Re: pthread_mutex_trylock crashes because _get_curthread() returns null
Date: Sun, 12 Jan 2025 11:31:08 UTC
On Sat, Jan 11, 2025 at 07:35:51PM -0800, Yuri wrote:
> I am trying to understand this crash:
>
> 614│ int
> 615│ __Tthr_mutex_trylock(pthread_mutex_t *mutex)
> 616│ {
> 617│ struct pthread *curthread;
> 618│ struct pthread_mutex *m;
> 619│ uint32_t id;
> 620│ int ret, robust;
> 621│
> 622│ ret = check_and_init_mutex(mutex, &m);
> 623│ if (ret != 0)
> 624│ return (ret);
> 625│ curthread = _get_curthread();
> 626│ id = TID(curthread);
> 627│ if (m->m_flags & PMUTEX_FLAG_PRIVATE)
> 628├───────────────> THR_CRITICAL_ENTER(curthread);
>
> Program received signal SIGSEGV, Segmentation fault.
> Address not mapped to object.
> __Tthr_mutex_trylock (mutex=<optimized out>) at
> /disk-samsung/freebsd-src/lib/libthr/thread/thr_mutex.c:628
> 628 THR_CRITICAL_ENTER(curthread);
>
> The crash occurs at the program startup within the pthread_mutex_trylock()
> function.
>
> The immediate cause is that _get_curthread() returned null in curthread.
>
>
> Testcase: the port databases/qdrant at rev.
> e7cee8d22daf5b6360238cad603ca9f96ecd87fd (at version 1.12.5).
>
>
> How can it be that _get_curthread() returns null?
>
> Shouldn't current thread be always defined?
The following patch should help, please check
commit fb77577e7a4995f038a5d28f42d4c3771e536fdb
Author: Konstantin Belousov <kib@FreeBSD.org>
Date: Sun Jan 12 13:28:52 2025 +0200
pthread_mutex_trylock(): init libthr if needed
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
diff --git a/lib/libthr/thread/thr_mutex.c b/lib/libthr/thread/thr_mutex.c
index ca8971cc720a..32bdc4afe65f 100644
--- a/lib/libthr/thread/thr_mutex.c
+++ b/lib/libthr/thread/thr_mutex.c
@@ -619,6 +619,7 @@ __Tthr_mutex_trylock(pthread_mutex_t *mutex)
uint32_t id;
int ret, robust;
+ _thr_check_init();
ret = check_and_init_mutex(mutex, &m);
if (ret != 0)
return (ret);