dlm: Fix kobject memleak

Currently the error return path from kobject_init_and_add() is not
followed by a call to kobject_put() - which means we are leaking
the kobject.

Set do_unreg = 1 before kobject_init_and_add() to ensure that
kobject_put() can be called in its error patch.

Fixes: 901195e ("Kobject: change GFS2 to use kobject_init_and_add")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Wang Hai <wanghai38@huawei.com>
Signed-off-by: David Teigland <teigland@redhat.com>

fs/dlm/lockspace.c

@@ -622,16 +622,16 @@ static int new_lockspace(const char *name, const char *cluster,
 	wait_event(ls->ls_recover_lock_wait,
 		   test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags));
 
+	/* let kobject handle freeing of ls if there's an error */
+	do_unreg = 1;
+
 	ls->ls_kobj.kset = dlm_kset;
 	error = kobject_init_and_add(&ls->ls_kobj, &dlm_ktype, NULL,
 				     "%s", ls->ls_name);
 	if (error)
 		goto out_recoverd;
 	kobject_uevent(&ls->ls_kobj, KOBJ_ADD);
 
-	/* let kobject handle freeing of ls if there's an error */
-	do_unreg = 1;
-
 	/* This uevent triggers dlm_controld in userspace to add us to the
 	   group of nodes that are members of this lockspace (managed by the
 	   cluster infrastructure.)  Once it's done that, it tells us who the