2 * Fast Userspace Mutexes (which I call "Futexes!").
3 * (C) Rusty Russell, IBM 2002
5 * Generalized futexes, futex requeueing, misc fixes by Ingo Molnar
6 * (C) Copyright 2003 Red Hat Inc, All Rights Reserved
8 * Removed page pinning, fix privately mapped COW pages and other cleanups
9 * (C) Copyright 2003, 2004 Jamie Lokier
11 * Robust futex support started by Ingo Molnar
12 * (C) Copyright 2006 Red Hat Inc, All Rights Reserved
13 * Thanks to Thomas Gleixner for suggestions, analysis and fixes.
15 * PI-futex support started by Ingo Molnar and Thomas Gleixner
16 * Copyright (C) 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
17 * Copyright (C) 2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
19 * PRIVATE futexes by Eric Dumazet
20 * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com>
22 * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly
23 * enough at me, Linus for the original (flawed) idea, Matthew
24 * Kirkwood for proof-of-concept implementation.
26 * "The futexes are also cursed."
27 * "But they come in a choice of three flavours!"
29 * This program is free software; you can redistribute it and/or modify
30 * it under the terms of the GNU General Public License as published by
31 * the Free Software Foundation; either version 2 of the License, or
32 * (at your option) any later version.
34 * This program is distributed in the hope that it will be useful,
35 * but WITHOUT ANY WARRANTY; without even the implied warranty of
36 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
37 * GNU General Public License for more details.
39 * You should have received a copy of the GNU General Public License
40 * along with this program; if not, write to the Free Software
41 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
43 #include <linux/slab.h>
44 #include <linux/poll.h>
46 #include <linux/file.h>
47 #include <linux/jhash.h>
48 #include <linux/init.h>
49 #include <linux/futex.h>
50 #include <linux/mount.h>
51 #include <linux/pagemap.h>
52 #include <linux/syscalls.h>
53 #include <linux/signal.h>
54 #include <linux/module.h>
55 #include <asm/futex.h>
57 #include "rtmutex_common.h"
59 #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
62 * Priority Inheritance state:
64 struct futex_pi_state
{
66 * list of 'owned' pi_state instances - these have to be
67 * cleaned up in do_exit() if the task exits prematurely:
69 struct list_head list
;
74 struct rt_mutex pi_mutex
;
76 struct task_struct
*owner
;
83 * We use this hashed waitqueue instead of a normal wait_queue_t, so
84 * we can wake only the relevant ones (hashed queues may be shared).
86 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
87 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
88 * The order of wakup is always to make the first condition true, then
89 * wake up q->waiters, then make the second condition true.
92 struct plist_node list
;
93 wait_queue_head_t waiters
;
95 /* Which hash list lock to use: */
98 /* Key which the futex is hashed on: */
101 /* For fd, sigio sent using these: */
105 /* Optional priority inheritance state: */
106 struct futex_pi_state
*pi_state
;
107 struct task_struct
*task
;
111 * Split the global futex_lock into every hash list lock.
113 struct futex_hash_bucket
{
115 struct plist_head chain
;
118 static struct futex_hash_bucket futex_queues
[1<<FUTEX_HASHBITS
];
120 /* Futex-fs vfsmount entry: */
121 static struct vfsmount
*futex_mnt
;
124 * Take mm->mmap_sem, when futex is shared
126 static inline void futex_lock_mm(struct rw_semaphore
*fshared
)
133 * Release mm->mmap_sem, when the futex is shared
135 static inline void futex_unlock_mm(struct rw_semaphore
*fshared
)
142 * We hash on the keys returned from get_futex_key (see below).
144 static struct futex_hash_bucket
*hash_futex(union futex_key
*key
)
146 u32 hash
= jhash2((u32
*)&key
->both
.word
,
147 (sizeof(key
->both
.word
)+sizeof(key
->both
.ptr
))/4,
149 return &futex_queues
[hash
& ((1 << FUTEX_HASHBITS
)-1)];
153 * Return 1 if two futex_keys are equal, 0 otherwise.
155 static inline int match_futex(union futex_key
*key1
, union futex_key
*key2
)
157 return (key1
->both
.word
== key2
->both
.word
158 && key1
->both
.ptr
== key2
->both
.ptr
159 && key1
->both
.offset
== key2
->both
.offset
);
163 * get_futex_key - Get parameters which are the keys for a futex.
164 * @uaddr: virtual address of the futex
165 * @shared: NULL for a PROCESS_PRIVATE futex,
166 * ¤t->mm->mmap_sem for a PROCESS_SHARED futex
167 * @key: address where result is stored.
169 * Returns a negative error code or 0
170 * The key words are stored in *key on success.
172 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
173 * offset_within_page). For private mappings, it's (uaddr, current->mm).
174 * We can usually work out the index without swapping in the page.
176 * fshared is NULL for PROCESS_PRIVATE futexes
177 * For other futexes, it points to ¤t->mm->mmap_sem and
178 * caller must have taken the reader lock. but NOT any spinlocks.
180 int get_futex_key(u32 __user
*uaddr
, struct rw_semaphore
*fshared
,
181 union futex_key
*key
)
183 unsigned long address
= (unsigned long)uaddr
;
184 struct mm_struct
*mm
= current
->mm
;
185 struct vm_area_struct
*vma
;
190 * The futex address must be "naturally" aligned.
192 key
->both
.offset
= address
% PAGE_SIZE
;
193 if (unlikely((address
% sizeof(u32
)) != 0))
195 address
-= key
->both
.offset
;
198 * PROCESS_PRIVATE futexes are fast.
199 * As the mm cannot disappear under us and the 'key' only needs
200 * virtual address, we dont even have to find the underlying vma.
201 * Note : We do have to check 'uaddr' is a valid user address,
202 * but access_ok() should be faster than find_vma()
205 if (unlikely(!access_ok(VERIFY_WRITE
, uaddr
, sizeof(u32
))))
207 key
->private.mm
= mm
;
208 key
->private.address
= address
;
212 * The futex is hashed differently depending on whether
213 * it's in a shared or private mapping. So check vma first.
215 vma
= find_extend_vma(mm
, address
);
222 if (unlikely((vma
->vm_flags
& (VM_IO
|VM_READ
)) != VM_READ
))
223 return (vma
->vm_flags
& VM_IO
) ? -EPERM
: -EACCES
;
226 * Private mappings are handled in a simple way.
228 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
229 * it's a read-only handle, it's expected that futexes attach to
230 * the object not the particular process. Therefore we use
231 * VM_MAYSHARE here, not VM_SHARED which is restricted to shared
232 * mappings of _writable_ handles.
234 if (likely(!(vma
->vm_flags
& VM_MAYSHARE
))) {
235 key
->both
.offset
|= FUT_OFF_MMSHARED
; /* reference taken on mm */
236 key
->private.mm
= mm
;
237 key
->private.address
= address
;
242 * Linear file mappings are also simple.
244 key
->shared
.inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
245 key
->both
.offset
|= FUT_OFF_INODE
; /* inode-based key. */
246 if (likely(!(vma
->vm_flags
& VM_NONLINEAR
))) {
247 key
->shared
.pgoff
= (((address
- vma
->vm_start
) >> PAGE_SHIFT
)
253 * We could walk the page table to read the non-linear
254 * pte, and get the page index without fetching the page
255 * from swap. But that's a lot of code to duplicate here
256 * for a rare case, so we simply fetch the page.
258 err
= get_user_pages(current
, mm
, address
, 1, 0, 0, &page
, NULL
);
261 page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
267 EXPORT_SYMBOL_GPL(get_futex_key
);
270 * Take a reference to the resource addressed by a key.
271 * Can be called while holding spinlocks.
274 inline void get_futex_key_refs(union futex_key
*key
)
276 if (key
->both
.ptr
== 0)
278 switch (key
->both
.offset
& (FUT_OFF_INODE
|FUT_OFF_MMSHARED
)) {
280 atomic_inc(&key
->shared
.inode
->i_count
);
282 case FUT_OFF_MMSHARED
:
283 atomic_inc(&key
->private.mm
->mm_count
);
287 EXPORT_SYMBOL_GPL(get_futex_key_refs
);
290 * Drop a reference to the resource addressed by a key.
291 * The hash bucket spinlock must not be held.
293 void drop_futex_key_refs(union futex_key
*key
)
295 if (key
->both
.ptr
== 0)
297 switch (key
->both
.offset
& (FUT_OFF_INODE
|FUT_OFF_MMSHARED
)) {
299 iput(key
->shared
.inode
);
301 case FUT_OFF_MMSHARED
:
302 mmdrop(key
->private.mm
);
306 EXPORT_SYMBOL_GPL(drop_futex_key_refs
);
308 static u32
cmpxchg_futex_value_locked(u32 __user
*uaddr
, u32 uval
, u32 newval
)
313 curval
= futex_atomic_cmpxchg_inatomic(uaddr
, uval
, newval
);
319 static int get_futex_value_locked(u32
*dest
, u32 __user
*from
)
324 ret
= __copy_from_user_inatomic(dest
, from
, sizeof(u32
));
327 return ret
? -EFAULT
: 0;
332 * if fshared is non NULL, current->mm->mmap_sem is already held
334 static int futex_handle_fault(unsigned long address
,
335 struct rw_semaphore
*fshared
, int attempt
)
337 struct vm_area_struct
* vma
;
338 struct mm_struct
*mm
= current
->mm
;
345 down_read(&mm
->mmap_sem
);
346 vma
= find_vma(mm
, address
);
347 if (vma
&& address
>= vma
->vm_start
&&
348 (vma
->vm_flags
& VM_WRITE
)) {
349 switch (handle_mm_fault(mm
, vma
, address
, 1)) {
361 up_read(&mm
->mmap_sem
);
368 static int refill_pi_state_cache(void)
370 struct futex_pi_state
*pi_state
;
372 if (likely(current
->pi_state_cache
))
375 pi_state
= kzalloc(sizeof(*pi_state
), GFP_KERNEL
);
380 INIT_LIST_HEAD(&pi_state
->list
);
381 /* pi_mutex gets initialized later */
382 pi_state
->owner
= NULL
;
383 atomic_set(&pi_state
->refcount
, 1);
385 current
->pi_state_cache
= pi_state
;
390 static struct futex_pi_state
* alloc_pi_state(void)
392 struct futex_pi_state
*pi_state
= current
->pi_state_cache
;
395 current
->pi_state_cache
= NULL
;
400 static void free_pi_state(struct futex_pi_state
*pi_state
)
402 if (!atomic_dec_and_test(&pi_state
->refcount
))
406 * If pi_state->owner is NULL, the owner is most probably dying
407 * and has cleaned up the pi_state already
409 if (pi_state
->owner
) {
410 spin_lock_irq(&pi_state
->owner
->pi_lock
);
411 list_del_init(&pi_state
->list
);
412 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
414 rt_mutex_proxy_unlock(&pi_state
->pi_mutex
, pi_state
->owner
);
417 if (current
->pi_state_cache
)
421 * pi_state->list is already empty.
422 * clear pi_state->owner.
423 * refcount is at 0 - put it back to 1.
425 pi_state
->owner
= NULL
;
426 atomic_set(&pi_state
->refcount
, 1);
427 current
->pi_state_cache
= pi_state
;
432 * Look up the task based on what TID userspace gave us.
435 static struct task_struct
* futex_find_get_task(pid_t pid
)
437 struct task_struct
*p
;
440 p
= find_task_by_pid(pid
);
442 if (!p
|| ((current
->euid
!= p
->euid
) && (current
->euid
!= p
->uid
)))
453 * This task is holding PI mutexes at exit time => bad.
454 * Kernel cleans up PI-state, but userspace is likely hosed.
455 * (Robust-futex cleanup is separate and might save the day for userspace.)
457 void exit_pi_state_list(struct task_struct
*curr
)
459 struct list_head
*next
, *head
= &curr
->pi_state_list
;
460 struct futex_pi_state
*pi_state
;
461 struct futex_hash_bucket
*hb
;
465 * We are a ZOMBIE and nobody can enqueue itself on
466 * pi_state_list anymore, but we have to be careful
467 * versus waiters unqueueing themselves:
469 spin_lock_irq(&curr
->pi_lock
);
470 while (!list_empty(head
)) {
473 pi_state
= list_entry(next
, struct futex_pi_state
, list
);
475 hb
= hash_futex(&key
);
476 spin_unlock_irq(&curr
->pi_lock
);
478 spin_lock(&hb
->lock
);
480 spin_lock_irq(&curr
->pi_lock
);
482 * We dropped the pi-lock, so re-check whether this
483 * task still owns the PI-state:
485 if (head
->next
!= next
) {
486 spin_unlock(&hb
->lock
);
490 WARN_ON(pi_state
->owner
!= curr
);
491 WARN_ON(list_empty(&pi_state
->list
));
492 list_del_init(&pi_state
->list
);
493 pi_state
->owner
= NULL
;
494 spin_unlock_irq(&curr
->pi_lock
);
496 rt_mutex_unlock(&pi_state
->pi_mutex
);
498 spin_unlock(&hb
->lock
);
500 spin_lock_irq(&curr
->pi_lock
);
502 spin_unlock_irq(&curr
->pi_lock
);
506 lookup_pi_state(u32 uval
, struct futex_hash_bucket
*hb
,
507 union futex_key
*key
, struct futex_pi_state
**ps
)
509 struct futex_pi_state
*pi_state
= NULL
;
510 struct futex_q
*this, *next
;
511 struct plist_head
*head
;
512 struct task_struct
*p
;
513 pid_t pid
= uval
& FUTEX_TID_MASK
;
517 plist_for_each_entry_safe(this, next
, head
, list
) {
518 if (match_futex(&this->key
, key
)) {
520 * Another waiter already exists - bump up
521 * the refcount and return its pi_state:
523 pi_state
= this->pi_state
;
525 * Userspace might have messed up non PI and PI futexes
527 if (unlikely(!pi_state
))
530 WARN_ON(!atomic_read(&pi_state
->refcount
));
531 WARN_ON(pid
&& pi_state
->owner
&&
532 pi_state
->owner
->pid
!= pid
);
534 atomic_inc(&pi_state
->refcount
);
542 * We are the first waiter - try to look up the real owner and attach
543 * the new pi_state to it, but bail out when TID = 0
547 p
= futex_find_get_task(pid
);
552 * We need to look at the task state flags to figure out,
553 * whether the task is exiting. To protect against the do_exit
554 * change of the task flags, we do this protected by
557 spin_lock_irq(&p
->pi_lock
);
558 if (unlikely(p
->flags
& PF_EXITING
)) {
560 * The task is on the way out. When PF_EXITPIDONE is
561 * set, we know that the task has finished the
564 int ret
= (p
->flags
& PF_EXITPIDONE
) ? -ESRCH
: -EAGAIN
;
566 spin_unlock_irq(&p
->pi_lock
);
571 pi_state
= alloc_pi_state();
574 * Initialize the pi_mutex in locked state and make 'p'
577 rt_mutex_init_proxy_locked(&pi_state
->pi_mutex
, p
);
579 /* Store the key for possible exit cleanups: */
580 pi_state
->key
= *key
;
582 WARN_ON(!list_empty(&pi_state
->list
));
583 list_add(&pi_state
->list
, &p
->pi_state_list
);
585 spin_unlock_irq(&p
->pi_lock
);
595 * The hash bucket lock must be held when this is called.
596 * Afterwards, the futex_q must not be accessed.
598 static void wake_futex(struct futex_q
*q
)
600 plist_del(&q
->list
, &q
->list
.plist
);
602 send_sigio(&q
->filp
->f_owner
, q
->fd
, POLL_IN
);
604 * The lock in wake_up_all() is a crucial memory barrier after the
605 * plist_del() and also before assigning to q->lock_ptr.
607 wake_up_all(&q
->waiters
);
609 * The waiting task can free the futex_q as soon as this is written,
610 * without taking any locks. This must come last.
612 * A memory barrier is required here to prevent the following store
613 * to lock_ptr from getting ahead of the wakeup. Clearing the lock
614 * at the end of wake_up_all() does not prevent this store from
621 static int wake_futex_pi(u32 __user
*uaddr
, u32 uval
, struct futex_q
*this)
623 struct task_struct
*new_owner
;
624 struct futex_pi_state
*pi_state
= this->pi_state
;
630 spin_lock(&pi_state
->pi_mutex
.wait_lock
);
631 new_owner
= rt_mutex_next_owner(&pi_state
->pi_mutex
);
634 * This happens when we have stolen the lock and the original
635 * pending owner did not enqueue itself back on the rt_mutex.
636 * Thats not a tragedy. We know that way, that a lock waiter
637 * is on the fly. We make the futex_q waiter the pending owner.
640 new_owner
= this->task
;
643 * We pass it to the next owner. (The WAITERS bit is always
644 * kept enabled while there is PI state around. We must also
645 * preserve the owner died bit.)
647 if (!(uval
& FUTEX_OWNER_DIED
)) {
650 newval
= FUTEX_WAITERS
| new_owner
->pid
;
652 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
654 if (curval
== -EFAULT
)
659 spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
664 spin_lock_irq(&pi_state
->owner
->pi_lock
);
665 WARN_ON(list_empty(&pi_state
->list
));
666 list_del_init(&pi_state
->list
);
667 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
669 spin_lock_irq(&new_owner
->pi_lock
);
670 WARN_ON(!list_empty(&pi_state
->list
));
671 list_add(&pi_state
->list
, &new_owner
->pi_state_list
);
672 pi_state
->owner
= new_owner
;
673 spin_unlock_irq(&new_owner
->pi_lock
);
675 spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
676 rt_mutex_unlock(&pi_state
->pi_mutex
);
681 static int unlock_futex_pi(u32 __user
*uaddr
, u32 uval
)
686 * There is no waiter, so we unlock the futex. The owner died
687 * bit has not to be preserved here. We are the owner:
689 oldval
= cmpxchg_futex_value_locked(uaddr
, uval
, 0);
691 if (oldval
== -EFAULT
)
700 * Express the locking dependencies for lockdep:
703 double_lock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
706 spin_lock(&hb1
->lock
);
708 spin_lock_nested(&hb2
->lock
, SINGLE_DEPTH_NESTING
);
709 } else { /* hb1 > hb2 */
710 spin_lock(&hb2
->lock
);
711 spin_lock_nested(&hb1
->lock
, SINGLE_DEPTH_NESTING
);
716 * Wake up all waiters hashed on the physical page that is mapped
717 * to this virtual address:
719 static int futex_wake(u32 __user
*uaddr
, struct rw_semaphore
*fshared
,
722 struct futex_hash_bucket
*hb
;
723 struct futex_q
*this, *next
;
724 struct plist_head
*head
;
728 futex_lock_mm(fshared
);
730 ret
= get_futex_key(uaddr
, fshared
, &key
);
731 if (unlikely(ret
!= 0))
734 hb
= hash_futex(&key
);
735 spin_lock(&hb
->lock
);
738 plist_for_each_entry_safe(this, next
, head
, list
) {
739 if (match_futex (&this->key
, &key
)) {
740 if (this->pi_state
) {
745 if (++ret
>= nr_wake
)
750 spin_unlock(&hb
->lock
);
752 futex_unlock_mm(fshared
);
757 * Wake up all waiters hashed on the physical page that is mapped
758 * to this virtual address:
761 futex_wake_op(u32 __user
*uaddr1
, struct rw_semaphore
*fshared
,
763 int nr_wake
, int nr_wake2
, int op
)
765 union futex_key key1
, key2
;
766 struct futex_hash_bucket
*hb1
, *hb2
;
767 struct plist_head
*head
;
768 struct futex_q
*this, *next
;
769 int ret
, op_ret
, attempt
= 0;
772 futex_lock_mm(fshared
);
774 ret
= get_futex_key(uaddr1
, fshared
, &key1
);
775 if (unlikely(ret
!= 0))
777 ret
= get_futex_key(uaddr2
, fshared
, &key2
);
778 if (unlikely(ret
!= 0))
781 hb1
= hash_futex(&key1
);
782 hb2
= hash_futex(&key2
);
785 double_lock_hb(hb1
, hb2
);
787 op_ret
= futex_atomic_op_inuser(op
, uaddr2
);
788 if (unlikely(op_ret
< 0)) {
791 spin_unlock(&hb1
->lock
);
793 spin_unlock(&hb2
->lock
);
797 * we don't get EFAULT from MMU faults if we don't have an MMU,
798 * but we might get them from range checking
804 if (unlikely(op_ret
!= -EFAULT
)) {
810 * futex_atomic_op_inuser needs to both read and write
811 * *(int __user *)uaddr2, but we can't modify it
812 * non-atomically. Therefore, if get_user below is not
813 * enough, we need to handle the fault ourselves, while
814 * still holding the mmap_sem.
817 ret
= futex_handle_fault((unsigned long)uaddr2
,
825 * If we would have faulted, release mmap_sem,
826 * fault it in and start all over again.
828 futex_unlock_mm(fshared
);
830 ret
= get_user(dummy
, uaddr2
);
839 plist_for_each_entry_safe(this, next
, head
, list
) {
840 if (match_futex (&this->key
, &key1
)) {
842 if (++ret
>= nr_wake
)
851 plist_for_each_entry_safe(this, next
, head
, list
) {
852 if (match_futex (&this->key
, &key2
)) {
854 if (++op_ret
>= nr_wake2
)
861 spin_unlock(&hb1
->lock
);
863 spin_unlock(&hb2
->lock
);
865 futex_unlock_mm(fshared
);
871 * Requeue all waiters hashed on one physical page to another
874 static int futex_requeue(u32 __user
*uaddr1
, struct rw_semaphore
*fshared
,
876 int nr_wake
, int nr_requeue
, u32
*cmpval
)
878 union futex_key key1
, key2
;
879 struct futex_hash_bucket
*hb1
, *hb2
;
880 struct plist_head
*head1
;
881 struct futex_q
*this, *next
;
882 int ret
, drop_count
= 0;
885 futex_lock_mm(fshared
);
887 ret
= get_futex_key(uaddr1
, fshared
, &key1
);
888 if (unlikely(ret
!= 0))
890 ret
= get_futex_key(uaddr2
, fshared
, &key2
);
891 if (unlikely(ret
!= 0))
894 hb1
= hash_futex(&key1
);
895 hb2
= hash_futex(&key2
);
897 double_lock_hb(hb1
, hb2
);
899 if (likely(cmpval
!= NULL
)) {
902 ret
= get_futex_value_locked(&curval
, uaddr1
);
905 spin_unlock(&hb1
->lock
);
907 spin_unlock(&hb2
->lock
);
910 * If we would have faulted, release mmap_sem, fault
911 * it in and start all over again.
913 futex_unlock_mm(fshared
);
915 ret
= get_user(curval
, uaddr1
);
922 if (curval
!= *cmpval
) {
929 plist_for_each_entry_safe(this, next
, head1
, list
) {
930 if (!match_futex (&this->key
, &key1
))
932 if (++ret
<= nr_wake
) {
936 * If key1 and key2 hash to the same bucket, no need to
939 if (likely(head1
!= &hb2
->chain
)) {
940 plist_del(&this->list
, &hb1
->chain
);
941 plist_add(&this->list
, &hb2
->chain
);
942 this->lock_ptr
= &hb2
->lock
;
943 #ifdef CONFIG_DEBUG_PI_LIST
944 this->list
.plist
.lock
= &hb2
->lock
;
948 get_futex_key_refs(&key2
);
951 if (ret
- nr_wake
>= nr_requeue
)
957 spin_unlock(&hb1
->lock
);
959 spin_unlock(&hb2
->lock
);
961 /* drop_futex_key_refs() must be called outside the spinlocks. */
962 while (--drop_count
>= 0)
963 drop_futex_key_refs(&key1
);
966 futex_unlock_mm(fshared
);
970 /* The key must be already stored in q->key. */
971 static inline struct futex_hash_bucket
*
972 queue_lock(struct futex_q
*q
, int fd
, struct file
*filp
)
974 struct futex_hash_bucket
*hb
;
979 init_waitqueue_head(&q
->waiters
);
981 get_futex_key_refs(&q
->key
);
982 hb
= hash_futex(&q
->key
);
983 q
->lock_ptr
= &hb
->lock
;
985 spin_lock(&hb
->lock
);
989 static inline void __queue_me(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
994 * The priority used to register this element is
995 * - either the real thread-priority for the real-time threads
996 * (i.e. threads with a priority lower than MAX_RT_PRIO)
997 * - or MAX_RT_PRIO for non-RT threads.
998 * Thus, all RT-threads are woken first in priority order, and
999 * the others are woken last, in FIFO order.
1001 prio
= min(current
->normal_prio
, MAX_RT_PRIO
);
1003 plist_node_init(&q
->list
, prio
);
1004 #ifdef CONFIG_DEBUG_PI_LIST
1005 q
->list
.plist
.lock
= &hb
->lock
;
1007 plist_add(&q
->list
, &hb
->chain
);
1009 spin_unlock(&hb
->lock
);
1013 queue_unlock(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
1015 spin_unlock(&hb
->lock
);
1016 drop_futex_key_refs(&q
->key
);
1020 * queue_me and unqueue_me must be called as a pair, each
1021 * exactly once. They are called with the hashed spinlock held.
1024 /* The key must be already stored in q->key. */
1025 static void queue_me(struct futex_q
*q
, int fd
, struct file
*filp
)
1027 struct futex_hash_bucket
*hb
;
1029 hb
= queue_lock(q
, fd
, filp
);
1033 /* Return 1 if we were still queued (ie. 0 means we were woken) */
1034 static int unqueue_me(struct futex_q
*q
)
1036 spinlock_t
*lock_ptr
;
1039 /* In the common case we don't take the spinlock, which is nice. */
1041 lock_ptr
= q
->lock_ptr
;
1043 if (lock_ptr
!= 0) {
1044 spin_lock(lock_ptr
);
1046 * q->lock_ptr can change between reading it and
1047 * spin_lock(), causing us to take the wrong lock. This
1048 * corrects the race condition.
1050 * Reasoning goes like this: if we have the wrong lock,
1051 * q->lock_ptr must have changed (maybe several times)
1052 * between reading it and the spin_lock(). It can
1053 * change again after the spin_lock() but only if it was
1054 * already changed before the spin_lock(). It cannot,
1055 * however, change back to the original value. Therefore
1056 * we can detect whether we acquired the correct lock.
1058 if (unlikely(lock_ptr
!= q
->lock_ptr
)) {
1059 spin_unlock(lock_ptr
);
1062 WARN_ON(plist_node_empty(&q
->list
));
1063 plist_del(&q
->list
, &q
->list
.plist
);
1065 BUG_ON(q
->pi_state
);
1067 spin_unlock(lock_ptr
);
1071 drop_futex_key_refs(&q
->key
);
1076 * PI futexes can not be requeued and must remove themself from the
1077 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1080 static void unqueue_me_pi(struct futex_q
*q
)
1082 WARN_ON(plist_node_empty(&q
->list
));
1083 plist_del(&q
->list
, &q
->list
.plist
);
1085 BUG_ON(!q
->pi_state
);
1086 free_pi_state(q
->pi_state
);
1089 spin_unlock(q
->lock_ptr
);
1091 drop_futex_key_refs(&q
->key
);
1095 * Fixup the pi_state owner with current.
1097 * Must be called with hash bucket lock held and mm->sem held for non
1100 static int fixup_pi_state_owner(u32 __user
*uaddr
, struct futex_q
*q
,
1101 struct task_struct
*curr
)
1103 u32 newtid
= curr
->pid
| FUTEX_WAITERS
;
1104 struct futex_pi_state
*pi_state
= q
->pi_state
;
1105 u32 uval
, curval
, newval
;
1109 if (pi_state
->owner
!= NULL
) {
1110 spin_lock_irq(&pi_state
->owner
->pi_lock
);
1111 WARN_ON(list_empty(&pi_state
->list
));
1112 list_del_init(&pi_state
->list
);
1113 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
1115 newtid
|= FUTEX_OWNER_DIED
;
1117 pi_state
->owner
= curr
;
1119 spin_lock_irq(&curr
->pi_lock
);
1120 WARN_ON(!list_empty(&pi_state
->list
));
1121 list_add(&pi_state
->list
, &curr
->pi_state_list
);
1122 spin_unlock_irq(&curr
->pi_lock
);
1125 * We own it, so we have to replace the pending owner
1126 * TID. This must be atomic as we have preserve the
1127 * owner died bit here.
1129 ret
= get_futex_value_locked(&uval
, uaddr
);
1132 newval
= (uval
& FUTEX_OWNER_DIED
) | newtid
;
1134 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
1136 if (curval
== -EFAULT
)
1146 * In case we must use restart_block to restart a futex_wait,
1147 * we encode in the 'arg3' shared capability
1149 #define ARG3_SHARED 1
1151 static long futex_wait_restart(struct restart_block
*restart
);
1153 static int futex_wait(u32 __user
*uaddr
, struct rw_semaphore
*fshared
,
1154 u32 val
, ktime_t
*abs_time
)
1156 struct task_struct
*curr
= current
;
1157 DECLARE_WAITQUEUE(wait
, curr
);
1158 struct futex_hash_bucket
*hb
;
1162 struct hrtimer_sleeper t
;
1167 futex_lock_mm(fshared
);
1169 ret
= get_futex_key(uaddr
, fshared
, &q
.key
);
1170 if (unlikely(ret
!= 0))
1171 goto out_release_sem
;
1173 hb
= queue_lock(&q
, -1, NULL
);
1176 * Access the page AFTER the futex is queued.
1177 * Order is important:
1179 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1180 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1182 * The basic logical guarantee of a futex is that it blocks ONLY
1183 * if cond(var) is known to be true at the time of blocking, for
1184 * any cond. If we queued after testing *uaddr, that would open
1185 * a race condition where we could block indefinitely with
1186 * cond(var) false, which would violate the guarantee.
1188 * A consequence is that futex_wait() can return zero and absorb
1189 * a wakeup when *uaddr != val on entry to the syscall. This is
1192 * for shared futexes, we hold the mmap semaphore, so the mapping
1193 * cannot have changed since we looked it up in get_futex_key.
1195 ret
= get_futex_value_locked(&uval
, uaddr
);
1197 if (unlikely(ret
)) {
1198 queue_unlock(&q
, hb
);
1201 * If we would have faulted, release mmap_sem, fault it in and
1202 * start all over again.
1204 futex_unlock_mm(fshared
);
1206 ret
= get_user(uval
, uaddr
);
1214 goto out_unlock_release_sem
;
1216 /* Only actually queue if *uaddr contained val. */
1220 * Now the futex is queued and we have checked the data, we
1221 * don't want to hold mmap_sem while we sleep.
1223 futex_unlock_mm(fshared
);
1226 * There might have been scheduling since the queue_me(), as we
1227 * cannot hold a spinlock across the get_user() in case it
1228 * faults, and we cannot just set TASK_INTERRUPTIBLE state when
1229 * queueing ourselves into the futex hash. This code thus has to
1230 * rely on the futex_wake() code removing us from hash when it
1234 /* add_wait_queue is the barrier after __set_current_state. */
1235 __set_current_state(TASK_INTERRUPTIBLE
);
1236 add_wait_queue(&q
.waiters
, &wait
);
1238 * !plist_node_empty() is safe here without any lock.
1239 * q.lock_ptr != 0 is not safe, because of ordering against wakeup.
1241 if (likely(!plist_node_empty(&q
.list
))) {
1245 hrtimer_init(&t
.timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_ABS
);
1246 hrtimer_init_sleeper(&t
, current
);
1247 t
.timer
.expires
= *abs_time
;
1249 hrtimer_start(&t
.timer
, t
.timer
.expires
, HRTIMER_MODE_ABS
);
1252 * the timer could have already expired, in which
1253 * case current would be flagged for rescheduling.
1254 * Don't bother calling schedule.
1259 hrtimer_cancel(&t
.timer
);
1261 /* Flag if a timeout occured */
1262 rem
= (t
.task
== NULL
);
1265 __set_current_state(TASK_RUNNING
);
1268 * NOTE: we don't remove ourselves from the waitqueue because
1269 * we are the only user of it.
1272 /* If we were woken (and unqueued), we succeeded, whatever. */
1273 if (!unqueue_me(&q
))
1279 * We expect signal_pending(current), but another thread may
1280 * have handled it for us already.
1283 return -ERESTARTSYS
;
1285 struct restart_block
*restart
;
1286 restart
= ¤t_thread_info()->restart_block
;
1287 restart
->fn
= futex_wait_restart
;
1288 restart
->arg0
= (unsigned long)uaddr
;
1289 restart
->arg1
= (unsigned long)val
;
1290 restart
->arg2
= (unsigned long)abs_time
;
1293 restart
->arg3
|= ARG3_SHARED
;
1294 return -ERESTART_RESTARTBLOCK
;
1297 out_unlock_release_sem
:
1298 queue_unlock(&q
, hb
);
1301 futex_unlock_mm(fshared
);
1306 static long futex_wait_restart(struct restart_block
*restart
)
1308 u32 __user
*uaddr
= (u32 __user
*)restart
->arg0
;
1309 u32 val
= (u32
)restart
->arg1
;
1310 ktime_t
*abs_time
= (ktime_t
*)restart
->arg2
;
1311 struct rw_semaphore
*fshared
= NULL
;
1313 restart
->fn
= do_no_restart_syscall
;
1314 if (restart
->arg3
& ARG3_SHARED
)
1315 fshared
= ¤t
->mm
->mmap_sem
;
1316 return (long)futex_wait(uaddr
, fshared
, val
, abs_time
);
1321 * Userspace tried a 0 -> TID atomic transition of the futex value
1322 * and failed. The kernel side here does the whole locking operation:
1323 * if there are waiters then it will block, it does PI, etc. (Due to
1324 * races the kernel might see a 0 value of the futex too.)
1326 static int futex_lock_pi(u32 __user
*uaddr
, struct rw_semaphore
*fshared
,
1327 int detect
, ktime_t
*time
, int trylock
)
1329 struct hrtimer_sleeper timeout
, *to
= NULL
;
1330 struct task_struct
*curr
= current
;
1331 struct futex_hash_bucket
*hb
;
1332 u32 uval
, newval
, curval
;
1334 int ret
, lock_taken
, ownerdied
= 0, attempt
= 0;
1336 if (refill_pi_state_cache())
1341 hrtimer_init(&to
->timer
, CLOCK_REALTIME
, HRTIMER_MODE_ABS
);
1342 hrtimer_init_sleeper(to
, current
);
1343 to
->timer
.expires
= *time
;
1348 futex_lock_mm(fshared
);
1350 ret
= get_futex_key(uaddr
, fshared
, &q
.key
);
1351 if (unlikely(ret
!= 0))
1352 goto out_release_sem
;
1355 hb
= queue_lock(&q
, -1, NULL
);
1358 ret
= lock_taken
= 0;
1361 * To avoid races, we attempt to take the lock here again
1362 * (by doing a 0 -> TID atomic cmpxchg), while holding all
1363 * the locks. It will most likely not succeed.
1365 newval
= current
->pid
;
1367 curval
= cmpxchg_futex_value_locked(uaddr
, 0, newval
);
1369 if (unlikely(curval
== -EFAULT
))
1373 * Detect deadlocks. In case of REQUEUE_PI this is a valid
1374 * situation and we return success to user space.
1376 if (unlikely((curval
& FUTEX_TID_MASK
) == current
->pid
)) {
1378 goto out_unlock_release_sem
;
1382 * Surprise - we got the lock. Just return to userspace:
1384 if (unlikely(!curval
))
1385 goto out_unlock_release_sem
;
1390 * Set the WAITERS flag, so the owner will know it has someone
1391 * to wake at next unlock
1393 newval
= curval
| FUTEX_WAITERS
;
1396 * There are two cases, where a futex might have no owner (the
1397 * owner TID is 0): OWNER_DIED. We take over the futex in this
1398 * case. We also do an unconditional take over, when the owner
1399 * of the futex died.
1401 * This is safe as we are protected by the hash bucket lock !
1403 if (unlikely(ownerdied
|| !(curval
& FUTEX_TID_MASK
))) {
1404 /* Keep the OWNER_DIED bit */
1405 newval
= (curval
& ~FUTEX_TID_MASK
) | current
->pid
;
1410 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
1412 if (unlikely(curval
== -EFAULT
))
1414 if (unlikely(curval
!= uval
))
1418 * We took the lock due to owner died take over.
1420 if (unlikely(lock_taken
))
1421 goto out_unlock_release_sem
;
1424 * We dont have the lock. Look up the PI state (or create it if
1425 * we are the first waiter):
1427 ret
= lookup_pi_state(uval
, hb
, &q
.key
, &q
.pi_state
);
1429 if (unlikely(ret
)) {
1434 * Task is exiting and we just wait for the
1437 queue_unlock(&q
, hb
);
1438 futex_unlock_mm(fshared
);
1444 * No owner found for this futex. Check if the
1445 * OWNER_DIED bit is set to figure out whether
1446 * this is a robust futex or not.
1448 if (get_futex_value_locked(&curval
, uaddr
))
1452 * We simply start over in case of a robust
1453 * futex. The code above will take the futex
1456 if (curval
& FUTEX_OWNER_DIED
) {
1461 goto out_unlock_release_sem
;
1466 * Only actually queue now that the atomic ops are done:
1471 * Now the futex is queued and we have checked the data, we
1472 * don't want to hold mmap_sem while we sleep.
1474 futex_unlock_mm(fshared
);
1476 WARN_ON(!q
.pi_state
);
1478 * Block on the PI mutex:
1481 ret
= rt_mutex_timed_lock(&q
.pi_state
->pi_mutex
, to
, 1);
1483 ret
= rt_mutex_trylock(&q
.pi_state
->pi_mutex
);
1484 /* Fixup the trylock return value: */
1485 ret
= ret
? 0 : -EWOULDBLOCK
;
1488 futex_lock_mm(fshared
);
1489 spin_lock(q
.lock_ptr
);
1493 * Got the lock. We might not be the anticipated owner
1494 * if we did a lock-steal - fix up the PI-state in
1497 if (q
.pi_state
->owner
!= curr
)
1498 ret
= fixup_pi_state_owner(uaddr
, &q
, curr
);
1501 * Catch the rare case, where the lock was released
1502 * when we were on the way back before we locked the
1505 if (q
.pi_state
->owner
== curr
&&
1506 rt_mutex_trylock(&q
.pi_state
->pi_mutex
)) {
1510 * Paranoia check. If we did not take the lock
1511 * in the trylock above, then we should not be
1512 * the owner of the rtmutex, neither the real
1513 * nor the pending one:
1515 if (rt_mutex_owner(&q
.pi_state
->pi_mutex
) == curr
)
1516 printk(KERN_ERR
"futex_lock_pi: ret = %d "
1517 "pi-mutex: %p pi-state %p\n", ret
,
1518 q
.pi_state
->pi_mutex
.owner
,
1523 /* Unqueue and drop the lock */
1525 futex_unlock_mm(fshared
);
1527 return ret
!= -EINTR
? ret
: -ERESTARTNOINTR
;
1529 out_unlock_release_sem
:
1530 queue_unlock(&q
, hb
);
1533 futex_unlock_mm(fshared
);
1538 * We have to r/w *(int __user *)uaddr, but we can't modify it
1539 * non-atomically. Therefore, if get_user below is not
1540 * enough, we need to handle the fault ourselves, while
1541 * still holding the mmap_sem.
1543 * ... and hb->lock. :-) --ANK
1545 queue_unlock(&q
, hb
);
1548 ret
= futex_handle_fault((unsigned long)uaddr
, fshared
,
1551 goto out_release_sem
;
1552 goto retry_unlocked
;
1555 futex_unlock_mm(fshared
);
1557 ret
= get_user(uval
, uaddr
);
1558 if (!ret
&& (uval
!= -EFAULT
))
1565 * Userspace attempted a TID -> 0 atomic transition, and failed.
1566 * This is the in-kernel slowpath: we look up the PI state (if any),
1567 * and do the rt-mutex unlock.
1569 static int futex_unlock_pi(u32 __user
*uaddr
, struct rw_semaphore
*fshared
)
1571 struct futex_hash_bucket
*hb
;
1572 struct futex_q
*this, *next
;
1574 struct plist_head
*head
;
1575 union futex_key key
;
1576 int ret
, attempt
= 0;
1579 if (get_user(uval
, uaddr
))
1582 * We release only a lock we actually own:
1584 if ((uval
& FUTEX_TID_MASK
) != current
->pid
)
1587 * First take all the futex related locks:
1589 futex_lock_mm(fshared
);
1591 ret
= get_futex_key(uaddr
, fshared
, &key
);
1592 if (unlikely(ret
!= 0))
1595 hb
= hash_futex(&key
);
1597 spin_lock(&hb
->lock
);
1600 * To avoid races, try to do the TID -> 0 atomic transition
1601 * again. If it succeeds then we can return without waking
1604 if (!(uval
& FUTEX_OWNER_DIED
))
1605 uval
= cmpxchg_futex_value_locked(uaddr
, current
->pid
, 0);
1608 if (unlikely(uval
== -EFAULT
))
1611 * Rare case: we managed to release the lock atomically,
1612 * no need to wake anyone else up:
1614 if (unlikely(uval
== current
->pid
))
1618 * Ok, other tasks may need to be woken up - check waiters
1619 * and do the wakeup if necessary:
1623 plist_for_each_entry_safe(this, next
, head
, list
) {
1624 if (!match_futex (&this->key
, &key
))
1626 ret
= wake_futex_pi(uaddr
, uval
, this);
1628 * The atomic access to the futex value
1629 * generated a pagefault, so retry the
1630 * user-access and the wakeup:
1637 * No waiters - kernel unlocks the futex:
1639 if (!(uval
& FUTEX_OWNER_DIED
)) {
1640 ret
= unlock_futex_pi(uaddr
, uval
);
1646 spin_unlock(&hb
->lock
);
1648 futex_unlock_mm(fshared
);
1654 * We have to r/w *(int __user *)uaddr, but we can't modify it
1655 * non-atomically. Therefore, if get_user below is not
1656 * enough, we need to handle the fault ourselves, while
1657 * still holding the mmap_sem.
1659 * ... and hb->lock. --ANK
1661 spin_unlock(&hb
->lock
);
1664 ret
= futex_handle_fault((unsigned long)uaddr
, fshared
,
1668 goto retry_unlocked
;
1671 futex_unlock_mm(fshared
);
1673 ret
= get_user(uval
, uaddr
);
1674 if (!ret
&& (uval
!= -EFAULT
))
1680 static int futex_close(struct inode
*inode
, struct file
*filp
)
1682 struct futex_q
*q
= filp
->private_data
;
1690 /* This is one-shot: once it's gone off you need a new fd */
1691 static unsigned int futex_poll(struct file
*filp
,
1692 struct poll_table_struct
*wait
)
1694 struct futex_q
*q
= filp
->private_data
;
1697 poll_wait(filp
, &q
->waiters
, wait
);
1700 * plist_node_empty() is safe here without any lock.
1701 * q->lock_ptr != 0 is not safe, because of ordering against wakeup.
1703 if (plist_node_empty(&q
->list
))
1704 ret
= POLLIN
| POLLRDNORM
;
1709 static const struct file_operations futex_fops
= {
1710 .release
= futex_close
,
1715 * Signal allows caller to avoid the race which would occur if they
1716 * set the sigio stuff up afterwards.
1718 static int futex_fd(u32 __user
*uaddr
, int signal
)
1723 struct rw_semaphore
*fshared
;
1724 static unsigned long printk_interval
;
1726 if (printk_timed_ratelimit(&printk_interval
, 60 * 60 * 1000)) {
1727 printk(KERN_WARNING
"Process `%s' used FUTEX_FD, which "
1728 "will be removed from the kernel in June 2007\n",
1733 if (!valid_signal(signal
))
1736 ret
= get_unused_fd();
1739 filp
= get_empty_filp();
1745 filp
->f_op
= &futex_fops
;
1746 filp
->f_path
.mnt
= mntget(futex_mnt
);
1747 filp
->f_path
.dentry
= dget(futex_mnt
->mnt_root
);
1748 filp
->f_mapping
= filp
->f_path
.dentry
->d_inode
->i_mapping
;
1751 err
= __f_setown(filp
, task_pid(current
), PIDTYPE_PID
, 1);
1755 filp
->f_owner
.signum
= signal
;
1758 q
= kmalloc(sizeof(*q
), GFP_KERNEL
);
1765 fshared
= ¤t
->mm
->mmap_sem
;
1767 err
= get_futex_key(uaddr
, fshared
, &q
->key
);
1769 if (unlikely(err
!= 0)) {
1776 * queue_me() must be called before releasing mmap_sem, because
1777 * key->shared.inode needs to be referenced while holding it.
1779 filp
->private_data
= q
;
1781 queue_me(q
, ret
, filp
);
1784 /* Now we map fd to filp, so userspace can access it */
1785 fd_install(ret
, filp
);
1796 * Support for robust futexes: the kernel cleans up held futexes at
1799 * Implementation: user-space maintains a per-thread list of locks it
1800 * is holding. Upon do_exit(), the kernel carefully walks this list,
1801 * and marks all locks that are owned by this thread with the
1802 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
1803 * always manipulated with the lock held, so the list is private and
1804 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
1805 * field, to allow the kernel to clean up if the thread dies after
1806 * acquiring the lock, but just before it could have added itself to
1807 * the list. There can only be one such pending lock.
1811 * sys_set_robust_list - set the robust-futex list head of a task
1812 * @head: pointer to the list-head
1813 * @len: length of the list-head, as userspace expects
1816 sys_set_robust_list(struct robust_list_head __user
*head
,
1820 * The kernel knows only one size for now:
1822 if (unlikely(len
!= sizeof(*head
)))
1825 current
->robust_list
= head
;
1831 * sys_get_robust_list - get the robust-futex list head of a task
1832 * @pid: pid of the process [zero for current task]
1833 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
1834 * @len_ptr: pointer to a length field, the kernel fills in the header size
1837 sys_get_robust_list(int pid
, struct robust_list_head __user
* __user
*head_ptr
,
1838 size_t __user
*len_ptr
)
1840 struct robust_list_head __user
*head
;
1844 head
= current
->robust_list
;
1846 struct task_struct
*p
;
1850 p
= find_task_by_pid(pid
);
1854 if ((current
->euid
!= p
->euid
) && (current
->euid
!= p
->uid
) &&
1855 !capable(CAP_SYS_PTRACE
))
1857 head
= p
->robust_list
;
1861 if (put_user(sizeof(*head
), len_ptr
))
1863 return put_user(head
, head_ptr
);
1872 * Process a futex-list entry, check whether it's owned by the
1873 * dying task, and do notification if so:
1875 int handle_futex_death(u32 __user
*uaddr
, struct task_struct
*curr
, int pi
)
1877 u32 uval
, nval
, mval
;
1880 if (get_user(uval
, uaddr
))
1883 if ((uval
& FUTEX_TID_MASK
) == curr
->pid
) {
1885 * Ok, this dying thread is truly holding a futex
1886 * of interest. Set the OWNER_DIED bit atomically
1887 * via cmpxchg, and if the value had FUTEX_WAITERS
1888 * set, wake up a waiter (if any). (We have to do a
1889 * futex_wake() even if OWNER_DIED is already set -
1890 * to handle the rare but possible case of recursive
1891 * thread-death.) The rest of the cleanup is done in
1894 mval
= (uval
& FUTEX_WAITERS
) | FUTEX_OWNER_DIED
;
1895 nval
= futex_atomic_cmpxchg_inatomic(uaddr
, uval
, mval
);
1897 if (nval
== -EFAULT
)
1904 * Wake robust non-PI futexes here. The wakeup of
1905 * PI futexes happens in exit_pi_state():
1907 if (!pi
&& (uval
& FUTEX_WAITERS
))
1908 futex_wake(uaddr
, &curr
->mm
->mmap_sem
, 1);
1914 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
1916 static inline int fetch_robust_entry(struct robust_list __user
**entry
,
1917 struct robust_list __user
* __user
*head
,
1920 unsigned long uentry
;
1922 if (get_user(uentry
, (unsigned long __user
*)head
))
1925 *entry
= (void __user
*)(uentry
& ~1UL);
1932 * Walk curr->robust_list (very carefully, it's a userspace list!)
1933 * and mark any locks found there dead, and notify any waiters.
1935 * We silently return on any sign of list-walking problem.
1937 void exit_robust_list(struct task_struct
*curr
)
1939 struct robust_list_head __user
*head
= curr
->robust_list
;
1940 struct robust_list __user
*entry
, *pending
;
1941 unsigned int limit
= ROBUST_LIST_LIMIT
, pi
, pip
;
1942 unsigned long futex_offset
;
1945 * Fetch the list head (which was registered earlier, via
1946 * sys_set_robust_list()):
1948 if (fetch_robust_entry(&entry
, &head
->list
.next
, &pi
))
1951 * Fetch the relative futex offset:
1953 if (get_user(futex_offset
, &head
->futex_offset
))
1956 * Fetch any possibly pending lock-add first, and handle it
1959 if (fetch_robust_entry(&pending
, &head
->list_op_pending
, &pip
))
1963 handle_futex_death((void __user
*)pending
+ futex_offset
,
1966 while (entry
!= &head
->list
) {
1968 * A pending lock might already be on the list, so
1969 * don't process it twice:
1971 if (entry
!= pending
)
1972 if (handle_futex_death((void __user
*)entry
+ futex_offset
,
1976 * Fetch the next entry in the list:
1978 if (fetch_robust_entry(&entry
, &entry
->next
, &pi
))
1981 * Avoid excessively long or circular lists:
1990 long do_futex(u32 __user
*uaddr
, int op
, u32 val
, ktime_t
*timeout
,
1991 u32 __user
*uaddr2
, u32 val2
, u32 val3
)
1994 int cmd
= op
& FUTEX_CMD_MASK
;
1995 struct rw_semaphore
*fshared
= NULL
;
1997 if (!(op
& FUTEX_PRIVATE_FLAG
))
1998 fshared
= ¤t
->mm
->mmap_sem
;
2002 ret
= futex_wait(uaddr
, fshared
, val
, timeout
);
2005 ret
= futex_wake(uaddr
, fshared
, val
);
2008 /* non-zero val means F_SETOWN(getpid()) & F_SETSIG(val) */
2009 ret
= futex_fd(uaddr
, val
);
2012 ret
= futex_requeue(uaddr
, fshared
, uaddr2
, val
, val2
, NULL
);
2014 case FUTEX_CMP_REQUEUE
:
2015 ret
= futex_requeue(uaddr
, fshared
, uaddr2
, val
, val2
, &val3
);
2018 ret
= futex_wake_op(uaddr
, fshared
, uaddr2
, val
, val2
, val3
);
2021 ret
= futex_lock_pi(uaddr
, fshared
, val
, timeout
, 0);
2023 case FUTEX_UNLOCK_PI
:
2024 ret
= futex_unlock_pi(uaddr
, fshared
);
2026 case FUTEX_TRYLOCK_PI
:
2027 ret
= futex_lock_pi(uaddr
, fshared
, 0, timeout
, 1);
2036 asmlinkage
long sys_futex(u32 __user
*uaddr
, int op
, u32 val
,
2037 struct timespec __user
*utime
, u32 __user
*uaddr2
,
2041 ktime_t t
, *tp
= NULL
;
2043 int cmd
= op
& FUTEX_CMD_MASK
;
2045 if (utime
&& (cmd
== FUTEX_WAIT
|| cmd
== FUTEX_LOCK_PI
)) {
2046 if (copy_from_user(&ts
, utime
, sizeof(ts
)) != 0)
2048 if (!timespec_valid(&ts
))
2051 t
= timespec_to_ktime(ts
);
2052 if (cmd
== FUTEX_WAIT
)
2053 t
= ktime_add(ktime_get(), t
);
2057 * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE.
2059 if (cmd
== FUTEX_REQUEUE
|| cmd
== FUTEX_CMP_REQUEUE
)
2060 val2
= (u32
) (unsigned long) utime
;
2062 return do_futex(uaddr
, op
, val
, tp
, uaddr2
, val2
, val3
);
2065 static int futexfs_get_sb(struct file_system_type
*fs_type
,
2066 int flags
, const char *dev_name
, void *data
,
2067 struct vfsmount
*mnt
)
2069 return get_sb_pseudo(fs_type
, "futex", NULL
, 0xBAD1DEA, mnt
);
2072 static struct file_system_type futex_fs_type
= {
2074 .get_sb
= futexfs_get_sb
,
2075 .kill_sb
= kill_anon_super
,
2078 static int __init
init(void)
2080 int i
= register_filesystem(&futex_fs_type
);
2085 futex_mnt
= kern_mount(&futex_fs_type
);
2086 if (IS_ERR(futex_mnt
)) {
2087 unregister_filesystem(&futex_fs_type
);
2088 return PTR_ERR(futex_mnt
);
2091 for (i
= 0; i
< ARRAY_SIZE(futex_queues
); i
++) {
2092 plist_head_init(&futex_queues
[i
].chain
, &futex_queues
[i
].lock
);
2093 spin_lock_init(&futex_queues
[i
].lock
);