futex: Fix errors in nested key ref-counting
Darren Hart [Sun, 17 Oct 2010 15:35:04 +0000 (08:35 -0700)]
futex_wait() is leaking key references due to futex_wait_setup()
acquiring an additional reference via the queue_lock() routine. The
nested key ref-counting has been masking bugs and complicating code
analysis. queue_lock() is only called with a previously ref-counted
key, so remove the additional ref-counting from the queue_(un)lock()
functions.

Also futex_wait_requeue_pi() drops one key reference too many in
unqueue_me_pi(). Remove the key reference handling from
unqueue_me_pi(). This was paired with a queue_lock() in
futex_lock_pi(), so the count remains unchanged.

Document remaining nested key ref-counting sites.

Signed-off-by: Darren Hart <dvhart@linux.intel.com>
Reported-and-tested-by: Matthieu Fertré<matthieu.fertre@kerlabs.com>
Reported-by: Louis Rilling<louis.rilling@kerlabs.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: John Kacur <jkacur@redhat.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
LKML-Reference: <4CBB17A8.70401@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@kernel.org

kernel/futex.c

index 6a3a5fa..e328f57 100644 (file)
@@ -1363,7 +1363,6 @@ static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
 {
        struct futex_hash_bucket *hb;
 
-       get_futex_key_refs(&q->key);
        hb = hash_futex(&q->key);
        q->lock_ptr = &hb->lock;
 
@@ -1375,7 +1374,6 @@ static inline void
 queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
 {
        spin_unlock(&hb->lock);
-       drop_futex_key_refs(&q->key);
 }
 
 /**
@@ -1480,8 +1478,6 @@ static void unqueue_me_pi(struct futex_q *q)
        q->pi_state = NULL;
 
        spin_unlock(q->lock_ptr);
-
-       drop_futex_key_refs(&q->key);
 }
 
 /*
@@ -1812,7 +1808,10 @@ static int futex_wait(u32 __user *uaddr, int fshared,
        }
 
 retry:
-       /* Prepare to wait on uaddr. */
+       /*
+        * Prepare to wait on uaddr. On success, holds hb lock and increments
+        * q.key refs.
+        */
        ret = futex_wait_setup(uaddr, val, fshared, &q, &hb);
        if (ret)
                goto out;
@@ -1822,24 +1821,23 @@ retry:
 
        /* If we were woken (and unqueued), we succeeded, whatever. */
        ret = 0;
+       /* unqueue_me() drops q.key ref */
        if (!unqueue_me(&q))
-               goto out_put_key;
+               goto out;
        ret = -ETIMEDOUT;
        if (to && !to->task)
-               goto out_put_key;
+               goto out;
 
        /*
         * We expect signal_pending(current), but we might be the
         * victim of a spurious wakeup as well.
         */
-       if (!signal_pending(current)) {
-               put_futex_key(fshared, &q.key);
+       if (!signal_pending(current))
                goto retry;
-       }
 
        ret = -ERESTARTSYS;
        if (!abs_time)
-               goto out_put_key;
+               goto out;
 
        restart = &current_thread_info()->restart_block;
        restart->fn = futex_wait_restart;
@@ -1856,8 +1854,6 @@ retry:
 
        ret = -ERESTART_RESTARTBLOCK;
 
-out_put_key:
-       put_futex_key(fshared, &q.key);
 out:
        if (to) {
                hrtimer_cancel(&to->timer);
@@ -2236,7 +2232,10 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared,
        q.rt_waiter = &rt_waiter;
        q.requeue_pi_key = &key2;
 
-       /* Prepare to wait on uaddr. */
+       /*
+        * Prepare to wait on uaddr. On success, increments q.key (key1) ref
+        * count.
+        */
        ret = futex_wait_setup(uaddr, val, fshared, &q, &hb);
        if (ret)
                goto out_key2;
@@ -2254,7 +2253,9 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared,
         * In order for us to be here, we know our q.key == key2, and since
         * we took the hb->lock above, we also know that futex_requeue() has
         * completed and we no longer have to concern ourselves with a wakeup
-        * race with the atomic proxy lock acquition by the requeue code.
+        * race with the atomic proxy lock acquisition by the requeue code. The
+        * futex_requeue dropped our key1 reference and incremented our key2
+        * reference count.
         */
 
        /* Check if the requeue code acquired the second futex for us. */