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>
67 #include <asm/futex.h>
69 #include "rtmutex_common.h"
71 #ifndef CONFIG_HAVE_FUTEX_CMPXCHG
72 int __read_mostly futex_cmpxchg_enabled
;
75 #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
78 * Futex flags used to encode options to functions and preserve them across
81 #define FLAGS_SHARED 0x01
82 #define FLAGS_CLOCKRT 0x02
83 #define FLAGS_HAS_TIMEOUT 0x04
86 * Priority Inheritance state:
88 struct futex_pi_state
{
90 * list of 'owned' pi_state instances - these have to be
91 * cleaned up in do_exit() if the task exits prematurely:
93 struct list_head list
;
98 struct rt_mutex pi_mutex
;
100 struct task_struct
*owner
;
107 * struct futex_q - The hashed futex queue entry, one per waiting task
108 * @list: priority-sorted list of tasks waiting on this futex
109 * @task: the task waiting on the futex
110 * @lock_ptr: the hash bucket lock
111 * @key: the key the futex is hashed on
112 * @pi_state: optional priority inheritance state
113 * @rt_waiter: rt_waiter storage for use with requeue_pi
114 * @requeue_pi_key: the requeue_pi target futex key
115 * @bitset: bitset for the optional bitmasked wakeup
117 * We use this hashed waitqueue, instead of a normal wait_queue_t, so
118 * we can wake only the relevant ones (hashed queues may be shared).
120 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
121 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
122 * The order of wakeup is always to make the first condition true, then
125 * PI futexes are typically woken before they are removed from the hash list via
126 * the rt_mutex code. See unqueue_me_pi().
129 struct plist_node list
;
131 struct task_struct
*task
;
132 spinlock_t
*lock_ptr
;
134 struct futex_pi_state
*pi_state
;
135 struct rt_mutex_waiter
*rt_waiter
;
136 union futex_key
*requeue_pi_key
;
140 static const struct futex_q futex_q_init
= {
141 /* list gets initialized in queue_me()*/
142 .key
= FUTEX_KEY_INIT
,
143 .bitset
= FUTEX_BITSET_MATCH_ANY
147 * Hash buckets are shared by all the futex_keys that hash to the same
148 * location. Each key may have multiple futex_q structures, one for each task
149 * waiting on a futex.
151 struct futex_hash_bucket
{
153 struct plist_head chain
;
156 static struct futex_hash_bucket futex_queues
[1<<FUTEX_HASHBITS
];
159 * We hash on the keys returned from get_futex_key (see below).
161 static struct futex_hash_bucket
*hash_futex(union futex_key
*key
)
163 u32 hash
= jhash2((u32
*)&key
->both
.word
,
164 (sizeof(key
->both
.word
)+sizeof(key
->both
.ptr
))/4,
166 return &futex_queues
[hash
& ((1 << FUTEX_HASHBITS
)-1)];
170 * Return 1 if two futex_keys are equal, 0 otherwise.
172 static inline int match_futex(union futex_key
*key1
, union futex_key
*key2
)
175 && key1
->both
.word
== key2
->both
.word
176 && key1
->both
.ptr
== key2
->both
.ptr
177 && key1
->both
.offset
== key2
->both
.offset
);
181 * Take a reference to the resource addressed by a key.
182 * Can be called while holding spinlocks.
185 static void get_futex_key_refs(union futex_key
*key
)
190 switch (key
->both
.offset
& (FUT_OFF_INODE
|FUT_OFF_MMSHARED
)) {
192 ihold(key
->shared
.inode
);
194 case FUT_OFF_MMSHARED
:
195 atomic_inc(&key
->private.mm
->mm_count
);
201 * Drop a reference to the resource addressed by a key.
202 * The hash bucket spinlock must not be held.
204 static void drop_futex_key_refs(union futex_key
*key
)
206 if (!key
->both
.ptr
) {
207 /* If we're here then we tried to put a key we failed to get */
212 switch (key
->both
.offset
& (FUT_OFF_INODE
|FUT_OFF_MMSHARED
)) {
214 iput(key
->shared
.inode
);
216 case FUT_OFF_MMSHARED
:
217 mmdrop(key
->private.mm
);
223 * get_futex_key() - Get parameters which are the keys for a futex
224 * @uaddr: virtual address of the futex
225 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
226 * @key: address where result is stored.
227 * @rw: mapping needs to be read/write (values: VERIFY_READ,
230 * Return: a negative error code or 0
232 * The key words are stored in *key on success.
234 * For shared mappings, it's (page->index, file_inode(vma->vm_file),
235 * offset_within_page). For private mappings, it's (uaddr, current->mm).
236 * We can usually work out the index without swapping in the page.
238 * lock_page() might sleep, the caller should not hold a spinlock.
241 get_futex_key(u32 __user
*uaddr
, int fshared
, union futex_key
*key
, int rw
)
243 unsigned long address
= (unsigned long)uaddr
;
244 struct mm_struct
*mm
= current
->mm
;
245 struct page
*page
, *page_head
;
249 * The futex address must be "naturally" aligned.
251 key
->both
.offset
= address
% PAGE_SIZE
;
252 if (unlikely((address
% sizeof(u32
)) != 0))
254 address
-= key
->both
.offset
;
257 * PROCESS_PRIVATE futexes are fast.
258 * As the mm cannot disappear under us and the 'key' only needs
259 * virtual address, we dont even have to find the underlying vma.
260 * Note : We do have to check 'uaddr' is a valid user address,
261 * but access_ok() should be faster than find_vma()
264 if (unlikely(!access_ok(VERIFY_WRITE
, uaddr
, sizeof(u32
))))
266 key
->private.mm
= mm
;
267 key
->private.address
= address
;
268 get_futex_key_refs(key
);
273 err
= get_user_pages_fast(address
, 1, 1, &page
);
275 * If write access is not required (eg. FUTEX_WAIT), try
276 * and get read-only access.
278 if (err
== -EFAULT
&& rw
== VERIFY_READ
) {
279 err
= get_user_pages_fast(address
, 1, 0, &page
);
287 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
289 if (unlikely(PageTail(page
))) {
291 /* serialize against __split_huge_page_splitting() */
293 if (likely(__get_user_pages_fast(address
, 1, !ro
, &page
) == 1)) {
294 page_head
= compound_head(page
);
296 * page_head is valid pointer but we must pin
297 * it before taking the PG_lock and/or
298 * PG_compound_lock. The moment we re-enable
299 * irqs __split_huge_page_splitting() can
300 * return and the head page can be freed from
301 * under us. We can't take the PG_lock and/or
302 * PG_compound_lock on a page that could be
303 * freed from under us.
305 if (page
!= page_head
) {
316 page_head
= compound_head(page
);
317 if (page
!= page_head
) {
323 lock_page(page_head
);
326 * If page_head->mapping is NULL, then it cannot be a PageAnon
327 * page; but it might be the ZERO_PAGE or in the gate area or
328 * in a special mapping (all cases which we are happy to fail);
329 * or it may have been a good file page when get_user_pages_fast
330 * found it, but truncated or holepunched or subjected to
331 * invalidate_complete_page2 before we got the page lock (also
332 * cases which we are happy to fail). And we hold a reference,
333 * so refcount care in invalidate_complete_page's remove_mapping
334 * prevents drop_caches from setting mapping to NULL beneath us.
336 * The case we do have to guard against is when memory pressure made
337 * shmem_writepage move it from filecache to swapcache beneath us:
338 * an unlikely race, but we do need to retry for page_head->mapping.
340 if (!page_head
->mapping
) {
341 int shmem_swizzled
= PageSwapCache(page_head
);
342 unlock_page(page_head
);
350 * Private mappings are handled in a simple way.
352 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
353 * it's a read-only handle, it's expected that futexes attach to
354 * the object not the particular process.
356 if (PageAnon(page_head
)) {
358 * A RO anonymous page will never change and thus doesn't make
359 * sense for futex operations.
366 key
->both
.offset
|= FUT_OFF_MMSHARED
; /* ref taken on mm */
367 key
->private.mm
= mm
;
368 key
->private.address
= address
;
370 key
->both
.offset
|= FUT_OFF_INODE
; /* inode-based key */
371 key
->shared
.inode
= page_head
->mapping
->host
;
372 key
->shared
.pgoff
= basepage_index(page
);
375 get_futex_key_refs(key
);
378 unlock_page(page_head
);
383 static inline void put_futex_key(union futex_key
*key
)
385 drop_futex_key_refs(key
);
389 * fault_in_user_writeable() - Fault in user address and verify RW access
390 * @uaddr: pointer to faulting user space address
392 * Slow path to fixup the fault we just took in the atomic write
395 * We have no generic implementation of a non-destructive write to the
396 * user address. We know that we faulted in the atomic pagefault
397 * disabled section so we can as well avoid the #PF overhead by
398 * calling get_user_pages() right away.
400 static int fault_in_user_writeable(u32 __user
*uaddr
)
402 struct mm_struct
*mm
= current
->mm
;
405 down_read(&mm
->mmap_sem
);
406 ret
= fixup_user_fault(current
, mm
, (unsigned long)uaddr
,
408 up_read(&mm
->mmap_sem
);
410 return ret
< 0 ? ret
: 0;
414 * futex_top_waiter() - Return the highest priority waiter on a futex
415 * @hb: the hash bucket the futex_q's reside in
416 * @key: the futex key (to distinguish it from other futex futex_q's)
418 * Must be called with the hb lock held.
420 static struct futex_q
*futex_top_waiter(struct futex_hash_bucket
*hb
,
421 union futex_key
*key
)
423 struct futex_q
*this;
425 plist_for_each_entry(this, &hb
->chain
, list
) {
426 if (match_futex(&this->key
, key
))
432 static int cmpxchg_futex_value_locked(u32
*curval
, u32 __user
*uaddr
,
433 u32 uval
, u32 newval
)
438 ret
= futex_atomic_cmpxchg_inatomic(curval
, uaddr
, uval
, newval
);
444 static int get_futex_value_locked(u32
*dest
, u32 __user
*from
)
449 ret
= __copy_from_user_inatomic(dest
, from
, sizeof(u32
));
452 return ret
? -EFAULT
: 0;
459 static int refill_pi_state_cache(void)
461 struct futex_pi_state
*pi_state
;
463 if (likely(current
->pi_state_cache
))
466 pi_state
= kzalloc(sizeof(*pi_state
), GFP_KERNEL
);
471 INIT_LIST_HEAD(&pi_state
->list
);
472 /* pi_mutex gets initialized later */
473 pi_state
->owner
= NULL
;
474 atomic_set(&pi_state
->refcount
, 1);
475 pi_state
->key
= FUTEX_KEY_INIT
;
477 current
->pi_state_cache
= pi_state
;
482 static struct futex_pi_state
* alloc_pi_state(void)
484 struct futex_pi_state
*pi_state
= current
->pi_state_cache
;
487 current
->pi_state_cache
= NULL
;
492 static void free_pi_state(struct futex_pi_state
*pi_state
)
494 if (!atomic_dec_and_test(&pi_state
->refcount
))
498 * If pi_state->owner is NULL, the owner is most probably dying
499 * and has cleaned up the pi_state already
501 if (pi_state
->owner
) {
502 raw_spin_lock_irq(&pi_state
->owner
->pi_lock
);
503 list_del_init(&pi_state
->list
);
504 raw_spin_unlock_irq(&pi_state
->owner
->pi_lock
);
506 rt_mutex_proxy_unlock(&pi_state
->pi_mutex
, pi_state
->owner
);
509 if (current
->pi_state_cache
)
513 * pi_state->list is already empty.
514 * clear pi_state->owner.
515 * refcount is at 0 - put it back to 1.
517 pi_state
->owner
= NULL
;
518 atomic_set(&pi_state
->refcount
, 1);
519 current
->pi_state_cache
= pi_state
;
524 * Look up the task based on what TID userspace gave us.
527 static struct task_struct
* futex_find_get_task(pid_t pid
)
529 struct task_struct
*p
;
532 p
= find_task_by_vpid(pid
);
542 * This task is holding PI mutexes at exit time => bad.
543 * Kernel cleans up PI-state, but userspace is likely hosed.
544 * (Robust-futex cleanup is separate and might save the day for userspace.)
546 void exit_pi_state_list(struct task_struct
*curr
)
548 struct list_head
*next
, *head
= &curr
->pi_state_list
;
549 struct futex_pi_state
*pi_state
;
550 struct futex_hash_bucket
*hb
;
551 union futex_key key
= FUTEX_KEY_INIT
;
553 if (!futex_cmpxchg_enabled
)
556 * We are a ZOMBIE and nobody can enqueue itself on
557 * pi_state_list anymore, but we have to be careful
558 * versus waiters unqueueing themselves:
560 raw_spin_lock_irq(&curr
->pi_lock
);
561 while (!list_empty(head
)) {
564 pi_state
= list_entry(next
, struct futex_pi_state
, list
);
566 hb
= hash_futex(&key
);
567 raw_spin_unlock_irq(&curr
->pi_lock
);
569 spin_lock(&hb
->lock
);
571 raw_spin_lock_irq(&curr
->pi_lock
);
573 * We dropped the pi-lock, so re-check whether this
574 * task still owns the PI-state:
576 if (head
->next
!= next
) {
577 spin_unlock(&hb
->lock
);
581 WARN_ON(pi_state
->owner
!= curr
);
582 WARN_ON(list_empty(&pi_state
->list
));
583 list_del_init(&pi_state
->list
);
584 pi_state
->owner
= NULL
;
585 raw_spin_unlock_irq(&curr
->pi_lock
);
587 rt_mutex_unlock(&pi_state
->pi_mutex
);
589 spin_unlock(&hb
->lock
);
591 raw_spin_lock_irq(&curr
->pi_lock
);
593 raw_spin_unlock_irq(&curr
->pi_lock
);
597 * We need to check the following states:
599 * Waiter | pi_state | pi->owner | uTID | uODIED | ?
601 * [1] NULL | --- | --- | 0 | 0/1 | Valid
602 * [2] NULL | --- | --- | >0 | 0/1 | Valid
604 * [3] Found | NULL | -- | Any | 0/1 | Invalid
606 * [4] Found | Found | NULL | 0 | 1 | Valid
607 * [5] Found | Found | NULL | >0 | 1 | Invalid
609 * [6] Found | Found | task | 0 | 1 | Valid
611 * [7] Found | Found | NULL | Any | 0 | Invalid
613 * [8] Found | Found | task | ==taskTID | 0/1 | Valid
614 * [9] Found | Found | task | 0 | 0 | Invalid
615 * [10] Found | Found | task | !=taskTID | 0/1 | Invalid
617 * [1] Indicates that the kernel can acquire the futex atomically. We
618 * came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit.
620 * [2] Valid, if TID does not belong to a kernel thread. If no matching
621 * thread is found then it indicates that the owner TID has died.
623 * [3] Invalid. The waiter is queued on a non PI futex
625 * [4] Valid state after exit_robust_list(), which sets the user space
626 * value to FUTEX_WAITERS | FUTEX_OWNER_DIED.
628 * [5] The user space value got manipulated between exit_robust_list()
629 * and exit_pi_state_list()
631 * [6] Valid state after exit_pi_state_list() which sets the new owner in
632 * the pi_state but cannot access the user space value.
634 * [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set.
636 * [8] Owner and user space value match
638 * [9] There is no transient state which sets the user space TID to 0
639 * except exit_robust_list(), but this is indicated by the
640 * FUTEX_OWNER_DIED bit. See [4]
642 * [10] There is no transient state which leaves owner and user space
646 lookup_pi_state(u32 uval
, struct futex_hash_bucket
*hb
,
647 union futex_key
*key
, struct futex_pi_state
**ps
)
649 struct futex_pi_state
*pi_state
= NULL
;
650 struct futex_q
*this, *next
;
651 struct plist_head
*head
;
652 struct task_struct
*p
;
653 pid_t pid
= uval
& FUTEX_TID_MASK
;
657 plist_for_each_entry_safe(this, next
, head
, list
) {
658 if (match_futex(&this->key
, key
)) {
660 * Sanity check the waiter before increasing
661 * the refcount and attaching to it.
663 pi_state
= this->pi_state
;
665 * Userspace might have messed up non-PI and
668 if (unlikely(!pi_state
))
671 WARN_ON(!atomic_read(&pi_state
->refcount
));
674 * Handle the owner died case:
676 if (uval
& FUTEX_OWNER_DIED
) {
678 * exit_pi_state_list sets owner to NULL and
679 * wakes the topmost waiter. The task which
680 * acquires the pi_state->rt_mutex will fixup
683 if (!pi_state
->owner
) {
685 * No pi state owner, but the user
686 * space TID is not 0. Inconsistent
692 * Take a ref on the state and
699 * If TID is 0, then either the dying owner
700 * has not yet executed exit_pi_state_list()
701 * or some waiter acquired the rtmutex in the
702 * pi state, but did not yet fixup the TID in
705 * Take a ref on the state and return. [6]
711 * If the owner died bit is not set,
712 * then the pi_state must have an
715 if (!pi_state
->owner
)
720 * Bail out if user space manipulated the
721 * futex value. If pi state exists then the
722 * owner TID must be the same as the user
725 if (pid
!= task_pid_vnr(pi_state
->owner
))
729 atomic_inc(&pi_state
->refcount
);
736 * We are the first waiter - try to look up the real owner and attach
737 * the new pi_state to it, but bail out when TID = 0 [1]
741 p
= futex_find_get_task(pid
);
751 * We need to look at the task state flags to figure out,
752 * whether the task is exiting. To protect against the do_exit
753 * change of the task flags, we do this protected by
756 raw_spin_lock_irq(&p
->pi_lock
);
757 if (unlikely(p
->flags
& PF_EXITING
)) {
759 * The task is on the way out. When PF_EXITPIDONE is
760 * set, we know that the task has finished the
763 int ret
= (p
->flags
& PF_EXITPIDONE
) ? -ESRCH
: -EAGAIN
;
765 raw_spin_unlock_irq(&p
->pi_lock
);
771 * No existing pi state. First waiter. [2]
773 pi_state
= alloc_pi_state();
776 * Initialize the pi_mutex in locked state and make 'p'
779 rt_mutex_init_proxy_locked(&pi_state
->pi_mutex
, p
);
781 /* Store the key for possible exit cleanups: */
782 pi_state
->key
= *key
;
784 WARN_ON(!list_empty(&pi_state
->list
));
785 list_add(&pi_state
->list
, &p
->pi_state_list
);
787 raw_spin_unlock_irq(&p
->pi_lock
);
797 * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
798 * @uaddr: the pi futex user address
799 * @hb: the pi futex hash bucket
800 * @key: the futex key associated with uaddr and hb
801 * @ps: the pi_state pointer where we store the result of the
803 * @task: the task to perform the atomic lock work for. This will
804 * be "current" except in the case of requeue pi.
805 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
809 * 1 - acquired the lock;
812 * The hb->lock and futex_key refs shall be held by the caller.
814 static int futex_lock_pi_atomic(u32 __user
*uaddr
, struct futex_hash_bucket
*hb
,
815 union futex_key
*key
,
816 struct futex_pi_state
**ps
,
817 struct task_struct
*task
, int set_waiters
)
819 int lock_taken
, ret
, force_take
= 0;
820 u32 uval
, newval
, curval
, vpid
= task_pid_vnr(task
);
823 ret
= lock_taken
= 0;
826 * To avoid races, we attempt to take the lock here again
827 * (by doing a 0 -> TID atomic cmpxchg), while holding all
828 * the locks. It will most likely not succeed.
832 newval
|= FUTEX_WAITERS
;
834 if (unlikely(cmpxchg_futex_value_locked(&curval
, uaddr
, 0, newval
)))
840 if ((unlikely((curval
& FUTEX_TID_MASK
) == vpid
)))
844 * Surprise - we got the lock, but we do not trust user space at all.
846 if (unlikely(!curval
)) {
848 * We verify whether there is kernel state for this
849 * futex. If not, we can safely assume, that the 0 ->
850 * TID transition is correct. If state exists, we do
851 * not bother to fixup the user space state as it was
854 return futex_top_waiter(hb
, key
) ? -EINVAL
: 1;
860 * Set the FUTEX_WAITERS flag, so the owner will know it has someone
861 * to wake at the next unlock.
863 newval
= curval
| FUTEX_WAITERS
;
866 * Should we force take the futex? See below.
868 if (unlikely(force_take
)) {
870 * Keep the OWNER_DIED and the WAITERS bit and set the
873 newval
= (curval
& ~FUTEX_TID_MASK
) | vpid
;
878 if (unlikely(cmpxchg_futex_value_locked(&curval
, uaddr
, uval
, newval
)))
880 if (unlikely(curval
!= uval
))
884 * We took the lock due to forced take over.
886 if (unlikely(lock_taken
))
890 * We dont have the lock. Look up the PI state (or create it if
891 * we are the first waiter):
893 ret
= lookup_pi_state(uval
, hb
, key
, ps
);
899 * We failed to find an owner for this
900 * futex. So we have no pi_state to block
901 * on. This can happen in two cases:
904 * 2) A stale FUTEX_WAITERS bit
906 * Re-read the futex value.
908 if (get_futex_value_locked(&curval
, uaddr
))
912 * If the owner died or we have a stale
913 * WAITERS bit the owner TID in the user space
916 if (!(curval
& FUTEX_TID_MASK
)) {
929 * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket
930 * @q: The futex_q to unqueue
932 * The q->lock_ptr must not be NULL and must be held by the caller.
934 static void __unqueue_futex(struct futex_q
*q
)
936 struct futex_hash_bucket
*hb
;
938 if (WARN_ON_SMP(!q
->lock_ptr
|| !spin_is_locked(q
->lock_ptr
))
939 || WARN_ON(plist_node_empty(&q
->list
)))
942 hb
= container_of(q
->lock_ptr
, struct futex_hash_bucket
, lock
);
943 plist_del(&q
->list
, &hb
->chain
);
947 * The hash bucket lock must be held when this is called.
948 * Afterwards, the futex_q must not be accessed.
950 static void wake_futex(struct futex_q
*q
)
952 struct task_struct
*p
= q
->task
;
954 if (WARN(q
->pi_state
|| q
->rt_waiter
, "refusing to wake PI futex\n"))
958 * We set q->lock_ptr = NULL _before_ we wake up the task. If
959 * a non-futex wake up happens on another CPU then the task
960 * might exit and p would dereference a non-existing task
961 * struct. Prevent this by holding a reference on p across the
968 * The waiting task can free the futex_q as soon as
969 * q->lock_ptr = NULL is written, without taking any locks. A
970 * memory barrier is required here to prevent the following
971 * store to lock_ptr from getting ahead of the plist_del.
976 wake_up_state(p
, TASK_NORMAL
);
980 static int wake_futex_pi(u32 __user
*uaddr
, u32 uval
, struct futex_q
*this)
982 struct task_struct
*new_owner
;
983 struct futex_pi_state
*pi_state
= this->pi_state
;
984 u32
uninitialized_var(curval
), newval
;
991 * If current does not own the pi_state then the futex is
992 * inconsistent and user space fiddled with the futex value.
994 if (pi_state
->owner
!= current
)
997 raw_spin_lock(&pi_state
->pi_mutex
.wait_lock
);
998 new_owner
= rt_mutex_next_owner(&pi_state
->pi_mutex
);
1001 * It is possible that the next waiter (the one that brought
1002 * this owner to the kernel) timed out and is no longer
1003 * waiting on the lock.
1006 new_owner
= this->task
;
1009 * We pass it to the next owner. The WAITERS bit is always
1010 * kept enabled while there is PI state around. We cleanup the
1011 * owner died bit, because we are the owner.
1013 newval
= FUTEX_WAITERS
| task_pid_vnr(new_owner
);
1015 if (cmpxchg_futex_value_locked(&curval
, uaddr
, uval
, newval
))
1017 else if (curval
!= uval
)
1020 raw_spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
1024 raw_spin_lock_irq(&pi_state
->owner
->pi_lock
);
1025 WARN_ON(list_empty(&pi_state
->list
));
1026 list_del_init(&pi_state
->list
);
1027 raw_spin_unlock_irq(&pi_state
->owner
->pi_lock
);
1029 raw_spin_lock_irq(&new_owner
->pi_lock
);
1030 WARN_ON(!list_empty(&pi_state
->list
));
1031 list_add(&pi_state
->list
, &new_owner
->pi_state_list
);
1032 pi_state
->owner
= new_owner
;
1033 raw_spin_unlock_irq(&new_owner
->pi_lock
);
1035 raw_spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
1036 rt_mutex_unlock(&pi_state
->pi_mutex
);
1041 static int unlock_futex_pi(u32 __user
*uaddr
, u32 uval
)
1043 u32
uninitialized_var(oldval
);
1046 * There is no waiter, so we unlock the futex. The owner died
1047 * bit has not to be preserved here. We are the owner:
1049 if (cmpxchg_futex_value_locked(&oldval
, uaddr
, uval
, 0))
1058 * Express the locking dependencies for lockdep:
1061 double_lock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
1064 spin_lock(&hb1
->lock
);
1066 spin_lock_nested(&hb2
->lock
, SINGLE_DEPTH_NESTING
);
1067 } else { /* hb1 > hb2 */
1068 spin_lock(&hb2
->lock
);
1069 spin_lock_nested(&hb1
->lock
, SINGLE_DEPTH_NESTING
);
1074 double_unlock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
1076 spin_unlock(&hb1
->lock
);
1078 spin_unlock(&hb2
->lock
);
1082 * Wake up waiters matching bitset queued on this futex (uaddr).
1085 futex_wake(u32 __user
*uaddr
, unsigned int flags
, int nr_wake
, u32 bitset
)
1087 struct futex_hash_bucket
*hb
;
1088 struct futex_q
*this, *next
;
1089 struct plist_head
*head
;
1090 union futex_key key
= FUTEX_KEY_INIT
;
1096 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &key
, VERIFY_READ
);
1097 if (unlikely(ret
!= 0))
1100 hb
= hash_futex(&key
);
1101 spin_lock(&hb
->lock
);
1104 plist_for_each_entry_safe(this, next
, head
, list
) {
1105 if (match_futex (&this->key
, &key
)) {
1106 if (this->pi_state
|| this->rt_waiter
) {
1111 /* Check if one of the bits is set in both bitsets */
1112 if (!(this->bitset
& bitset
))
1116 if (++ret
>= nr_wake
)
1121 spin_unlock(&hb
->lock
);
1122 put_futex_key(&key
);
1128 * Wake up all waiters hashed on the physical page that is mapped
1129 * to this virtual address:
1132 futex_wake_op(u32 __user
*uaddr1
, unsigned int flags
, u32 __user
*uaddr2
,
1133 int nr_wake
, int nr_wake2
, int op
)
1135 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
1136 struct futex_hash_bucket
*hb1
, *hb2
;
1137 struct plist_head
*head
;
1138 struct futex_q
*this, *next
;
1142 ret
= get_futex_key(uaddr1
, flags
& FLAGS_SHARED
, &key1
, VERIFY_READ
);
1143 if (unlikely(ret
!= 0))
1145 ret
= get_futex_key(uaddr2
, flags
& FLAGS_SHARED
, &key2
, VERIFY_WRITE
);
1146 if (unlikely(ret
!= 0))
1149 hb1
= hash_futex(&key1
);
1150 hb2
= hash_futex(&key2
);
1153 double_lock_hb(hb1
, hb2
);
1154 op_ret
= futex_atomic_op_inuser(op
, uaddr2
);
1155 if (unlikely(op_ret
< 0)) {
1157 double_unlock_hb(hb1
, hb2
);
1161 * we don't get EFAULT from MMU faults if we don't have an MMU,
1162 * but we might get them from range checking
1168 if (unlikely(op_ret
!= -EFAULT
)) {
1173 ret
= fault_in_user_writeable(uaddr2
);
1177 if (!(flags
& FLAGS_SHARED
))
1180 put_futex_key(&key2
);
1181 put_futex_key(&key1
);
1187 plist_for_each_entry_safe(this, next
, head
, list
) {
1188 if (match_futex (&this->key
, &key1
)) {
1189 if (this->pi_state
|| this->rt_waiter
) {
1194 if (++ret
>= nr_wake
)
1203 plist_for_each_entry_safe(this, next
, head
, list
) {
1204 if (match_futex (&this->key
, &key2
)) {
1205 if (this->pi_state
|| this->rt_waiter
) {
1210 if (++op_ret
>= nr_wake2
)
1218 double_unlock_hb(hb1
, hb2
);
1220 put_futex_key(&key2
);
1222 put_futex_key(&key1
);
1228 * requeue_futex() - Requeue a futex_q from one hb to another
1229 * @q: the futex_q to requeue
1230 * @hb1: the source hash_bucket
1231 * @hb2: the target hash_bucket
1232 * @key2: the new key for the requeued futex_q
1235 void requeue_futex(struct futex_q
*q
, struct futex_hash_bucket
*hb1
,
1236 struct futex_hash_bucket
*hb2
, union futex_key
*key2
)
1240 * If key1 and key2 hash to the same bucket, no need to
1243 if (likely(&hb1
->chain
!= &hb2
->chain
)) {
1244 plist_del(&q
->list
, &hb1
->chain
);
1245 plist_add(&q
->list
, &hb2
->chain
);
1246 q
->lock_ptr
= &hb2
->lock
;
1248 get_futex_key_refs(key2
);
1253 * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
1255 * @key: the key of the requeue target futex
1256 * @hb: the hash_bucket of the requeue target futex
1258 * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1259 * target futex if it is uncontended or via a lock steal. Set the futex_q key
1260 * to the requeue target futex so the waiter can detect the wakeup on the right
1261 * futex, but remove it from the hb and NULL the rt_waiter so it can detect
1262 * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1263 * to protect access to the pi_state to fixup the owner later. Must be called
1264 * with both q->lock_ptr and hb->lock held.
1267 void requeue_pi_wake_futex(struct futex_q
*q
, union futex_key
*key
,
1268 struct futex_hash_bucket
*hb
)
1270 get_futex_key_refs(key
);
1275 WARN_ON(!q
->rt_waiter
);
1276 q
->rt_waiter
= NULL
;
1278 q
->lock_ptr
= &hb
->lock
;
1280 wake_up_state(q
->task
, TASK_NORMAL
);
1284 * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
1285 * @pifutex: the user address of the to futex
1286 * @hb1: the from futex hash bucket, must be locked by the caller
1287 * @hb2: the to futex hash bucket, must be locked by the caller
1288 * @key1: the from futex key
1289 * @key2: the to futex key
1290 * @ps: address to store the pi_state pointer
1291 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
1293 * Try and get the lock on behalf of the top waiter if we can do it atomically.
1294 * Wake the top waiter if we succeed. If the caller specified set_waiters,
1295 * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1296 * hb1 and hb2 must be held by the caller.
1299 * 0 - failed to acquire the lock atomically;
1300 * >0 - acquired the lock, return value is vpid of the top_waiter
1303 static int futex_proxy_trylock_atomic(u32 __user
*pifutex
,
1304 struct futex_hash_bucket
*hb1
,
1305 struct futex_hash_bucket
*hb2
,
1306 union futex_key
*key1
, union futex_key
*key2
,
1307 struct futex_pi_state
**ps
, int set_waiters
)
1309 struct futex_q
*top_waiter
= NULL
;
1313 if (get_futex_value_locked(&curval
, pifutex
))
1317 * Find the top_waiter and determine if there are additional waiters.
1318 * If the caller intends to requeue more than 1 waiter to pifutex,
1319 * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1320 * as we have means to handle the possible fault. If not, don't set
1321 * the bit unecessarily as it will force the subsequent unlock to enter
1324 top_waiter
= futex_top_waiter(hb1
, key1
);
1326 /* There are no waiters, nothing for us to do. */
1330 /* Ensure we requeue to the expected futex. */
1331 if (!match_futex(top_waiter
->requeue_pi_key
, key2
))
1335 * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1336 * the contended case or if set_waiters is 1. The pi_state is returned
1337 * in ps in contended cases.
1339 vpid
= task_pid_vnr(top_waiter
->task
);
1340 ret
= futex_lock_pi_atomic(pifutex
, hb2
, key2
, ps
, top_waiter
->task
,
1343 requeue_pi_wake_futex(top_waiter
, key2
, hb2
);
1350 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
1351 * @uaddr1: source futex user address
1352 * @flags: futex flags (FLAGS_SHARED, etc.)
1353 * @uaddr2: target futex user address
1354 * @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1355 * @nr_requeue: number of waiters to requeue (0-INT_MAX)
1356 * @cmpval: @uaddr1 expected value (or %NULL)
1357 * @requeue_pi: if we are attempting to requeue from a non-pi futex to a
1358 * pi futex (pi to pi requeue is not supported)
1360 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1361 * uaddr2 atomically on behalf of the top waiter.
1364 * >=0 - on success, the number of tasks requeued or woken;
1367 static int futex_requeue(u32 __user
*uaddr1
, unsigned int flags
,
1368 u32 __user
*uaddr2
, int nr_wake
, int nr_requeue
,
1369 u32
*cmpval
, int requeue_pi
)
1371 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
1372 int drop_count
= 0, task_count
= 0, ret
;
1373 struct futex_pi_state
*pi_state
= NULL
;
1374 struct futex_hash_bucket
*hb1
, *hb2
;
1375 struct plist_head
*head1
;
1376 struct futex_q
*this, *next
;
1380 * Requeue PI only works on two distinct uaddrs. This
1381 * check is only valid for private futexes. See below.
1383 if (uaddr1
== uaddr2
)
1387 * requeue_pi requires a pi_state, try to allocate it now
1388 * without any locks in case it fails.
1390 if (refill_pi_state_cache())
1393 * requeue_pi must wake as many tasks as it can, up to nr_wake
1394 * + nr_requeue, since it acquires the rt_mutex prior to
1395 * returning to userspace, so as to not leave the rt_mutex with
1396 * waiters and no owner. However, second and third wake-ups
1397 * cannot be predicted as they involve race conditions with the
1398 * first wake and a fault while looking up the pi_state. Both
1399 * pthread_cond_signal() and pthread_cond_broadcast() should
1407 if (pi_state
!= NULL
) {
1409 * We will have to lookup the pi_state again, so free this one
1410 * to keep the accounting correct.
1412 free_pi_state(pi_state
);
1416 ret
= get_futex_key(uaddr1
, flags
& FLAGS_SHARED
, &key1
, VERIFY_READ
);
1417 if (unlikely(ret
!= 0))
1419 ret
= get_futex_key(uaddr2
, flags
& FLAGS_SHARED
, &key2
,
1420 requeue_pi
? VERIFY_WRITE
: VERIFY_READ
);
1421 if (unlikely(ret
!= 0))
1425 * The check above which compares uaddrs is not sufficient for
1426 * shared futexes. We need to compare the keys:
1428 if (requeue_pi
&& match_futex(&key1
, &key2
)) {
1433 hb1
= hash_futex(&key1
);
1434 hb2
= hash_futex(&key2
);
1437 double_lock_hb(hb1
, hb2
);
1439 if (likely(cmpval
!= NULL
)) {
1442 ret
= get_futex_value_locked(&curval
, uaddr1
);
1444 if (unlikely(ret
)) {
1445 double_unlock_hb(hb1
, hb2
);
1447 ret
= get_user(curval
, uaddr1
);
1451 if (!(flags
& FLAGS_SHARED
))
1454 put_futex_key(&key2
);
1455 put_futex_key(&key1
);
1458 if (curval
!= *cmpval
) {
1464 if (requeue_pi
&& (task_count
- nr_wake
< nr_requeue
)) {
1466 * Attempt to acquire uaddr2 and wake the top waiter. If we
1467 * intend to requeue waiters, force setting the FUTEX_WAITERS
1468 * bit. We force this here where we are able to easily handle
1469 * faults rather in the requeue loop below.
1471 ret
= futex_proxy_trylock_atomic(uaddr2
, hb1
, hb2
, &key1
,
1472 &key2
, &pi_state
, nr_requeue
);
1475 * At this point the top_waiter has either taken uaddr2 or is
1476 * waiting on it. If the former, then the pi_state will not
1477 * exist yet, look it up one more time to ensure we have a
1478 * reference to it. If the lock was taken, ret contains the
1479 * vpid of the top waiter task.
1486 * If we acquired the lock, then the user
1487 * space value of uaddr2 should be vpid. It
1488 * cannot be changed by the top waiter as it
1489 * is blocked on hb2 lock if it tries to do
1490 * so. If something fiddled with it behind our
1491 * back the pi state lookup might unearth
1492 * it. So we rather use the known value than
1493 * rereading and handing potential crap to
1496 ret
= lookup_pi_state(ret
, hb2
, &key2
, &pi_state
);
1503 double_unlock_hb(hb1
, hb2
);
1504 put_futex_key(&key2
);
1505 put_futex_key(&key1
);
1506 ret
= fault_in_user_writeable(uaddr2
);
1511 /* The owner was exiting, try again. */
1512 double_unlock_hb(hb1
, hb2
);
1513 put_futex_key(&key2
);
1514 put_futex_key(&key1
);
1522 head1
= &hb1
->chain
;
1523 plist_for_each_entry_safe(this, next
, head1
, list
) {
1524 if (task_count
- nr_wake
>= nr_requeue
)
1527 if (!match_futex(&this->key
, &key1
))
1531 * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
1532 * be paired with each other and no other futex ops.
1534 * We should never be requeueing a futex_q with a pi_state,
1535 * which is awaiting a futex_unlock_pi().
1537 if ((requeue_pi
&& !this->rt_waiter
) ||
1538 (!requeue_pi
&& this->rt_waiter
) ||
1545 * Wake nr_wake waiters. For requeue_pi, if we acquired the
1546 * lock, we already woke the top_waiter. If not, it will be
1547 * woken by futex_unlock_pi().
1549 if (++task_count
<= nr_wake
&& !requeue_pi
) {
1554 /* Ensure we requeue to the expected futex for requeue_pi. */
1555 if (requeue_pi
&& !match_futex(this->requeue_pi_key
, &key2
)) {
1561 * Requeue nr_requeue waiters and possibly one more in the case
1562 * of requeue_pi if we couldn't acquire the lock atomically.
1565 /* Prepare the waiter to take the rt_mutex. */
1566 atomic_inc(&pi_state
->refcount
);
1567 this->pi_state
= pi_state
;
1568 ret
= rt_mutex_start_proxy_lock(&pi_state
->pi_mutex
,
1572 /* We got the lock. */
1573 requeue_pi_wake_futex(this, &key2
, hb2
);
1578 this->pi_state
= NULL
;
1579 free_pi_state(pi_state
);
1583 requeue_futex(this, hb1
, hb2
, &key2
);
1588 double_unlock_hb(hb1
, hb2
);
1591 * drop_futex_key_refs() must be called outside the spinlocks. During
1592 * the requeue we moved futex_q's from the hash bucket at key1 to the
1593 * one at key2 and updated their key pointer. We no longer need to
1594 * hold the references to key1.
1596 while (--drop_count
>= 0)
1597 drop_futex_key_refs(&key1
);
1600 put_futex_key(&key2
);
1602 put_futex_key(&key1
);
1604 if (pi_state
!= NULL
)
1605 free_pi_state(pi_state
);
1606 return ret
? ret
: task_count
;
1609 /* The key must be already stored in q->key. */
1610 static inline struct futex_hash_bucket
*queue_lock(struct futex_q
*q
)
1611 __acquires(&hb
->lock
)
1613 struct futex_hash_bucket
*hb
;
1615 hb
= hash_futex(&q
->key
);
1616 q
->lock_ptr
= &hb
->lock
;
1618 spin_lock(&hb
->lock
);
1623 queue_unlock(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
1624 __releases(&hb
->lock
)
1626 spin_unlock(&hb
->lock
);
1630 * queue_me() - Enqueue the futex_q on the futex_hash_bucket
1631 * @q: The futex_q to enqueue
1632 * @hb: The destination hash bucket
1634 * The hb->lock must be held by the caller, and is released here. A call to
1635 * queue_me() is typically paired with exactly one call to unqueue_me(). The
1636 * exceptions involve the PI related operations, which may use unqueue_me_pi()
1637 * or nothing if the unqueue is done as part of the wake process and the unqueue
1638 * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
1641 static inline void queue_me(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
1642 __releases(&hb
->lock
)
1647 * The priority used to register this element is
1648 * - either the real thread-priority for the real-time threads
1649 * (i.e. threads with a priority lower than MAX_RT_PRIO)
1650 * - or MAX_RT_PRIO for non-RT threads.
1651 * Thus, all RT-threads are woken first in priority order, and
1652 * the others are woken last, in FIFO order.
1654 prio
= min(current
->normal_prio
, MAX_RT_PRIO
);
1656 plist_node_init(&q
->list
, prio
);
1657 plist_add(&q
->list
, &hb
->chain
);
1659 spin_unlock(&hb
->lock
);
1663 * unqueue_me() - Remove the futex_q from its futex_hash_bucket
1664 * @q: The futex_q to unqueue
1666 * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must
1667 * be paired with exactly one earlier call to queue_me().
1670 * 1 - if the futex_q was still queued (and we removed unqueued it);
1671 * 0 - if the futex_q was already removed by the waking thread
1673 static int unqueue_me(struct futex_q
*q
)
1675 spinlock_t
*lock_ptr
;
1678 /* In the common case we don't take the spinlock, which is nice. */
1680 lock_ptr
= q
->lock_ptr
;
1682 if (lock_ptr
!= NULL
) {
1683 spin_lock(lock_ptr
);
1685 * q->lock_ptr can change between reading it and
1686 * spin_lock(), causing us to take the wrong lock. This
1687 * corrects the race condition.
1689 * Reasoning goes like this: if we have the wrong lock,
1690 * q->lock_ptr must have changed (maybe several times)
1691 * between reading it and the spin_lock(). It can
1692 * change again after the spin_lock() but only if it was
1693 * already changed before the spin_lock(). It cannot,
1694 * however, change back to the original value. Therefore
1695 * we can detect whether we acquired the correct lock.
1697 if (unlikely(lock_ptr
!= q
->lock_ptr
)) {
1698 spin_unlock(lock_ptr
);
1703 BUG_ON(q
->pi_state
);
1705 spin_unlock(lock_ptr
);
1709 drop_futex_key_refs(&q
->key
);
1714 * PI futexes can not be requeued and must remove themself from the
1715 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1718 static void unqueue_me_pi(struct futex_q
*q
)
1719 __releases(q
->lock_ptr
)
1723 BUG_ON(!q
->pi_state
);
1724 free_pi_state(q
->pi_state
);
1727 spin_unlock(q
->lock_ptr
);
1731 * Fixup the pi_state owner with the new owner.
1733 * Must be called with hash bucket lock held and mm->sem held for non
1736 static int fixup_pi_state_owner(u32 __user
*uaddr
, struct futex_q
*q
,
1737 struct task_struct
*newowner
)
1739 u32 newtid
= task_pid_vnr(newowner
) | FUTEX_WAITERS
;
1740 struct futex_pi_state
*pi_state
= q
->pi_state
;
1741 struct task_struct
*oldowner
= pi_state
->owner
;
1742 u32 uval
, uninitialized_var(curval
), newval
;
1746 if (!pi_state
->owner
)
1747 newtid
|= FUTEX_OWNER_DIED
;
1750 * We are here either because we stole the rtmutex from the
1751 * previous highest priority waiter or we are the highest priority
1752 * waiter but failed to get the rtmutex the first time.
1753 * We have to replace the newowner TID in the user space variable.
1754 * This must be atomic as we have to preserve the owner died bit here.
1756 * Note: We write the user space value _before_ changing the pi_state
1757 * because we can fault here. Imagine swapped out pages or a fork
1758 * that marked all the anonymous memory readonly for cow.
1760 * Modifying pi_state _before_ the user space value would
1761 * leave the pi_state in an inconsistent state when we fault
1762 * here, because we need to drop the hash bucket lock to
1763 * handle the fault. This might be observed in the PID check
1764 * in lookup_pi_state.
1767 if (get_futex_value_locked(&uval
, uaddr
))
1771 newval
= (uval
& FUTEX_OWNER_DIED
) | newtid
;
1773 if (cmpxchg_futex_value_locked(&curval
, uaddr
, uval
, newval
))
1781 * We fixed up user space. Now we need to fix the pi_state
1784 if (pi_state
->owner
!= NULL
) {
1785 raw_spin_lock_irq(&pi_state
->owner
->pi_lock
);
1786 WARN_ON(list_empty(&pi_state
->list
));
1787 list_del_init(&pi_state
->list
);
1788 raw_spin_unlock_irq(&pi_state
->owner
->pi_lock
);
1791 pi_state
->owner
= newowner
;
1793 raw_spin_lock_irq(&newowner
->pi_lock
);
1794 WARN_ON(!list_empty(&pi_state
->list
));
1795 list_add(&pi_state
->list
, &newowner
->pi_state_list
);
1796 raw_spin_unlock_irq(&newowner
->pi_lock
);
1800 * To handle the page fault we need to drop the hash bucket
1801 * lock here. That gives the other task (either the highest priority
1802 * waiter itself or the task which stole the rtmutex) the
1803 * chance to try the fixup of the pi_state. So once we are
1804 * back from handling the fault we need to check the pi_state
1805 * after reacquiring the hash bucket lock and before trying to
1806 * do another fixup. When the fixup has been done already we
1810 spin_unlock(q
->lock_ptr
);
1812 ret
= fault_in_user_writeable(uaddr
);
1814 spin_lock(q
->lock_ptr
);
1817 * Check if someone else fixed it for us:
1819 if (pi_state
->owner
!= oldowner
)
1828 static long futex_wait_restart(struct restart_block
*restart
);
1831 * fixup_owner() - Post lock pi_state and corner case management
1832 * @uaddr: user address of the futex
1833 * @q: futex_q (contains pi_state and access to the rt_mutex)
1834 * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
1836 * After attempting to lock an rt_mutex, this function is called to cleanup
1837 * the pi_state owner as well as handle race conditions that may allow us to
1838 * acquire the lock. Must be called with the hb lock held.
1841 * 1 - success, lock taken;
1842 * 0 - success, lock not taken;
1843 * <0 - on error (-EFAULT)
1845 static int fixup_owner(u32 __user
*uaddr
, struct futex_q
*q
, int locked
)
1847 struct task_struct
*owner
;
1852 * Got the lock. We might not be the anticipated owner if we
1853 * did a lock-steal - fix up the PI-state in that case:
1855 if (q
->pi_state
->owner
!= current
)
1856 ret
= fixup_pi_state_owner(uaddr
, q
, current
);
1861 * Catch the rare case, where the lock was released when we were on the
1862 * way back before we locked the hash bucket.
1864 if (q
->pi_state
->owner
== current
) {
1866 * Try to get the rt_mutex now. This might fail as some other
1867 * task acquired the rt_mutex after we removed ourself from the
1868 * rt_mutex waiters list.
1870 if (rt_mutex_trylock(&q
->pi_state
->pi_mutex
)) {
1876 * pi_state is incorrect, some other task did a lock steal and
1877 * we returned due to timeout or signal without taking the
1878 * rt_mutex. Too late.
1880 raw_spin_lock(&q
->pi_state
->pi_mutex
.wait_lock
);
1881 owner
= rt_mutex_owner(&q
->pi_state
->pi_mutex
);
1883 owner
= rt_mutex_next_owner(&q
->pi_state
->pi_mutex
);
1884 raw_spin_unlock(&q
->pi_state
->pi_mutex
.wait_lock
);
1885 ret
= fixup_pi_state_owner(uaddr
, q
, owner
);
1890 * Paranoia check. If we did not take the lock, then we should not be
1891 * the owner of the rt_mutex.
1893 if (rt_mutex_owner(&q
->pi_state
->pi_mutex
) == current
)
1894 printk(KERN_ERR
"fixup_owner: ret = %d pi-mutex: %p "
1895 "pi-state %p\n", ret
,
1896 q
->pi_state
->pi_mutex
.owner
,
1897 q
->pi_state
->owner
);
1900 return ret
? ret
: locked
;
1904 * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
1905 * @hb: the futex hash bucket, must be locked by the caller
1906 * @q: the futex_q to queue up on
1907 * @timeout: the prepared hrtimer_sleeper, or null for no timeout
1909 static void futex_wait_queue_me(struct futex_hash_bucket
*hb
, struct futex_q
*q
,
1910 struct hrtimer_sleeper
*timeout
)
1913 * The task state is guaranteed to be set before another task can
1914 * wake it. set_current_state() is implemented using set_mb() and
1915 * queue_me() calls spin_unlock() upon completion, both serializing
1916 * access to the hash list and forcing another memory barrier.
1918 set_current_state(TASK_INTERRUPTIBLE
);
1923 hrtimer_start_expires(&timeout
->timer
, HRTIMER_MODE_ABS
);
1924 if (!hrtimer_active(&timeout
->timer
))
1925 timeout
->task
= NULL
;
1929 * If we have been removed from the hash list, then another task
1930 * has tried to wake us, and we can skip the call to schedule().
1932 if (likely(!plist_node_empty(&q
->list
))) {
1934 * If the timer has already expired, current will already be
1935 * flagged for rescheduling. Only call schedule if there
1936 * is no timeout, or if it has yet to expire.
1938 if (!timeout
|| timeout
->task
)
1939 freezable_schedule();
1941 __set_current_state(TASK_RUNNING
);
1945 * futex_wait_setup() - Prepare to wait on a futex
1946 * @uaddr: the futex userspace address
1947 * @val: the expected value
1948 * @flags: futex flags (FLAGS_SHARED, etc.)
1949 * @q: the associated futex_q
1950 * @hb: storage for hash_bucket pointer to be returned to caller
1952 * Setup the futex_q and locate the hash_bucket. Get the futex value and
1953 * compare it with the expected value. Handle atomic faults internally.
1954 * Return with the hb lock held and a q.key reference on success, and unlocked
1955 * with no q.key reference on failure.
1958 * 0 - uaddr contains val and hb has been locked;
1959 * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked
1961 static int futex_wait_setup(u32 __user
*uaddr
, u32 val
, unsigned int flags
,
1962 struct futex_q
*q
, struct futex_hash_bucket
**hb
)
1968 * Access the page AFTER the hash-bucket is locked.
1969 * Order is important:
1971 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1972 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1974 * The basic logical guarantee of a futex is that it blocks ONLY
1975 * if cond(var) is known to be true at the time of blocking, for
1976 * any cond. If we locked the hash-bucket after testing *uaddr, that
1977 * would open a race condition where we could block indefinitely with
1978 * cond(var) false, which would violate the guarantee.
1980 * On the other hand, we insert q and release the hash-bucket only
1981 * after testing *uaddr. This guarantees that futex_wait() will NOT
1982 * absorb a wakeup if *uaddr does not match the desired values
1983 * while the syscall executes.
1986 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &q
->key
, VERIFY_READ
);
1987 if (unlikely(ret
!= 0))
1991 *hb
= queue_lock(q
);
1993 ret
= get_futex_value_locked(&uval
, uaddr
);
1996 queue_unlock(q
, *hb
);
1998 ret
= get_user(uval
, uaddr
);
2002 if (!(flags
& FLAGS_SHARED
))
2005 put_futex_key(&q
->key
);
2010 queue_unlock(q
, *hb
);
2016 put_futex_key(&q
->key
);
2020 static int futex_wait(u32 __user
*uaddr
, unsigned int flags
, u32 val
,
2021 ktime_t
*abs_time
, u32 bitset
)
2023 struct hrtimer_sleeper timeout
, *to
= NULL
;
2024 struct restart_block
*restart
;
2025 struct futex_hash_bucket
*hb
;
2026 struct futex_q q
= futex_q_init
;
2036 hrtimer_init_on_stack(&to
->timer
, (flags
& FLAGS_CLOCKRT
) ?
2037 CLOCK_REALTIME
: CLOCK_MONOTONIC
,
2039 hrtimer_init_sleeper(to
, current
);
2040 hrtimer_set_expires_range_ns(&to
->timer
, *abs_time
,
2041 current
->timer_slack_ns
);
2046 * Prepare to wait on uaddr. On success, holds hb lock and increments
2049 ret
= futex_wait_setup(uaddr
, val
, flags
, &q
, &hb
);
2053 /* queue_me and wait for wakeup, timeout, or a signal. */
2054 futex_wait_queue_me(hb
, &q
, to
);
2056 /* If we were woken (and unqueued), we succeeded, whatever. */
2058 /* unqueue_me() drops q.key ref */
2059 if (!unqueue_me(&q
))
2062 if (to
&& !to
->task
)
2066 * We expect signal_pending(current), but we might be the
2067 * victim of a spurious wakeup as well.
2069 if (!signal_pending(current
))
2076 restart
= ¤t_thread_info()->restart_block
;
2077 restart
->fn
= futex_wait_restart
;
2078 restart
->futex
.uaddr
= uaddr
;
2079 restart
->futex
.val
= val
;
2080 restart
->futex
.time
= abs_time
->tv64
;
2081 restart
->futex
.bitset
= bitset
;
2082 restart
->futex
.flags
= flags
| FLAGS_HAS_TIMEOUT
;
2084 ret
= -ERESTART_RESTARTBLOCK
;
2088 hrtimer_cancel(&to
->timer
);
2089 destroy_hrtimer_on_stack(&to
->timer
);
2095 static long futex_wait_restart(struct restart_block
*restart
)
2097 u32 __user
*uaddr
= restart
->futex
.uaddr
;
2098 ktime_t t
, *tp
= NULL
;
2100 if (restart
->futex
.flags
& FLAGS_HAS_TIMEOUT
) {
2101 t
.tv64
= restart
->futex
.time
;
2104 restart
->fn
= do_no_restart_syscall
;
2106 return (long)futex_wait(uaddr
, restart
->futex
.flags
,
2107 restart
->futex
.val
, tp
, restart
->futex
.bitset
);
2112 * Userspace tried a 0 -> TID atomic transition of the futex value
2113 * and failed. The kernel side here does the whole locking operation:
2114 * if there are waiters then it will block, it does PI, etc. (Due to
2115 * races the kernel might see a 0 value of the futex too.)
2117 static int futex_lock_pi(u32 __user
*uaddr
, unsigned int flags
, int detect
,
2118 ktime_t
*time
, int trylock
)
2120 struct hrtimer_sleeper timeout
, *to
= NULL
;
2121 struct futex_hash_bucket
*hb
;
2122 struct futex_q q
= futex_q_init
;
2125 if (refill_pi_state_cache())
2130 hrtimer_init_on_stack(&to
->timer
, CLOCK_REALTIME
,
2132 hrtimer_init_sleeper(to
, current
);
2133 hrtimer_set_expires(&to
->timer
, *time
);
2137 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &q
.key
, VERIFY_WRITE
);
2138 if (unlikely(ret
!= 0))
2142 hb
= queue_lock(&q
);
2144 ret
= futex_lock_pi_atomic(uaddr
, hb
, &q
.key
, &q
.pi_state
, current
, 0);
2145 if (unlikely(ret
)) {
2148 /* We got the lock. */
2150 goto out_unlock_put_key
;
2155 * Task is exiting and we just wait for the
2158 queue_unlock(&q
, hb
);
2159 put_futex_key(&q
.key
);
2163 goto out_unlock_put_key
;
2168 * Only actually queue now that the atomic ops are done:
2172 WARN_ON(!q
.pi_state
);
2174 * Block on the PI mutex:
2177 ret
= rt_mutex_timed_lock(&q
.pi_state
->pi_mutex
, to
, 1);
2179 ret
= rt_mutex_trylock(&q
.pi_state
->pi_mutex
);
2180 /* Fixup the trylock return value: */
2181 ret
= ret
? 0 : -EWOULDBLOCK
;
2184 spin_lock(q
.lock_ptr
);
2186 * Fixup the pi_state owner and possibly acquire the lock if we
2189 res
= fixup_owner(uaddr
, &q
, !ret
);
2191 * If fixup_owner() returned an error, proprogate that. If it acquired
2192 * the lock, clear our -ETIMEDOUT or -EINTR.
2195 ret
= (res
< 0) ? res
: 0;
2198 * If fixup_owner() faulted and was unable to handle the fault, unlock
2199 * it and return the fault to userspace.
2201 if (ret
&& (rt_mutex_owner(&q
.pi_state
->pi_mutex
) == current
))
2202 rt_mutex_unlock(&q
.pi_state
->pi_mutex
);
2204 /* Unqueue and drop the lock */
2210 queue_unlock(&q
, hb
);
2213 put_futex_key(&q
.key
);
2216 destroy_hrtimer_on_stack(&to
->timer
);
2217 return ret
!= -EINTR
? ret
: -ERESTARTNOINTR
;
2220 queue_unlock(&q
, hb
);
2222 ret
= fault_in_user_writeable(uaddr
);
2226 if (!(flags
& FLAGS_SHARED
))
2229 put_futex_key(&q
.key
);
2234 * Userspace attempted a TID -> 0 atomic transition, and failed.
2235 * This is the in-kernel slowpath: we look up the PI state (if any),
2236 * and do the rt-mutex unlock.
2238 static int futex_unlock_pi(u32 __user
*uaddr
, unsigned int flags
)
2240 struct futex_hash_bucket
*hb
;
2241 struct futex_q
*this, *next
;
2242 struct plist_head
*head
;
2243 union futex_key key
= FUTEX_KEY_INIT
;
2244 u32 uval
, vpid
= task_pid_vnr(current
);
2248 if (get_user(uval
, uaddr
))
2251 * We release only a lock we actually own:
2253 if ((uval
& FUTEX_TID_MASK
) != vpid
)
2256 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &key
, VERIFY_WRITE
);
2257 if (unlikely(ret
!= 0))
2260 hb
= hash_futex(&key
);
2261 spin_lock(&hb
->lock
);
2264 * To avoid races, try to do the TID -> 0 atomic transition
2265 * again. If it succeeds then we can return without waking
2266 * anyone else up. We only try this if neither the waiters nor
2267 * the owner died bit are set.
2269 if (!(uval
& ~FUTEX_TID_MASK
) &&
2270 cmpxchg_futex_value_locked(&uval
, uaddr
, vpid
, 0))
2273 * Rare case: we managed to release the lock atomically,
2274 * no need to wake anyone else up:
2276 if (unlikely(uval
== vpid
))
2280 * Ok, other tasks may need to be woken up - check waiters
2281 * and do the wakeup if necessary:
2285 plist_for_each_entry_safe(this, next
, head
, list
) {
2286 if (!match_futex (&this->key
, &key
))
2288 ret
= wake_futex_pi(uaddr
, uval
, this);
2290 * The atomic access to the futex value
2291 * generated a pagefault, so retry the
2292 * user-access and the wakeup:
2299 * No waiters - kernel unlocks the futex:
2301 ret
= unlock_futex_pi(uaddr
, uval
);
2306 spin_unlock(&hb
->lock
);
2307 put_futex_key(&key
);
2313 spin_unlock(&hb
->lock
);
2314 put_futex_key(&key
);
2316 ret
= fault_in_user_writeable(uaddr
);
2324 * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2325 * @hb: the hash_bucket futex_q was original enqueued on
2326 * @q: the futex_q woken while waiting to be requeued
2327 * @key2: the futex_key of the requeue target futex
2328 * @timeout: the timeout associated with the wait (NULL if none)
2330 * Detect if the task was woken on the initial futex as opposed to the requeue
2331 * target futex. If so, determine if it was a timeout or a signal that caused
2332 * the wakeup and return the appropriate error code to the caller. Must be
2333 * called with the hb lock held.
2336 * 0 = no early wakeup detected;
2337 * <0 = -ETIMEDOUT or -ERESTARTNOINTR
2340 int handle_early_requeue_pi_wakeup(struct futex_hash_bucket
*hb
,
2341 struct futex_q
*q
, union futex_key
*key2
,
2342 struct hrtimer_sleeper
*timeout
)
2347 * With the hb lock held, we avoid races while we process the wakeup.
2348 * We only need to hold hb (and not hb2) to ensure atomicity as the
2349 * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2350 * It can't be requeued from uaddr2 to something else since we don't
2351 * support a PI aware source futex for requeue.
2353 if (!match_futex(&q
->key
, key2
)) {
2354 WARN_ON(q
->lock_ptr
&& (&hb
->lock
!= q
->lock_ptr
));
2356 * We were woken prior to requeue by a timeout or a signal.
2357 * Unqueue the futex_q and determine which it was.
2359 plist_del(&q
->list
, &hb
->chain
);
2361 /* Handle spurious wakeups gracefully */
2363 if (timeout
&& !timeout
->task
)
2365 else if (signal_pending(current
))
2366 ret
= -ERESTARTNOINTR
;
2372 * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
2373 * @uaddr: the futex we initially wait on (non-pi)
2374 * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be
2375 * the same type, no requeueing from private to shared, etc.
2376 * @val: the expected value of uaddr
2377 * @abs_time: absolute timeout
2378 * @bitset: 32 bit wakeup bitset set by userspace, defaults to all
2379 * @uaddr2: the pi futex we will take prior to returning to user-space
2381 * The caller will wait on uaddr and will be requeued by futex_requeue() to
2382 * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake
2383 * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to
2384 * userspace. This ensures the rt_mutex maintains an owner when it has waiters;
2385 * without one, the pi logic would not know which task to boost/deboost, if
2386 * there was a need to.
2388 * We call schedule in futex_wait_queue_me() when we enqueue and return there
2389 * via the following--
2390 * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
2391 * 2) wakeup on uaddr2 after a requeue
2395 * If 3, cleanup and return -ERESTARTNOINTR.
2397 * If 2, we may then block on trying to take the rt_mutex and return via:
2398 * 5) successful lock
2401 * 8) other lock acquisition failure
2403 * If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
2405 * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2411 static int futex_wait_requeue_pi(u32 __user
*uaddr
, unsigned int flags
,
2412 u32 val
, ktime_t
*abs_time
, u32 bitset
,
2415 struct hrtimer_sleeper timeout
, *to
= NULL
;
2416 struct rt_mutex_waiter rt_waiter
;
2417 struct futex_hash_bucket
*hb
;
2418 union futex_key key2
= FUTEX_KEY_INIT
;
2419 struct futex_q q
= futex_q_init
;
2422 if (uaddr
== uaddr2
)
2430 hrtimer_init_on_stack(&to
->timer
, (flags
& FLAGS_CLOCKRT
) ?
2431 CLOCK_REALTIME
: CLOCK_MONOTONIC
,
2433 hrtimer_init_sleeper(to
, current
);
2434 hrtimer_set_expires_range_ns(&to
->timer
, *abs_time
,
2435 current
->timer_slack_ns
);
2439 * The waiter is allocated on our stack, manipulated by the requeue
2440 * code while we sleep on uaddr.
2442 debug_rt_mutex_init_waiter(&rt_waiter
);
2443 rt_waiter
.task
= NULL
;
2445 ret
= get_futex_key(uaddr2
, flags
& FLAGS_SHARED
, &key2
, VERIFY_WRITE
);
2446 if (unlikely(ret
!= 0))
2450 q
.rt_waiter
= &rt_waiter
;
2451 q
.requeue_pi_key
= &key2
;
2454 * Prepare to wait on uaddr. On success, increments q.key (key1) ref
2457 ret
= futex_wait_setup(uaddr
, val
, flags
, &q
, &hb
);
2462 * The check above which compares uaddrs is not sufficient for
2463 * shared futexes. We need to compare the keys:
2465 if (match_futex(&q
.key
, &key2
)) {
2470 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
2471 futex_wait_queue_me(hb
, &q
, to
);
2473 spin_lock(&hb
->lock
);
2474 ret
= handle_early_requeue_pi_wakeup(hb
, &q
, &key2
, to
);
2475 spin_unlock(&hb
->lock
);
2480 * In order for us to be here, we know our q.key == key2, and since
2481 * we took the hb->lock above, we also know that futex_requeue() has
2482 * completed and we no longer have to concern ourselves with a wakeup
2483 * race with the atomic proxy lock acquisition by the requeue code. The
2484 * futex_requeue dropped our key1 reference and incremented our key2
2488 /* Check if the requeue code acquired the second futex for us. */
2491 * Got the lock. We might not be the anticipated owner if we
2492 * did a lock-steal - fix up the PI-state in that case.
2494 if (q
.pi_state
&& (q
.pi_state
->owner
!= current
)) {
2495 spin_lock(q
.lock_ptr
);
2496 ret
= fixup_pi_state_owner(uaddr2
, &q
, current
);
2497 if (ret
&& rt_mutex_owner(&q
.pi_state
->pi_mutex
) == current
)
2498 rt_mutex_unlock(&q
.pi_state
->pi_mutex
);
2500 * Drop the reference to the pi state which
2501 * the requeue_pi() code acquired for us.
2503 free_pi_state(q
.pi_state
);
2504 spin_unlock(q
.lock_ptr
);
2507 struct rt_mutex
*pi_mutex
;
2510 * We have been woken up by futex_unlock_pi(), a timeout, or a
2511 * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2514 WARN_ON(!q
.pi_state
);
2515 pi_mutex
= &q
.pi_state
->pi_mutex
;
2516 ret
= rt_mutex_finish_proxy_lock(pi_mutex
, to
, &rt_waiter
, 1);
2517 debug_rt_mutex_free_waiter(&rt_waiter
);
2519 spin_lock(q
.lock_ptr
);
2521 * Fixup the pi_state owner and possibly acquire the lock if we
2524 res
= fixup_owner(uaddr2
, &q
, !ret
);
2526 * If fixup_owner() returned an error, proprogate that. If it
2527 * acquired the lock, clear -ETIMEDOUT or -EINTR.
2530 ret
= (res
< 0) ? res
: 0;
2533 * If fixup_pi_state_owner() faulted and was unable to handle
2534 * the fault, unlock the rt_mutex and return the fault to
2537 if (ret
&& rt_mutex_owner(pi_mutex
) == current
)
2538 rt_mutex_unlock(pi_mutex
);
2540 /* Unqueue and drop the lock. */
2544 if (ret
== -EINTR
) {
2546 * We've already been requeued, but cannot restart by calling
2547 * futex_lock_pi() directly. We could restart this syscall, but
2548 * it would detect that the user space "val" changed and return
2549 * -EWOULDBLOCK. Save the overhead of the restart and return
2550 * -EWOULDBLOCK directly.
2556 put_futex_key(&q
.key
);
2558 put_futex_key(&key2
);
2562 hrtimer_cancel(&to
->timer
);
2563 destroy_hrtimer_on_stack(&to
->timer
);
2569 * Support for robust futexes: the kernel cleans up held futexes at
2572 * Implementation: user-space maintains a per-thread list of locks it
2573 * is holding. Upon do_exit(), the kernel carefully walks this list,
2574 * and marks all locks that are owned by this thread with the
2575 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
2576 * always manipulated with the lock held, so the list is private and
2577 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2578 * field, to allow the kernel to clean up if the thread dies after
2579 * acquiring the lock, but just before it could have added itself to
2580 * the list. There can only be one such pending lock.
2584 * sys_set_robust_list() - Set the robust-futex list head of a task
2585 * @head: pointer to the list-head
2586 * @len: length of the list-head, as userspace expects
2588 SYSCALL_DEFINE2(set_robust_list
, struct robust_list_head __user
*, head
,
2591 if (!futex_cmpxchg_enabled
)
2594 * The kernel knows only one size for now:
2596 if (unlikely(len
!= sizeof(*head
)))
2599 current
->robust_list
= head
;
2605 * sys_get_robust_list() - Get the robust-futex list head of a task
2606 * @pid: pid of the process [zero for current task]
2607 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2608 * @len_ptr: pointer to a length field, the kernel fills in the header size
2610 SYSCALL_DEFINE3(get_robust_list
, int, pid
,
2611 struct robust_list_head __user
* __user
*, head_ptr
,
2612 size_t __user
*, len_ptr
)
2614 struct robust_list_head __user
*head
;
2616 struct task_struct
*p
;
2618 if (!futex_cmpxchg_enabled
)
2627 p
= find_task_by_vpid(pid
);
2633 if (!ptrace_may_access(p
, PTRACE_MODE_READ_REALCREDS
))
2636 head
= p
->robust_list
;
2639 if (put_user(sizeof(*head
), len_ptr
))
2641 return put_user(head
, head_ptr
);
2650 * Process a futex-list entry, check whether it's owned by the
2651 * dying task, and do notification if so:
2653 int handle_futex_death(u32 __user
*uaddr
, struct task_struct
*curr
, int pi
)
2655 u32 uval
, uninitialized_var(nval
), mval
;
2658 if (get_user(uval
, uaddr
))
2661 if ((uval
& FUTEX_TID_MASK
) == task_pid_vnr(curr
)) {
2663 * Ok, this dying thread is truly holding a futex
2664 * of interest. Set the OWNER_DIED bit atomically
2665 * via cmpxchg, and if the value had FUTEX_WAITERS
2666 * set, wake up a waiter (if any). (We have to do a
2667 * futex_wake() even if OWNER_DIED is already set -
2668 * to handle the rare but possible case of recursive
2669 * thread-death.) The rest of the cleanup is done in
2672 mval
= (uval
& FUTEX_WAITERS
) | FUTEX_OWNER_DIED
;
2674 * We are not holding a lock here, but we want to have
2675 * the pagefault_disable/enable() protection because
2676 * we want to handle the fault gracefully. If the
2677 * access fails we try to fault in the futex with R/W
2678 * verification via get_user_pages. get_user() above
2679 * does not guarantee R/W access. If that fails we
2680 * give up and leave the futex locked.
2682 if (cmpxchg_futex_value_locked(&nval
, uaddr
, uval
, mval
)) {
2683 if (fault_in_user_writeable(uaddr
))
2691 * Wake robust non-PI futexes here. The wakeup of
2692 * PI futexes happens in exit_pi_state():
2694 if (!pi
&& (uval
& FUTEX_WAITERS
))
2695 futex_wake(uaddr
, 1, 1, FUTEX_BITSET_MATCH_ANY
);
2701 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2703 static inline int fetch_robust_entry(struct robust_list __user
**entry
,
2704 struct robust_list __user
* __user
*head
,
2707 unsigned long uentry
;
2709 if (get_user(uentry
, (unsigned long __user
*)head
))
2712 *entry
= (void __user
*)(uentry
& ~1UL);
2719 * Walk curr->robust_list (very carefully, it's a userspace list!)
2720 * and mark any locks found there dead, and notify any waiters.
2722 * We silently return on any sign of list-walking problem.
2724 void exit_robust_list(struct task_struct
*curr
)
2726 struct robust_list_head __user
*head
= curr
->robust_list
;
2727 struct robust_list __user
*entry
, *next_entry
, *pending
;
2728 unsigned int limit
= ROBUST_LIST_LIMIT
, pi
, pip
;
2729 unsigned int uninitialized_var(next_pi
);
2730 unsigned long futex_offset
;
2733 if (!futex_cmpxchg_enabled
)
2737 * Fetch the list head (which was registered earlier, via
2738 * sys_set_robust_list()):
2740 if (fetch_robust_entry(&entry
, &head
->list
.next
, &pi
))
2743 * Fetch the relative futex offset:
2745 if (get_user(futex_offset
, &head
->futex_offset
))
2748 * Fetch any possibly pending lock-add first, and handle it
2751 if (fetch_robust_entry(&pending
, &head
->list_op_pending
, &pip
))
2754 next_entry
= NULL
; /* avoid warning with gcc */
2755 while (entry
!= &head
->list
) {
2757 * Fetch the next entry in the list before calling
2758 * handle_futex_death:
2760 rc
= fetch_robust_entry(&next_entry
, &entry
->next
, &next_pi
);
2762 * A pending lock might already be on the list, so
2763 * don't process it twice:
2765 if (entry
!= pending
)
2766 if (handle_futex_death((void __user
*)entry
+ futex_offset
,
2774 * Avoid excessively long or circular lists:
2783 handle_futex_death((void __user
*)pending
+ futex_offset
,
2787 long do_futex(u32 __user
*uaddr
, int op
, u32 val
, ktime_t
*timeout
,
2788 u32 __user
*uaddr2
, u32 val2
, u32 val3
)
2790 int cmd
= op
& FUTEX_CMD_MASK
;
2791 unsigned int flags
= 0;
2793 if (!(op
& FUTEX_PRIVATE_FLAG
))
2794 flags
|= FLAGS_SHARED
;
2796 if (op
& FUTEX_CLOCK_REALTIME
) {
2797 flags
|= FLAGS_CLOCKRT
;
2798 if (cmd
!= FUTEX_WAIT_BITSET
&& cmd
!= FUTEX_WAIT_REQUEUE_PI
)
2804 case FUTEX_UNLOCK_PI
:
2805 case FUTEX_TRYLOCK_PI
:
2806 case FUTEX_WAIT_REQUEUE_PI
:
2807 case FUTEX_CMP_REQUEUE_PI
:
2808 if (!futex_cmpxchg_enabled
)
2814 val3
= FUTEX_BITSET_MATCH_ANY
;
2815 case FUTEX_WAIT_BITSET
:
2816 return futex_wait(uaddr
, flags
, val
, timeout
, val3
);
2818 val3
= FUTEX_BITSET_MATCH_ANY
;
2819 case FUTEX_WAKE_BITSET
:
2820 return futex_wake(uaddr
, flags
, val
, val3
);
2822 return futex_requeue(uaddr
, flags
, uaddr2
, val
, val2
, NULL
, 0);
2823 case FUTEX_CMP_REQUEUE
:
2824 return futex_requeue(uaddr
, flags
, uaddr2
, val
, val2
, &val3
, 0);
2826 return futex_wake_op(uaddr
, flags
, uaddr2
, val
, val2
, val3
);
2828 return futex_lock_pi(uaddr
, flags
, val
, timeout
, 0);
2829 case FUTEX_UNLOCK_PI
:
2830 return futex_unlock_pi(uaddr
, flags
);
2831 case FUTEX_TRYLOCK_PI
:
2832 return futex_lock_pi(uaddr
, flags
, 0, timeout
, 1);
2833 case FUTEX_WAIT_REQUEUE_PI
:
2834 val3
= FUTEX_BITSET_MATCH_ANY
;
2835 return futex_wait_requeue_pi(uaddr
, flags
, val
, timeout
, val3
,
2837 case FUTEX_CMP_REQUEUE_PI
:
2838 return futex_requeue(uaddr
, flags
, uaddr2
, val
, val2
, &val3
, 1);
2844 SYSCALL_DEFINE6(futex
, u32 __user
*, uaddr
, int, op
, u32
, val
,
2845 struct timespec __user
*, utime
, u32 __user
*, uaddr2
,
2849 ktime_t t
, *tp
= NULL
;
2851 int cmd
= op
& FUTEX_CMD_MASK
;
2853 if (utime
&& (cmd
== FUTEX_WAIT
|| cmd
== FUTEX_LOCK_PI
||
2854 cmd
== FUTEX_WAIT_BITSET
||
2855 cmd
== FUTEX_WAIT_REQUEUE_PI
)) {
2856 if (copy_from_user(&ts
, utime
, sizeof(ts
)) != 0)
2858 if (!timespec_valid(&ts
))
2861 t
= timespec_to_ktime(ts
);
2862 if (cmd
== FUTEX_WAIT
)
2863 t
= ktime_add_safe(ktime_get(), t
);
2867 * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
2868 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
2870 if (cmd
== FUTEX_REQUEUE
|| cmd
== FUTEX_CMP_REQUEUE
||
2871 cmd
== FUTEX_CMP_REQUEUE_PI
|| cmd
== FUTEX_WAKE_OP
)
2872 val2
= (u32
) (unsigned long) utime
;
2874 return do_futex(uaddr
, op
, val
, tp
, uaddr2
, val2
, val3
);
2877 static void __init
futex_detect_cmpxchg(void)
2879 #ifndef CONFIG_HAVE_FUTEX_CMPXCHG
2883 * This will fail and we want it. Some arch implementations do
2884 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2885 * functionality. We want to know that before we call in any
2886 * of the complex code paths. Also we want to prevent
2887 * registration of robust lists in that case. NULL is
2888 * guaranteed to fault and we get -EFAULT on functional
2889 * implementation, the non-functional ones will return
2892 if (cmpxchg_futex_value_locked(&curval
, NULL
, 0, 0) == -EFAULT
)
2893 futex_cmpxchg_enabled
= 1;
2897 static int __init
futex_init(void)
2901 futex_detect_cmpxchg();
2903 for (i
= 0; i
< ARRAY_SIZE(futex_queues
); i
++) {
2904 plist_head_init(&futex_queues
[i
].chain
);
2905 spin_lock_init(&futex_queues
[i
].lock
);
2910 core_initcall(futex_init
);