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 * Requeue-PI support by Darren Hart <dvhltc@us.ibm.com>
23 * Copyright (C) IBM Corporation, 2009
24 * Thanks to Thomas Gleixner for conceptual design and careful reviews.
26 * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly
27 * enough at me, Linus for the original (flawed) idea, Matthew
28 * Kirkwood for proof-of-concept implementation.
30 * "The futexes are also cursed."
31 * "But they come in a choice of three flavours!"
33 * This program is free software; you can redistribute it and/or modify
34 * it under the terms of the GNU General Public License as published by
35 * the Free Software Foundation; either version 2 of the License, or
36 * (at your option) any later version.
38 * This program is distributed in the hope that it will be useful,
39 * but WITHOUT ANY WARRANTY; without even the implied warranty of
40 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
41 * GNU General Public License for more details.
43 * You should have received a copy of the GNU General Public License
44 * along with this program; if not, write to the Free Software
45 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
47 #include <linux/slab.h>
48 #include <linux/poll.h>
50 #include <linux/file.h>
51 #include <linux/jhash.h>
52 #include <linux/init.h>
53 #include <linux/futex.h>
54 #include <linux/mount.h>
55 #include <linux/pagemap.h>
56 #include <linux/syscalls.h>
57 #include <linux/signal.h>
58 #include <linux/export.h>
59 #include <linux/magic.h>
60 #include <linux/pid.h>
61 #include <linux/nsproxy.h>
62 #include <linux/ptrace.h>
63 #include <linux/sched/rt.h>
64 #include <linux/hugetlb.h>
65 #include <linux/freezer.h>
66 #include <linux/bootmem.h>
68 #include <asm/futex.h>
70 #include "locking/rtmutex_common.h"
73 * Basic futex operation and ordering guarantees:
75 * The waiter reads the futex value in user space and calls
76 * futex_wait(). This function computes the hash bucket and acquires
77 * the hash bucket lock. After that it reads the futex user space value
78 * again and verifies that the data has not changed. If it has not changed
79 * it enqueues itself into the hash bucket, releases the hash bucket lock
82 * The waker side modifies the user space value of the futex and calls
83 * futex_wake(). This function computes the hash bucket and acquires the
84 * hash bucket lock. Then it looks for waiters on that futex in the hash
85 * bucket and wakes them.
87 * In futex wake up scenarios where no tasks are blocked on a futex, taking
88 * the hb spinlock can be avoided and simply return. In order for this
89 * optimization to work, ordering guarantees must exist so that the waiter
90 * being added to the list is acknowledged when the list is concurrently being
91 * checked by the waker, avoiding scenarios like the following:
95 * sys_futex(WAIT, futex, val);
96 * futex_wait(futex, val);
99 * sys_futex(WAKE, futex);
104 * lock(hash_bucket(futex));
106 * unlock(hash_bucket(futex));
109 * This would cause the waiter on CPU 0 to wait forever because it
110 * missed the transition of the user space value from val to newval
111 * and the waker did not find the waiter in the hash bucket queue.
113 * The correct serialization ensures that a waiter either observes
114 * the changed user space value before blocking or is woken by a
119 * sys_futex(WAIT, futex, val);
120 * futex_wait(futex, val);
123 * mb(); (A) <-- paired with -.
125 * lock(hash_bucket(futex)); |
129 * | sys_futex(WAKE, futex);
130 * | futex_wake(futex);
132 * `-------> mb(); (B)
135 * unlock(hash_bucket(futex));
136 * schedule(); if (waiters)
137 * lock(hash_bucket(futex));
138 * wake_waiters(futex);
139 * unlock(hash_bucket(futex));
141 * Where (A) orders the waiters increment and the futex value read -- this
142 * is guaranteed by the head counter in the hb spinlock; and where (B)
143 * orders the write to futex and the waiters read -- this is done by the
144 * barriers in get_futex_key_refs(), through either ihold or atomic_inc,
145 * depending on the futex type.
147 * This yields the following case (where X:=waiters, Y:=futex):
155 * Which guarantees that x==0 && y==0 is impossible; which translates back into
156 * the guarantee that we cannot both miss the futex variable change and the
160 int __read_mostly futex_cmpxchg_enabled
;
163 * Futex flags used to encode options to functions and preserve them across
166 #define FLAGS_SHARED 0x01
167 #define FLAGS_CLOCKRT 0x02
168 #define FLAGS_HAS_TIMEOUT 0x04
171 * Priority Inheritance state:
173 struct futex_pi_state
{
175 * list of 'owned' pi_state instances - these have to be
176 * cleaned up in do_exit() if the task exits prematurely:
178 struct list_head list
;
183 struct rt_mutex pi_mutex
;
185 struct task_struct
*owner
;
192 * struct futex_q - The hashed futex queue entry, one per waiting task
193 * @list: priority-sorted list of tasks waiting on this futex
194 * @task: the task waiting on the futex
195 * @lock_ptr: the hash bucket lock
196 * @key: the key the futex is hashed on
197 * @pi_state: optional priority inheritance state
198 * @rt_waiter: rt_waiter storage for use with requeue_pi
199 * @requeue_pi_key: the requeue_pi target futex key
200 * @bitset: bitset for the optional bitmasked wakeup
202 * We use this hashed waitqueue, instead of a normal wait_queue_t, so
203 * we can wake only the relevant ones (hashed queues may be shared).
205 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
206 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
207 * The order of wakeup is always to make the first condition true, then
210 * PI futexes are typically woken before they are removed from the hash list via
211 * the rt_mutex code. See unqueue_me_pi().
214 struct plist_node list
;
216 struct task_struct
*task
;
217 spinlock_t
*lock_ptr
;
219 struct futex_pi_state
*pi_state
;
220 struct rt_mutex_waiter
*rt_waiter
;
221 union futex_key
*requeue_pi_key
;
225 static const struct futex_q futex_q_init
= {
226 /* list gets initialized in queue_me()*/
227 .key
= FUTEX_KEY_INIT
,
228 .bitset
= FUTEX_BITSET_MATCH_ANY
232 * Hash buckets are shared by all the futex_keys that hash to the same
233 * location. Each key may have multiple futex_q structures, one for each task
234 * waiting on a futex.
236 struct futex_hash_bucket
{
239 struct plist_head chain
;
240 } ____cacheline_aligned_in_smp
;
242 static unsigned long __read_mostly futex_hashsize
;
244 static struct futex_hash_bucket
*futex_queues
;
246 static inline void futex_get_mm(union futex_key
*key
)
248 atomic_inc(&key
->private.mm
->mm_count
);
250 * Ensure futex_get_mm() implies a full barrier such that
251 * get_futex_key() implies a full barrier. This is relied upon
252 * as full barrier (B), see the ordering comment above.
254 smp_mb__after_atomic_inc();
258 * Reflects a new waiter being added to the waitqueue.
260 static inline void hb_waiters_inc(struct futex_hash_bucket
*hb
)
263 atomic_inc(&hb
->waiters
);
265 * Full barrier (A), see the ordering comment above.
267 smp_mb__after_atomic_inc();
272 * Reflects a waiter being removed from the waitqueue by wakeup
275 static inline void hb_waiters_dec(struct futex_hash_bucket
*hb
)
278 atomic_dec(&hb
->waiters
);
282 static inline int hb_waiters_pending(struct futex_hash_bucket
*hb
)
285 return atomic_read(&hb
->waiters
);
292 * We hash on the keys returned from get_futex_key (see below).
294 static struct futex_hash_bucket
*hash_futex(union futex_key
*key
)
296 u32 hash
= jhash2((u32
*)&key
->both
.word
,
297 (sizeof(key
->both
.word
)+sizeof(key
->both
.ptr
))/4,
299 return &futex_queues
[hash
& (futex_hashsize
- 1)];
303 * Return 1 if two futex_keys are equal, 0 otherwise.
305 static inline int match_futex(union futex_key
*key1
, union futex_key
*key2
)
308 && key1
->both
.word
== key2
->both
.word
309 && key1
->both
.ptr
== key2
->both
.ptr
310 && key1
->both
.offset
== key2
->both
.offset
);
314 * Take a reference to the resource addressed by a key.
315 * Can be called while holding spinlocks.
318 static void get_futex_key_refs(union futex_key
*key
)
323 switch (key
->both
.offset
& (FUT_OFF_INODE
|FUT_OFF_MMSHARED
)) {
325 ihold(key
->shared
.inode
); /* implies MB (B) */
327 case FUT_OFF_MMSHARED
:
328 futex_get_mm(key
); /* implies MB (B) */
334 * Drop a reference to the resource addressed by a key.
335 * The hash bucket spinlock must not be held.
337 static void drop_futex_key_refs(union futex_key
*key
)
339 if (!key
->both
.ptr
) {
340 /* If we're here then we tried to put a key we failed to get */
345 switch (key
->both
.offset
& (FUT_OFF_INODE
|FUT_OFF_MMSHARED
)) {
347 iput(key
->shared
.inode
);
349 case FUT_OFF_MMSHARED
:
350 mmdrop(key
->private.mm
);
356 * get_futex_key() - Get parameters which are the keys for a futex
357 * @uaddr: virtual address of the futex
358 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
359 * @key: address where result is stored.
360 * @rw: mapping needs to be read/write (values: VERIFY_READ,
363 * Return: a negative error code or 0
365 * The key words are stored in *key on success.
367 * For shared mappings, it's (page->index, file_inode(vma->vm_file),
368 * offset_within_page). For private mappings, it's (uaddr, current->mm).
369 * We can usually work out the index without swapping in the page.
371 * lock_page() might sleep, the caller should not hold a spinlock.
374 get_futex_key(u32 __user
*uaddr
, int fshared
, union futex_key
*key
, int rw
)
376 unsigned long address
= (unsigned long)uaddr
;
377 struct mm_struct
*mm
= current
->mm
;
378 struct page
*page
, *page_head
;
382 * The futex address must be "naturally" aligned.
384 key
->both
.offset
= address
% PAGE_SIZE
;
385 if (unlikely((address
% sizeof(u32
)) != 0))
387 address
-= key
->both
.offset
;
389 if (unlikely(!access_ok(rw
, uaddr
, sizeof(u32
))))
393 * PROCESS_PRIVATE futexes are fast.
394 * As the mm cannot disappear under us and the 'key' only needs
395 * virtual address, we dont even have to find the underlying vma.
396 * Note : We do have to check 'uaddr' is a valid user address,
397 * but access_ok() should be faster than find_vma()
400 key
->private.mm
= mm
;
401 key
->private.address
= address
;
402 get_futex_key_refs(key
); /* implies MB (B) */
407 err
= get_user_pages_fast(address
, 1, 1, &page
);
409 * If write access is not required (eg. FUTEX_WAIT), try
410 * and get read-only access.
412 if (err
== -EFAULT
&& rw
== VERIFY_READ
) {
413 err
= get_user_pages_fast(address
, 1, 0, &page
);
421 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
423 if (unlikely(PageTail(page
))) {
425 /* serialize against __split_huge_page_splitting() */
427 if (likely(__get_user_pages_fast(address
, 1, !ro
, &page
) == 1)) {
428 page_head
= compound_head(page
);
430 * page_head is valid pointer but we must pin
431 * it before taking the PG_lock and/or
432 * PG_compound_lock. The moment we re-enable
433 * irqs __split_huge_page_splitting() can
434 * return and the head page can be freed from
435 * under us. We can't take the PG_lock and/or
436 * PG_compound_lock on a page that could be
437 * freed from under us.
439 if (page
!= page_head
) {
450 page_head
= compound_head(page
);
451 if (page
!= page_head
) {
457 lock_page(page_head
);
460 * If page_head->mapping is NULL, then it cannot be a PageAnon
461 * page; but it might be the ZERO_PAGE or in the gate area or
462 * in a special mapping (all cases which we are happy to fail);
463 * or it may have been a good file page when get_user_pages_fast
464 * found it, but truncated or holepunched or subjected to
465 * invalidate_complete_page2 before we got the page lock (also
466 * cases which we are happy to fail). And we hold a reference,
467 * so refcount care in invalidate_complete_page's remove_mapping
468 * prevents drop_caches from setting mapping to NULL beneath us.
470 * The case we do have to guard against is when memory pressure made
471 * shmem_writepage move it from filecache to swapcache beneath us:
472 * an unlikely race, but we do need to retry for page_head->mapping.
474 if (!page_head
->mapping
) {
475 int shmem_swizzled
= PageSwapCache(page_head
);
476 unlock_page(page_head
);
484 * Private mappings are handled in a simple way.
486 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
487 * it's a read-only handle, it's expected that futexes attach to
488 * the object not the particular process.
490 if (PageAnon(page_head
)) {
492 * A RO anonymous page will never change and thus doesn't make
493 * sense for futex operations.
500 key
->both
.offset
|= FUT_OFF_MMSHARED
; /* ref taken on mm */
501 key
->private.mm
= mm
;
502 key
->private.address
= address
;
504 key
->both
.offset
|= FUT_OFF_INODE
; /* inode-based key */
505 key
->shared
.inode
= page_head
->mapping
->host
;
506 key
->shared
.pgoff
= basepage_index(page
);
509 get_futex_key_refs(key
); /* implies MB (B) */
512 unlock_page(page_head
);
517 static inline void put_futex_key(union futex_key
*key
)
519 drop_futex_key_refs(key
);
523 * fault_in_user_writeable() - Fault in user address and verify RW access
524 * @uaddr: pointer to faulting user space address
526 * Slow path to fixup the fault we just took in the atomic write
529 * We have no generic implementation of a non-destructive write to the
530 * user address. We know that we faulted in the atomic pagefault
531 * disabled section so we can as well avoid the #PF overhead by
532 * calling get_user_pages() right away.
534 static int fault_in_user_writeable(u32 __user
*uaddr
)
536 struct mm_struct
*mm
= current
->mm
;
539 down_read(&mm
->mmap_sem
);
540 ret
= fixup_user_fault(current
, mm
, (unsigned long)uaddr
,
542 up_read(&mm
->mmap_sem
);
544 return ret
< 0 ? ret
: 0;
548 * futex_top_waiter() - Return the highest priority waiter on a futex
549 * @hb: the hash bucket the futex_q's reside in
550 * @key: the futex key (to distinguish it from other futex futex_q's)
552 * Must be called with the hb lock held.
554 static struct futex_q
*futex_top_waiter(struct futex_hash_bucket
*hb
,
555 union futex_key
*key
)
557 struct futex_q
*this;
559 plist_for_each_entry(this, &hb
->chain
, list
) {
560 if (match_futex(&this->key
, key
))
566 static int cmpxchg_futex_value_locked(u32
*curval
, u32 __user
*uaddr
,
567 u32 uval
, u32 newval
)
572 ret
= futex_atomic_cmpxchg_inatomic(curval
, uaddr
, uval
, newval
);
578 static int get_futex_value_locked(u32
*dest
, u32 __user
*from
)
583 ret
= __copy_from_user_inatomic(dest
, from
, sizeof(u32
));
586 return ret
? -EFAULT
: 0;
593 static int refill_pi_state_cache(void)
595 struct futex_pi_state
*pi_state
;
597 if (likely(current
->pi_state_cache
))
600 pi_state
= kzalloc(sizeof(*pi_state
), GFP_KERNEL
);
605 INIT_LIST_HEAD(&pi_state
->list
);
606 /* pi_mutex gets initialized later */
607 pi_state
->owner
= NULL
;
608 atomic_set(&pi_state
->refcount
, 1);
609 pi_state
->key
= FUTEX_KEY_INIT
;
611 current
->pi_state_cache
= pi_state
;
616 static struct futex_pi_state
* alloc_pi_state(void)
618 struct futex_pi_state
*pi_state
= current
->pi_state_cache
;
621 current
->pi_state_cache
= NULL
;
626 static void free_pi_state(struct futex_pi_state
*pi_state
)
628 if (!atomic_dec_and_test(&pi_state
->refcount
))
632 * If pi_state->owner is NULL, the owner is most probably dying
633 * and has cleaned up the pi_state already
635 if (pi_state
->owner
) {
636 raw_spin_lock_irq(&pi_state
->owner
->pi_lock
);
637 list_del_init(&pi_state
->list
);
638 raw_spin_unlock_irq(&pi_state
->owner
->pi_lock
);
640 rt_mutex_proxy_unlock(&pi_state
->pi_mutex
, pi_state
->owner
);
643 if (current
->pi_state_cache
)
647 * pi_state->list is already empty.
648 * clear pi_state->owner.
649 * refcount is at 0 - put it back to 1.
651 pi_state
->owner
= NULL
;
652 atomic_set(&pi_state
->refcount
, 1);
653 current
->pi_state_cache
= pi_state
;
658 * Look up the task based on what TID userspace gave us.
661 static struct task_struct
* futex_find_get_task(pid_t pid
)
663 struct task_struct
*p
;
666 p
= find_task_by_vpid(pid
);
676 * This task is holding PI mutexes at exit time => bad.
677 * Kernel cleans up PI-state, but userspace is likely hosed.
678 * (Robust-futex cleanup is separate and might save the day for userspace.)
680 void exit_pi_state_list(struct task_struct
*curr
)
682 struct list_head
*next
, *head
= &curr
->pi_state_list
;
683 struct futex_pi_state
*pi_state
;
684 struct futex_hash_bucket
*hb
;
685 union futex_key key
= FUTEX_KEY_INIT
;
687 if (!futex_cmpxchg_enabled
)
690 * We are a ZOMBIE and nobody can enqueue itself on
691 * pi_state_list anymore, but we have to be careful
692 * versus waiters unqueueing themselves:
694 raw_spin_lock_irq(&curr
->pi_lock
);
695 while (!list_empty(head
)) {
698 pi_state
= list_entry(next
, struct futex_pi_state
, list
);
700 hb
= hash_futex(&key
);
701 raw_spin_unlock_irq(&curr
->pi_lock
);
703 spin_lock(&hb
->lock
);
705 raw_spin_lock_irq(&curr
->pi_lock
);
707 * We dropped the pi-lock, so re-check whether this
708 * task still owns the PI-state:
710 if (head
->next
!= next
) {
711 spin_unlock(&hb
->lock
);
715 WARN_ON(pi_state
->owner
!= curr
);
716 WARN_ON(list_empty(&pi_state
->list
));
717 list_del_init(&pi_state
->list
);
718 pi_state
->owner
= NULL
;
719 raw_spin_unlock_irq(&curr
->pi_lock
);
721 rt_mutex_unlock(&pi_state
->pi_mutex
);
723 spin_unlock(&hb
->lock
);
725 raw_spin_lock_irq(&curr
->pi_lock
);
727 raw_spin_unlock_irq(&curr
->pi_lock
);
731 lookup_pi_state(u32 uval
, struct futex_hash_bucket
*hb
,
732 union futex_key
*key
, struct futex_pi_state
**ps
)
734 struct futex_pi_state
*pi_state
= NULL
;
735 struct futex_q
*this, *next
;
736 struct task_struct
*p
;
737 pid_t pid
= uval
& FUTEX_TID_MASK
;
739 plist_for_each_entry_safe(this, next
, &hb
->chain
, list
) {
740 if (match_futex(&this->key
, key
)) {
742 * Another waiter already exists - bump up
743 * the refcount and return its pi_state:
745 pi_state
= this->pi_state
;
747 * Userspace might have messed up non-PI and PI futexes
749 if (unlikely(!pi_state
))
752 WARN_ON(!atomic_read(&pi_state
->refcount
));
755 * When pi_state->owner is NULL then the owner died
756 * and another waiter is on the fly. pi_state->owner
757 * is fixed up by the task which acquires
758 * pi_state->rt_mutex.
760 * We do not check for pid == 0 which can happen when
761 * the owner died and robust_list_exit() cleared the
764 if (pid
&& pi_state
->owner
) {
766 * Bail out if user space manipulated the
769 if (pid
!= task_pid_vnr(pi_state
->owner
))
773 atomic_inc(&pi_state
->refcount
);
781 * We are the first waiter - try to look up the real owner and attach
782 * the new pi_state to it, but bail out when TID = 0
786 p
= futex_find_get_task(pid
);
791 * We need to look at the task state flags to figure out,
792 * whether the task is exiting. To protect against the do_exit
793 * change of the task flags, we do this protected by
796 raw_spin_lock_irq(&p
->pi_lock
);
797 if (unlikely(p
->flags
& PF_EXITING
)) {
799 * The task is on the way out. When PF_EXITPIDONE is
800 * set, we know that the task has finished the
803 int ret
= (p
->flags
& PF_EXITPIDONE
) ? -ESRCH
: -EAGAIN
;
805 raw_spin_unlock_irq(&p
->pi_lock
);
810 pi_state
= alloc_pi_state();
813 * Initialize the pi_mutex in locked state and make 'p'
816 rt_mutex_init_proxy_locked(&pi_state
->pi_mutex
, p
);
818 /* Store the key for possible exit cleanups: */
819 pi_state
->key
= *key
;
821 WARN_ON(!list_empty(&pi_state
->list
));
822 list_add(&pi_state
->list
, &p
->pi_state_list
);
824 raw_spin_unlock_irq(&p
->pi_lock
);
834 * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
835 * @uaddr: the pi futex user address
836 * @hb: the pi futex hash bucket
837 * @key: the futex key associated with uaddr and hb
838 * @ps: the pi_state pointer where we store the result of the
840 * @task: the task to perform the atomic lock work for. This will
841 * be "current" except in the case of requeue pi.
842 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
846 * 1 - acquired the lock;
849 * The hb->lock and futex_key refs shall be held by the caller.
851 static int futex_lock_pi_atomic(u32 __user
*uaddr
, struct futex_hash_bucket
*hb
,
852 union futex_key
*key
,
853 struct futex_pi_state
**ps
,
854 struct task_struct
*task
, int set_waiters
)
856 int lock_taken
, ret
, force_take
= 0;
857 u32 uval
, newval
, curval
, vpid
= task_pid_vnr(task
);
860 ret
= lock_taken
= 0;
863 * To avoid races, we attempt to take the lock here again
864 * (by doing a 0 -> TID atomic cmpxchg), while holding all
865 * the locks. It will most likely not succeed.
869 newval
|= FUTEX_WAITERS
;
871 if (unlikely(cmpxchg_futex_value_locked(&curval
, uaddr
, 0, newval
)))
877 if ((unlikely((curval
& FUTEX_TID_MASK
) == vpid
)))
881 * Surprise - we got the lock. Just return to userspace:
883 if (unlikely(!curval
))
889 * Set the FUTEX_WAITERS flag, so the owner will know it has someone
890 * to wake at the next unlock.
892 newval
= curval
| FUTEX_WAITERS
;
895 * Should we force take the futex? See below.
897 if (unlikely(force_take
)) {
899 * Keep the OWNER_DIED and the WAITERS bit and set the
902 newval
= (curval
& ~FUTEX_TID_MASK
) | vpid
;
907 if (unlikely(cmpxchg_futex_value_locked(&curval
, uaddr
, uval
, newval
)))
909 if (unlikely(curval
!= uval
))
913 * We took the lock due to forced take over.
915 if (unlikely(lock_taken
))
919 * We dont have the lock. Look up the PI state (or create it if
920 * we are the first waiter):
922 ret
= lookup_pi_state(uval
, hb
, key
, ps
);
928 * We failed to find an owner for this
929 * futex. So we have no pi_state to block
930 * on. This can happen in two cases:
933 * 2) A stale FUTEX_WAITERS bit
935 * Re-read the futex value.
937 if (get_futex_value_locked(&curval
, uaddr
))
941 * If the owner died or we have a stale
942 * WAITERS bit the owner TID in the user space
945 if (!(curval
& FUTEX_TID_MASK
)) {
958 * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket
959 * @q: The futex_q to unqueue
961 * The q->lock_ptr must not be NULL and must be held by the caller.
963 static void __unqueue_futex(struct futex_q
*q
)
965 struct futex_hash_bucket
*hb
;
967 if (WARN_ON_SMP(!q
->lock_ptr
|| !spin_is_locked(q
->lock_ptr
))
968 || WARN_ON(plist_node_empty(&q
->list
)))
971 hb
= container_of(q
->lock_ptr
, struct futex_hash_bucket
, lock
);
972 plist_del(&q
->list
, &hb
->chain
);
977 * The hash bucket lock must be held when this is called.
978 * Afterwards, the futex_q must not be accessed.
980 static void wake_futex(struct futex_q
*q
)
982 struct task_struct
*p
= q
->task
;
984 if (WARN(q
->pi_state
|| q
->rt_waiter
, "refusing to wake PI futex\n"))
988 * We set q->lock_ptr = NULL _before_ we wake up the task. If
989 * a non-futex wake up happens on another CPU then the task
990 * might exit and p would dereference a non-existing task
991 * struct. Prevent this by holding a reference on p across the
998 * The waiting task can free the futex_q as soon as
999 * q->lock_ptr = NULL is written, without taking any locks. A
1000 * memory barrier is required here to prevent the following
1001 * store to lock_ptr from getting ahead of the plist_del.
1006 wake_up_state(p
, TASK_NORMAL
);
1010 static int wake_futex_pi(u32 __user
*uaddr
, u32 uval
, struct futex_q
*this)
1012 struct task_struct
*new_owner
;
1013 struct futex_pi_state
*pi_state
= this->pi_state
;
1014 u32
uninitialized_var(curval
), newval
;
1020 * If current does not own the pi_state then the futex is
1021 * inconsistent and user space fiddled with the futex value.
1023 if (pi_state
->owner
!= current
)
1026 raw_spin_lock(&pi_state
->pi_mutex
.wait_lock
);
1027 new_owner
= rt_mutex_next_owner(&pi_state
->pi_mutex
);
1030 * It is possible that the next waiter (the one that brought
1031 * this owner to the kernel) timed out and is no longer
1032 * waiting on the lock.
1035 new_owner
= this->task
;
1038 * We pass it to the next owner. (The WAITERS bit is always
1039 * kept enabled while there is PI state around. We must also
1040 * preserve the owner died bit.)
1042 if (!(uval
& FUTEX_OWNER_DIED
)) {
1045 newval
= FUTEX_WAITERS
| task_pid_vnr(new_owner
);
1047 if (cmpxchg_futex_value_locked(&curval
, uaddr
, uval
, newval
))
1049 else if (curval
!= uval
)
1052 raw_spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
1057 raw_spin_lock_irq(&pi_state
->owner
->pi_lock
);
1058 WARN_ON(list_empty(&pi_state
->list
));
1059 list_del_init(&pi_state
->list
);
1060 raw_spin_unlock_irq(&pi_state
->owner
->pi_lock
);
1062 raw_spin_lock_irq(&new_owner
->pi_lock
);
1063 WARN_ON(!list_empty(&pi_state
->list
));
1064 list_add(&pi_state
->list
, &new_owner
->pi_state_list
);
1065 pi_state
->owner
= new_owner
;
1066 raw_spin_unlock_irq(&new_owner
->pi_lock
);
1068 raw_spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
1069 rt_mutex_unlock(&pi_state
->pi_mutex
);
1074 static int unlock_futex_pi(u32 __user
*uaddr
, u32 uval
)
1076 u32
uninitialized_var(oldval
);
1079 * There is no waiter, so we unlock the futex. The owner died
1080 * bit has not to be preserved here. We are the owner:
1082 if (cmpxchg_futex_value_locked(&oldval
, uaddr
, uval
, 0))
1091 * Express the locking dependencies for lockdep:
1094 double_lock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
1097 spin_lock(&hb1
->lock
);
1099 spin_lock_nested(&hb2
->lock
, SINGLE_DEPTH_NESTING
);
1100 } else { /* hb1 > hb2 */
1101 spin_lock(&hb2
->lock
);
1102 spin_lock_nested(&hb1
->lock
, SINGLE_DEPTH_NESTING
);
1107 double_unlock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
1109 spin_unlock(&hb1
->lock
);
1111 spin_unlock(&hb2
->lock
);
1115 * Wake up waiters matching bitset queued on this futex (uaddr).
1118 futex_wake(u32 __user
*uaddr
, unsigned int flags
, int nr_wake
, u32 bitset
)
1120 struct futex_hash_bucket
*hb
;
1121 struct futex_q
*this, *next
;
1122 union futex_key key
= FUTEX_KEY_INIT
;
1128 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &key
, VERIFY_READ
);
1129 if (unlikely(ret
!= 0))
1132 hb
= hash_futex(&key
);
1134 /* Make sure we really have tasks to wakeup */
1135 if (!hb_waiters_pending(hb
))
1138 spin_lock(&hb
->lock
);
1140 plist_for_each_entry_safe(this, next
, &hb
->chain
, list
) {
1141 if (match_futex (&this->key
, &key
)) {
1142 if (this->pi_state
|| this->rt_waiter
) {
1147 /* Check if one of the bits is set in both bitsets */
1148 if (!(this->bitset
& bitset
))
1152 if (++ret
>= nr_wake
)
1157 spin_unlock(&hb
->lock
);
1159 put_futex_key(&key
);
1165 * Wake up all waiters hashed on the physical page that is mapped
1166 * to this virtual address:
1169 futex_wake_op(u32 __user
*uaddr1
, unsigned int flags
, u32 __user
*uaddr2
,
1170 int nr_wake
, int nr_wake2
, int op
)
1172 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
1173 struct futex_hash_bucket
*hb1
, *hb2
;
1174 struct futex_q
*this, *next
;
1178 ret
= get_futex_key(uaddr1
, flags
& FLAGS_SHARED
, &key1
, VERIFY_READ
);
1179 if (unlikely(ret
!= 0))
1181 ret
= get_futex_key(uaddr2
, flags
& FLAGS_SHARED
, &key2
, VERIFY_WRITE
);
1182 if (unlikely(ret
!= 0))
1185 hb1
= hash_futex(&key1
);
1186 hb2
= hash_futex(&key2
);
1189 double_lock_hb(hb1
, hb2
);
1190 op_ret
= futex_atomic_op_inuser(op
, uaddr2
);
1191 if (unlikely(op_ret
< 0)) {
1193 double_unlock_hb(hb1
, hb2
);
1197 * we don't get EFAULT from MMU faults if we don't have an MMU,
1198 * but we might get them from range checking
1204 if (unlikely(op_ret
!= -EFAULT
)) {
1209 ret
= fault_in_user_writeable(uaddr2
);
1213 if (!(flags
& FLAGS_SHARED
))
1216 put_futex_key(&key2
);
1217 put_futex_key(&key1
);
1221 plist_for_each_entry_safe(this, next
, &hb1
->chain
, list
) {
1222 if (match_futex (&this->key
, &key1
)) {
1223 if (this->pi_state
|| this->rt_waiter
) {
1228 if (++ret
>= nr_wake
)
1235 plist_for_each_entry_safe(this, next
, &hb2
->chain
, list
) {
1236 if (match_futex (&this->key
, &key2
)) {
1237 if (this->pi_state
|| this->rt_waiter
) {
1242 if (++op_ret
>= nr_wake2
)
1250 double_unlock_hb(hb1
, hb2
);
1252 put_futex_key(&key2
);
1254 put_futex_key(&key1
);
1260 * requeue_futex() - Requeue a futex_q from one hb to another
1261 * @q: the futex_q to requeue
1262 * @hb1: the source hash_bucket
1263 * @hb2: the target hash_bucket
1264 * @key2: the new key for the requeued futex_q
1267 void requeue_futex(struct futex_q
*q
, struct futex_hash_bucket
*hb1
,
1268 struct futex_hash_bucket
*hb2
, union futex_key
*key2
)
1272 * If key1 and key2 hash to the same bucket, no need to
1275 if (likely(&hb1
->chain
!= &hb2
->chain
)) {
1276 plist_del(&q
->list
, &hb1
->chain
);
1277 hb_waiters_dec(hb1
);
1278 plist_add(&q
->list
, &hb2
->chain
);
1279 hb_waiters_inc(hb2
);
1280 q
->lock_ptr
= &hb2
->lock
;
1282 get_futex_key_refs(key2
);
1287 * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
1289 * @key: the key of the requeue target futex
1290 * @hb: the hash_bucket of the requeue target futex
1292 * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1293 * target futex if it is uncontended or via a lock steal. Set the futex_q key
1294 * to the requeue target futex so the waiter can detect the wakeup on the right
1295 * futex, but remove it from the hb and NULL the rt_waiter so it can detect
1296 * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1297 * to protect access to the pi_state to fixup the owner later. Must be called
1298 * with both q->lock_ptr and hb->lock held.
1301 void requeue_pi_wake_futex(struct futex_q
*q
, union futex_key
*key
,
1302 struct futex_hash_bucket
*hb
)
1304 get_futex_key_refs(key
);
1309 WARN_ON(!q
->rt_waiter
);
1310 q
->rt_waiter
= NULL
;
1312 q
->lock_ptr
= &hb
->lock
;
1314 wake_up_state(q
->task
, TASK_NORMAL
);
1318 * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
1319 * @pifutex: the user address of the to futex
1320 * @hb1: the from futex hash bucket, must be locked by the caller
1321 * @hb2: the to futex hash bucket, must be locked by the caller
1322 * @key1: the from futex key
1323 * @key2: the to futex key
1324 * @ps: address to store the pi_state pointer
1325 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
1327 * Try and get the lock on behalf of the top waiter if we can do it atomically.
1328 * Wake the top waiter if we succeed. If the caller specified set_waiters,
1329 * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1330 * hb1 and hb2 must be held by the caller.
1333 * 0 - failed to acquire the lock atomically;
1334 * 1 - acquired the lock;
1337 static int futex_proxy_trylock_atomic(u32 __user
*pifutex
,
1338 struct futex_hash_bucket
*hb1
,
1339 struct futex_hash_bucket
*hb2
,
1340 union futex_key
*key1
, union futex_key
*key2
,
1341 struct futex_pi_state
**ps
, int set_waiters
)
1343 struct futex_q
*top_waiter
= NULL
;
1347 if (get_futex_value_locked(&curval
, pifutex
))
1351 * Find the top_waiter and determine if there are additional waiters.
1352 * If the caller intends to requeue more than 1 waiter to pifutex,
1353 * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1354 * as we have means to handle the possible fault. If not, don't set
1355 * the bit unecessarily as it will force the subsequent unlock to enter
1358 top_waiter
= futex_top_waiter(hb1
, key1
);
1360 /* There are no waiters, nothing for us to do. */
1364 /* Ensure we requeue to the expected futex. */
1365 if (!match_futex(top_waiter
->requeue_pi_key
, key2
))
1369 * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1370 * the contended case or if set_waiters is 1. The pi_state is returned
1371 * in ps in contended cases.
1373 ret
= futex_lock_pi_atomic(pifutex
, hb2
, key2
, ps
, top_waiter
->task
,
1376 requeue_pi_wake_futex(top_waiter
, key2
, hb2
);
1382 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
1383 * @uaddr1: source futex user address
1384 * @flags: futex flags (FLAGS_SHARED, etc.)
1385 * @uaddr2: target futex user address
1386 * @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1387 * @nr_requeue: number of waiters to requeue (0-INT_MAX)
1388 * @cmpval: @uaddr1 expected value (or %NULL)
1389 * @requeue_pi: if we are attempting to requeue from a non-pi futex to a
1390 * pi futex (pi to pi requeue is not supported)
1392 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1393 * uaddr2 atomically on behalf of the top waiter.
1396 * >=0 - on success, the number of tasks requeued or woken;
1399 static int futex_requeue(u32 __user
*uaddr1
, unsigned int flags
,
1400 u32 __user
*uaddr2
, int nr_wake
, int nr_requeue
,
1401 u32
*cmpval
, int requeue_pi
)
1403 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
1404 int drop_count
= 0, task_count
= 0, ret
;
1405 struct futex_pi_state
*pi_state
= NULL
;
1406 struct futex_hash_bucket
*hb1
, *hb2
;
1407 struct futex_q
*this, *next
;
1412 * requeue_pi requires a pi_state, try to allocate it now
1413 * without any locks in case it fails.
1415 if (refill_pi_state_cache())
1418 * requeue_pi must wake as many tasks as it can, up to nr_wake
1419 * + nr_requeue, since it acquires the rt_mutex prior to
1420 * returning to userspace, so as to not leave the rt_mutex with
1421 * waiters and no owner. However, second and third wake-ups
1422 * cannot be predicted as they involve race conditions with the
1423 * first wake and a fault while looking up the pi_state. Both
1424 * pthread_cond_signal() and pthread_cond_broadcast() should
1432 if (pi_state
!= NULL
) {
1434 * We will have to lookup the pi_state again, so free this one
1435 * to keep the accounting correct.
1437 free_pi_state(pi_state
);
1441 ret
= get_futex_key(uaddr1
, flags
& FLAGS_SHARED
, &key1
, VERIFY_READ
);
1442 if (unlikely(ret
!= 0))
1444 ret
= get_futex_key(uaddr2
, flags
& FLAGS_SHARED
, &key2
,
1445 requeue_pi
? VERIFY_WRITE
: VERIFY_READ
);
1446 if (unlikely(ret
!= 0))
1449 hb1
= hash_futex(&key1
);
1450 hb2
= hash_futex(&key2
);
1453 double_lock_hb(hb1
, hb2
);
1455 if (likely(cmpval
!= NULL
)) {
1458 ret
= get_futex_value_locked(&curval
, uaddr1
);
1460 if (unlikely(ret
)) {
1461 double_unlock_hb(hb1
, hb2
);
1463 ret
= get_user(curval
, uaddr1
);
1467 if (!(flags
& FLAGS_SHARED
))
1470 put_futex_key(&key2
);
1471 put_futex_key(&key1
);
1474 if (curval
!= *cmpval
) {
1480 if (requeue_pi
&& (task_count
- nr_wake
< nr_requeue
)) {
1482 * Attempt to acquire uaddr2 and wake the top waiter. If we
1483 * intend to requeue waiters, force setting the FUTEX_WAITERS
1484 * bit. We force this here where we are able to easily handle
1485 * faults rather in the requeue loop below.
1487 ret
= futex_proxy_trylock_atomic(uaddr2
, hb1
, hb2
, &key1
,
1488 &key2
, &pi_state
, nr_requeue
);
1491 * At this point the top_waiter has either taken uaddr2 or is
1492 * waiting on it. If the former, then the pi_state will not
1493 * exist yet, look it up one more time to ensure we have a
1500 ret
= get_futex_value_locked(&curval2
, uaddr2
);
1502 ret
= lookup_pi_state(curval2
, hb2
, &key2
,
1510 double_unlock_hb(hb1
, hb2
);
1511 put_futex_key(&key2
);
1512 put_futex_key(&key1
);
1513 ret
= fault_in_user_writeable(uaddr2
);
1518 /* The owner was exiting, try again. */
1519 double_unlock_hb(hb1
, hb2
);
1520 put_futex_key(&key2
);
1521 put_futex_key(&key1
);
1529 plist_for_each_entry_safe(this, next
, &hb1
->chain
, list
) {
1530 if (task_count
- nr_wake
>= nr_requeue
)
1533 if (!match_futex(&this->key
, &key1
))
1537 * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
1538 * be paired with each other and no other futex ops.
1540 * We should never be requeueing a futex_q with a pi_state,
1541 * which is awaiting a futex_unlock_pi().
1543 if ((requeue_pi
&& !this->rt_waiter
) ||
1544 (!requeue_pi
&& this->rt_waiter
) ||
1551 * Wake nr_wake waiters. For requeue_pi, if we acquired the
1552 * lock, we already woke the top_waiter. If not, it will be
1553 * woken by futex_unlock_pi().
1555 if (++task_count
<= nr_wake
&& !requeue_pi
) {
1560 /* Ensure we requeue to the expected futex for requeue_pi. */
1561 if (requeue_pi
&& !match_futex(this->requeue_pi_key
, &key2
)) {
1567 * Requeue nr_requeue waiters and possibly one more in the case
1568 * of requeue_pi if we couldn't acquire the lock atomically.
1571 /* Prepare the waiter to take the rt_mutex. */
1572 atomic_inc(&pi_state
->refcount
);
1573 this->pi_state
= pi_state
;
1574 ret
= rt_mutex_start_proxy_lock(&pi_state
->pi_mutex
,
1578 /* We got the lock. */
1579 requeue_pi_wake_futex(this, &key2
, hb2
);
1584 this->pi_state
= NULL
;
1585 free_pi_state(pi_state
);
1589 requeue_futex(this, hb1
, hb2
, &key2
);
1594 double_unlock_hb(hb1
, hb2
);
1597 * drop_futex_key_refs() must be called outside the spinlocks. During
1598 * the requeue we moved futex_q's from the hash bucket at key1 to the
1599 * one at key2 and updated their key pointer. We no longer need to
1600 * hold the references to key1.
1602 while (--drop_count
>= 0)
1603 drop_futex_key_refs(&key1
);
1606 put_futex_key(&key2
);
1608 put_futex_key(&key1
);
1610 if (pi_state
!= NULL
)
1611 free_pi_state(pi_state
);
1612 return ret
? ret
: task_count
;
1615 /* The key must be already stored in q->key. */
1616 static inline struct futex_hash_bucket
*queue_lock(struct futex_q
*q
)
1617 __acquires(&hb
->lock
)
1619 struct futex_hash_bucket
*hb
;
1621 hb
= hash_futex(&q
->key
);
1624 * Increment the counter before taking the lock so that
1625 * a potential waker won't miss a to-be-slept task that is
1626 * waiting for the spinlock. This is safe as all queue_lock()
1627 * users end up calling queue_me(). Similarly, for housekeeping,
1628 * decrement the counter at queue_unlock() when some error has
1629 * occurred and we don't end up adding the task to the list.
1633 q
->lock_ptr
= &hb
->lock
;
1635 spin_lock(&hb
->lock
); /* implies MB (A) */
1640 queue_unlock(struct futex_hash_bucket
*hb
)
1641 __releases(&hb
->lock
)
1643 spin_unlock(&hb
->lock
);
1648 * queue_me() - Enqueue the futex_q on the futex_hash_bucket
1649 * @q: The futex_q to enqueue
1650 * @hb: The destination hash bucket
1652 * The hb->lock must be held by the caller, and is released here. A call to
1653 * queue_me() is typically paired with exactly one call to unqueue_me(). The
1654 * exceptions involve the PI related operations, which may use unqueue_me_pi()
1655 * or nothing if the unqueue is done as part of the wake process and the unqueue
1656 * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
1659 static inline void queue_me(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
1660 __releases(&hb
->lock
)
1665 * The priority used to register this element is
1666 * - either the real thread-priority for the real-time threads
1667 * (i.e. threads with a priority lower than MAX_RT_PRIO)
1668 * - or MAX_RT_PRIO for non-RT threads.
1669 * Thus, all RT-threads are woken first in priority order, and
1670 * the others are woken last, in FIFO order.
1672 prio
= min(current
->normal_prio
, MAX_RT_PRIO
);
1674 plist_node_init(&q
->list
, prio
);
1675 plist_add(&q
->list
, &hb
->chain
);
1677 spin_unlock(&hb
->lock
);
1681 * unqueue_me() - Remove the futex_q from its futex_hash_bucket
1682 * @q: The futex_q to unqueue
1684 * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must
1685 * be paired with exactly one earlier call to queue_me().
1688 * 1 - if the futex_q was still queued (and we removed unqueued it);
1689 * 0 - if the futex_q was already removed by the waking thread
1691 static int unqueue_me(struct futex_q
*q
)
1693 spinlock_t
*lock_ptr
;
1696 /* In the common case we don't take the spinlock, which is nice. */
1698 lock_ptr
= q
->lock_ptr
;
1700 if (lock_ptr
!= NULL
) {
1701 spin_lock(lock_ptr
);
1703 * q->lock_ptr can change between reading it and
1704 * spin_lock(), causing us to take the wrong lock. This
1705 * corrects the race condition.
1707 * Reasoning goes like this: if we have the wrong lock,
1708 * q->lock_ptr must have changed (maybe several times)
1709 * between reading it and the spin_lock(). It can
1710 * change again after the spin_lock() but only if it was
1711 * already changed before the spin_lock(). It cannot,
1712 * however, change back to the original value. Therefore
1713 * we can detect whether we acquired the correct lock.
1715 if (unlikely(lock_ptr
!= q
->lock_ptr
)) {
1716 spin_unlock(lock_ptr
);
1721 BUG_ON(q
->pi_state
);
1723 spin_unlock(lock_ptr
);
1727 drop_futex_key_refs(&q
->key
);
1732 * PI futexes can not be requeued and must remove themself from the
1733 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1736 static void unqueue_me_pi(struct futex_q
*q
)
1737 __releases(q
->lock_ptr
)
1741 BUG_ON(!q
->pi_state
);
1742 free_pi_state(q
->pi_state
);
1745 spin_unlock(q
->lock_ptr
);
1749 * Fixup the pi_state owner with the new owner.
1751 * Must be called with hash bucket lock held and mm->sem held for non
1754 static int fixup_pi_state_owner(u32 __user
*uaddr
, struct futex_q
*q
,
1755 struct task_struct
*newowner
)
1757 u32 newtid
= task_pid_vnr(newowner
) | FUTEX_WAITERS
;
1758 struct futex_pi_state
*pi_state
= q
->pi_state
;
1759 struct task_struct
*oldowner
= pi_state
->owner
;
1760 u32 uval
, uninitialized_var(curval
), newval
;
1764 if (!pi_state
->owner
)
1765 newtid
|= FUTEX_OWNER_DIED
;
1768 * We are here either because we stole the rtmutex from the
1769 * previous highest priority waiter or we are the highest priority
1770 * waiter but failed to get the rtmutex the first time.
1771 * We have to replace the newowner TID in the user space variable.
1772 * This must be atomic as we have to preserve the owner died bit here.
1774 * Note: We write the user space value _before_ changing the pi_state
1775 * because we can fault here. Imagine swapped out pages or a fork
1776 * that marked all the anonymous memory readonly for cow.
1778 * Modifying pi_state _before_ the user space value would
1779 * leave the pi_state in an inconsistent state when we fault
1780 * here, because we need to drop the hash bucket lock to
1781 * handle the fault. This might be observed in the PID check
1782 * in lookup_pi_state.
1785 if (get_futex_value_locked(&uval
, uaddr
))
1789 newval
= (uval
& FUTEX_OWNER_DIED
) | newtid
;
1791 if (cmpxchg_futex_value_locked(&curval
, uaddr
, uval
, newval
))
1799 * We fixed up user space. Now we need to fix the pi_state
1802 if (pi_state
->owner
!= NULL
) {
1803 raw_spin_lock_irq(&pi_state
->owner
->pi_lock
);
1804 WARN_ON(list_empty(&pi_state
->list
));
1805 list_del_init(&pi_state
->list
);
1806 raw_spin_unlock_irq(&pi_state
->owner
->pi_lock
);
1809 pi_state
->owner
= newowner
;
1811 raw_spin_lock_irq(&newowner
->pi_lock
);
1812 WARN_ON(!list_empty(&pi_state
->list
));
1813 list_add(&pi_state
->list
, &newowner
->pi_state_list
);
1814 raw_spin_unlock_irq(&newowner
->pi_lock
);
1818 * To handle the page fault we need to drop the hash bucket
1819 * lock here. That gives the other task (either the highest priority
1820 * waiter itself or the task which stole the rtmutex) the
1821 * chance to try the fixup of the pi_state. So once we are
1822 * back from handling the fault we need to check the pi_state
1823 * after reacquiring the hash bucket lock and before trying to
1824 * do another fixup. When the fixup has been done already we
1828 spin_unlock(q
->lock_ptr
);
1830 ret
= fault_in_user_writeable(uaddr
);
1832 spin_lock(q
->lock_ptr
);
1835 * Check if someone else fixed it for us:
1837 if (pi_state
->owner
!= oldowner
)
1846 static long futex_wait_restart(struct restart_block
*restart
);
1849 * fixup_owner() - Post lock pi_state and corner case management
1850 * @uaddr: user address of the futex
1851 * @q: futex_q (contains pi_state and access to the rt_mutex)
1852 * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
1854 * After attempting to lock an rt_mutex, this function is called to cleanup
1855 * the pi_state owner as well as handle race conditions that may allow us to
1856 * acquire the lock. Must be called with the hb lock held.
1859 * 1 - success, lock taken;
1860 * 0 - success, lock not taken;
1861 * <0 - on error (-EFAULT)
1863 static int fixup_owner(u32 __user
*uaddr
, struct futex_q
*q
, int locked
)
1865 struct task_struct
*owner
;
1870 * Got the lock. We might not be the anticipated owner if we
1871 * did a lock-steal - fix up the PI-state in that case:
1873 if (q
->pi_state
->owner
!= current
)
1874 ret
= fixup_pi_state_owner(uaddr
, q
, current
);
1879 * Catch the rare case, where the lock was released when we were on the
1880 * way back before we locked the hash bucket.
1882 if (q
->pi_state
->owner
== current
) {
1884 * Try to get the rt_mutex now. This might fail as some other
1885 * task acquired the rt_mutex after we removed ourself from the
1886 * rt_mutex waiters list.
1888 if (rt_mutex_trylock(&q
->pi_state
->pi_mutex
)) {
1894 * pi_state is incorrect, some other task did a lock steal and
1895 * we returned due to timeout or signal without taking the
1896 * rt_mutex. Too late.
1898 raw_spin_lock(&q
->pi_state
->pi_mutex
.wait_lock
);
1899 owner
= rt_mutex_owner(&q
->pi_state
->pi_mutex
);
1901 owner
= rt_mutex_next_owner(&q
->pi_state
->pi_mutex
);
1902 raw_spin_unlock(&q
->pi_state
->pi_mutex
.wait_lock
);
1903 ret
= fixup_pi_state_owner(uaddr
, q
, owner
);
1908 * Paranoia check. If we did not take the lock, then we should not be
1909 * the owner of the rt_mutex.
1911 if (rt_mutex_owner(&q
->pi_state
->pi_mutex
) == current
)
1912 printk(KERN_ERR
"fixup_owner: ret = %d pi-mutex: %p "
1913 "pi-state %p\n", ret
,
1914 q
->pi_state
->pi_mutex
.owner
,
1915 q
->pi_state
->owner
);
1918 return ret
? ret
: locked
;
1922 * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
1923 * @hb: the futex hash bucket, must be locked by the caller
1924 * @q: the futex_q to queue up on
1925 * @timeout: the prepared hrtimer_sleeper, or null for no timeout
1927 static void futex_wait_queue_me(struct futex_hash_bucket
*hb
, struct futex_q
*q
,
1928 struct hrtimer_sleeper
*timeout
)
1931 * The task state is guaranteed to be set before another task can
1932 * wake it. set_current_state() is implemented using set_mb() and
1933 * queue_me() calls spin_unlock() upon completion, both serializing
1934 * access to the hash list and forcing another memory barrier.
1936 set_current_state(TASK_INTERRUPTIBLE
);
1941 hrtimer_start_expires(&timeout
->timer
, HRTIMER_MODE_ABS
);
1942 if (!hrtimer_active(&timeout
->timer
))
1943 timeout
->task
= NULL
;
1947 * If we have been removed from the hash list, then another task
1948 * has tried to wake us, and we can skip the call to schedule().
1950 if (likely(!plist_node_empty(&q
->list
))) {
1952 * If the timer has already expired, current will already be
1953 * flagged for rescheduling. Only call schedule if there
1954 * is no timeout, or if it has yet to expire.
1956 if (!timeout
|| timeout
->task
)
1957 freezable_schedule();
1959 __set_current_state(TASK_RUNNING
);
1963 * futex_wait_setup() - Prepare to wait on a futex
1964 * @uaddr: the futex userspace address
1965 * @val: the expected value
1966 * @flags: futex flags (FLAGS_SHARED, etc.)
1967 * @q: the associated futex_q
1968 * @hb: storage for hash_bucket pointer to be returned to caller
1970 * Setup the futex_q and locate the hash_bucket. Get the futex value and
1971 * compare it with the expected value. Handle atomic faults internally.
1972 * Return with the hb lock held and a q.key reference on success, and unlocked
1973 * with no q.key reference on failure.
1976 * 0 - uaddr contains val and hb has been locked;
1977 * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked
1979 static int futex_wait_setup(u32 __user
*uaddr
, u32 val
, unsigned int flags
,
1980 struct futex_q
*q
, struct futex_hash_bucket
**hb
)
1986 * Access the page AFTER the hash-bucket is locked.
1987 * Order is important:
1989 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1990 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1992 * The basic logical guarantee of a futex is that it blocks ONLY
1993 * if cond(var) is known to be true at the time of blocking, for
1994 * any cond. If we locked the hash-bucket after testing *uaddr, that
1995 * would open a race condition where we could block indefinitely with
1996 * cond(var) false, which would violate the guarantee.
1998 * On the other hand, we insert q and release the hash-bucket only
1999 * after testing *uaddr. This guarantees that futex_wait() will NOT
2000 * absorb a wakeup if *uaddr does not match the desired values
2001 * while the syscall executes.
2004 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &q
->key
, VERIFY_READ
);
2005 if (unlikely(ret
!= 0))
2009 *hb
= queue_lock(q
);
2011 ret
= get_futex_value_locked(&uval
, uaddr
);
2016 ret
= get_user(uval
, uaddr
);
2020 if (!(flags
& FLAGS_SHARED
))
2023 put_futex_key(&q
->key
);
2034 put_futex_key(&q
->key
);
2038 static int futex_wait(u32 __user
*uaddr
, unsigned int flags
, u32 val
,
2039 ktime_t
*abs_time
, u32 bitset
)
2041 struct hrtimer_sleeper timeout
, *to
= NULL
;
2042 struct restart_block
*restart
;
2043 struct futex_hash_bucket
*hb
;
2044 struct futex_q q
= futex_q_init
;
2054 hrtimer_init_on_stack(&to
->timer
, (flags
& FLAGS_CLOCKRT
) ?
2055 CLOCK_REALTIME
: CLOCK_MONOTONIC
,
2057 hrtimer_init_sleeper(to
, current
);
2058 hrtimer_set_expires_range_ns(&to
->timer
, *abs_time
,
2059 current
->timer_slack_ns
);
2064 * Prepare to wait on uaddr. On success, holds hb lock and increments
2067 ret
= futex_wait_setup(uaddr
, val
, flags
, &q
, &hb
);
2071 /* queue_me and wait for wakeup, timeout, or a signal. */
2072 futex_wait_queue_me(hb
, &q
, to
);
2074 /* If we were woken (and unqueued), we succeeded, whatever. */
2076 /* unqueue_me() drops q.key ref */
2077 if (!unqueue_me(&q
))
2080 if (to
&& !to
->task
)
2084 * We expect signal_pending(current), but we might be the
2085 * victim of a spurious wakeup as well.
2087 if (!signal_pending(current
))
2094 restart
= ¤t_thread_info()->restart_block
;
2095 restart
->fn
= futex_wait_restart
;
2096 restart
->futex
.uaddr
= uaddr
;
2097 restart
->futex
.val
= val
;
2098 restart
->futex
.time
= abs_time
->tv64
;
2099 restart
->futex
.bitset
= bitset
;
2100 restart
->futex
.flags
= flags
| FLAGS_HAS_TIMEOUT
;
2102 ret
= -ERESTART_RESTARTBLOCK
;
2106 hrtimer_cancel(&to
->timer
);
2107 destroy_hrtimer_on_stack(&to
->timer
);
2113 static long futex_wait_restart(struct restart_block
*restart
)
2115 u32 __user
*uaddr
= restart
->futex
.uaddr
;
2116 ktime_t t
, *tp
= NULL
;
2118 if (restart
->futex
.flags
& FLAGS_HAS_TIMEOUT
) {
2119 t
.tv64
= restart
->futex
.time
;
2122 restart
->fn
= do_no_restart_syscall
;
2124 return (long)futex_wait(uaddr
, restart
->futex
.flags
,
2125 restart
->futex
.val
, tp
, restart
->futex
.bitset
);
2130 * Userspace tried a 0 -> TID atomic transition of the futex value
2131 * and failed. The kernel side here does the whole locking operation:
2132 * if there are waiters then it will block, it does PI, etc. (Due to
2133 * races the kernel might see a 0 value of the futex too.)
2135 static int futex_lock_pi(u32 __user
*uaddr
, unsigned int flags
, int detect
,
2136 ktime_t
*time
, int trylock
)
2138 struct hrtimer_sleeper timeout
, *to
= NULL
;
2139 struct futex_hash_bucket
*hb
;
2140 struct futex_q q
= futex_q_init
;
2143 if (refill_pi_state_cache())
2148 hrtimer_init_on_stack(&to
->timer
, CLOCK_REALTIME
,
2150 hrtimer_init_sleeper(to
, current
);
2151 hrtimer_set_expires(&to
->timer
, *time
);
2155 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &q
.key
, VERIFY_WRITE
);
2156 if (unlikely(ret
!= 0))
2160 hb
= queue_lock(&q
);
2162 ret
= futex_lock_pi_atomic(uaddr
, hb
, &q
.key
, &q
.pi_state
, current
, 0);
2163 if (unlikely(ret
)) {
2166 /* We got the lock. */
2168 goto out_unlock_put_key
;
2173 * Task is exiting and we just wait for the
2177 put_futex_key(&q
.key
);
2181 goto out_unlock_put_key
;
2186 * Only actually queue now that the atomic ops are done:
2190 WARN_ON(!q
.pi_state
);
2192 * Block on the PI mutex:
2195 ret
= rt_mutex_timed_lock(&q
.pi_state
->pi_mutex
, to
, 1);
2197 ret
= rt_mutex_trylock(&q
.pi_state
->pi_mutex
);
2198 /* Fixup the trylock return value: */
2199 ret
= ret
? 0 : -EWOULDBLOCK
;
2202 spin_lock(q
.lock_ptr
);
2204 * Fixup the pi_state owner and possibly acquire the lock if we
2207 res
= fixup_owner(uaddr
, &q
, !ret
);
2209 * If fixup_owner() returned an error, proprogate that. If it acquired
2210 * the lock, clear our -ETIMEDOUT or -EINTR.
2213 ret
= (res
< 0) ? res
: 0;
2216 * If fixup_owner() faulted and was unable to handle the fault, unlock
2217 * it and return the fault to userspace.
2219 if (ret
&& (rt_mutex_owner(&q
.pi_state
->pi_mutex
) == current
))
2220 rt_mutex_unlock(&q
.pi_state
->pi_mutex
);
2222 /* Unqueue and drop the lock */
2231 put_futex_key(&q
.key
);
2234 destroy_hrtimer_on_stack(&to
->timer
);
2235 return ret
!= -EINTR
? ret
: -ERESTARTNOINTR
;
2240 ret
= fault_in_user_writeable(uaddr
);
2244 if (!(flags
& FLAGS_SHARED
))
2247 put_futex_key(&q
.key
);
2252 * Userspace attempted a TID -> 0 atomic transition, and failed.
2253 * This is the in-kernel slowpath: we look up the PI state (if any),
2254 * and do the rt-mutex unlock.
2256 static int futex_unlock_pi(u32 __user
*uaddr
, unsigned int flags
)
2258 struct futex_hash_bucket
*hb
;
2259 struct futex_q
*this, *next
;
2260 union futex_key key
= FUTEX_KEY_INIT
;
2261 u32 uval
, vpid
= task_pid_vnr(current
);
2265 if (get_user(uval
, uaddr
))
2268 * We release only a lock we actually own:
2270 if ((uval
& FUTEX_TID_MASK
) != vpid
)
2273 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &key
, VERIFY_WRITE
);
2274 if (unlikely(ret
!= 0))
2277 hb
= hash_futex(&key
);
2278 spin_lock(&hb
->lock
);
2281 * To avoid races, try to do the TID -> 0 atomic transition
2282 * again. If it succeeds then we can return without waking
2285 if (!(uval
& FUTEX_OWNER_DIED
) &&
2286 cmpxchg_futex_value_locked(&uval
, uaddr
, vpid
, 0))
2289 * Rare case: we managed to release the lock atomically,
2290 * no need to wake anyone else up:
2292 if (unlikely(uval
== vpid
))
2296 * Ok, other tasks may need to be woken up - check waiters
2297 * and do the wakeup if necessary:
2299 plist_for_each_entry_safe(this, next
, &hb
->chain
, list
) {
2300 if (!match_futex (&this->key
, &key
))
2302 ret
= wake_futex_pi(uaddr
, uval
, this);
2304 * The atomic access to the futex value
2305 * generated a pagefault, so retry the
2306 * user-access and the wakeup:
2313 * No waiters - kernel unlocks the futex:
2315 if (!(uval
& FUTEX_OWNER_DIED
)) {
2316 ret
= unlock_futex_pi(uaddr
, uval
);
2322 spin_unlock(&hb
->lock
);
2323 put_futex_key(&key
);
2329 spin_unlock(&hb
->lock
);
2330 put_futex_key(&key
);
2332 ret
= fault_in_user_writeable(uaddr
);
2340 * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2341 * @hb: the hash_bucket futex_q was original enqueued on
2342 * @q: the futex_q woken while waiting to be requeued
2343 * @key2: the futex_key of the requeue target futex
2344 * @timeout: the timeout associated with the wait (NULL if none)
2346 * Detect if the task was woken on the initial futex as opposed to the requeue
2347 * target futex. If so, determine if it was a timeout or a signal that caused
2348 * the wakeup and return the appropriate error code to the caller. Must be
2349 * called with the hb lock held.
2352 * 0 = no early wakeup detected;
2353 * <0 = -ETIMEDOUT or -ERESTARTNOINTR
2356 int handle_early_requeue_pi_wakeup(struct futex_hash_bucket
*hb
,
2357 struct futex_q
*q
, union futex_key
*key2
,
2358 struct hrtimer_sleeper
*timeout
)
2363 * With the hb lock held, we avoid races while we process the wakeup.
2364 * We only need to hold hb (and not hb2) to ensure atomicity as the
2365 * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2366 * It can't be requeued from uaddr2 to something else since we don't
2367 * support a PI aware source futex for requeue.
2369 if (!match_futex(&q
->key
, key2
)) {
2370 WARN_ON(q
->lock_ptr
&& (&hb
->lock
!= q
->lock_ptr
));
2372 * We were woken prior to requeue by a timeout or a signal.
2373 * Unqueue the futex_q and determine which it was.
2375 plist_del(&q
->list
, &hb
->chain
);
2378 /* Handle spurious wakeups gracefully */
2380 if (timeout
&& !timeout
->task
)
2382 else if (signal_pending(current
))
2383 ret
= -ERESTARTNOINTR
;
2389 * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
2390 * @uaddr: the futex we initially wait on (non-pi)
2391 * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be
2392 * the same type, no requeueing from private to shared, etc.
2393 * @val: the expected value of uaddr
2394 * @abs_time: absolute timeout
2395 * @bitset: 32 bit wakeup bitset set by userspace, defaults to all
2396 * @uaddr2: the pi futex we will take prior to returning to user-space
2398 * The caller will wait on uaddr and will be requeued by futex_requeue() to
2399 * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake
2400 * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to
2401 * userspace. This ensures the rt_mutex maintains an owner when it has waiters;
2402 * without one, the pi logic would not know which task to boost/deboost, if
2403 * there was a need to.
2405 * We call schedule in futex_wait_queue_me() when we enqueue and return there
2406 * via the following--
2407 * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
2408 * 2) wakeup on uaddr2 after a requeue
2412 * If 3, cleanup and return -ERESTARTNOINTR.
2414 * If 2, we may then block on trying to take the rt_mutex and return via:
2415 * 5) successful lock
2418 * 8) other lock acquisition failure
2420 * If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
2422 * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2428 static int futex_wait_requeue_pi(u32 __user
*uaddr
, unsigned int flags
,
2429 u32 val
, ktime_t
*abs_time
, u32 bitset
,
2432 struct hrtimer_sleeper timeout
, *to
= NULL
;
2433 struct rt_mutex_waiter rt_waiter
;
2434 struct rt_mutex
*pi_mutex
= NULL
;
2435 struct futex_hash_bucket
*hb
;
2436 union futex_key key2
= FUTEX_KEY_INIT
;
2437 struct futex_q q
= futex_q_init
;
2440 if (uaddr
== uaddr2
)
2448 hrtimer_init_on_stack(&to
->timer
, (flags
& FLAGS_CLOCKRT
) ?
2449 CLOCK_REALTIME
: CLOCK_MONOTONIC
,
2451 hrtimer_init_sleeper(to
, current
);
2452 hrtimer_set_expires_range_ns(&to
->timer
, *abs_time
,
2453 current
->timer_slack_ns
);
2457 * The waiter is allocated on our stack, manipulated by the requeue
2458 * code while we sleep on uaddr.
2460 debug_rt_mutex_init_waiter(&rt_waiter
);
2461 RB_CLEAR_NODE(&rt_waiter
.pi_tree_entry
);
2462 RB_CLEAR_NODE(&rt_waiter
.tree_entry
);
2463 rt_waiter
.task
= NULL
;
2465 ret
= get_futex_key(uaddr2
, flags
& FLAGS_SHARED
, &key2
, VERIFY_WRITE
);
2466 if (unlikely(ret
!= 0))
2470 q
.rt_waiter
= &rt_waiter
;
2471 q
.requeue_pi_key
= &key2
;
2474 * Prepare to wait on uaddr. On success, increments q.key (key1) ref
2477 ret
= futex_wait_setup(uaddr
, val
, flags
, &q
, &hb
);
2481 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
2482 futex_wait_queue_me(hb
, &q
, to
);
2484 spin_lock(&hb
->lock
);
2485 ret
= handle_early_requeue_pi_wakeup(hb
, &q
, &key2
, to
);
2486 spin_unlock(&hb
->lock
);
2491 * In order for us to be here, we know our q.key == key2, and since
2492 * we took the hb->lock above, we also know that futex_requeue() has
2493 * completed and we no longer have to concern ourselves with a wakeup
2494 * race with the atomic proxy lock acquisition by the requeue code. The
2495 * futex_requeue dropped our key1 reference and incremented our key2
2499 /* Check if the requeue code acquired the second futex for us. */
2502 * Got the lock. We might not be the anticipated owner if we
2503 * did a lock-steal - fix up the PI-state in that case.
2505 if (q
.pi_state
&& (q
.pi_state
->owner
!= current
)) {
2506 spin_lock(q
.lock_ptr
);
2507 ret
= fixup_pi_state_owner(uaddr2
, &q
, current
);
2508 spin_unlock(q
.lock_ptr
);
2512 * We have been woken up by futex_unlock_pi(), a timeout, or a
2513 * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2516 WARN_ON(!q
.pi_state
);
2517 pi_mutex
= &q
.pi_state
->pi_mutex
;
2518 ret
= rt_mutex_finish_proxy_lock(pi_mutex
, to
, &rt_waiter
, 1);
2519 debug_rt_mutex_free_waiter(&rt_waiter
);
2521 spin_lock(q
.lock_ptr
);
2523 * Fixup the pi_state owner and possibly acquire the lock if we
2526 res
= fixup_owner(uaddr2
, &q
, !ret
);
2528 * If fixup_owner() returned an error, proprogate that. If it
2529 * acquired the lock, clear -ETIMEDOUT or -EINTR.
2532 ret
= (res
< 0) ? res
: 0;
2534 /* Unqueue and drop the lock. */
2539 * If fixup_pi_state_owner() faulted and was unable to handle the
2540 * fault, unlock the rt_mutex and return the fault to userspace.
2542 if (ret
== -EFAULT
) {
2543 if (pi_mutex
&& rt_mutex_owner(pi_mutex
) == current
)
2544 rt_mutex_unlock(pi_mutex
);
2545 } else if (ret
== -EINTR
) {
2547 * We've already been requeued, but cannot restart by calling
2548 * futex_lock_pi() directly. We could restart this syscall, but
2549 * it would detect that the user space "val" changed and return
2550 * -EWOULDBLOCK. Save the overhead of the restart and return
2551 * -EWOULDBLOCK directly.
2557 put_futex_key(&q
.key
);
2559 put_futex_key(&key2
);
2563 hrtimer_cancel(&to
->timer
);
2564 destroy_hrtimer_on_stack(&to
->timer
);
2570 * Support for robust futexes: the kernel cleans up held futexes at
2573 * Implementation: user-space maintains a per-thread list of locks it
2574 * is holding. Upon do_exit(), the kernel carefully walks this list,
2575 * and marks all locks that are owned by this thread with the
2576 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
2577 * always manipulated with the lock held, so the list is private and
2578 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2579 * field, to allow the kernel to clean up if the thread dies after
2580 * acquiring the lock, but just before it could have added itself to
2581 * the list. There can only be one such pending lock.
2585 * sys_set_robust_list() - Set the robust-futex list head of a task
2586 * @head: pointer to the list-head
2587 * @len: length of the list-head, as userspace expects
2589 SYSCALL_DEFINE2(set_robust_list
, struct robust_list_head __user
*, head
,
2592 if (!futex_cmpxchg_enabled
)
2595 * The kernel knows only one size for now:
2597 if (unlikely(len
!= sizeof(*head
)))
2600 current
->robust_list
= head
;
2606 * sys_get_robust_list() - Get the robust-futex list head of a task
2607 * @pid: pid of the process [zero for current task]
2608 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2609 * @len_ptr: pointer to a length field, the kernel fills in the header size
2611 SYSCALL_DEFINE3(get_robust_list
, int, pid
,
2612 struct robust_list_head __user
* __user
*, head_ptr
,
2613 size_t __user
*, len_ptr
)
2615 struct robust_list_head __user
*head
;
2617 struct task_struct
*p
;
2619 if (!futex_cmpxchg_enabled
)
2628 p
= find_task_by_vpid(pid
);
2634 if (!ptrace_may_access(p
, PTRACE_MODE_READ
))
2637 head
= p
->robust_list
;
2640 if (put_user(sizeof(*head
), len_ptr
))
2642 return put_user(head
, head_ptr
);
2651 * Process a futex-list entry, check whether it's owned by the
2652 * dying task, and do notification if so:
2654 int handle_futex_death(u32 __user
*uaddr
, struct task_struct
*curr
, int pi
)
2656 u32 uval
, uninitialized_var(nval
), mval
;
2659 if (get_user(uval
, uaddr
))
2662 if ((uval
& FUTEX_TID_MASK
) == task_pid_vnr(curr
)) {
2664 * Ok, this dying thread is truly holding a futex
2665 * of interest. Set the OWNER_DIED bit atomically
2666 * via cmpxchg, and if the value had FUTEX_WAITERS
2667 * set, wake up a waiter (if any). (We have to do a
2668 * futex_wake() even if OWNER_DIED is already set -
2669 * to handle the rare but possible case of recursive
2670 * thread-death.) The rest of the cleanup is done in
2673 mval
= (uval
& FUTEX_WAITERS
) | FUTEX_OWNER_DIED
;
2675 * We are not holding a lock here, but we want to have
2676 * the pagefault_disable/enable() protection because
2677 * we want to handle the fault gracefully. If the
2678 * access fails we try to fault in the futex with R/W
2679 * verification via get_user_pages. get_user() above
2680 * does not guarantee R/W access. If that fails we
2681 * give up and leave the futex locked.
2683 if (cmpxchg_futex_value_locked(&nval
, uaddr
, uval
, mval
)) {
2684 if (fault_in_user_writeable(uaddr
))
2692 * Wake robust non-PI futexes here. The wakeup of
2693 * PI futexes happens in exit_pi_state():
2695 if (!pi
&& (uval
& FUTEX_WAITERS
))
2696 futex_wake(uaddr
, 1, 1, FUTEX_BITSET_MATCH_ANY
);
2702 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2704 static inline int fetch_robust_entry(struct robust_list __user
**entry
,
2705 struct robust_list __user
* __user
*head
,
2708 unsigned long uentry
;
2710 if (get_user(uentry
, (unsigned long __user
*)head
))
2713 *entry
= (void __user
*)(uentry
& ~1UL);
2720 * Walk curr->robust_list (very carefully, it's a userspace list!)
2721 * and mark any locks found there dead, and notify any waiters.
2723 * We silently return on any sign of list-walking problem.
2725 void exit_robust_list(struct task_struct
*curr
)
2727 struct robust_list_head __user
*head
= curr
->robust_list
;
2728 struct robust_list __user
*entry
, *next_entry
, *pending
;
2729 unsigned int limit
= ROBUST_LIST_LIMIT
, pi
, pip
;
2730 unsigned int uninitialized_var(next_pi
);
2731 unsigned long futex_offset
;
2734 if (!futex_cmpxchg_enabled
)
2738 * Fetch the list head (which was registered earlier, via
2739 * sys_set_robust_list()):
2741 if (fetch_robust_entry(&entry
, &head
->list
.next
, &pi
))
2744 * Fetch the relative futex offset:
2746 if (get_user(futex_offset
, &head
->futex_offset
))
2749 * Fetch any possibly pending lock-add first, and handle it
2752 if (fetch_robust_entry(&pending
, &head
->list_op_pending
, &pip
))
2755 next_entry
= NULL
; /* avoid warning with gcc */
2756 while (entry
!= &head
->list
) {
2758 * Fetch the next entry in the list before calling
2759 * handle_futex_death:
2761 rc
= fetch_robust_entry(&next_entry
, &entry
->next
, &next_pi
);
2763 * A pending lock might already be on the list, so
2764 * don't process it twice:
2766 if (entry
!= pending
)
2767 if (handle_futex_death((void __user
*)entry
+ futex_offset
,
2775 * Avoid excessively long or circular lists:
2784 handle_futex_death((void __user
*)pending
+ futex_offset
,
2788 long do_futex(u32 __user
*uaddr
, int op
, u32 val
, ktime_t
*timeout
,
2789 u32 __user
*uaddr2
, u32 val2
, u32 val3
)
2791 int cmd
= op
& FUTEX_CMD_MASK
;
2792 unsigned int flags
= 0;
2794 if (!(op
& FUTEX_PRIVATE_FLAG
))
2795 flags
|= FLAGS_SHARED
;
2797 if (op
& FUTEX_CLOCK_REALTIME
) {
2798 flags
|= FLAGS_CLOCKRT
;
2799 if (cmd
!= FUTEX_WAIT_BITSET
&& cmd
!= FUTEX_WAIT_REQUEUE_PI
)
2805 case FUTEX_UNLOCK_PI
:
2806 case FUTEX_TRYLOCK_PI
:
2807 case FUTEX_WAIT_REQUEUE_PI
:
2808 case FUTEX_CMP_REQUEUE_PI
:
2809 if (!futex_cmpxchg_enabled
)
2815 val3
= FUTEX_BITSET_MATCH_ANY
;
2816 case FUTEX_WAIT_BITSET
:
2817 return futex_wait(uaddr
, flags
, val
, timeout
, val3
);
2819 val3
= FUTEX_BITSET_MATCH_ANY
;
2820 case FUTEX_WAKE_BITSET
:
2821 return futex_wake(uaddr
, flags
, val
, val3
);
2823 return futex_requeue(uaddr
, flags
, uaddr2
, val
, val2
, NULL
, 0);
2824 case FUTEX_CMP_REQUEUE
:
2825 return futex_requeue(uaddr
, flags
, uaddr2
, val
, val2
, &val3
, 0);
2827 return futex_wake_op(uaddr
, flags
, uaddr2
, val
, val2
, val3
);
2829 return futex_lock_pi(uaddr
, flags
, val
, timeout
, 0);
2830 case FUTEX_UNLOCK_PI
:
2831 return futex_unlock_pi(uaddr
, flags
);
2832 case FUTEX_TRYLOCK_PI
:
2833 return futex_lock_pi(uaddr
, flags
, 0, timeout
, 1);
2834 case FUTEX_WAIT_REQUEUE_PI
:
2835 val3
= FUTEX_BITSET_MATCH_ANY
;
2836 return futex_wait_requeue_pi(uaddr
, flags
, val
, timeout
, val3
,
2838 case FUTEX_CMP_REQUEUE_PI
:
2839 return futex_requeue(uaddr
, flags
, uaddr2
, val
, val2
, &val3
, 1);
2845 SYSCALL_DEFINE6(futex
, u32 __user
*, uaddr
, int, op
, u32
, val
,
2846 struct timespec __user
*, utime
, u32 __user
*, uaddr2
,
2850 ktime_t t
, *tp
= NULL
;
2852 int cmd
= op
& FUTEX_CMD_MASK
;
2854 if (utime
&& (cmd
== FUTEX_WAIT
|| cmd
== FUTEX_LOCK_PI
||
2855 cmd
== FUTEX_WAIT_BITSET
||
2856 cmd
== FUTEX_WAIT_REQUEUE_PI
)) {
2857 if (copy_from_user(&ts
, utime
, sizeof(ts
)) != 0)
2859 if (!timespec_valid(&ts
))
2862 t
= timespec_to_ktime(ts
);
2863 if (cmd
== FUTEX_WAIT
)
2864 t
= ktime_add_safe(ktime_get(), t
);
2868 * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
2869 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
2871 if (cmd
== FUTEX_REQUEUE
|| cmd
== FUTEX_CMP_REQUEUE
||
2872 cmd
== FUTEX_CMP_REQUEUE_PI
|| cmd
== FUTEX_WAKE_OP
)
2873 val2
= (u32
) (unsigned long) utime
;
2875 return do_futex(uaddr
, op
, val
, tp
, uaddr2
, val2
, val3
);
2878 static int __init
futex_init(void)
2881 unsigned int futex_shift
;
2884 #if CONFIG_BASE_SMALL
2885 futex_hashsize
= 16;
2887 futex_hashsize
= roundup_pow_of_two(256 * num_possible_cpus());
2890 futex_queues
= alloc_large_system_hash("futex", sizeof(*futex_queues
),
2892 futex_hashsize
< 256 ? HASH_SMALL
: 0,
2894 futex_hashsize
, futex_hashsize
);
2895 futex_hashsize
= 1UL << futex_shift
;
2897 * This will fail and we want it. Some arch implementations do
2898 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2899 * functionality. We want to know that before we call in any
2900 * of the complex code paths. Also we want to prevent
2901 * registration of robust lists in that case. NULL is
2902 * guaranteed to fault and we get -EFAULT on functional
2903 * implementation, the non-functional ones will return
2906 if (cmpxchg_futex_value_locked(&curval
, NULL
, 0, 0) == -EFAULT
)
2907 futex_cmpxchg_enabled
= 1;
2909 for (i
= 0; i
< futex_hashsize
; i
++) {
2910 atomic_set(&futex_queues
[i
].waiters
, 0);
2911 plist_head_init(&futex_queues
[i
].chain
);
2912 spin_lock_init(&futex_queues
[i
].lock
);
2917 __initcall(futex_init
);