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 * We hash on the keys returned from get_futex_key (see below).
126 static struct futex_hash_bucket
*hash_futex(union futex_key
*key
)
128 u32 hash
= jhash2((u32
*)&key
->both
.word
,
129 (sizeof(key
->both
.word
)+sizeof(key
->both
.ptr
))/4,
131 return &futex_queues
[hash
& ((1 << FUTEX_HASHBITS
)-1)];
135 * Return 1 if two futex_keys are equal, 0 otherwise.
137 static inline int match_futex(union futex_key
*key1
, union futex_key
*key2
)
139 return (key1
->both
.word
== key2
->both
.word
140 && key1
->both
.ptr
== key2
->both
.ptr
141 && key1
->both
.offset
== key2
->both
.offset
);
145 * get_futex_key - Get parameters which are the keys for a futex.
146 * @uaddr: virtual address of the futex
147 * @shared: NULL for a PROCESS_PRIVATE futex,
148 * ¤t->mm->mmap_sem for a PROCESS_SHARED futex
149 * @key: address where result is stored.
151 * Returns a negative error code or 0
152 * The key words are stored in *key on success.
154 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
155 * offset_within_page). For private mappings, it's (uaddr, current->mm).
156 * We can usually work out the index without swapping in the page.
158 * fshared is NULL for PROCESS_PRIVATE futexes
159 * For other futexes, it points to ¤t->mm->mmap_sem and
160 * caller must have taken the reader lock. but NOT any spinlocks.
162 int get_futex_key(u32 __user
*uaddr
, struct rw_semaphore
*fshared
,
163 union futex_key
*key
)
165 unsigned long address
= (unsigned long)uaddr
;
166 struct mm_struct
*mm
= current
->mm
;
167 struct vm_area_struct
*vma
;
172 * The futex address must be "naturally" aligned.
174 key
->both
.offset
= address
% PAGE_SIZE
;
175 if (unlikely((address
% sizeof(u32
)) != 0))
177 address
-= key
->both
.offset
;
180 * PROCESS_PRIVATE futexes are fast.
181 * As the mm cannot disappear under us and the 'key' only needs
182 * virtual address, we dont even have to find the underlying vma.
183 * Note : We do have to check 'uaddr' is a valid user address,
184 * but access_ok() should be faster than find_vma()
187 if (unlikely(!access_ok(VERIFY_WRITE
, uaddr
, sizeof(u32
))))
189 key
->private.mm
= mm
;
190 key
->private.address
= address
;
194 * The futex is hashed differently depending on whether
195 * it's in a shared or private mapping. So check vma first.
197 vma
= find_extend_vma(mm
, address
);
204 if (unlikely((vma
->vm_flags
& (VM_IO
|VM_READ
)) != VM_READ
))
205 return (vma
->vm_flags
& VM_IO
) ? -EPERM
: -EACCES
;
208 * Private mappings are handled in a simple way.
210 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
211 * it's a read-only handle, it's expected that futexes attach to
212 * the object not the particular process. Therefore we use
213 * VM_MAYSHARE here, not VM_SHARED which is restricted to shared
214 * mappings of _writable_ handles.
216 if (likely(!(vma
->vm_flags
& VM_MAYSHARE
))) {
217 key
->both
.offset
|= FUT_OFF_MMSHARED
; /* reference taken on mm */
218 key
->private.mm
= mm
;
219 key
->private.address
= address
;
224 * Linear file mappings are also simple.
226 key
->shared
.inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
227 key
->both
.offset
|= FUT_OFF_INODE
; /* inode-based key. */
228 if (likely(!(vma
->vm_flags
& VM_NONLINEAR
))) {
229 key
->shared
.pgoff
= (((address
- vma
->vm_start
) >> PAGE_SHIFT
)
235 * We could walk the page table to read the non-linear
236 * pte, and get the page index without fetching the page
237 * from swap. But that's a lot of code to duplicate here
238 * for a rare case, so we simply fetch the page.
240 err
= get_user_pages(current
, mm
, address
, 1, 0, 0, &page
, NULL
);
243 page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
249 EXPORT_SYMBOL_GPL(get_futex_key
);
252 * Take a reference to the resource addressed by a key.
253 * Can be called while holding spinlocks.
256 inline void get_futex_key_refs(union futex_key
*key
)
258 if (key
->both
.ptr
== 0)
260 switch (key
->both
.offset
& (FUT_OFF_INODE
|FUT_OFF_MMSHARED
)) {
262 atomic_inc(&key
->shared
.inode
->i_count
);
264 case FUT_OFF_MMSHARED
:
265 atomic_inc(&key
->private.mm
->mm_count
);
269 EXPORT_SYMBOL_GPL(get_futex_key_refs
);
272 * Drop a reference to the resource addressed by a key.
273 * The hash bucket spinlock must not be held.
275 void drop_futex_key_refs(union futex_key
*key
)
277 if (key
->both
.ptr
== 0)
279 switch (key
->both
.offset
& (FUT_OFF_INODE
|FUT_OFF_MMSHARED
)) {
281 iput(key
->shared
.inode
);
283 case FUT_OFF_MMSHARED
:
284 mmdrop(key
->private.mm
);
288 EXPORT_SYMBOL_GPL(drop_futex_key_refs
);
290 static inline int get_futex_value_locked(u32
*dest
, u32 __user
*from
)
295 ret
= __copy_from_user_inatomic(dest
, from
, sizeof(u32
));
298 return ret
? -EFAULT
: 0;
303 * if fshared is non NULL, current->mm->mmap_sem is already held
305 static int futex_handle_fault(unsigned long address
,
306 struct rw_semaphore
*fshared
, int attempt
)
308 struct vm_area_struct
* vma
;
309 struct mm_struct
*mm
= current
->mm
;
316 down_read(&mm
->mmap_sem
);
317 vma
= find_vma(mm
, address
);
318 if (vma
&& address
>= vma
->vm_start
&&
319 (vma
->vm_flags
& VM_WRITE
)) {
320 switch (handle_mm_fault(mm
, vma
, address
, 1)) {
332 up_read(&mm
->mmap_sem
);
339 static int refill_pi_state_cache(void)
341 struct futex_pi_state
*pi_state
;
343 if (likely(current
->pi_state_cache
))
346 pi_state
= kzalloc(sizeof(*pi_state
), GFP_KERNEL
);
351 INIT_LIST_HEAD(&pi_state
->list
);
352 /* pi_mutex gets initialized later */
353 pi_state
->owner
= NULL
;
354 atomic_set(&pi_state
->refcount
, 1);
356 current
->pi_state_cache
= pi_state
;
361 static struct futex_pi_state
* alloc_pi_state(void)
363 struct futex_pi_state
*pi_state
= current
->pi_state_cache
;
366 current
->pi_state_cache
= NULL
;
371 static void free_pi_state(struct futex_pi_state
*pi_state
)
373 if (!atomic_dec_and_test(&pi_state
->refcount
))
377 * If pi_state->owner is NULL, the owner is most probably dying
378 * and has cleaned up the pi_state already
380 if (pi_state
->owner
) {
381 spin_lock_irq(&pi_state
->owner
->pi_lock
);
382 list_del_init(&pi_state
->list
);
383 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
385 rt_mutex_proxy_unlock(&pi_state
->pi_mutex
, pi_state
->owner
);
388 if (current
->pi_state_cache
)
392 * pi_state->list is already empty.
393 * clear pi_state->owner.
394 * refcount is at 0 - put it back to 1.
396 pi_state
->owner
= NULL
;
397 atomic_set(&pi_state
->refcount
, 1);
398 current
->pi_state_cache
= pi_state
;
403 * Look up the task based on what TID userspace gave us.
406 static struct task_struct
* futex_find_get_task(pid_t pid
)
408 struct task_struct
*p
;
411 p
= find_task_by_pid(pid
);
414 if ((current
->euid
!= p
->euid
) && (current
->euid
!= p
->uid
)) {
426 * This task is holding PI mutexes at exit time => bad.
427 * Kernel cleans up PI-state, but userspace is likely hosed.
428 * (Robust-futex cleanup is separate and might save the day for userspace.)
430 void exit_pi_state_list(struct task_struct
*curr
)
432 struct list_head
*next
, *head
= &curr
->pi_state_list
;
433 struct futex_pi_state
*pi_state
;
434 struct futex_hash_bucket
*hb
;
438 * We are a ZOMBIE and nobody can enqueue itself on
439 * pi_state_list anymore, but we have to be careful
440 * versus waiters unqueueing themselves:
442 spin_lock_irq(&curr
->pi_lock
);
443 while (!list_empty(head
)) {
446 pi_state
= list_entry(next
, struct futex_pi_state
, list
);
448 hb
= hash_futex(&key
);
449 spin_unlock_irq(&curr
->pi_lock
);
451 spin_lock(&hb
->lock
);
453 spin_lock_irq(&curr
->pi_lock
);
455 * We dropped the pi-lock, so re-check whether this
456 * task still owns the PI-state:
458 if (head
->next
!= next
) {
459 spin_unlock(&hb
->lock
);
463 WARN_ON(pi_state
->owner
!= curr
);
464 WARN_ON(list_empty(&pi_state
->list
));
465 list_del_init(&pi_state
->list
);
466 pi_state
->owner
= NULL
;
467 spin_unlock_irq(&curr
->pi_lock
);
469 rt_mutex_unlock(&pi_state
->pi_mutex
);
471 spin_unlock(&hb
->lock
);
473 spin_lock_irq(&curr
->pi_lock
);
475 spin_unlock_irq(&curr
->pi_lock
);
479 lookup_pi_state(u32 uval
, struct futex_hash_bucket
*hb
,
480 union futex_key
*key
, struct futex_pi_state
**ps
)
482 struct futex_pi_state
*pi_state
= NULL
;
483 struct futex_q
*this, *next
;
484 struct plist_head
*head
;
485 struct task_struct
*p
;
486 pid_t pid
= uval
& FUTEX_TID_MASK
;
490 plist_for_each_entry_safe(this, next
, head
, list
) {
491 if (match_futex(&this->key
, key
)) {
493 * Another waiter already exists - bump up
494 * the refcount and return its pi_state:
496 pi_state
= this->pi_state
;
498 * Userspace might have messed up non PI and PI futexes
500 if (unlikely(!pi_state
))
503 WARN_ON(!atomic_read(&pi_state
->refcount
));
504 WARN_ON(pid
&& pi_state
->owner
&&
505 pi_state
->owner
->pid
!= pid
);
507 atomic_inc(&pi_state
->refcount
);
515 * We are the first waiter - try to look up the real owner and attach
516 * the new pi_state to it, but bail out when TID = 0
520 p
= futex_find_get_task(pid
);
525 * We need to look at the task state flags to figure out,
526 * whether the task is exiting. To protect against the do_exit
527 * change of the task flags, we do this protected by
530 spin_lock_irq(&p
->pi_lock
);
531 if (unlikely(p
->flags
& PF_EXITING
)) {
533 * The task is on the way out. When PF_EXITPIDONE is
534 * set, we know that the task has finished the
537 int ret
= (p
->flags
& PF_EXITPIDONE
) ? -ESRCH
: -EAGAIN
;
539 spin_unlock_irq(&p
->pi_lock
);
544 pi_state
= alloc_pi_state();
547 * Initialize the pi_mutex in locked state and make 'p'
550 rt_mutex_init_proxy_locked(&pi_state
->pi_mutex
, p
);
552 /* Store the key for possible exit cleanups: */
553 pi_state
->key
= *key
;
555 WARN_ON(!list_empty(&pi_state
->list
));
556 list_add(&pi_state
->list
, &p
->pi_state_list
);
558 spin_unlock_irq(&p
->pi_lock
);
568 * The hash bucket lock must be held when this is called.
569 * Afterwards, the futex_q must not be accessed.
571 static void wake_futex(struct futex_q
*q
)
573 plist_del(&q
->list
, &q
->list
.plist
);
575 send_sigio(&q
->filp
->f_owner
, q
->fd
, POLL_IN
);
577 * The lock in wake_up_all() is a crucial memory barrier after the
578 * plist_del() and also before assigning to q->lock_ptr.
580 wake_up_all(&q
->waiters
);
582 * The waiting task can free the futex_q as soon as this is written,
583 * without taking any locks. This must come last.
585 * A memory barrier is required here to prevent the following store
586 * to lock_ptr from getting ahead of the wakeup. Clearing the lock
587 * at the end of wake_up_all() does not prevent this store from
594 static int wake_futex_pi(u32 __user
*uaddr
, u32 uval
, struct futex_q
*this)
596 struct task_struct
*new_owner
;
597 struct futex_pi_state
*pi_state
= this->pi_state
;
603 spin_lock(&pi_state
->pi_mutex
.wait_lock
);
604 new_owner
= rt_mutex_next_owner(&pi_state
->pi_mutex
);
607 * This happens when we have stolen the lock and the original
608 * pending owner did not enqueue itself back on the rt_mutex.
609 * Thats not a tragedy. We know that way, that a lock waiter
610 * is on the fly. We make the futex_q waiter the pending owner.
613 new_owner
= this->task
;
616 * We pass it to the next owner. (The WAITERS bit is always
617 * kept enabled while there is PI state around. We must also
618 * preserve the owner died bit.)
620 if (!(uval
& FUTEX_OWNER_DIED
)) {
623 newval
= FUTEX_WAITERS
| new_owner
->pid
;
626 curval
= futex_atomic_cmpxchg_inatomic(uaddr
, uval
, newval
);
629 if (curval
== -EFAULT
)
634 spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
639 spin_lock_irq(&pi_state
->owner
->pi_lock
);
640 WARN_ON(list_empty(&pi_state
->list
));
641 list_del_init(&pi_state
->list
);
642 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
644 spin_lock_irq(&new_owner
->pi_lock
);
645 WARN_ON(!list_empty(&pi_state
->list
));
646 list_add(&pi_state
->list
, &new_owner
->pi_state_list
);
647 pi_state
->owner
= new_owner
;
648 spin_unlock_irq(&new_owner
->pi_lock
);
650 spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
651 rt_mutex_unlock(&pi_state
->pi_mutex
);
656 static int unlock_futex_pi(u32 __user
*uaddr
, u32 uval
)
661 * There is no waiter, so we unlock the futex. The owner died
662 * bit has not to be preserved here. We are the owner:
665 oldval
= futex_atomic_cmpxchg_inatomic(uaddr
, uval
, 0);
668 if (oldval
== -EFAULT
)
677 * Express the locking dependencies for lockdep:
680 double_lock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
683 spin_lock(&hb1
->lock
);
685 spin_lock_nested(&hb2
->lock
, SINGLE_DEPTH_NESTING
);
686 } else { /* hb1 > hb2 */
687 spin_lock(&hb2
->lock
);
688 spin_lock_nested(&hb1
->lock
, SINGLE_DEPTH_NESTING
);
693 * Wake up all waiters hashed on the physical page that is mapped
694 * to this virtual address:
696 static int futex_wake(u32 __user
*uaddr
, struct rw_semaphore
*fshared
,
699 struct futex_hash_bucket
*hb
;
700 struct futex_q
*this, *next
;
701 struct plist_head
*head
;
708 ret
= get_futex_key(uaddr
, fshared
, &key
);
709 if (unlikely(ret
!= 0))
712 hb
= hash_futex(&key
);
713 spin_lock(&hb
->lock
);
716 plist_for_each_entry_safe(this, next
, head
, list
) {
717 if (match_futex (&this->key
, &key
)) {
718 if (this->pi_state
) {
723 if (++ret
>= nr_wake
)
728 spin_unlock(&hb
->lock
);
736 * Wake up all waiters hashed on the physical page that is mapped
737 * to this virtual address:
740 futex_wake_op(u32 __user
*uaddr1
, struct rw_semaphore
*fshared
,
742 int nr_wake
, int nr_wake2
, int op
)
744 union futex_key key1
, key2
;
745 struct futex_hash_bucket
*hb1
, *hb2
;
746 struct plist_head
*head
;
747 struct futex_q
*this, *next
;
748 int ret
, op_ret
, attempt
= 0;
754 ret
= get_futex_key(uaddr1
, fshared
, &key1
);
755 if (unlikely(ret
!= 0))
757 ret
= get_futex_key(uaddr2
, fshared
, &key2
);
758 if (unlikely(ret
!= 0))
761 hb1
= hash_futex(&key1
);
762 hb2
= hash_futex(&key2
);
765 double_lock_hb(hb1
, hb2
);
767 op_ret
= futex_atomic_op_inuser(op
, uaddr2
);
768 if (unlikely(op_ret
< 0)) {
771 spin_unlock(&hb1
->lock
);
773 spin_unlock(&hb2
->lock
);
777 * we don't get EFAULT from MMU faults if we don't have an MMU,
778 * but we might get them from range checking
784 if (unlikely(op_ret
!= -EFAULT
)) {
790 * futex_atomic_op_inuser needs to both read and write
791 * *(int __user *)uaddr2, but we can't modify it
792 * non-atomically. Therefore, if get_user below is not
793 * enough, we need to handle the fault ourselves, while
794 * still holding the mmap_sem.
797 ret
= futex_handle_fault((unsigned long)uaddr2
,
805 * If we would have faulted, release mmap_sem,
806 * fault it in and start all over again.
811 ret
= get_user(dummy
, uaddr2
);
820 plist_for_each_entry_safe(this, next
, head
, list
) {
821 if (match_futex (&this->key
, &key1
)) {
823 if (++ret
>= nr_wake
)
832 plist_for_each_entry_safe(this, next
, head
, list
) {
833 if (match_futex (&this->key
, &key2
)) {
835 if (++op_ret
>= nr_wake2
)
842 spin_unlock(&hb1
->lock
);
844 spin_unlock(&hb2
->lock
);
852 * Requeue all waiters hashed on one physical page to another
855 static int futex_requeue(u32 __user
*uaddr1
, struct rw_semaphore
*fshared
,
857 int nr_wake
, int nr_requeue
, u32
*cmpval
)
859 union futex_key key1
, key2
;
860 struct futex_hash_bucket
*hb1
, *hb2
;
861 struct plist_head
*head1
;
862 struct futex_q
*this, *next
;
863 int ret
, drop_count
= 0;
869 ret
= get_futex_key(uaddr1
, fshared
, &key1
);
870 if (unlikely(ret
!= 0))
872 ret
= get_futex_key(uaddr2
, fshared
, &key2
);
873 if (unlikely(ret
!= 0))
876 hb1
= hash_futex(&key1
);
877 hb2
= hash_futex(&key2
);
879 double_lock_hb(hb1
, hb2
);
881 if (likely(cmpval
!= NULL
)) {
884 ret
= get_futex_value_locked(&curval
, uaddr1
);
887 spin_unlock(&hb1
->lock
);
889 spin_unlock(&hb2
->lock
);
892 * If we would have faulted, release mmap_sem, fault
893 * it in and start all over again.
898 ret
= get_user(curval
, uaddr1
);
905 if (curval
!= *cmpval
) {
912 plist_for_each_entry_safe(this, next
, head1
, list
) {
913 if (!match_futex (&this->key
, &key1
))
915 if (++ret
<= nr_wake
) {
919 * If key1 and key2 hash to the same bucket, no need to
922 if (likely(head1
!= &hb2
->chain
)) {
923 plist_del(&this->list
, &hb1
->chain
);
924 plist_add(&this->list
, &hb2
->chain
);
925 this->lock_ptr
= &hb2
->lock
;
926 #ifdef CONFIG_DEBUG_PI_LIST
927 this->list
.plist
.lock
= &hb2
->lock
;
931 get_futex_key_refs(&key2
);
934 if (ret
- nr_wake
>= nr_requeue
)
940 spin_unlock(&hb1
->lock
);
942 spin_unlock(&hb2
->lock
);
944 /* drop_futex_key_refs() must be called outside the spinlocks. */
945 while (--drop_count
>= 0)
946 drop_futex_key_refs(&key1
);
954 /* The key must be already stored in q->key. */
955 static inline struct futex_hash_bucket
*
956 queue_lock(struct futex_q
*q
, int fd
, struct file
*filp
)
958 struct futex_hash_bucket
*hb
;
963 init_waitqueue_head(&q
->waiters
);
965 get_futex_key_refs(&q
->key
);
966 hb
= hash_futex(&q
->key
);
967 q
->lock_ptr
= &hb
->lock
;
969 spin_lock(&hb
->lock
);
973 static inline void __queue_me(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
978 * The priority used to register this element is
979 * - either the real thread-priority for the real-time threads
980 * (i.e. threads with a priority lower than MAX_RT_PRIO)
981 * - or MAX_RT_PRIO for non-RT threads.
982 * Thus, all RT-threads are woken first in priority order, and
983 * the others are woken last, in FIFO order.
985 prio
= min(current
->normal_prio
, MAX_RT_PRIO
);
987 plist_node_init(&q
->list
, prio
);
988 #ifdef CONFIG_DEBUG_PI_LIST
989 q
->list
.plist
.lock
= &hb
->lock
;
991 plist_add(&q
->list
, &hb
->chain
);
993 spin_unlock(&hb
->lock
);
997 queue_unlock(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
999 spin_unlock(&hb
->lock
);
1000 drop_futex_key_refs(&q
->key
);
1004 * queue_me and unqueue_me must be called as a pair, each
1005 * exactly once. They are called with the hashed spinlock held.
1008 /* The key must be already stored in q->key. */
1009 static void queue_me(struct futex_q
*q
, int fd
, struct file
*filp
)
1011 struct futex_hash_bucket
*hb
;
1013 hb
= queue_lock(q
, fd
, filp
);
1017 /* Return 1 if we were still queued (ie. 0 means we were woken) */
1018 static int unqueue_me(struct futex_q
*q
)
1020 spinlock_t
*lock_ptr
;
1023 /* In the common case we don't take the spinlock, which is nice. */
1025 lock_ptr
= q
->lock_ptr
;
1027 if (lock_ptr
!= 0) {
1028 spin_lock(lock_ptr
);
1030 * q->lock_ptr can change between reading it and
1031 * spin_lock(), causing us to take the wrong lock. This
1032 * corrects the race condition.
1034 * Reasoning goes like this: if we have the wrong lock,
1035 * q->lock_ptr must have changed (maybe several times)
1036 * between reading it and the spin_lock(). It can
1037 * change again after the spin_lock() but only if it was
1038 * already changed before the spin_lock(). It cannot,
1039 * however, change back to the original value. Therefore
1040 * we can detect whether we acquired the correct lock.
1042 if (unlikely(lock_ptr
!= q
->lock_ptr
)) {
1043 spin_unlock(lock_ptr
);
1046 WARN_ON(plist_node_empty(&q
->list
));
1047 plist_del(&q
->list
, &q
->list
.plist
);
1049 BUG_ON(q
->pi_state
);
1051 spin_unlock(lock_ptr
);
1055 drop_futex_key_refs(&q
->key
);
1060 * PI futexes can not be requeued and must remove themself from the
1061 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1064 static void unqueue_me_pi(struct futex_q
*q
)
1066 WARN_ON(plist_node_empty(&q
->list
));
1067 plist_del(&q
->list
, &q
->list
.plist
);
1069 BUG_ON(!q
->pi_state
);
1070 free_pi_state(q
->pi_state
);
1073 spin_unlock(q
->lock_ptr
);
1075 drop_futex_key_refs(&q
->key
);
1079 * Fixup the pi_state owner with current.
1081 * Must be called with hash bucket lock held and mm->sem held for non
1084 static int fixup_pi_state_owner(u32 __user
*uaddr
, struct futex_q
*q
,
1085 struct task_struct
*curr
)
1087 u32 newtid
= curr
->pid
| FUTEX_WAITERS
;
1088 struct futex_pi_state
*pi_state
= q
->pi_state
;
1089 u32 uval
, curval
, newval
;
1093 if (pi_state
->owner
!= NULL
) {
1094 spin_lock_irq(&pi_state
->owner
->pi_lock
);
1095 WARN_ON(list_empty(&pi_state
->list
));
1096 list_del_init(&pi_state
->list
);
1097 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
1099 newtid
|= FUTEX_OWNER_DIED
;
1101 pi_state
->owner
= curr
;
1103 spin_lock_irq(&curr
->pi_lock
);
1104 WARN_ON(!list_empty(&pi_state
->list
));
1105 list_add(&pi_state
->list
, &curr
->pi_state_list
);
1106 spin_unlock_irq(&curr
->pi_lock
);
1109 * We own it, so we have to replace the pending owner
1110 * TID. This must be atomic as we have preserve the
1111 * owner died bit here.
1113 ret
= get_futex_value_locked(&uval
, uaddr
);
1116 newval
= (uval
& FUTEX_OWNER_DIED
) | newtid
;
1118 pagefault_disable();
1119 curval
= futex_atomic_cmpxchg_inatomic(uaddr
,
1123 if (curval
== -EFAULT
)
1133 * In case we must use restart_block to restart a futex_wait,
1134 * we encode in the 'arg3' shared capability
1136 #define ARG3_SHARED 1
1138 static long futex_wait_restart(struct restart_block
*restart
);
1139 static int futex_wait(u32 __user
*uaddr
, struct rw_semaphore
*fshared
,
1140 u32 val
, ktime_t
*abs_time
)
1142 struct task_struct
*curr
= current
;
1143 DECLARE_WAITQUEUE(wait
, curr
);
1144 struct futex_hash_bucket
*hb
;
1148 struct hrtimer_sleeper t
;
1156 ret
= get_futex_key(uaddr
, fshared
, &q
.key
);
1157 if (unlikely(ret
!= 0))
1158 goto out_release_sem
;
1160 hb
= queue_lock(&q
, -1, NULL
);
1163 * Access the page AFTER the futex is queued.
1164 * Order is important:
1166 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1167 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1169 * The basic logical guarantee of a futex is that it blocks ONLY
1170 * if cond(var) is known to be true at the time of blocking, for
1171 * any cond. If we queued after testing *uaddr, that would open
1172 * a race condition where we could block indefinitely with
1173 * cond(var) false, which would violate the guarantee.
1175 * A consequence is that futex_wait() can return zero and absorb
1176 * a wakeup when *uaddr != val on entry to the syscall. This is
1179 * for shared futexes, we hold the mmap semaphore, so the mapping
1180 * cannot have changed since we looked it up in get_futex_key.
1182 ret
= get_futex_value_locked(&uval
, uaddr
);
1184 if (unlikely(ret
)) {
1185 queue_unlock(&q
, hb
);
1188 * If we would have faulted, release mmap_sem, fault it in and
1189 * start all over again.
1194 ret
= get_user(uval
, uaddr
);
1202 goto out_unlock_release_sem
;
1204 /* Only actually queue if *uaddr contained val. */
1208 * Now the futex is queued and we have checked the data, we
1209 * don't want to hold mmap_sem while we sleep.
1215 * There might have been scheduling since the queue_me(), as we
1216 * cannot hold a spinlock across the get_user() in case it
1217 * faults, and we cannot just set TASK_INTERRUPTIBLE state when
1218 * queueing ourselves into the futex hash. This code thus has to
1219 * rely on the futex_wake() code removing us from hash when it
1223 /* add_wait_queue is the barrier after __set_current_state. */
1224 __set_current_state(TASK_INTERRUPTIBLE
);
1225 add_wait_queue(&q
.waiters
, &wait
);
1227 * !plist_node_empty() is safe here without any lock.
1228 * q.lock_ptr != 0 is not safe, because of ordering against wakeup.
1230 if (likely(!plist_node_empty(&q
.list
))) {
1234 hrtimer_init(&t
.timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_ABS
);
1235 hrtimer_init_sleeper(&t
, current
);
1236 t
.timer
.expires
= *abs_time
;
1238 hrtimer_start(&t
.timer
, t
.timer
.expires
, HRTIMER_MODE_ABS
);
1241 * the timer could have already expired, in which
1242 * case current would be flagged for rescheduling.
1243 * Don't bother calling schedule.
1248 hrtimer_cancel(&t
.timer
);
1250 /* Flag if a timeout occured */
1251 rem
= (t
.task
== NULL
);
1254 __set_current_state(TASK_RUNNING
);
1257 * NOTE: we don't remove ourselves from the waitqueue because
1258 * we are the only user of it.
1261 /* If we were woken (and unqueued), we succeeded, whatever. */
1262 if (!unqueue_me(&q
))
1268 * We expect signal_pending(current), but another thread may
1269 * have handled it for us already.
1272 return -ERESTARTSYS
;
1274 struct restart_block
*restart
;
1275 restart
= ¤t_thread_info()->restart_block
;
1276 restart
->fn
= futex_wait_restart
;
1277 restart
->arg0
= (unsigned long)uaddr
;
1278 restart
->arg1
= (unsigned long)val
;
1279 restart
->arg2
= (unsigned long)abs_time
;
1282 restart
->arg3
|= ARG3_SHARED
;
1283 return -ERESTART_RESTARTBLOCK
;
1286 out_unlock_release_sem
:
1287 queue_unlock(&q
, hb
);
1296 static long futex_wait_restart(struct restart_block
*restart
)
1298 u32 __user
*uaddr
= (u32 __user
*)restart
->arg0
;
1299 u32 val
= (u32
)restart
->arg1
;
1300 ktime_t
*abs_time
= (ktime_t
*)restart
->arg2
;
1301 struct rw_semaphore
*fshared
= NULL
;
1303 restart
->fn
= do_no_restart_syscall
;
1304 if (restart
->arg3
& ARG3_SHARED
)
1305 fshared
= ¤t
->mm
->mmap_sem
;
1306 return (long)futex_wait(uaddr
, fshared
, val
, abs_time
);
1311 * Userspace tried a 0 -> TID atomic transition of the futex value
1312 * and failed. The kernel side here does the whole locking operation:
1313 * if there are waiters then it will block, it does PI, etc. (Due to
1314 * races the kernel might see a 0 value of the futex too.)
1316 static int futex_lock_pi(u32 __user
*uaddr
, struct rw_semaphore
*fshared
,
1317 int detect
, ktime_t
*time
, int trylock
)
1319 struct hrtimer_sleeper timeout
, *to
= NULL
;
1320 struct task_struct
*curr
= current
;
1321 struct futex_hash_bucket
*hb
;
1322 u32 uval
, newval
, curval
;
1324 int ret
, lock_taken
, ownerdied
= 0, attempt
= 0;
1326 if (refill_pi_state_cache())
1331 hrtimer_init(&to
->timer
, CLOCK_REALTIME
, HRTIMER_MODE_ABS
);
1332 hrtimer_init_sleeper(to
, current
);
1333 to
->timer
.expires
= *time
;
1341 ret
= get_futex_key(uaddr
, fshared
, &q
.key
);
1342 if (unlikely(ret
!= 0))
1343 goto out_release_sem
;
1346 hb
= queue_lock(&q
, -1, NULL
);
1349 ret
= lock_taken
= 0;
1352 * To avoid races, we attempt to take the lock here again
1353 * (by doing a 0 -> TID atomic cmpxchg), while holding all
1354 * the locks. It will most likely not succeed.
1356 newval
= current
->pid
;
1358 pagefault_disable();
1359 curval
= futex_atomic_cmpxchg_inatomic(uaddr
, 0, newval
);
1362 if (unlikely(curval
== -EFAULT
))
1366 * Detect deadlocks. In case of REQUEUE_PI this is a valid
1367 * situation and we return success to user space.
1369 if (unlikely((curval
& FUTEX_TID_MASK
) == current
->pid
)) {
1371 goto out_unlock_release_sem
;
1375 * Surprise - we got the lock. Just return to userspace:
1377 if (unlikely(!curval
))
1378 goto out_unlock_release_sem
;
1383 * Set the WAITERS flag, so the owner will know it has someone
1384 * to wake at next unlock
1386 newval
= curval
| FUTEX_WAITERS
;
1389 * There are two cases, where a futex might have no owner (the
1390 * owner TID is 0): OWNER_DIED. We take over the futex in this
1391 * case. We also do an unconditional take over, when the owner
1392 * of the futex died.
1394 * This is safe as we are protected by the hash bucket lock !
1396 if (unlikely(ownerdied
|| !(curval
& FUTEX_TID_MASK
))) {
1397 /* Keep the OWNER_DIED bit */
1398 newval
= (curval
& ~FUTEX_TID_MASK
) | current
->pid
;
1403 pagefault_disable();
1404 curval
= futex_atomic_cmpxchg_inatomic(uaddr
, uval
, newval
);
1407 if (unlikely(curval
== -EFAULT
))
1409 if (unlikely(curval
!= uval
))
1413 * We took the lock due to owner died take over.
1415 if (unlikely(lock_taken
))
1416 goto out_unlock_release_sem
;
1419 * We dont have the lock. Look up the PI state (or create it if
1420 * we are the first waiter):
1422 ret
= lookup_pi_state(uval
, hb
, &q
.key
, &q
.pi_state
);
1424 if (unlikely(ret
)) {
1429 * Task is exiting and we just wait for the
1432 queue_unlock(&q
, hb
);
1440 * No owner found for this futex. Check if the
1441 * OWNER_DIED bit is set to figure out whether
1442 * this is a robust futex or not.
1444 if (get_futex_value_locked(&curval
, uaddr
))
1448 * We simply start over in case of a robust
1449 * futex. The code above will take the futex
1452 if (curval
& FUTEX_OWNER_DIED
) {
1457 goto out_unlock_release_sem
;
1462 * Only actually queue now that the atomic ops are done:
1467 * Now the futex is queued and we have checked the data, we
1468 * don't want to hold mmap_sem while we sleep.
1473 WARN_ON(!q
.pi_state
);
1475 * Block on the PI mutex:
1478 ret
= rt_mutex_timed_lock(&q
.pi_state
->pi_mutex
, to
, 1);
1480 ret
= rt_mutex_trylock(&q
.pi_state
->pi_mutex
);
1481 /* Fixup the trylock return value: */
1482 ret
= ret
? 0 : -EWOULDBLOCK
;
1487 spin_lock(q
.lock_ptr
);
1491 * Got the lock. We might not be the anticipated owner
1492 * if we did a lock-steal - fix up the PI-state in
1495 if (q
.pi_state
->owner
!= curr
)
1496 ret
= fixup_pi_state_owner(uaddr
, &q
, curr
);
1499 * Catch the rare case, where the lock was released
1500 * when we were on the way back before we locked the
1503 if (q
.pi_state
->owner
== curr
&&
1504 rt_mutex_trylock(&q
.pi_state
->pi_mutex
)) {
1508 * Paranoia check. If we did not take the lock
1509 * in the trylock above, then we should not be
1510 * the owner of the rtmutex, neither the real
1511 * nor the pending one:
1513 if (rt_mutex_owner(&q
.pi_state
->pi_mutex
) == curr
)
1514 printk(KERN_ERR
"futex_lock_pi: ret = %d "
1515 "pi-mutex: %p pi-state %p\n", ret
,
1516 q
.pi_state
->pi_mutex
.owner
,
1521 /* Unqueue and drop the lock */
1526 return ret
!= -EINTR
? ret
: -ERESTARTNOINTR
;
1528 out_unlock_release_sem
:
1529 queue_unlock(&q
, hb
);
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
;
1558 ret
= get_user(uval
, uaddr
);
1559 if (!ret
&& (uval
!= -EFAULT
))
1566 * Userspace attempted a TID -> 0 atomic transition, and failed.
1567 * This is the in-kernel slowpath: we look up the PI state (if any),
1568 * and do the rt-mutex unlock.
1570 static int futex_unlock_pi(u32 __user
*uaddr
, struct rw_semaphore
*fshared
)
1572 struct futex_hash_bucket
*hb
;
1573 struct futex_q
*this, *next
;
1575 struct plist_head
*head
;
1576 union futex_key key
;
1577 int ret
, attempt
= 0;
1580 if (get_user(uval
, uaddr
))
1583 * We release only a lock we actually own:
1585 if ((uval
& FUTEX_TID_MASK
) != current
->pid
)
1588 * First take all the futex related locks:
1593 ret
= get_futex_key(uaddr
, fshared
, &key
);
1594 if (unlikely(ret
!= 0))
1597 hb
= hash_futex(&key
);
1599 spin_lock(&hb
->lock
);
1602 * To avoid races, try to do the TID -> 0 atomic transition
1603 * again. If it succeeds then we can return without waking
1606 if (!(uval
& FUTEX_OWNER_DIED
)) {
1607 pagefault_disable();
1608 uval
= futex_atomic_cmpxchg_inatomic(uaddr
, current
->pid
, 0);
1612 if (unlikely(uval
== -EFAULT
))
1615 * Rare case: we managed to release the lock atomically,
1616 * no need to wake anyone else up:
1618 if (unlikely(uval
== current
->pid
))
1622 * Ok, other tasks may need to be woken up - check waiters
1623 * and do the wakeup if necessary:
1627 plist_for_each_entry_safe(this, next
, head
, list
) {
1628 if (!match_futex (&this->key
, &key
))
1630 ret
= wake_futex_pi(uaddr
, uval
, this);
1632 * The atomic access to the futex value
1633 * generated a pagefault, so retry the
1634 * user-access and the wakeup:
1641 * No waiters - kernel unlocks the futex:
1643 if (!(uval
& FUTEX_OWNER_DIED
)) {
1644 ret
= unlock_futex_pi(uaddr
, uval
);
1650 spin_unlock(&hb
->lock
);
1659 * We have to r/w *(int __user *)uaddr, but we can't modify it
1660 * non-atomically. Therefore, if get_user below is not
1661 * enough, we need to handle the fault ourselves, while
1662 * still holding the mmap_sem.
1664 * ... and hb->lock. --ANK
1666 spin_unlock(&hb
->lock
);
1669 ret
= futex_handle_fault((unsigned long)uaddr
, fshared
,
1673 goto retry_unlocked
;
1679 ret
= get_user(uval
, uaddr
);
1680 if (!ret
&& (uval
!= -EFAULT
))
1686 static int futex_close(struct inode
*inode
, struct file
*filp
)
1688 struct futex_q
*q
= filp
->private_data
;
1696 /* This is one-shot: once it's gone off you need a new fd */
1697 static unsigned int futex_poll(struct file
*filp
,
1698 struct poll_table_struct
*wait
)
1700 struct futex_q
*q
= filp
->private_data
;
1703 poll_wait(filp
, &q
->waiters
, wait
);
1706 * plist_node_empty() is safe here without any lock.
1707 * q->lock_ptr != 0 is not safe, because of ordering against wakeup.
1709 if (plist_node_empty(&q
->list
))
1710 ret
= POLLIN
| POLLRDNORM
;
1715 static const struct file_operations futex_fops
= {
1716 .release
= futex_close
,
1721 * Signal allows caller to avoid the race which would occur if they
1722 * set the sigio stuff up afterwards.
1724 static int futex_fd(u32 __user
*uaddr
, int signal
)
1729 struct rw_semaphore
*fshared
;
1730 static unsigned long printk_interval
;
1732 if (printk_timed_ratelimit(&printk_interval
, 60 * 60 * 1000)) {
1733 printk(KERN_WARNING
"Process `%s' used FUTEX_FD, which "
1734 "will be removed from the kernel in June 2007\n",
1739 if (!valid_signal(signal
))
1742 ret
= get_unused_fd();
1745 filp
= get_empty_filp();
1751 filp
->f_op
= &futex_fops
;
1752 filp
->f_path
.mnt
= mntget(futex_mnt
);
1753 filp
->f_path
.dentry
= dget(futex_mnt
->mnt_root
);
1754 filp
->f_mapping
= filp
->f_path
.dentry
->d_inode
->i_mapping
;
1757 err
= __f_setown(filp
, task_pid(current
), PIDTYPE_PID
, 1);
1761 filp
->f_owner
.signum
= signal
;
1764 q
= kmalloc(sizeof(*q
), GFP_KERNEL
);
1771 fshared
= ¤t
->mm
->mmap_sem
;
1773 err
= get_futex_key(uaddr
, fshared
, &q
->key
);
1775 if (unlikely(err
!= 0)) {
1782 * queue_me() must be called before releasing mmap_sem, because
1783 * key->shared.inode needs to be referenced while holding it.
1785 filp
->private_data
= q
;
1787 queue_me(q
, ret
, filp
);
1790 /* Now we map fd to filp, so userspace can access it */
1791 fd_install(ret
, filp
);
1802 * Support for robust futexes: the kernel cleans up held futexes at
1805 * Implementation: user-space maintains a per-thread list of locks it
1806 * is holding. Upon do_exit(), the kernel carefully walks this list,
1807 * and marks all locks that are owned by this thread with the
1808 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
1809 * always manipulated with the lock held, so the list is private and
1810 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
1811 * field, to allow the kernel to clean up if the thread dies after
1812 * acquiring the lock, but just before it could have added itself to
1813 * the list. There can only be one such pending lock.
1817 * sys_set_robust_list - set the robust-futex list head of a task
1818 * @head: pointer to the list-head
1819 * @len: length of the list-head, as userspace expects
1822 sys_set_robust_list(struct robust_list_head __user
*head
,
1826 * The kernel knows only one size for now:
1828 if (unlikely(len
!= sizeof(*head
)))
1831 current
->robust_list
= head
;
1837 * sys_get_robust_list - get the robust-futex list head of a task
1838 * @pid: pid of the process [zero for current task]
1839 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
1840 * @len_ptr: pointer to a length field, the kernel fills in the header size
1843 sys_get_robust_list(int pid
, struct robust_list_head __user
* __user
*head_ptr
,
1844 size_t __user
*len_ptr
)
1846 struct robust_list_head __user
*head
;
1850 head
= current
->robust_list
;
1852 struct task_struct
*p
;
1856 p
= find_task_by_pid(pid
);
1860 if ((current
->euid
!= p
->euid
) && (current
->euid
!= p
->uid
) &&
1861 !capable(CAP_SYS_PTRACE
))
1863 head
= p
->robust_list
;
1867 if (put_user(sizeof(*head
), len_ptr
))
1869 return put_user(head
, head_ptr
);
1878 * Process a futex-list entry, check whether it's owned by the
1879 * dying task, and do notification if so:
1881 int handle_futex_death(u32 __user
*uaddr
, struct task_struct
*curr
, int pi
)
1883 u32 uval
, nval
, mval
;
1886 if (get_user(uval
, uaddr
))
1889 if ((uval
& FUTEX_TID_MASK
) == curr
->pid
) {
1891 * Ok, this dying thread is truly holding a futex
1892 * of interest. Set the OWNER_DIED bit atomically
1893 * via cmpxchg, and if the value had FUTEX_WAITERS
1894 * set, wake up a waiter (if any). (We have to do a
1895 * futex_wake() even if OWNER_DIED is already set -
1896 * to handle the rare but possible case of recursive
1897 * thread-death.) The rest of the cleanup is done in
1900 mval
= (uval
& FUTEX_WAITERS
) | FUTEX_OWNER_DIED
;
1901 nval
= futex_atomic_cmpxchg_inatomic(uaddr
, uval
, mval
);
1903 if (nval
== -EFAULT
)
1910 * Wake robust non-PI futexes here. The wakeup of
1911 * PI futexes happens in exit_pi_state():
1914 if (uval
& FUTEX_WAITERS
)
1915 futex_wake(uaddr
, &curr
->mm
->mmap_sem
, 1);
1922 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
1924 static inline int fetch_robust_entry(struct robust_list __user
**entry
,
1925 struct robust_list __user
* __user
*head
,
1928 unsigned long uentry
;
1930 if (get_user(uentry
, (unsigned long __user
*)head
))
1933 *entry
= (void __user
*)(uentry
& ~1UL);
1940 * Walk curr->robust_list (very carefully, it's a userspace list!)
1941 * and mark any locks found there dead, and notify any waiters.
1943 * We silently return on any sign of list-walking problem.
1945 void exit_robust_list(struct task_struct
*curr
)
1947 struct robust_list_head __user
*head
= curr
->robust_list
;
1948 struct robust_list __user
*entry
, *pending
;
1949 unsigned int limit
= ROBUST_LIST_LIMIT
, pi
, pip
;
1950 unsigned long futex_offset
;
1953 * Fetch the list head (which was registered earlier, via
1954 * sys_set_robust_list()):
1956 if (fetch_robust_entry(&entry
, &head
->list
.next
, &pi
))
1959 * Fetch the relative futex offset:
1961 if (get_user(futex_offset
, &head
->futex_offset
))
1964 * Fetch any possibly pending lock-add first, and handle it
1967 if (fetch_robust_entry(&pending
, &head
->list_op_pending
, &pip
))
1971 handle_futex_death((void __user
*)pending
+ futex_offset
,
1974 while (entry
!= &head
->list
) {
1976 * A pending lock might already be on the list, so
1977 * don't process it twice:
1979 if (entry
!= pending
)
1980 if (handle_futex_death((void __user
*)entry
+ futex_offset
,
1984 * Fetch the next entry in the list:
1986 if (fetch_robust_entry(&entry
, &entry
->next
, &pi
))
1989 * Avoid excessively long or circular lists:
1998 long do_futex(u32 __user
*uaddr
, int op
, u32 val
, ktime_t
*timeout
,
1999 u32 __user
*uaddr2
, u32 val2
, u32 val3
)
2002 int cmd
= op
& FUTEX_CMD_MASK
;
2003 struct rw_semaphore
*fshared
= NULL
;
2005 if (!(op
& FUTEX_PRIVATE_FLAG
))
2006 fshared
= ¤t
->mm
->mmap_sem
;
2010 ret
= futex_wait(uaddr
, fshared
, val
, timeout
);
2013 ret
= futex_wake(uaddr
, fshared
, val
);
2016 /* non-zero val means F_SETOWN(getpid()) & F_SETSIG(val) */
2017 ret
= futex_fd(uaddr
, val
);
2020 ret
= futex_requeue(uaddr
, fshared
, uaddr2
, val
, val2
, NULL
);
2022 case FUTEX_CMP_REQUEUE
:
2023 ret
= futex_requeue(uaddr
, fshared
, uaddr2
, val
, val2
, &val3
);
2026 ret
= futex_wake_op(uaddr
, fshared
, uaddr2
, val
, val2
, val3
);
2029 ret
= futex_lock_pi(uaddr
, fshared
, val
, timeout
, 0);
2031 case FUTEX_UNLOCK_PI
:
2032 ret
= futex_unlock_pi(uaddr
, fshared
);
2034 case FUTEX_TRYLOCK_PI
:
2035 ret
= futex_lock_pi(uaddr
, fshared
, 0, timeout
, 1);
2044 asmlinkage
long sys_futex(u32 __user
*uaddr
, int op
, u32 val
,
2045 struct timespec __user
*utime
, u32 __user
*uaddr2
,
2049 ktime_t t
, *tp
= NULL
;
2051 int cmd
= op
& FUTEX_CMD_MASK
;
2053 if (utime
&& (cmd
== FUTEX_WAIT
|| cmd
== FUTEX_LOCK_PI
)) {
2054 if (copy_from_user(&ts
, utime
, sizeof(ts
)) != 0)
2056 if (!timespec_valid(&ts
))
2059 t
= timespec_to_ktime(ts
);
2060 if (cmd
== FUTEX_WAIT
)
2061 t
= ktime_add(ktime_get(), t
);
2065 * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE.
2067 if (cmd
== FUTEX_REQUEUE
|| cmd
== FUTEX_CMP_REQUEUE
)
2068 val2
= (u32
) (unsigned long) utime
;
2070 return do_futex(uaddr
, op
, val
, tp
, uaddr2
, val2
, val3
);
2073 static int futexfs_get_sb(struct file_system_type
*fs_type
,
2074 int flags
, const char *dev_name
, void *data
,
2075 struct vfsmount
*mnt
)
2077 return get_sb_pseudo(fs_type
, "futex", NULL
, 0xBAD1DEA, mnt
);
2080 static struct file_system_type futex_fs_type
= {
2082 .get_sb
= futexfs_get_sb
,
2083 .kill_sb
= kill_anon_super
,
2086 static int __init
init(void)
2088 int i
= register_filesystem(&futex_fs_type
);
2093 futex_mnt
= kern_mount(&futex_fs_type
);
2094 if (IS_ERR(futex_mnt
)) {
2095 unregister_filesystem(&futex_fs_type
);
2096 return PTR_ERR(futex_mnt
);
2099 for (i
= 0; i
< ARRAY_SIZE(futex_queues
); i
++) {
2100 plist_head_init(&futex_queues
[i
].chain
, &futex_queues
[i
].lock
);
2101 spin_lock_init(&futex_queues
[i
].lock
);