From owner-freebsd-threads@FreeBSD.ORG  Wed Mar  2 14:56:24 2005
Return-Path: <owner-freebsd-threads@FreeBSD.ORG>
Delivered-To: freebsd-threads@freebsd.org
Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125])
	by hub.freebsd.org (Postfix) with ESMTP id 87EF416A4CE
	for <threads@freebsd.org>; Wed,  2 Mar 2005 14:56:24 +0000 (GMT)
Received: from mail.emict.com (brig.emict.com [212.90.172.42])
	by mx1.FreeBSD.org (Postfix) with ESMTP id 9900143D39
	for <threads@freebsd.org>; Wed,  2 Mar 2005 14:56:20 +0000 (GMT)
	(envelope-from andrit@ukr.net)
Received: from BORJA (unknown [203.199.120.221])
	by mail.emict.com (Postfix) with ESMTP id EBCD733390
	for <threads@freebsd.org>; Wed,  2 Mar 2005 16:56:09 +0200 (EET)
Message-ID: <000e01c51f37$fb3dc980$090210ac@BORJA>
From: "Andriy Tkachuk" <andrit@ukr.net>
To: <threads@freebsd.org>
Date: Wed, 2 Mar 2005 20:26:02 +0530
MIME-Version: 1.0
X-Priority: 3
X-MSMail-Priority: Normal
X-Mailer: Microsoft Outlook Express 6.00.2800.1158
X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2800.1165
Content-Type: text/plain;
	charset="koi8-r"
Content-Transfer-Encoding: quoted-printable
X-Content-Filtered-By: Mailman/MimeDel 2.1.1
Subject: patch for threads/76690 - critical - fork hang in child for -lc_r
X-BeenThere: freebsd-threads@freebsd.org
X-Mailman-Version: 2.1.1
Precedence: list
List-Id: Threading on FreeBSD <freebsd-threads.freebsd.org>
List-Unsubscribe: <http://lists.freebsd.org/mailman/listinfo/freebsd-threads>,
	<mailto:freebsd-threads-request@freebsd.org?subject=unsubscribe>
List-Archive: <http://lists.freebsd.org/pipermail/freebsd-threads>
List-Post: <mailto:freebsd-threads@freebsd.org>
List-Help: <mailto:freebsd-threads-request@freebsd.org?subject=help>
List-Subscribe: <http://lists.freebsd.org/mailman/listinfo/freebsd-threads>,
	<mailto:freebsd-threads-request@freebsd.org?subject=subscribe>
X-List-Received-Date: Wed, 02 Mar 2005 14:56:24 -0000

Hi folks.

I spent some time on the problem in $subj and found some
solution that seems to be working but i'm not sure about it's
architectural correctness because libc was changed little bit ;)

here it is:

# diff -u -r lib lib.patched
diff -u -r lib/libc/stdlib/malloc.c lib.patched/libc/stdlib/malloc.c
--- lib/libc/stdlib/malloc.c    Sun Feb 27 22:46:16 2005
+++ lib.patched/libc/stdlib/malloc.c    Wed Mar  2 19:55:24 2005
@@ -74,7 +74,7 @@
      */
 #   include "libc_private.h"
 #   include "spinlock.h"
-    static spinlock_t thread_lock      =3D _SPINLOCK_INITIALIZER;
+    spinlock_t thread_lock     =3D _SPINLOCK_INITIALIZER;
     spinlock_t *__malloc_lock          =3D &thread_lock;
 #   define _MALLOC_LOCK()              if (__isthreaded) =
_SPINLOCK(&thread_lock);
 #   define _MALLOC_UNLOCK()            if (__isthreaded) =
_SPINUNLOCK(&thread_lock);
diff -u -r lib/libc_r/uthread/uthread_fork.c =
lib.patched/libc_r/uthread/uthread_fork.c
--- lib/libc_r/uthread/uthread_fork.c   Fri Dec 10 09:06:45 2004
+++ lib.patched/libc_r/uthread/uthread_fork.c   Wed Mar  2 20:10:49 2005
@@ -61,6 +61,8 @@
        _thread_kern_sig_defer();

        _pthread_mutex_lock(&_atfork_mutex);
+       extern spinlock_t thread_lock;
+       _SPINLOCK(&thread_lock);

        /* Run down atfork prepare handlers. */
        TAILQ_FOREACH_REVERSE(af, &_atfork_list, atfork_head, qe) {
@@ -70,6 +72,8 @@

        /* Fork a new process: */
        if ((ret =3D __sys_fork()) !=3D 0) {
+               _SPINUNLOCK(&thread_lock);
+
                /* Run down atfork parent handlers. */
                TAILQ_FOREACH(af, &_atfork_list, qe) {
                        if (af->parent !=3D NULL)
@@ -78,6 +82,8 @@
                _pthread_mutex_unlock(&_atfork_mutex);

        } else {
+               _SPINUNLOCK(&thread_lock);
+
                /* Close the pthread kernel pipe: */
                __sys_close(_thread_kern_pipe[0]);
                __sys_close(_thread_kern_pipe[1]);


The main problem with it is that after fork child inherits
from the parent inconsist heap. So the obvious solution is
to synchronyze the fork and heap manipulations. I didn't find
the easyest way how to get the lock variable from malloc lib
as to open this variable deleting static spec.

So what do you think guys about this patch?

Thanks,
  Andriy Tkachuk.