diff --git a/lib/libc/gen/Symbol.map b/lib/libc/gen/Symbol.map
index 6e8bfce..2e64313 100644
--- a/lib/libc/gen/Symbol.map
+++ b/lib/libc/gen/Symbol.map
@@ -432,6 +432,8 @@ FBSDprivate_1.0 {
 	_spinlock;
 	_spinlock_debug;
 	_spinunlock;
+	_rtld_atfork_pre;
+	_rtld_atfork_post;
 	_rtld_error;		/* for private use */
 	_rtld_thread_init;	/* for private use */
 	_err;
diff --git a/lib/libc/gen/dlfcn.c b/lib/libc/gen/dlfcn.c
index 39ba141..eac3f1a 100644
--- a/lib/libc/gen/dlfcn.c
+++ b/lib/libc/gen/dlfcn.c
@@ -137,3 +137,15 @@ dl_iterate_phdr(int (*callback)(struct dl_phdr_info *, size_t, void *),
 	_rtld_error(sorry);
 	return 0;
 }
+
+#pragma weak _rtld_atfork_pre
+void
+_rtld_atfork_pre(int *locks)
+{
+}
+
+#pragma weak _rtld_atfork_post
+void
+_rtld_atfork_post(int *locks)
+{
+}
diff --git a/lib/libc/stdlib/malloc.c b/lib/libc/stdlib/malloc.c
index 55bc413..b407cba 100644
--- a/lib/libc/stdlib/malloc.c
+++ b/lib/libc/stdlib/malloc.c
@@ -3114,6 +3114,8 @@ arena_lock_balance(arena_t *arena)
 {
 	unsigned contention;
 
+	if (narenas > 1)
+		malloc_spin_lock(&arenas_lock);
 	contention = malloc_spin_lock(&arena->lock);
 	if (narenas > 1) {
 		/*
@@ -3126,6 +3128,7 @@ arena_lock_balance(arena_t *arena)
 		    + (uint64_t)contention) >> BALANCE_ALPHA_INV_2POW;
 		if (arena->contention >= opt_balance_threshold)
 			arena_lock_balance_hard(arena);
+		malloc_spin_unlock(&arenas_lock);
 	}
 }
 
@@ -3142,12 +3145,10 @@ arena_lock_balance_hard(arena_t *arena)
 	if (arenas[ind] != NULL)
 		arenas_map = arenas[ind];
 	else {
-		malloc_spin_lock(&arenas_lock);
 		if (arenas[ind] != NULL)
 			arenas_map = arenas[ind];
 		else
 			arenas_map = arenas_extend(ind);
-		malloc_spin_unlock(&arenas_lock);
 	}
 }
 #endif
diff --git a/lib/libthr/thread/thr_fork.c b/lib/libthr/thread/thr_fork.c
index 713e0b5..4ddd93a 100644
--- a/lib/libthr/thread/thr_fork.c
+++ b/lib/libthr/thread/thr_fork.c
@@ -67,6 +67,7 @@
 #include "un-namespace.h"
 
 #include "libc_private.h"
+#include "rtld_lock.h"
 #include "thr_private.h"
 
 __weak_reference(_pthread_atfork, pthread_atfork);
@@ -105,6 +106,7 @@ _fork(void)
 	pid_t ret;
 	int errsave;
 	int unlock_malloc;
+	int rtld_locks[16];
 
 	if (!_thr_is_inited())
 		return (__sys_fork());
@@ -126,7 +128,8 @@ _fork(void)
 	 */
 	if (_thr_isthreaded() != 0) {
 		unlock_malloc = 1;
-		_malloc_prefork();
+		_malloc_prefork();	
+		_rtld_atfork_pre(rtld_locks);
 	} else {
 		unlock_malloc = 0;
 	}
@@ -167,6 +170,11 @@ _fork(void)
 		/* Ready to continue, unblock signals. */ 
 		_thr_signal_unblock(curthread);
 
+		if (unlock_malloc) {
+			_rtld_atfork_post(rtld_locks);
+			_malloc_postfork();
+		}
+
 		/* Run down atfork child handlers. */
 		TAILQ_FOREACH(af, &_thr_atfork_list, qe) {
 			if (af->child != NULL)
@@ -179,8 +187,10 @@ _fork(void)
 		/* Ready to continue, unblock signals. */ 
 		_thr_signal_unblock(curthread);
 
-		if (unlock_malloc)
+		if (unlock_malloc) {
+			_rtld_atfork_post(rtld_locks);
 			_malloc_postfork();
+		}
 
 		/* Run down atfork parent handlers. */
 		TAILQ_FOREACH(af, &_thr_atfork_list, qe) {
diff --git a/lib/libthr/thread/thr_rtld.c b/lib/libthr/thread/thr_rtld.c
index a8cd61d..9101cc1 100644
--- a/lib/libthr/thread/thr_rtld.c
+++ b/lib/libthr/thread/thr_rtld.c
@@ -181,6 +181,7 @@ _thr_rtld_init(void)
 	struct RtldLockInfo	li;
 	struct pthread		*curthread;
 	long dummy;
+	void *dummy1;
 
 	curthread = _get_curthread();
 
@@ -190,6 +191,10 @@ _thr_rtld_init(void)
 	/* force to resolve errno() PLT */
 	__error();
 
+	/* force to resolve free() PLT */
+	dummy1 = malloc(1);
+	free(dummy1);
+
 	li.lock_create  = _thr_rtld_lock_create;
 	li.lock_destroy = _thr_rtld_lock_destroy;
 	li.rlock_acquire = _thr_rtld_rlock_acquire;
@@ -203,6 +208,7 @@ _thr_rtld_init(void)
 	_thr_signal_block(curthread);
 	_rtld_thread_init(&li);
 	_thr_signal_unblock(curthread);
+
 }
 
 void
diff --git a/libexec/rtld-elf/Symbol.map b/libexec/rtld-elf/Symbol.map
index 325362c..20f8d60 100644
--- a/libexec/rtld-elf/Symbol.map
+++ b/libexec/rtld-elf/Symbol.map
@@ -21,4 +21,6 @@ FBSDprivate_1.0 {
     _rtld_thread_init;
     _rtld_allocate_tls;
     _rtld_free_tls;
+    _rtld_atfork_pre;
+    _rtld_atfork_post;
 };
diff --git a/libexec/rtld-elf/rtld.c b/libexec/rtld-elf/rtld.c
index 144ea6c..f67093a 100644
--- a/libexec/rtld-elf/rtld.c
+++ b/libexec/rtld-elf/rtld.c
@@ -206,6 +206,8 @@ static func_ptr_type exports[] = {
     (func_ptr_type) &_rtld_allocate_tls,
     (func_ptr_type) &_rtld_free_tls,
     (func_ptr_type) &dl_iterate_phdr,
+    (func_ptr_type) &_rtld_atfork_pre,
+    (func_ptr_type) &_rtld_atfork_post,
     NULL
 };
 
diff --git a/libexec/rtld-elf/rtld_lock.c b/libexec/rtld-elf/rtld_lock.c
index 5bb891a..bf4caf7 100644
--- a/libexec/rtld-elf/rtld_lock.c
+++ b/libexec/rtld-elf/rtld_lock.c
@@ -316,3 +316,19 @@ _rtld_thread_init(struct RtldLockInfo *pli)
 	thread_mask_set(flags);
 	dbg("_rtld_thread_init: done");
 }
+
+void
+_rtld_atfork_pre(int *locks)
+{
+
+	locks[2] = wlock_acquire(rtld_phdr_lock);
+	locks[0] = rlock_acquire(rtld_bind_lock);
+}
+
+void
+_rtld_atfork_post(int *locks)
+{
+
+	rlock_release(rtld_bind_lock, locks[0]);
+	wlock_release(rtld_phdr_lock, locks[2]);
+}
diff --git a/libexec/rtld-elf/rtld_lock.h b/libexec/rtld-elf/rtld_lock.h
index 2e3f954..4c5d854 100644
--- a/libexec/rtld-elf/rtld_lock.h
+++ b/libexec/rtld-elf/rtld_lock.h
@@ -44,6 +44,8 @@ struct RtldLockInfo
 };
 
 extern void _rtld_thread_init(struct RtldLockInfo *);
+extern void _rtld_atfork_pre(int *);
+extern void _rtld_atfork_post(int *);
 
 #ifdef IN_RTLD