win32: use _endthreadex to terminate threads, not ExitThread

Commit Message

Seija Kijin Dec. 22, 2022, 8:59 p.m. UTC

From: Seija Kijin <[e-mail]>

This is a pretty serious bug actually:
Because we use the C runtime and
use _beginthreadex to create pthreads,
pthread_exit MUST use _endthreadex.

Otherwise, according to Microsoft:
"Failure to do so results in small
memory leaks when the thread
calls ExitThread."

Simply put, this is not the same as ExitThread.

Signed-off-by: Seija Kijin <[e-mail]>
---
    win32: use _endthreadex to terminate threads, not ExitThread
    
    This is a pretty serious bug actually: Because we use the C runtime and
    use _beginthread to create pthreads, pthread_exit MUST use _endthread.
    
    Otherwise, according to Microsoft: "Failure to do so results in small
    memory leaks when the thread calls ExitThread."
    
    Simply put, this is not the same as ExitThread.
    
    Signed-off-by: Seija Kijin [e-mail]

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-git-1414%2FAtariDreams%2Fsevere-bug-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-git-1414/AtariDreams/severe-bug-v1
Pull-Request: https://github.com/git/git/pull/1414

 compat/win32/pthread.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)


base-commit: 7c2ef319c52c4997256f5807564523dfd4acdfc7

Comments

Johannes Sixt Dec. 24, 2022, 8:01 a.m. UTC | #1

Am 22.12.22 um 21:59 schrieb Rose via GitGitGadget:
> From: Seija Kijin <[e-mail]>
> 
> This is a pretty serious bug actually:
> Because we use the C runtime and
> use _beginthreadex to create pthreads,
> pthread_exit MUST use _endthreadex.
> 
> Otherwise, according to Microsoft:
> "Failure to do so results in small
> memory leaks when the thread
> calls ExitThread."
> 
> Simply put, this is not the same as ExitThread.
> 
> Signed-off-by: Seija Kijin <[e-mail]>
> ---

>  compat/win32/pthread.h | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/compat/win32/pthread.h b/compat/win32/pthread.h
> index 737983d00ba..cc3221cb2c8 100644
> --- a/compat/win32/pthread.h
> +++ b/compat/win32/pthread.h
> @@ -66,7 +66,7 @@ pthread_t pthread_self(void);
>  
>  static inline void NORETURN pthread_exit(void *ret)
>  {
> -	ExitThread((DWORD)(intptr_t)ret);
> +	_endthreadex((unsigned)(uintptr_t)ret);
>  }
>  
>  typedef DWORD pthread_key_t;

Nice find! FWIW, this passes the test suite on Windows.

The patch text is:

Acked-by: Johannes Sixt <[e-mail]>

The commit message is highly exaggerated, though. The bug is by no means
serious. After all, we've been living with it for a decade. Notice that
pthread_exit() is only called when a thread runs into die(). Even though
we have a small memory leak, it does not occur in a loop because, after
one thread dies, we do not tend to keep starting many, many more that
all die.

-- Hannes

Patch

diff --git a/compat/win32/pthread.h b/compat/win32/pthread.h
index 737983d00ba..cc3221cb2c8 100644
--- a/compat/win32/pthread.h
+++ b/compat/win32/pthread.h
@@ -66,7 +66,7 @@  pthread_t pthread_self(void);
 
 static inline void NORETURN pthread_exit(void *ret)
 {
-	ExitThread((DWORD)(intptr_t)ret);
+	_endthreadex((unsigned)(uintptr_t)ret);
 }
 
 typedef DWORD pthread_key_t;