lockd: fix race in nlm_release()
authorJ. Bruce Fields <bfields@citi.umich.edu>
Wed, 20 Feb 2008 19:02:47 +0000 (14:02 -0500)
committerJ. Bruce Fields <bfields@citi.umich.edu>
Wed, 23 Apr 2008 20:13:39 +0000 (16:13 -0400)
The sm_count is decremented to zero but left on the nsm_handles list.
So in the space between decrementing sm_count and acquiring nsm_mutex,
it is possible for another task to find this nsm_handle, increment the
use count and then enter nsm_release itself.

Thus there's nothing to prevent the nsm being freed before we acquire
nsm_mutex here.

Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
fs/lockd/host.c

index c3f119426d832ea396ebffbcd2820e335477bd8b..960911c4a11c60d72ff957e72f4951ab328f2d34 100644 (file)
@@ -529,12 +529,10 @@ nsm_release(struct nsm_handle *nsm)
 {
        if (!nsm)
                return;
+       mutex_lock(&nsm_mutex);
        if (atomic_dec_and_test(&nsm->sm_count)) {
-               mutex_lock(&nsm_mutex);
-               if (atomic_read(&nsm->sm_count) == 0) {
-                       list_del(&nsm->sm_link);
-                       kfree(nsm);
-               }
-               mutex_unlock(&nsm_mutex);
+               list_del(&nsm->sm_link);
+               kfree(nsm);
        }
+       mutex_unlock(&nsm_mutex);
 }