1 files changed, 86 insertions, 36 deletions

diff --git a/kernel/futex.c b/kernel/futex.c
index cf0c8e21d1ab..c2b2e0b83abf 100644
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -415,15 +415,15 @@ out_unlock:
  */
 void exit_pi_state_list(struct task_struct *curr)
 {
-        struct futex_hash_bucket *hb;
         struct list_head *next, *head = &curr->pi_state_list;
         struct futex_pi_state *pi_state;
+        struct futex_hash_bucket *hb;
         union futex_key key;
 
         /*
          * We are a ZOMBIE and nobody can enqueue itself on
          * pi_state_list anymore, but we have to be careful
-         * versus waiters unqueueing themselfs
+         * versus waiters unqueueing themselves:
          */
         spin_lock_irq(&curr->pi_lock);
         while (!list_empty(head)) {
@@ -431,21 +431,24 @@ void exit_pi_state_list(struct task_struct *curr)
                 next = head->next;
                 pi_state = list_entry(next, struct futex_pi_state, list);
                 key = pi_state->key;
+                hb = hash_futex(&key);
                 spin_unlock_irq(&curr->pi_lock);
 
-                hb = hash_futex(&key);
                 spin_lock(&hb->lock);
 
                 spin_lock_irq(&curr->pi_lock);
+                /*
+                 * We dropped the pi-lock, so re-check whether this
+                 * task still owns the PI-state:
+                 */
                 if (head->next != next) {
                         spin_unlock(&hb->lock);
                         continue;
                 }
 
-                list_del_init(&pi_state->list);
-
                 WARN_ON(pi_state->owner != curr);
-
+                WARN_ON(list_empty(&pi_state->list));
+                list_del_init(&pi_state->list);
                 pi_state->owner = NULL;
                 spin_unlock_irq(&curr->pi_lock);
 
@@ -470,7 +473,7 @@ lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, struct futex_q *me)
         head = &hb->chain;
 
         list_for_each_entry_safe(this, next, head, list) {
-                if (match_futex (&this->key, &me->key)) {
+                if (match_futex(&this->key, &me->key)) {
                         /*
                          * Another waiter already exists - bump up
                          * the refcount and return its pi_state:
@@ -482,6 +485,8 @@ lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, struct futex_q *me)
                         if (unlikely(!pi_state))
                                 return -EINVAL;
 
+                        WARN_ON(!atomic_read(&pi_state->refcount));
+
                         atomic_inc(&pi_state->refcount);
                         me->pi_state = pi_state;
 
@@ -490,10 +495,13 @@ lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, struct futex_q *me)
         }
 
         /*
-         * We are the first waiter - try to look up the real owner and
-         * attach the new pi_state to it:
+         * We are the first waiter - try to look up the real owner and attach
+         * the new pi_state to it, but bail out when the owner died bit is set
+         * and TID = 0:
          */
         pid = uval & FUTEX_TID_MASK;
+        if (!pid && (uval & FUTEX_OWNER_DIED))
+                return -ESRCH;
         p = futex_find_get_task(pid);
         if (!p)
                 return -ESRCH;
@@ -510,6 +518,7 @@ lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, struct futex_q *me)
         pi_state->key = me->key;
 
         spin_lock_irq(&p->pi_lock);
+        WARN_ON(!list_empty(&pi_state->list));
         list_add(&pi_state->list, &p->pi_state_list);
         pi_state->owner = p;
         spin_unlock_irq(&p->pi_lock);
@@ -573,20 +582,29 @@ static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
          * kept enabled while there is PI state around. We must also
          * preserve the owner died bit.)
          */
-        newval = (uval & FUTEX_OWNER_DIED) | FUTEX_WAITERS | new_owner->pid;
+        if (!(uval & FUTEX_OWNER_DIED)) {
+                newval = FUTEX_WAITERS | new_owner->pid;
 
-        inc_preempt_count();
-        curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval);
-        dec_preempt_count();
+                inc_preempt_count();
+                curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval);
+                dec_preempt_count();
+                if (curval == -EFAULT)
+                        return -EFAULT;
+                if (curval != uval)
+                        return -EINVAL;
+        }
 
-        if (curval == -EFAULT)
-                return -EFAULT;
-        if (curval != uval)
-                return -EINVAL;
+        spin_lock_irq(&pi_state->owner->pi_lock);
+        WARN_ON(list_empty(&pi_state->list));
+        list_del_init(&pi_state->list);
+        spin_unlock_irq(&pi_state->owner->pi_lock);
 
-        list_del_init(&pi_state->owner->pi_state_list);
+        spin_lock_irq(&new_owner->pi_lock);
+        WARN_ON(!list_empty(&pi_state->list));
         list_add(&pi_state->list, &new_owner->pi_state_list);
         pi_state->owner = new_owner;
+        spin_unlock_irq(&new_owner->pi_lock);
+
         rt_mutex_unlock(&pi_state->pi_mutex);
 
         return 0;
@@ -930,6 +948,7 @@ static int unqueue_me(struct futex_q *q)
         /* In the common case we don't take the spinlock, which is nice. */
  retry:
         lock_ptr = q->lock_ptr;
+        barrier();
         if (lock_ptr != 0) {
                 spin_lock(lock_ptr);
                 /*
@@ -1236,6 +1255,7 @@ static int do_futex_lock_pi(u32 __user *uaddr, int detect, int trylock,
                 /* Owner died? */
                 if (q.pi_state->owner != NULL) {
                         spin_lock_irq(&q.pi_state->owner->pi_lock);
+                        WARN_ON(list_empty(&q.pi_state->list));
                         list_del_init(&q.pi_state->list);
                         spin_unlock_irq(&q.pi_state->owner->pi_lock);
                 } else
@@ -1244,6 +1264,7 @@ static int do_futex_lock_pi(u32 __user *uaddr, int detect, int trylock,
                 q.pi_state->owner = current;
 
                 spin_lock_irq(&current->pi_lock);
+                WARN_ON(!list_empty(&q.pi_state->list));
                 list_add(&q.pi_state->list, &current->pi_state_list);
                 spin_unlock_irq(&current->pi_lock);
 
@@ -1427,9 +1448,11 @@ retry_locked:
          * again. If it succeeds then we can return without waking
          * anyone else up:
          */
-        inc_preempt_count();
-        uval = futex_atomic_cmpxchg_inatomic(uaddr, current->pid, 0);
-        dec_preempt_count();
+        if (!(uval & FUTEX_OWNER_DIED)) {
+                inc_preempt_count();
+                uval = futex_atomic_cmpxchg_inatomic(uaddr, current->pid, 0);
+                dec_preempt_count();
+        }
 
         if (unlikely(uval == -EFAULT))
                 goto pi_faulted;
@@ -1462,9 +1485,11 @@ retry_locked:
         /*
          * No waiters - kernel unlocks the futex:
          */
-        ret = unlock_futex_pi(uaddr, uval);
-        if (ret == -EFAULT)
-                goto pi_faulted;
+        if (!(uval & FUTEX_OWNER_DIED)) {
+                ret = unlock_futex_pi(uaddr, uval);
+                if (ret == -EFAULT)
+                        goto pi_faulted;
+        }
 
 out_unlock:
         spin_unlock(&hb->lock);
@@ -1683,9 +1708,9 @@ err_unlock:
  * Process a futex-list entry, check whether it's owned by the
  * dying task, and do notification if so:
  */
-int handle_futex_death(u32 __user *uaddr, struct task_struct *curr)
+int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
 {
-        u32 uval, nval;
+        u32 uval, nval, mval;
 
 retry:
         if (get_user(uval, uaddr))
@@ -1702,21 +1727,45 @@ retry:
                  * thread-death.) The rest of the cleanup is done in
                  * userspace.
                  */
-                nval = futex_atomic_cmpxchg_inatomic(uaddr, uval,
-                                                     uval | FUTEX_OWNER_DIED);
+                mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
+                nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval);
+
                 if (nval == -EFAULT)
                         return -1;
 
                 if (nval != uval)
                         goto retry;
 
-                if (uval & FUTEX_WAITERS)
-                        futex_wake(uaddr, 1);
+                /*
+                 * Wake robust non-PI futexes here. The wakeup of
+                 * PI futexes happens in exit_pi_state():
+                 */
+                if (!pi) {
+                        if (uval & FUTEX_WAITERS)
+                                futex_wake(uaddr, 1);
+                }
         }
         return 0;
 }
 
 /*
+ * Fetch a robust-list pointer. Bit 0 signals PI futexes:
+ */
+static inline int fetch_robust_entry(struct robust_list __user **entry,
+                                     struct robust_list __user **head, int *pi)
+{
+        unsigned long uentry;
+
+        if (get_user(uentry, (unsigned long *)head))
+                return -EFAULT;
+
+        *entry = (void *)(uentry & ~1UL);
+        *pi = uentry & 1;
+
+        return 0;
+}
+
+/*
  * Walk curr->robust_list (very carefully, it's a userspace list!)
  * and mark any locks found there dead, and notify any waiters.
  *
@@ -1726,14 +1775,14 @@ void exit_robust_list(struct task_struct *curr)
 {
         struct robust_list_head __user *head = curr->robust_list;
         struct robust_list __user *entry, *pending;
-        unsigned int limit = ROBUST_LIST_LIMIT;
+        unsigned int limit = ROBUST_LIST_LIMIT, pi, pip;
         unsigned long futex_offset;
 
         /*
          * Fetch the list head (which was registered earlier, via
          * sys_set_robust_list()):
          */
-        if (get_user(entry, &head->list.next))
+        if (fetch_robust_entry(&entry, &head->list.next, &pi))
                 return;
         /*
          * Fetch the relative futex offset:
@@ -1744,10 +1793,11 @@ void exit_robust_list(struct task_struct *curr)
          * Fetch any possibly pending lock-add first, and handle it
          * if it exists:
          */
-        if (get_user(pending, &head->list_op_pending))
+        if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
                 return;
+
         if (pending)
-                handle_futex_death((void *)pending + futex_offset, curr);
+                handle_futex_death((void *)pending + futex_offset, curr, pip);
 
         while (entry != &head->list) {
                 /*
@@ -1756,12 +1806,12 @@ void exit_robust_list(struct task_struct *curr)
                  */
                 if (entry != pending)
                         if (handle_futex_death((void *)entry + futex_offset,
-                                                curr))
+                                                curr, pi))
                                 return;
                 /*
                  * Fetch the next entry in the list:
                  */
-                if (get_user(entry, &entry->next))
+                if (fetch_robust_entry(&entry, &entry->next, &pi))
                         return;
                 /*
                  * Avoid excessively long or circular lists: