]> git.ipfire.org Git - thirdparty/postgresql.git/commitdiff
Fix ancient bug in ecpg's pthread_once() emulation for Windows.
authorTom Lane <tgl@sss.pgh.pa.us>
Sat, 24 Oct 2020 17:12:08 +0000 (13:12 -0400)
committerTom Lane <tgl@sss.pgh.pa.us>
Sat, 24 Oct 2020 17:12:31 +0000 (13:12 -0400)
We must not set the "done" flag until after we've executed the
initialization function.  Otherwise, other threads can fall through
the initial unlocked test before initialization is really complete.

This has been seen to cause rare failures of ecpg's thread/descriptor
test, and it could presumably cause other sorts of misbehavior in
threaded ECPG-using applications, since ecpglib relies on
pthread_once() in several places.

Diagnosis and patch by me, based on investigation by Alexander Lakhin.
Back-patch to all supported branches (the bug dates to 2007).

Discussion: https://postgr.es/m/16685-d6cd241872c101d3@postgresql.org

src/interfaces/ecpg/ecpglib/misc.c

index be9cac6e7b4a46e92331cf7d8bdc507acb93ef17..85a308f537954e4322fae7de72b1bc0675a08c08 100644 (file)
@@ -475,8 +475,8 @@ win32_pthread_once(volatile pthread_once_t *once, void (*fn) (void))
                pthread_mutex_lock(&win32_pthread_once_lock);
                if (!*once)
                {
-                       *once = true;
                        fn();
+                       *once = true;
                }
                pthread_mutex_unlock(&win32_pthread_once_lock);
        }