsCorrectApartment(This, apt); - if (hr != S_OK) + if (!apt) { ERR("called from wrong apartment, should have been 0x%s\n", wine_dbgstr_longlong(This->oxid)); - if (apt) apartment_release(apt); return RPC_E_WRONG_THREAD; }
Could we come up with a test for this?
On Mon Apr 28 17:42:29 2025 +0000, Huw Davies wrote:
Could we come up with a test for this?
I tried, however I couldn't reproduce this in a test.
Observed behaviour is similar to what b4da14035741f287130002b512c7d0e9117da80a has fixed, and in that case creating a test also proved to be problematic.
This merge request was approved by Huw Davies.
On Mon Apr 28 17:42:29 2025 +0000, Dmitry Timoshkov wrote:
I tried, however I couldn't reproduce this in a test. Observed behaviour is similar to what b4da14035741f287130002b512c7d0e9117da80a has fixed, and in that case creating a test also proved to be problematic.
Right, ok. Makes sense.