[PATCH 0/1] MR705: combase: Omit thread ID from the stub manager ipid for MTA objects.

[Connor McAdams (＠cmcadams)]

Fixes a bug that occurs when:

- `CoInitializeEx(NULL, COINIT_MULTITHREADED);` is called on thread 1
- `CoInitializeEx(NULL, COINIT_MULTITHREADED);` is called on thread 2
- `CoUninitialize()` is called on thread 1.
- `CoInitializeEx(NULL, COINIT_APARTMENTTHREADED);` is called on thread 1.

After this, when attempting to marshal an interface on thread 2, when `ipid_to_ifstub` is called, `apartment_findfromtid()` will find thread 1's STA and not the MTA.

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/705

[PATCH 1/1] combase: Omit thread ID from the stub manager ipid for MTA objects.

[Connor McAdams]

From: Connor McAdams <cmcadams(a)codeweavers.com>

It is possible for a thread that creates an MTA to call
CoUninitialize() and not destroy the MTA if another thread has
entered the MTA in the meantime. If the original thread then creates
an STA, subsequent attempts to find the MTA with 'apartment_findfromtid'
will get the original thread's STA. To avoid this, don't set a TID value
in the stub manager IPID value to indicate that the stub resides in the
MTA.

Signed-off-by: Connor McAdams <cmcadams(a)codeweavers.com>
---
 dlls/combase/stubmanager.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/dlls/combase/stubmanager.c b/dlls/combase/stubmanager.c
index 59782b8c623..5bb87c60c1d 100644
--- a/dlls/combase/stubmanager.c
+++ b/dlls/combase/stubmanager.c
@@ -46,7 +46,7 @@ WINE_DEFAULT_DEBUG_CHANNEL(ole);
 
 /* generates an ipid in the following format (similar to native version):
  * Data1 = apartment-local ipid counter
- * Data2 = apartment creator thread ID
+ * Data2 = apartment creator thread ID, or 0 for an MTA.
  * Data3 = process ID
  * Data4 = random value
  */
@@ -62,7 +62,7 @@ static inline HRESULT generate_ipid(struct stub_manager *m, IPID *ipid)
     }
 
     ipid->Data1 = InterlockedIncrement(&m->apt->ipidc);
-    ipid->Data2 = (USHORT)m->apt->tid;
+    ipid->Data2 = !m->apt->multi_threaded ? (USHORT)m->apt->tid : 0;
     ipid->Data3 = (USHORT)GetCurrentProcessId();
     return S_OK;
 }
@@ -499,6 +499,8 @@ static HRESULT ipid_to_ifstub(const IPID *ipid, struct apartment **stub_apt,
     /* FIXME: hack for IRemUnknown */
     if (ipid->Data2 == 0xffff)
         *stub_apt = apartment_findfromoxid(*(const OXID *)ipid->Data4);
+    else if (!ipid->Data2)
+        *stub_apt = apartment_get_mta();
     else
         *stub_apt = apartment_findfromtid(ipid->Data2);
     if (!*stub_apt)
-- 
GitLab

https://gitlab.winehq.org/wine/wine/-/merge_requests/705

[PATCH 1/1] combase: Omit thread ID from the stub manager ipid for MTA objects.

[Zebediah Figura (she/her)]

On 8/24/22 13:42, Connor McAdams wrote:
> From: Connor McAdams <cmcadams(a)codeweavers.com>
>
> It is possible for a thread that creates an MTA to call
> CoUninitialize() and not destroy the MTA if another thread has
> entered the MTA in the meantime. If the original thread then creates
> an STA, subsequent attempts to find the MTA with 'apartment_findfromtid'
> will get the original thread's STA. To avoid this, don't set a TID value
> in the stub manager IPID value to indicate that the stub resides in the
> MTA.
>
> Signed-off-by: Connor McAdams <cmcadams(a)codeweavers.com>
> ---
>   dlls/combase/stubmanager.c | 6 ++++--
>   1 file changed, 4 insertions(+), 2 deletions(-)

Is it possible to add a test for this?

[Connor McAdams (＠cmcadams)]

On Wed Aug 24 20:19:17 2022 +0000, **** wrote:
> Zebediah Figura (she/her) replied on the mailing list:
> ```
> On 8/24/22 13:42, Connor McAdams wrote:
> > From: Connor McAdams <cmcadams(a)codeweavers.com>
> >
> > It is possible for a thread that creates an MTA to call
> > CoUninitialize() and not destroy the MTA if another thread has
> > entered the MTA in the meantime. If the original thread then creates
> > an STA, subsequent attempts to find the MTA with 'apartment_findfromtid'
> > will get the original thread's STA. To avoid this, don't set a TID value
> > in the stub manager IPID value to indicate that the stub resides in the
> > MTA.
> >
> > Signed-off-by: Connor McAdams <cmcadams(a)codeweavers.com>
> > ---
> >   dlls/combase/stubmanager.c | 6 ++++--
> >   1 file changed, 4 insertions(+), 2 deletions(-)
>
> Is it possible to add a test for this?
> ```
> Sure! For some reason I had only looked at `dlls/combase/tests/` and thought all the tests went there, but looks like I could add one to `dlls/ole32/tests/`. I'll do that tomorrow and send a v2.

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/705#note_6850