Loading...
1/*
2 * Fast Userspace Mutexes (which I call "Futexes!").
3 * (C) Rusty Russell, IBM 2002
4 *
5 * Generalized futexes, futex requeueing, misc fixes by Ingo Molnar
6 * (C) Copyright 2003 Red Hat Inc, All Rights Reserved
7 *
8 * Removed page pinning, fix privately mapped COW pages and other cleanups
9 * (C) Copyright 2003, 2004 Jamie Lokier
10 *
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.
14 *
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>
18 *
19 * PRIVATE futexes by Eric Dumazet
20 * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com>
21 *
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.
25 *
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.
29 *
30 * "The futexes are also cursed."
31 * "But they come in a choice of three flavours!"
32 *
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.
37 *
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.
42 *
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
46 */
47#include <linux/slab.h>
48#include <linux/poll.h>
49#include <linux/fs.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
64#include <asm/futex.h>
65
66#include "rtmutex_common.h"
67
68int __read_mostly futex_cmpxchg_enabled;
69
70#define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
71
72/*
73 * Futex flags used to encode options to functions and preserve them across
74 * restarts.
75 */
76#define FLAGS_SHARED 0x01
77#define FLAGS_CLOCKRT 0x02
78#define FLAGS_HAS_TIMEOUT 0x04
79
80/*
81 * Priority Inheritance state:
82 */
83struct futex_pi_state {
84 /*
85 * list of 'owned' pi_state instances - these have to be
86 * cleaned up in do_exit() if the task exits prematurely:
87 */
88 struct list_head list;
89
90 /*
91 * The PI object:
92 */
93 struct rt_mutex pi_mutex;
94
95 struct task_struct *owner;
96 atomic_t refcount;
97
98 union futex_key key;
99};
100
101/**
102 * struct futex_q - The hashed futex queue entry, one per waiting task
103 * @list: priority-sorted list of tasks waiting on this futex
104 * @task: the task waiting on the futex
105 * @lock_ptr: the hash bucket lock
106 * @key: the key the futex is hashed on
107 * @pi_state: optional priority inheritance state
108 * @rt_waiter: rt_waiter storage for use with requeue_pi
109 * @requeue_pi_key: the requeue_pi target futex key
110 * @bitset: bitset for the optional bitmasked wakeup
111 *
112 * We use this hashed waitqueue, instead of a normal wait_queue_t, so
113 * we can wake only the relevant ones (hashed queues may be shared).
114 *
115 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
116 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
117 * The order of wakeup is always to make the first condition true, then
118 * the second.
119 *
120 * PI futexes are typically woken before they are removed from the hash list via
121 * the rt_mutex code. See unqueue_me_pi().
122 */
123struct futex_q {
124 struct plist_node list;
125
126 struct task_struct *task;
127 spinlock_t *lock_ptr;
128 union futex_key key;
129 struct futex_pi_state *pi_state;
130 struct rt_mutex_waiter *rt_waiter;
131 union futex_key *requeue_pi_key;
132 u32 bitset;
133};
134
135static const struct futex_q futex_q_init = {
136 /* list gets initialized in queue_me()*/
137 .key = FUTEX_KEY_INIT,
138 .bitset = FUTEX_BITSET_MATCH_ANY
139};
140
141/*
142 * Hash buckets are shared by all the futex_keys that hash to the same
143 * location. Each key may have multiple futex_q structures, one for each task
144 * waiting on a futex.
145 */
146struct futex_hash_bucket {
147 spinlock_t lock;
148 struct plist_head chain;
149};
150
151static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS];
152
153/*
154 * We hash on the keys returned from get_futex_key (see below).
155 */
156static struct futex_hash_bucket *hash_futex(union futex_key *key)
157{
158 u32 hash = jhash2((u32*)&key->both.word,
159 (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
160 key->both.offset);
161 return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)];
162}
163
164/*
165 * Return 1 if two futex_keys are equal, 0 otherwise.
166 */
167static inline int match_futex(union futex_key *key1, union futex_key *key2)
168{
169 return (key1 && key2
170 && key1->both.word == key2->both.word
171 && key1->both.ptr == key2->both.ptr
172 && key1->both.offset == key2->both.offset);
173}
174
175/*
176 * Take a reference to the resource addressed by a key.
177 * Can be called while holding spinlocks.
178 *
179 */
180static void get_futex_key_refs(union futex_key *key)
181{
182 if (!key->both.ptr)
183 return;
184
185 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
186 case FUT_OFF_INODE:
187 ihold(key->shared.inode);
188 break;
189 case FUT_OFF_MMSHARED:
190 atomic_inc(&key->private.mm->mm_count);
191 break;
192 }
193}
194
195/*
196 * Drop a reference to the resource addressed by a key.
197 * The hash bucket spinlock must not be held.
198 */
199static void drop_futex_key_refs(union futex_key *key)
200{
201 if (!key->both.ptr) {
202 /* If we're here then we tried to put a key we failed to get */
203 WARN_ON_ONCE(1);
204 return;
205 }
206
207 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
208 case FUT_OFF_INODE:
209 iput(key->shared.inode);
210 break;
211 case FUT_OFF_MMSHARED:
212 mmdrop(key->private.mm);
213 break;
214 }
215}
216
217/**
218 * get_futex_key() - Get parameters which are the keys for a futex
219 * @uaddr: virtual address of the futex
220 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
221 * @key: address where result is stored.
222 * @rw: mapping needs to be read/write (values: VERIFY_READ,
223 * VERIFY_WRITE)
224 *
225 * Returns a negative error code or 0
226 * The key words are stored in *key on success.
227 *
228 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
229 * offset_within_page). For private mappings, it's (uaddr, current->mm).
230 * We can usually work out the index without swapping in the page.
231 *
232 * lock_page() might sleep, the caller should not hold a spinlock.
233 */
234static int
235get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw)
236{
237 unsigned long address = (unsigned long)uaddr;
238 struct mm_struct *mm = current->mm;
239 struct page *page, *page_head;
240 int err, ro = 0;
241
242 /*
243 * The futex address must be "naturally" aligned.
244 */
245 key->both.offset = address % PAGE_SIZE;
246 if (unlikely((address % sizeof(u32)) != 0))
247 return -EINVAL;
248 address -= key->both.offset;
249
250 /*
251 * PROCESS_PRIVATE futexes are fast.
252 * As the mm cannot disappear under us and the 'key' only needs
253 * virtual address, we dont even have to find the underlying vma.
254 * Note : We do have to check 'uaddr' is a valid user address,
255 * but access_ok() should be faster than find_vma()
256 */
257 if (!fshared) {
258 if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))))
259 return -EFAULT;
260 key->private.mm = mm;
261 key->private.address = address;
262 get_futex_key_refs(key);
263 return 0;
264 }
265
266again:
267 err = get_user_pages_fast(address, 1, 1, &page);
268 /*
269 * If write access is not required (eg. FUTEX_WAIT), try
270 * and get read-only access.
271 */
272 if (err == -EFAULT && rw == VERIFY_READ) {
273 err = get_user_pages_fast(address, 1, 0, &page);
274 ro = 1;
275 }
276 if (err < 0)
277 return err;
278 else
279 err = 0;
280
281#ifdef CONFIG_TRANSPARENT_HUGEPAGE
282 page_head = page;
283 if (unlikely(PageTail(page))) {
284 put_page(page);
285 /* serialize against __split_huge_page_splitting() */
286 local_irq_disable();
287 if (likely(__get_user_pages_fast(address, 1, 1, &page) == 1)) {
288 page_head = compound_head(page);
289 /*
290 * page_head is valid pointer but we must pin
291 * it before taking the PG_lock and/or
292 * PG_compound_lock. The moment we re-enable
293 * irqs __split_huge_page_splitting() can
294 * return and the head page can be freed from
295 * under us. We can't take the PG_lock and/or
296 * PG_compound_lock on a page that could be
297 * freed from under us.
298 */
299 if (page != page_head) {
300 get_page(page_head);
301 put_page(page);
302 }
303 local_irq_enable();
304 } else {
305 local_irq_enable();
306 goto again;
307 }
308 }
309#else
310 page_head = compound_head(page);
311 if (page != page_head) {
312 get_page(page_head);
313 put_page(page);
314 }
315#endif
316
317 lock_page(page_head);
318
319 /*
320 * If page_head->mapping is NULL, then it cannot be a PageAnon
321 * page; but it might be the ZERO_PAGE or in the gate area or
322 * in a special mapping (all cases which we are happy to fail);
323 * or it may have been a good file page when get_user_pages_fast
324 * found it, but truncated or holepunched or subjected to
325 * invalidate_complete_page2 before we got the page lock (also
326 * cases which we are happy to fail). And we hold a reference,
327 * so refcount care in invalidate_complete_page's remove_mapping
328 * prevents drop_caches from setting mapping to NULL beneath us.
329 *
330 * The case we do have to guard against is when memory pressure made
331 * shmem_writepage move it from filecache to swapcache beneath us:
332 * an unlikely race, but we do need to retry for page_head->mapping.
333 */
334 if (!page_head->mapping) {
335 int shmem_swizzled = PageSwapCache(page_head);
336 unlock_page(page_head);
337 put_page(page_head);
338 if (shmem_swizzled)
339 goto again;
340 return -EFAULT;
341 }
342
343 /*
344 * Private mappings are handled in a simple way.
345 *
346 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
347 * it's a read-only handle, it's expected that futexes attach to
348 * the object not the particular process.
349 */
350 if (PageAnon(page_head)) {
351 /*
352 * A RO anonymous page will never change and thus doesn't make
353 * sense for futex operations.
354 */
355 if (ro) {
356 err = -EFAULT;
357 goto out;
358 }
359
360 key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
361 key->private.mm = mm;
362 key->private.address = address;
363 } else {
364 key->both.offset |= FUT_OFF_INODE; /* inode-based key */
365 key->shared.inode = page_head->mapping->host;
366 key->shared.pgoff = page_head->index;
367 }
368
369 get_futex_key_refs(key);
370
371out:
372 unlock_page(page_head);
373 put_page(page_head);
374 return err;
375}
376
377static inline void put_futex_key(union futex_key *key)
378{
379 drop_futex_key_refs(key);
380}
381
382/**
383 * fault_in_user_writeable() - Fault in user address and verify RW access
384 * @uaddr: pointer to faulting user space address
385 *
386 * Slow path to fixup the fault we just took in the atomic write
387 * access to @uaddr.
388 *
389 * We have no generic implementation of a non-destructive write to the
390 * user address. We know that we faulted in the atomic pagefault
391 * disabled section so we can as well avoid the #PF overhead by
392 * calling get_user_pages() right away.
393 */
394static int fault_in_user_writeable(u32 __user *uaddr)
395{
396 struct mm_struct *mm = current->mm;
397 int ret;
398
399 down_read(&mm->mmap_sem);
400 ret = fixup_user_fault(current, mm, (unsigned long)uaddr,
401 FAULT_FLAG_WRITE);
402 up_read(&mm->mmap_sem);
403
404 return ret < 0 ? ret : 0;
405}
406
407/**
408 * futex_top_waiter() - Return the highest priority waiter on a futex
409 * @hb: the hash bucket the futex_q's reside in
410 * @key: the futex key (to distinguish it from other futex futex_q's)
411 *
412 * Must be called with the hb lock held.
413 */
414static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb,
415 union futex_key *key)
416{
417 struct futex_q *this;
418
419 plist_for_each_entry(this, &hb->chain, list) {
420 if (match_futex(&this->key, key))
421 return this;
422 }
423 return NULL;
424}
425
426static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr,
427 u32 uval, u32 newval)
428{
429 int ret;
430
431 pagefault_disable();
432 ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval);
433 pagefault_enable();
434
435 return ret;
436}
437
438static int get_futex_value_locked(u32 *dest, u32 __user *from)
439{
440 int ret;
441
442 pagefault_disable();
443 ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
444 pagefault_enable();
445
446 return ret ? -EFAULT : 0;
447}
448
449
450/*
451 * PI code:
452 */
453static int refill_pi_state_cache(void)
454{
455 struct futex_pi_state *pi_state;
456
457 if (likely(current->pi_state_cache))
458 return 0;
459
460 pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
461
462 if (!pi_state)
463 return -ENOMEM;
464
465 INIT_LIST_HEAD(&pi_state->list);
466 /* pi_mutex gets initialized later */
467 pi_state->owner = NULL;
468 atomic_set(&pi_state->refcount, 1);
469 pi_state->key = FUTEX_KEY_INIT;
470
471 current->pi_state_cache = pi_state;
472
473 return 0;
474}
475
476static struct futex_pi_state * alloc_pi_state(void)
477{
478 struct futex_pi_state *pi_state = current->pi_state_cache;
479
480 WARN_ON(!pi_state);
481 current->pi_state_cache = NULL;
482
483 return pi_state;
484}
485
486static void free_pi_state(struct futex_pi_state *pi_state)
487{
488 if (!atomic_dec_and_test(&pi_state->refcount))
489 return;
490
491 /*
492 * If pi_state->owner is NULL, the owner is most probably dying
493 * and has cleaned up the pi_state already
494 */
495 if (pi_state->owner) {
496 raw_spin_lock_irq(&pi_state->owner->pi_lock);
497 list_del_init(&pi_state->list);
498 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
499
500 rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
501 }
502
503 if (current->pi_state_cache)
504 kfree(pi_state);
505 else {
506 /*
507 * pi_state->list is already empty.
508 * clear pi_state->owner.
509 * refcount is at 0 - put it back to 1.
510 */
511 pi_state->owner = NULL;
512 atomic_set(&pi_state->refcount, 1);
513 current->pi_state_cache = pi_state;
514 }
515}
516
517/*
518 * Look up the task based on what TID userspace gave us.
519 * We dont trust it.
520 */
521static struct task_struct * futex_find_get_task(pid_t pid)
522{
523 struct task_struct *p;
524
525 rcu_read_lock();
526 p = find_task_by_vpid(pid);
527 if (p)
528 get_task_struct(p);
529
530 rcu_read_unlock();
531
532 return p;
533}
534
535/*
536 * This task is holding PI mutexes at exit time => bad.
537 * Kernel cleans up PI-state, but userspace is likely hosed.
538 * (Robust-futex cleanup is separate and might save the day for userspace.)
539 */
540void exit_pi_state_list(struct task_struct *curr)
541{
542 struct list_head *next, *head = &curr->pi_state_list;
543 struct futex_pi_state *pi_state;
544 struct futex_hash_bucket *hb;
545 union futex_key key = FUTEX_KEY_INIT;
546
547 if (!futex_cmpxchg_enabled)
548 return;
549 /*
550 * We are a ZOMBIE and nobody can enqueue itself on
551 * pi_state_list anymore, but we have to be careful
552 * versus waiters unqueueing themselves:
553 */
554 raw_spin_lock_irq(&curr->pi_lock);
555 while (!list_empty(head)) {
556
557 next = head->next;
558 pi_state = list_entry(next, struct futex_pi_state, list);
559 key = pi_state->key;
560 hb = hash_futex(&key);
561 raw_spin_unlock_irq(&curr->pi_lock);
562
563 spin_lock(&hb->lock);
564
565 raw_spin_lock_irq(&curr->pi_lock);
566 /*
567 * We dropped the pi-lock, so re-check whether this
568 * task still owns the PI-state:
569 */
570 if (head->next != next) {
571 spin_unlock(&hb->lock);
572 continue;
573 }
574
575 WARN_ON(pi_state->owner != curr);
576 WARN_ON(list_empty(&pi_state->list));
577 list_del_init(&pi_state->list);
578 pi_state->owner = NULL;
579 raw_spin_unlock_irq(&curr->pi_lock);
580
581 rt_mutex_unlock(&pi_state->pi_mutex);
582
583 spin_unlock(&hb->lock);
584
585 raw_spin_lock_irq(&curr->pi_lock);
586 }
587 raw_spin_unlock_irq(&curr->pi_lock);
588}
589
590static int
591lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
592 union futex_key *key, struct futex_pi_state **ps)
593{
594 struct futex_pi_state *pi_state = NULL;
595 struct futex_q *this, *next;
596 struct plist_head *head;
597 struct task_struct *p;
598 pid_t pid = uval & FUTEX_TID_MASK;
599
600 head = &hb->chain;
601
602 plist_for_each_entry_safe(this, next, head, list) {
603 if (match_futex(&this->key, key)) {
604 /*
605 * Another waiter already exists - bump up
606 * the refcount and return its pi_state:
607 */
608 pi_state = this->pi_state;
609 /*
610 * Userspace might have messed up non-PI and PI futexes
611 */
612 if (unlikely(!pi_state))
613 return -EINVAL;
614
615 WARN_ON(!atomic_read(&pi_state->refcount));
616
617 /*
618 * When pi_state->owner is NULL then the owner died
619 * and another waiter is on the fly. pi_state->owner
620 * is fixed up by the task which acquires
621 * pi_state->rt_mutex.
622 *
623 * We do not check for pid == 0 which can happen when
624 * the owner died and robust_list_exit() cleared the
625 * TID.
626 */
627 if (pid && pi_state->owner) {
628 /*
629 * Bail out if user space manipulated the
630 * futex value.
631 */
632 if (pid != task_pid_vnr(pi_state->owner))
633 return -EINVAL;
634 }
635
636 atomic_inc(&pi_state->refcount);
637 *ps = pi_state;
638
639 return 0;
640 }
641 }
642
643 /*
644 * We are the first waiter - try to look up the real owner and attach
645 * the new pi_state to it, but bail out when TID = 0
646 */
647 if (!pid)
648 return -ESRCH;
649 p = futex_find_get_task(pid);
650 if (!p)
651 return -ESRCH;
652
653 /*
654 * We need to look at the task state flags to figure out,
655 * whether the task is exiting. To protect against the do_exit
656 * change of the task flags, we do this protected by
657 * p->pi_lock:
658 */
659 raw_spin_lock_irq(&p->pi_lock);
660 if (unlikely(p->flags & PF_EXITING)) {
661 /*
662 * The task is on the way out. When PF_EXITPIDONE is
663 * set, we know that the task has finished the
664 * cleanup:
665 */
666 int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
667
668 raw_spin_unlock_irq(&p->pi_lock);
669 put_task_struct(p);
670 return ret;
671 }
672
673 pi_state = alloc_pi_state();
674
675 /*
676 * Initialize the pi_mutex in locked state and make 'p'
677 * the owner of it:
678 */
679 rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
680
681 /* Store the key for possible exit cleanups: */
682 pi_state->key = *key;
683
684 WARN_ON(!list_empty(&pi_state->list));
685 list_add(&pi_state->list, &p->pi_state_list);
686 pi_state->owner = p;
687 raw_spin_unlock_irq(&p->pi_lock);
688
689 put_task_struct(p);
690
691 *ps = pi_state;
692
693 return 0;
694}
695
696/**
697 * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
698 * @uaddr: the pi futex user address
699 * @hb: the pi futex hash bucket
700 * @key: the futex key associated with uaddr and hb
701 * @ps: the pi_state pointer where we store the result of the
702 * lookup
703 * @task: the task to perform the atomic lock work for. This will
704 * be "current" except in the case of requeue pi.
705 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
706 *
707 * Returns:
708 * 0 - ready to wait
709 * 1 - acquired the lock
710 * <0 - error
711 *
712 * The hb->lock and futex_key refs shall be held by the caller.
713 */
714static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb,
715 union futex_key *key,
716 struct futex_pi_state **ps,
717 struct task_struct *task, int set_waiters)
718{
719 int lock_taken, ret, ownerdied = 0;
720 u32 uval, newval, curval, vpid = task_pid_vnr(task);
721
722retry:
723 ret = lock_taken = 0;
724
725 /*
726 * To avoid races, we attempt to take the lock here again
727 * (by doing a 0 -> TID atomic cmpxchg), while holding all
728 * the locks. It will most likely not succeed.
729 */
730 newval = vpid;
731 if (set_waiters)
732 newval |= FUTEX_WAITERS;
733
734 if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval)))
735 return -EFAULT;
736
737 /*
738 * Detect deadlocks.
739 */
740 if ((unlikely((curval & FUTEX_TID_MASK) == vpid)))
741 return -EDEADLK;
742
743 /*
744 * Surprise - we got the lock. Just return to userspace:
745 */
746 if (unlikely(!curval))
747 return 1;
748
749 uval = curval;
750
751 /*
752 * Set the FUTEX_WAITERS flag, so the owner will know it has someone
753 * to wake at the next unlock.
754 */
755 newval = curval | FUTEX_WAITERS;
756
757 /*
758 * There are two cases, where a futex might have no owner (the
759 * owner TID is 0): OWNER_DIED. We take over the futex in this
760 * case. We also do an unconditional take over, when the owner
761 * of the futex died.
762 *
763 * This is safe as we are protected by the hash bucket lock !
764 */
765 if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) {
766 /* Keep the OWNER_DIED bit */
767 newval = (curval & ~FUTEX_TID_MASK) | vpid;
768 ownerdied = 0;
769 lock_taken = 1;
770 }
771
772 if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)))
773 return -EFAULT;
774 if (unlikely(curval != uval))
775 goto retry;
776
777 /*
778 * We took the lock due to owner died take over.
779 */
780 if (unlikely(lock_taken))
781 return 1;
782
783 /*
784 * We dont have the lock. Look up the PI state (or create it if
785 * we are the first waiter):
786 */
787 ret = lookup_pi_state(uval, hb, key, ps);
788
789 if (unlikely(ret)) {
790 switch (ret) {
791 case -ESRCH:
792 /*
793 * No owner found for this futex. Check if the
794 * OWNER_DIED bit is set to figure out whether
795 * this is a robust futex or not.
796 */
797 if (get_futex_value_locked(&curval, uaddr))
798 return -EFAULT;
799
800 /*
801 * We simply start over in case of a robust
802 * futex. The code above will take the futex
803 * and return happy.
804 */
805 if (curval & FUTEX_OWNER_DIED) {
806 ownerdied = 1;
807 goto retry;
808 }
809 default:
810 break;
811 }
812 }
813
814 return ret;
815}
816
817/**
818 * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket
819 * @q: The futex_q to unqueue
820 *
821 * The q->lock_ptr must not be NULL and must be held by the caller.
822 */
823static void __unqueue_futex(struct futex_q *q)
824{
825 struct futex_hash_bucket *hb;
826
827 if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr))
828 || WARN_ON(plist_node_empty(&q->list)))
829 return;
830
831 hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock);
832 plist_del(&q->list, &hb->chain);
833}
834
835/*
836 * The hash bucket lock must be held when this is called.
837 * Afterwards, the futex_q must not be accessed.
838 */
839static void wake_futex(struct futex_q *q)
840{
841 struct task_struct *p = q->task;
842
843 /*
844 * We set q->lock_ptr = NULL _before_ we wake up the task. If
845 * a non-futex wake up happens on another CPU then the task
846 * might exit and p would dereference a non-existing task
847 * struct. Prevent this by holding a reference on p across the
848 * wake up.
849 */
850 get_task_struct(p);
851
852 __unqueue_futex(q);
853 /*
854 * The waiting task can free the futex_q as soon as
855 * q->lock_ptr = NULL is written, without taking any locks. A
856 * memory barrier is required here to prevent the following
857 * store to lock_ptr from getting ahead of the plist_del.
858 */
859 smp_wmb();
860 q->lock_ptr = NULL;
861
862 wake_up_state(p, TASK_NORMAL);
863 put_task_struct(p);
864}
865
866static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
867{
868 struct task_struct *new_owner;
869 struct futex_pi_state *pi_state = this->pi_state;
870 u32 uninitialized_var(curval), newval;
871
872 if (!pi_state)
873 return -EINVAL;
874
875 /*
876 * If current does not own the pi_state then the futex is
877 * inconsistent and user space fiddled with the futex value.
878 */
879 if (pi_state->owner != current)
880 return -EINVAL;
881
882 raw_spin_lock(&pi_state->pi_mutex.wait_lock);
883 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
884
885 /*
886 * It is possible that the next waiter (the one that brought
887 * this owner to the kernel) timed out and is no longer
888 * waiting on the lock.
889 */
890 if (!new_owner)
891 new_owner = this->task;
892
893 /*
894 * We pass it to the next owner. (The WAITERS bit is always
895 * kept enabled while there is PI state around. We must also
896 * preserve the owner died bit.)
897 */
898 if (!(uval & FUTEX_OWNER_DIED)) {
899 int ret = 0;
900
901 newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
902
903 if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
904 ret = -EFAULT;
905 else if (curval != uval)
906 ret = -EINVAL;
907 if (ret) {
908 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
909 return ret;
910 }
911 }
912
913 raw_spin_lock_irq(&pi_state->owner->pi_lock);
914 WARN_ON(list_empty(&pi_state->list));
915 list_del_init(&pi_state->list);
916 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
917
918 raw_spin_lock_irq(&new_owner->pi_lock);
919 WARN_ON(!list_empty(&pi_state->list));
920 list_add(&pi_state->list, &new_owner->pi_state_list);
921 pi_state->owner = new_owner;
922 raw_spin_unlock_irq(&new_owner->pi_lock);
923
924 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
925 rt_mutex_unlock(&pi_state->pi_mutex);
926
927 return 0;
928}
929
930static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
931{
932 u32 uninitialized_var(oldval);
933
934 /*
935 * There is no waiter, so we unlock the futex. The owner died
936 * bit has not to be preserved here. We are the owner:
937 */
938 if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0))
939 return -EFAULT;
940 if (oldval != uval)
941 return -EAGAIN;
942
943 return 0;
944}
945
946/*
947 * Express the locking dependencies for lockdep:
948 */
949static inline void
950double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
951{
952 if (hb1 <= hb2) {
953 spin_lock(&hb1->lock);
954 if (hb1 < hb2)
955 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
956 } else { /* hb1 > hb2 */
957 spin_lock(&hb2->lock);
958 spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
959 }
960}
961
962static inline void
963double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
964{
965 spin_unlock(&hb1->lock);
966 if (hb1 != hb2)
967 spin_unlock(&hb2->lock);
968}
969
970/*
971 * Wake up waiters matching bitset queued on this futex (uaddr).
972 */
973static int
974futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
975{
976 struct futex_hash_bucket *hb;
977 struct futex_q *this, *next;
978 struct plist_head *head;
979 union futex_key key = FUTEX_KEY_INIT;
980 int ret;
981
982 if (!bitset)
983 return -EINVAL;
984
985 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ);
986 if (unlikely(ret != 0))
987 goto out;
988
989 hb = hash_futex(&key);
990 spin_lock(&hb->lock);
991 head = &hb->chain;
992
993 plist_for_each_entry_safe(this, next, head, list) {
994 if (match_futex (&this->key, &key)) {
995 if (this->pi_state || this->rt_waiter) {
996 ret = -EINVAL;
997 break;
998 }
999
1000 /* Check if one of the bits is set in both bitsets */
1001 if (!(this->bitset & bitset))
1002 continue;
1003
1004 wake_futex(this);
1005 if (++ret >= nr_wake)
1006 break;
1007 }
1008 }
1009
1010 spin_unlock(&hb->lock);
1011 put_futex_key(&key);
1012out:
1013 return ret;
1014}
1015
1016/*
1017 * Wake up all waiters hashed on the physical page that is mapped
1018 * to this virtual address:
1019 */
1020static int
1021futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
1022 int nr_wake, int nr_wake2, int op)
1023{
1024 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
1025 struct futex_hash_bucket *hb1, *hb2;
1026 struct plist_head *head;
1027 struct futex_q *this, *next;
1028 int ret, op_ret;
1029
1030retry:
1031 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
1032 if (unlikely(ret != 0))
1033 goto out;
1034 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
1035 if (unlikely(ret != 0))
1036 goto out_put_key1;
1037
1038 hb1 = hash_futex(&key1);
1039 hb2 = hash_futex(&key2);
1040
1041retry_private:
1042 double_lock_hb(hb1, hb2);
1043 op_ret = futex_atomic_op_inuser(op, uaddr2);
1044 if (unlikely(op_ret < 0)) {
1045
1046 double_unlock_hb(hb1, hb2);
1047
1048#ifndef CONFIG_MMU
1049 /*
1050 * we don't get EFAULT from MMU faults if we don't have an MMU,
1051 * but we might get them from range checking
1052 */
1053 ret = op_ret;
1054 goto out_put_keys;
1055#endif
1056
1057 if (unlikely(op_ret != -EFAULT)) {
1058 ret = op_ret;
1059 goto out_put_keys;
1060 }
1061
1062 ret = fault_in_user_writeable(uaddr2);
1063 if (ret)
1064 goto out_put_keys;
1065
1066 if (!(flags & FLAGS_SHARED))
1067 goto retry_private;
1068
1069 put_futex_key(&key2);
1070 put_futex_key(&key1);
1071 goto retry;
1072 }
1073
1074 head = &hb1->chain;
1075
1076 plist_for_each_entry_safe(this, next, head, list) {
1077 if (match_futex (&this->key, &key1)) {
1078 wake_futex(this);
1079 if (++ret >= nr_wake)
1080 break;
1081 }
1082 }
1083
1084 if (op_ret > 0) {
1085 head = &hb2->chain;
1086
1087 op_ret = 0;
1088 plist_for_each_entry_safe(this, next, head, list) {
1089 if (match_futex (&this->key, &key2)) {
1090 wake_futex(this);
1091 if (++op_ret >= nr_wake2)
1092 break;
1093 }
1094 }
1095 ret += op_ret;
1096 }
1097
1098 double_unlock_hb(hb1, hb2);
1099out_put_keys:
1100 put_futex_key(&key2);
1101out_put_key1:
1102 put_futex_key(&key1);
1103out:
1104 return ret;
1105}
1106
1107/**
1108 * requeue_futex() - Requeue a futex_q from one hb to another
1109 * @q: the futex_q to requeue
1110 * @hb1: the source hash_bucket
1111 * @hb2: the target hash_bucket
1112 * @key2: the new key for the requeued futex_q
1113 */
1114static inline
1115void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1,
1116 struct futex_hash_bucket *hb2, union futex_key *key2)
1117{
1118
1119 /*
1120 * If key1 and key2 hash to the same bucket, no need to
1121 * requeue.
1122 */
1123 if (likely(&hb1->chain != &hb2->chain)) {
1124 plist_del(&q->list, &hb1->chain);
1125 plist_add(&q->list, &hb2->chain);
1126 q->lock_ptr = &hb2->lock;
1127 }
1128 get_futex_key_refs(key2);
1129 q->key = *key2;
1130}
1131
1132/**
1133 * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
1134 * @q: the futex_q
1135 * @key: the key of the requeue target futex
1136 * @hb: the hash_bucket of the requeue target futex
1137 *
1138 * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1139 * target futex if it is uncontended or via a lock steal. Set the futex_q key
1140 * to the requeue target futex so the waiter can detect the wakeup on the right
1141 * futex, but remove it from the hb and NULL the rt_waiter so it can detect
1142 * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1143 * to protect access to the pi_state to fixup the owner later. Must be called
1144 * with both q->lock_ptr and hb->lock held.
1145 */
1146static inline
1147void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key,
1148 struct futex_hash_bucket *hb)
1149{
1150 get_futex_key_refs(key);
1151 q->key = *key;
1152
1153 __unqueue_futex(q);
1154
1155 WARN_ON(!q->rt_waiter);
1156 q->rt_waiter = NULL;
1157
1158 q->lock_ptr = &hb->lock;
1159
1160 wake_up_state(q->task, TASK_NORMAL);
1161}
1162
1163/**
1164 * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
1165 * @pifutex: the user address of the to futex
1166 * @hb1: the from futex hash bucket, must be locked by the caller
1167 * @hb2: the to futex hash bucket, must be locked by the caller
1168 * @key1: the from futex key
1169 * @key2: the to futex key
1170 * @ps: address to store the pi_state pointer
1171 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
1172 *
1173 * Try and get the lock on behalf of the top waiter if we can do it atomically.
1174 * Wake the top waiter if we succeed. If the caller specified set_waiters,
1175 * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1176 * hb1 and hb2 must be held by the caller.
1177 *
1178 * Returns:
1179 * 0 - failed to acquire the lock atomicly
1180 * 1 - acquired the lock
1181 * <0 - error
1182 */
1183static int futex_proxy_trylock_atomic(u32 __user *pifutex,
1184 struct futex_hash_bucket *hb1,
1185 struct futex_hash_bucket *hb2,
1186 union futex_key *key1, union futex_key *key2,
1187 struct futex_pi_state **ps, int set_waiters)
1188{
1189 struct futex_q *top_waiter = NULL;
1190 u32 curval;
1191 int ret;
1192
1193 if (get_futex_value_locked(&curval, pifutex))
1194 return -EFAULT;
1195
1196 /*
1197 * Find the top_waiter and determine if there are additional waiters.
1198 * If the caller intends to requeue more than 1 waiter to pifutex,
1199 * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1200 * as we have means to handle the possible fault. If not, don't set
1201 * the bit unecessarily as it will force the subsequent unlock to enter
1202 * the kernel.
1203 */
1204 top_waiter = futex_top_waiter(hb1, key1);
1205
1206 /* There are no waiters, nothing for us to do. */
1207 if (!top_waiter)
1208 return 0;
1209
1210 /* Ensure we requeue to the expected futex. */
1211 if (!match_futex(top_waiter->requeue_pi_key, key2))
1212 return -EINVAL;
1213
1214 /*
1215 * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1216 * the contended case or if set_waiters is 1. The pi_state is returned
1217 * in ps in contended cases.
1218 */
1219 ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task,
1220 set_waiters);
1221 if (ret == 1)
1222 requeue_pi_wake_futex(top_waiter, key2, hb2);
1223
1224 return ret;
1225}
1226
1227/**
1228 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
1229 * @uaddr1: source futex user address
1230 * @flags: futex flags (FLAGS_SHARED, etc.)
1231 * @uaddr2: target futex user address
1232 * @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1233 * @nr_requeue: number of waiters to requeue (0-INT_MAX)
1234 * @cmpval: @uaddr1 expected value (or %NULL)
1235 * @requeue_pi: if we are attempting to requeue from a non-pi futex to a
1236 * pi futex (pi to pi requeue is not supported)
1237 *
1238 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1239 * uaddr2 atomically on behalf of the top waiter.
1240 *
1241 * Returns:
1242 * >=0 - on success, the number of tasks requeued or woken
1243 * <0 - on error
1244 */
1245static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
1246 u32 __user *uaddr2, int nr_wake, int nr_requeue,
1247 u32 *cmpval, int requeue_pi)
1248{
1249 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
1250 int drop_count = 0, task_count = 0, ret;
1251 struct futex_pi_state *pi_state = NULL;
1252 struct futex_hash_bucket *hb1, *hb2;
1253 struct plist_head *head1;
1254 struct futex_q *this, *next;
1255 u32 curval2;
1256
1257 if (requeue_pi) {
1258 /*
1259 * requeue_pi requires a pi_state, try to allocate it now
1260 * without any locks in case it fails.
1261 */
1262 if (refill_pi_state_cache())
1263 return -ENOMEM;
1264 /*
1265 * requeue_pi must wake as many tasks as it can, up to nr_wake
1266 * + nr_requeue, since it acquires the rt_mutex prior to
1267 * returning to userspace, so as to not leave the rt_mutex with
1268 * waiters and no owner. However, second and third wake-ups
1269 * cannot be predicted as they involve race conditions with the
1270 * first wake and a fault while looking up the pi_state. Both
1271 * pthread_cond_signal() and pthread_cond_broadcast() should
1272 * use nr_wake=1.
1273 */
1274 if (nr_wake != 1)
1275 return -EINVAL;
1276 }
1277
1278retry:
1279 if (pi_state != NULL) {
1280 /*
1281 * We will have to lookup the pi_state again, so free this one
1282 * to keep the accounting correct.
1283 */
1284 free_pi_state(pi_state);
1285 pi_state = NULL;
1286 }
1287
1288 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
1289 if (unlikely(ret != 0))
1290 goto out;
1291 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2,
1292 requeue_pi ? VERIFY_WRITE : VERIFY_READ);
1293 if (unlikely(ret != 0))
1294 goto out_put_key1;
1295
1296 hb1 = hash_futex(&key1);
1297 hb2 = hash_futex(&key2);
1298
1299retry_private:
1300 double_lock_hb(hb1, hb2);
1301
1302 if (likely(cmpval != NULL)) {
1303 u32 curval;
1304
1305 ret = get_futex_value_locked(&curval, uaddr1);
1306
1307 if (unlikely(ret)) {
1308 double_unlock_hb(hb1, hb2);
1309
1310 ret = get_user(curval, uaddr1);
1311 if (ret)
1312 goto out_put_keys;
1313
1314 if (!(flags & FLAGS_SHARED))
1315 goto retry_private;
1316
1317 put_futex_key(&key2);
1318 put_futex_key(&key1);
1319 goto retry;
1320 }
1321 if (curval != *cmpval) {
1322 ret = -EAGAIN;
1323 goto out_unlock;
1324 }
1325 }
1326
1327 if (requeue_pi && (task_count - nr_wake < nr_requeue)) {
1328 /*
1329 * Attempt to acquire uaddr2 and wake the top waiter. If we
1330 * intend to requeue waiters, force setting the FUTEX_WAITERS
1331 * bit. We force this here where we are able to easily handle
1332 * faults rather in the requeue loop below.
1333 */
1334 ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1,
1335 &key2, &pi_state, nr_requeue);
1336
1337 /*
1338 * At this point the top_waiter has either taken uaddr2 or is
1339 * waiting on it. If the former, then the pi_state will not
1340 * exist yet, look it up one more time to ensure we have a
1341 * reference to it.
1342 */
1343 if (ret == 1) {
1344 WARN_ON(pi_state);
1345 drop_count++;
1346 task_count++;
1347 ret = get_futex_value_locked(&curval2, uaddr2);
1348 if (!ret)
1349 ret = lookup_pi_state(curval2, hb2, &key2,
1350 &pi_state);
1351 }
1352
1353 switch (ret) {
1354 case 0:
1355 break;
1356 case -EFAULT:
1357 double_unlock_hb(hb1, hb2);
1358 put_futex_key(&key2);
1359 put_futex_key(&key1);
1360 ret = fault_in_user_writeable(uaddr2);
1361 if (!ret)
1362 goto retry;
1363 goto out;
1364 case -EAGAIN:
1365 /* The owner was exiting, try again. */
1366 double_unlock_hb(hb1, hb2);
1367 put_futex_key(&key2);
1368 put_futex_key(&key1);
1369 cond_resched();
1370 goto retry;
1371 default:
1372 goto out_unlock;
1373 }
1374 }
1375
1376 head1 = &hb1->chain;
1377 plist_for_each_entry_safe(this, next, head1, list) {
1378 if (task_count - nr_wake >= nr_requeue)
1379 break;
1380
1381 if (!match_futex(&this->key, &key1))
1382 continue;
1383
1384 /*
1385 * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
1386 * be paired with each other and no other futex ops.
1387 */
1388 if ((requeue_pi && !this->rt_waiter) ||
1389 (!requeue_pi && this->rt_waiter)) {
1390 ret = -EINVAL;
1391 break;
1392 }
1393
1394 /*
1395 * Wake nr_wake waiters. For requeue_pi, if we acquired the
1396 * lock, we already woke the top_waiter. If not, it will be
1397 * woken by futex_unlock_pi().
1398 */
1399 if (++task_count <= nr_wake && !requeue_pi) {
1400 wake_futex(this);
1401 continue;
1402 }
1403
1404 /* Ensure we requeue to the expected futex for requeue_pi. */
1405 if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) {
1406 ret = -EINVAL;
1407 break;
1408 }
1409
1410 /*
1411 * Requeue nr_requeue waiters and possibly one more in the case
1412 * of requeue_pi if we couldn't acquire the lock atomically.
1413 */
1414 if (requeue_pi) {
1415 /* Prepare the waiter to take the rt_mutex. */
1416 atomic_inc(&pi_state->refcount);
1417 this->pi_state = pi_state;
1418 ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex,
1419 this->rt_waiter,
1420 this->task, 1);
1421 if (ret == 1) {
1422 /* We got the lock. */
1423 requeue_pi_wake_futex(this, &key2, hb2);
1424 drop_count++;
1425 continue;
1426 } else if (ret) {
1427 /* -EDEADLK */
1428 this->pi_state = NULL;
1429 free_pi_state(pi_state);
1430 goto out_unlock;
1431 }
1432 }
1433 requeue_futex(this, hb1, hb2, &key2);
1434 drop_count++;
1435 }
1436
1437out_unlock:
1438 double_unlock_hb(hb1, hb2);
1439
1440 /*
1441 * drop_futex_key_refs() must be called outside the spinlocks. During
1442 * the requeue we moved futex_q's from the hash bucket at key1 to the
1443 * one at key2 and updated their key pointer. We no longer need to
1444 * hold the references to key1.
1445 */
1446 while (--drop_count >= 0)
1447 drop_futex_key_refs(&key1);
1448
1449out_put_keys:
1450 put_futex_key(&key2);
1451out_put_key1:
1452 put_futex_key(&key1);
1453out:
1454 if (pi_state != NULL)
1455 free_pi_state(pi_state);
1456 return ret ? ret : task_count;
1457}
1458
1459/* The key must be already stored in q->key. */
1460static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
1461 __acquires(&hb->lock)
1462{
1463 struct futex_hash_bucket *hb;
1464
1465 hb = hash_futex(&q->key);
1466 q->lock_ptr = &hb->lock;
1467
1468 spin_lock(&hb->lock);
1469 return hb;
1470}
1471
1472static inline void
1473queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
1474 __releases(&hb->lock)
1475{
1476 spin_unlock(&hb->lock);
1477}
1478
1479/**
1480 * queue_me() - Enqueue the futex_q on the futex_hash_bucket
1481 * @q: The futex_q to enqueue
1482 * @hb: The destination hash bucket
1483 *
1484 * The hb->lock must be held by the caller, and is released here. A call to
1485 * queue_me() is typically paired with exactly one call to unqueue_me(). The
1486 * exceptions involve the PI related operations, which may use unqueue_me_pi()
1487 * or nothing if the unqueue is done as part of the wake process and the unqueue
1488 * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
1489 * an example).
1490 */
1491static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
1492 __releases(&hb->lock)
1493{
1494 int prio;
1495
1496 /*
1497 * The priority used to register this element is
1498 * - either the real thread-priority for the real-time threads
1499 * (i.e. threads with a priority lower than MAX_RT_PRIO)
1500 * - or MAX_RT_PRIO for non-RT threads.
1501 * Thus, all RT-threads are woken first in priority order, and
1502 * the others are woken last, in FIFO order.
1503 */
1504 prio = min(current->normal_prio, MAX_RT_PRIO);
1505
1506 plist_node_init(&q->list, prio);
1507 plist_add(&q->list, &hb->chain);
1508 q->task = current;
1509 spin_unlock(&hb->lock);
1510}
1511
1512/**
1513 * unqueue_me() - Remove the futex_q from its futex_hash_bucket
1514 * @q: The futex_q to unqueue
1515 *
1516 * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must
1517 * be paired with exactly one earlier call to queue_me().
1518 *
1519 * Returns:
1520 * 1 - if the futex_q was still queued (and we removed unqueued it)
1521 * 0 - if the futex_q was already removed by the waking thread
1522 */
1523static int unqueue_me(struct futex_q *q)
1524{
1525 spinlock_t *lock_ptr;
1526 int ret = 0;
1527
1528 /* In the common case we don't take the spinlock, which is nice. */
1529retry:
1530 lock_ptr = q->lock_ptr;
1531 barrier();
1532 if (lock_ptr != NULL) {
1533 spin_lock(lock_ptr);
1534 /*
1535 * q->lock_ptr can change between reading it and
1536 * spin_lock(), causing us to take the wrong lock. This
1537 * corrects the race condition.
1538 *
1539 * Reasoning goes like this: if we have the wrong lock,
1540 * q->lock_ptr must have changed (maybe several times)
1541 * between reading it and the spin_lock(). It can
1542 * change again after the spin_lock() but only if it was
1543 * already changed before the spin_lock(). It cannot,
1544 * however, change back to the original value. Therefore
1545 * we can detect whether we acquired the correct lock.
1546 */
1547 if (unlikely(lock_ptr != q->lock_ptr)) {
1548 spin_unlock(lock_ptr);
1549 goto retry;
1550 }
1551 __unqueue_futex(q);
1552
1553 BUG_ON(q->pi_state);
1554
1555 spin_unlock(lock_ptr);
1556 ret = 1;
1557 }
1558
1559 drop_futex_key_refs(&q->key);
1560 return ret;
1561}
1562
1563/*
1564 * PI futexes can not be requeued and must remove themself from the
1565 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1566 * and dropped here.
1567 */
1568static void unqueue_me_pi(struct futex_q *q)
1569 __releases(q->lock_ptr)
1570{
1571 __unqueue_futex(q);
1572
1573 BUG_ON(!q->pi_state);
1574 free_pi_state(q->pi_state);
1575 q->pi_state = NULL;
1576
1577 spin_unlock(q->lock_ptr);
1578}
1579
1580/*
1581 * Fixup the pi_state owner with the new owner.
1582 *
1583 * Must be called with hash bucket lock held and mm->sem held for non
1584 * private futexes.
1585 */
1586static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
1587 struct task_struct *newowner)
1588{
1589 u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;
1590 struct futex_pi_state *pi_state = q->pi_state;
1591 struct task_struct *oldowner = pi_state->owner;
1592 u32 uval, uninitialized_var(curval), newval;
1593 int ret;
1594
1595 /* Owner died? */
1596 if (!pi_state->owner)
1597 newtid |= FUTEX_OWNER_DIED;
1598
1599 /*
1600 * We are here either because we stole the rtmutex from the
1601 * previous highest priority waiter or we are the highest priority
1602 * waiter but failed to get the rtmutex the first time.
1603 * We have to replace the newowner TID in the user space variable.
1604 * This must be atomic as we have to preserve the owner died bit here.
1605 *
1606 * Note: We write the user space value _before_ changing the pi_state
1607 * because we can fault here. Imagine swapped out pages or a fork
1608 * that marked all the anonymous memory readonly for cow.
1609 *
1610 * Modifying pi_state _before_ the user space value would
1611 * leave the pi_state in an inconsistent state when we fault
1612 * here, because we need to drop the hash bucket lock to
1613 * handle the fault. This might be observed in the PID check
1614 * in lookup_pi_state.
1615 */
1616retry:
1617 if (get_futex_value_locked(&uval, uaddr))
1618 goto handle_fault;
1619
1620 while (1) {
1621 newval = (uval & FUTEX_OWNER_DIED) | newtid;
1622
1623 if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
1624 goto handle_fault;
1625 if (curval == uval)
1626 break;
1627 uval = curval;
1628 }
1629
1630 /*
1631 * We fixed up user space. Now we need to fix the pi_state
1632 * itself.
1633 */
1634 if (pi_state->owner != NULL) {
1635 raw_spin_lock_irq(&pi_state->owner->pi_lock);
1636 WARN_ON(list_empty(&pi_state->list));
1637 list_del_init(&pi_state->list);
1638 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
1639 }
1640
1641 pi_state->owner = newowner;
1642
1643 raw_spin_lock_irq(&newowner->pi_lock);
1644 WARN_ON(!list_empty(&pi_state->list));
1645 list_add(&pi_state->list, &newowner->pi_state_list);
1646 raw_spin_unlock_irq(&newowner->pi_lock);
1647 return 0;
1648
1649 /*
1650 * To handle the page fault we need to drop the hash bucket
1651 * lock here. That gives the other task (either the highest priority
1652 * waiter itself or the task which stole the rtmutex) the
1653 * chance to try the fixup of the pi_state. So once we are
1654 * back from handling the fault we need to check the pi_state
1655 * after reacquiring the hash bucket lock and before trying to
1656 * do another fixup. When the fixup has been done already we
1657 * simply return.
1658 */
1659handle_fault:
1660 spin_unlock(q->lock_ptr);
1661
1662 ret = fault_in_user_writeable(uaddr);
1663
1664 spin_lock(q->lock_ptr);
1665
1666 /*
1667 * Check if someone else fixed it for us:
1668 */
1669 if (pi_state->owner != oldowner)
1670 return 0;
1671
1672 if (ret)
1673 return ret;
1674
1675 goto retry;
1676}
1677
1678static long futex_wait_restart(struct restart_block *restart);
1679
1680/**
1681 * fixup_owner() - Post lock pi_state and corner case management
1682 * @uaddr: user address of the futex
1683 * @q: futex_q (contains pi_state and access to the rt_mutex)
1684 * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
1685 *
1686 * After attempting to lock an rt_mutex, this function is called to cleanup
1687 * the pi_state owner as well as handle race conditions that may allow us to
1688 * acquire the lock. Must be called with the hb lock held.
1689 *
1690 * Returns:
1691 * 1 - success, lock taken
1692 * 0 - success, lock not taken
1693 * <0 - on error (-EFAULT)
1694 */
1695static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked)
1696{
1697 struct task_struct *owner;
1698 int ret = 0;
1699
1700 if (locked) {
1701 /*
1702 * Got the lock. We might not be the anticipated owner if we
1703 * did a lock-steal - fix up the PI-state in that case:
1704 */
1705 if (q->pi_state->owner != current)
1706 ret = fixup_pi_state_owner(uaddr, q, current);
1707 goto out;
1708 }
1709
1710 /*
1711 * Catch the rare case, where the lock was released when we were on the
1712 * way back before we locked the hash bucket.
1713 */
1714 if (q->pi_state->owner == current) {
1715 /*
1716 * Try to get the rt_mutex now. This might fail as some other
1717 * task acquired the rt_mutex after we removed ourself from the
1718 * rt_mutex waiters list.
1719 */
1720 if (rt_mutex_trylock(&q->pi_state->pi_mutex)) {
1721 locked = 1;
1722 goto out;
1723 }
1724
1725 /*
1726 * pi_state is incorrect, some other task did a lock steal and
1727 * we returned due to timeout or signal without taking the
1728 * rt_mutex. Too late.
1729 */
1730 raw_spin_lock(&q->pi_state->pi_mutex.wait_lock);
1731 owner = rt_mutex_owner(&q->pi_state->pi_mutex);
1732 if (!owner)
1733 owner = rt_mutex_next_owner(&q->pi_state->pi_mutex);
1734 raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock);
1735 ret = fixup_pi_state_owner(uaddr, q, owner);
1736 goto out;
1737 }
1738
1739 /*
1740 * Paranoia check. If we did not take the lock, then we should not be
1741 * the owner of the rt_mutex.
1742 */
1743 if (rt_mutex_owner(&q->pi_state->pi_mutex) == current)
1744 printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p "
1745 "pi-state %p\n", ret,
1746 q->pi_state->pi_mutex.owner,
1747 q->pi_state->owner);
1748
1749out:
1750 return ret ? ret : locked;
1751}
1752
1753/**
1754 * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
1755 * @hb: the futex hash bucket, must be locked by the caller
1756 * @q: the futex_q to queue up on
1757 * @timeout: the prepared hrtimer_sleeper, or null for no timeout
1758 */
1759static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
1760 struct hrtimer_sleeper *timeout)
1761{
1762 /*
1763 * The task state is guaranteed to be set before another task can
1764 * wake it. set_current_state() is implemented using set_mb() and
1765 * queue_me() calls spin_unlock() upon completion, both serializing
1766 * access to the hash list and forcing another memory barrier.
1767 */
1768 set_current_state(TASK_INTERRUPTIBLE);
1769 queue_me(q, hb);
1770
1771 /* Arm the timer */
1772 if (timeout) {
1773 hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
1774 if (!hrtimer_active(&timeout->timer))
1775 timeout->task = NULL;
1776 }
1777
1778 /*
1779 * If we have been removed from the hash list, then another task
1780 * has tried to wake us, and we can skip the call to schedule().
1781 */
1782 if (likely(!plist_node_empty(&q->list))) {
1783 /*
1784 * If the timer has already expired, current will already be
1785 * flagged for rescheduling. Only call schedule if there
1786 * is no timeout, or if it has yet to expire.
1787 */
1788 if (!timeout || timeout->task)
1789 schedule();
1790 }
1791 __set_current_state(TASK_RUNNING);
1792}
1793
1794/**
1795 * futex_wait_setup() - Prepare to wait on a futex
1796 * @uaddr: the futex userspace address
1797 * @val: the expected value
1798 * @flags: futex flags (FLAGS_SHARED, etc.)
1799 * @q: the associated futex_q
1800 * @hb: storage for hash_bucket pointer to be returned to caller
1801 *
1802 * Setup the futex_q and locate the hash_bucket. Get the futex value and
1803 * compare it with the expected value. Handle atomic faults internally.
1804 * Return with the hb lock held and a q.key reference on success, and unlocked
1805 * with no q.key reference on failure.
1806 *
1807 * Returns:
1808 * 0 - uaddr contains val and hb has been locked
1809 * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked
1810 */
1811static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
1812 struct futex_q *q, struct futex_hash_bucket **hb)
1813{
1814 u32 uval;
1815 int ret;
1816
1817 /*
1818 * Access the page AFTER the hash-bucket is locked.
1819 * Order is important:
1820 *
1821 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1822 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1823 *
1824 * The basic logical guarantee of a futex is that it blocks ONLY
1825 * if cond(var) is known to be true at the time of blocking, for
1826 * any cond. If we locked the hash-bucket after testing *uaddr, that
1827 * would open a race condition where we could block indefinitely with
1828 * cond(var) false, which would violate the guarantee.
1829 *
1830 * On the other hand, we insert q and release the hash-bucket only
1831 * after testing *uaddr. This guarantees that futex_wait() will NOT
1832 * absorb a wakeup if *uaddr does not match the desired values
1833 * while the syscall executes.
1834 */
1835retry:
1836 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ);
1837 if (unlikely(ret != 0))
1838 return ret;
1839
1840retry_private:
1841 *hb = queue_lock(q);
1842
1843 ret = get_futex_value_locked(&uval, uaddr);
1844
1845 if (ret) {
1846 queue_unlock(q, *hb);
1847
1848 ret = get_user(uval, uaddr);
1849 if (ret)
1850 goto out;
1851
1852 if (!(flags & FLAGS_SHARED))
1853 goto retry_private;
1854
1855 put_futex_key(&q->key);
1856 goto retry;
1857 }
1858
1859 if (uval != val) {
1860 queue_unlock(q, *hb);
1861 ret = -EWOULDBLOCK;
1862 }
1863
1864out:
1865 if (ret)
1866 put_futex_key(&q->key);
1867 return ret;
1868}
1869
1870static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val,
1871 ktime_t *abs_time, u32 bitset)
1872{
1873 struct hrtimer_sleeper timeout, *to = NULL;
1874 struct restart_block *restart;
1875 struct futex_hash_bucket *hb;
1876 struct futex_q q = futex_q_init;
1877 int ret;
1878
1879 if (!bitset)
1880 return -EINVAL;
1881 q.bitset = bitset;
1882
1883 if (abs_time) {
1884 to = &timeout;
1885
1886 hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
1887 CLOCK_REALTIME : CLOCK_MONOTONIC,
1888 HRTIMER_MODE_ABS);
1889 hrtimer_init_sleeper(to, current);
1890 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
1891 current->timer_slack_ns);
1892 }
1893
1894retry:
1895 /*
1896 * Prepare to wait on uaddr. On success, holds hb lock and increments
1897 * q.key refs.
1898 */
1899 ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
1900 if (ret)
1901 goto out;
1902
1903 /* queue_me and wait for wakeup, timeout, or a signal. */
1904 futex_wait_queue_me(hb, &q, to);
1905
1906 /* If we were woken (and unqueued), we succeeded, whatever. */
1907 ret = 0;
1908 /* unqueue_me() drops q.key ref */
1909 if (!unqueue_me(&q))
1910 goto out;
1911 ret = -ETIMEDOUT;
1912 if (to && !to->task)
1913 goto out;
1914
1915 /*
1916 * We expect signal_pending(current), but we might be the
1917 * victim of a spurious wakeup as well.
1918 */
1919 if (!signal_pending(current))
1920 goto retry;
1921
1922 ret = -ERESTARTSYS;
1923 if (!abs_time)
1924 goto out;
1925
1926 restart = ¤t_thread_info()->restart_block;
1927 restart->fn = futex_wait_restart;
1928 restart->futex.uaddr = uaddr;
1929 restart->futex.val = val;
1930 restart->futex.time = abs_time->tv64;
1931 restart->futex.bitset = bitset;
1932 restart->futex.flags = flags | FLAGS_HAS_TIMEOUT;
1933
1934 ret = -ERESTART_RESTARTBLOCK;
1935
1936out:
1937 if (to) {
1938 hrtimer_cancel(&to->timer);
1939 destroy_hrtimer_on_stack(&to->timer);
1940 }
1941 return ret;
1942}
1943
1944
1945static long futex_wait_restart(struct restart_block *restart)
1946{
1947 u32 __user *uaddr = restart->futex.uaddr;
1948 ktime_t t, *tp = NULL;
1949
1950 if (restart->futex.flags & FLAGS_HAS_TIMEOUT) {
1951 t.tv64 = restart->futex.time;
1952 tp = &t;
1953 }
1954 restart->fn = do_no_restart_syscall;
1955
1956 return (long)futex_wait(uaddr, restart->futex.flags,
1957 restart->futex.val, tp, restart->futex.bitset);
1958}
1959
1960
1961/*
1962 * Userspace tried a 0 -> TID atomic transition of the futex value
1963 * and failed. The kernel side here does the whole locking operation:
1964 * if there are waiters then it will block, it does PI, etc. (Due to
1965 * races the kernel might see a 0 value of the futex too.)
1966 */
1967static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect,
1968 ktime_t *time, int trylock)
1969{
1970 struct hrtimer_sleeper timeout, *to = NULL;
1971 struct futex_hash_bucket *hb;
1972 struct futex_q q = futex_q_init;
1973 int res, ret;
1974
1975 if (refill_pi_state_cache())
1976 return -ENOMEM;
1977
1978 if (time) {
1979 to = &timeout;
1980 hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME,
1981 HRTIMER_MODE_ABS);
1982 hrtimer_init_sleeper(to, current);
1983 hrtimer_set_expires(&to->timer, *time);
1984 }
1985
1986retry:
1987 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE);
1988 if (unlikely(ret != 0))
1989 goto out;
1990
1991retry_private:
1992 hb = queue_lock(&q);
1993
1994 ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0);
1995 if (unlikely(ret)) {
1996 switch (ret) {
1997 case 1:
1998 /* We got the lock. */
1999 ret = 0;
2000 goto out_unlock_put_key;
2001 case -EFAULT:
2002 goto uaddr_faulted;
2003 case -EAGAIN:
2004 /*
2005 * Task is exiting and we just wait for the
2006 * exit to complete.
2007 */
2008 queue_unlock(&q, hb);
2009 put_futex_key(&q.key);
2010 cond_resched();
2011 goto retry;
2012 default:
2013 goto out_unlock_put_key;
2014 }
2015 }
2016
2017 /*
2018 * Only actually queue now that the atomic ops are done:
2019 */
2020 queue_me(&q, hb);
2021
2022 WARN_ON(!q.pi_state);
2023 /*
2024 * Block on the PI mutex:
2025 */
2026 if (!trylock)
2027 ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1);
2028 else {
2029 ret = rt_mutex_trylock(&q.pi_state->pi_mutex);
2030 /* Fixup the trylock return value: */
2031 ret = ret ? 0 : -EWOULDBLOCK;
2032 }
2033
2034 spin_lock(q.lock_ptr);
2035 /*
2036 * Fixup the pi_state owner and possibly acquire the lock if we
2037 * haven't already.
2038 */
2039 res = fixup_owner(uaddr, &q, !ret);
2040 /*
2041 * If fixup_owner() returned an error, proprogate that. If it acquired
2042 * the lock, clear our -ETIMEDOUT or -EINTR.
2043 */
2044 if (res)
2045 ret = (res < 0) ? res : 0;
2046
2047 /*
2048 * If fixup_owner() faulted and was unable to handle the fault, unlock
2049 * it and return the fault to userspace.
2050 */
2051 if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current))
2052 rt_mutex_unlock(&q.pi_state->pi_mutex);
2053
2054 /* Unqueue and drop the lock */
2055 unqueue_me_pi(&q);
2056
2057 goto out_put_key;
2058
2059out_unlock_put_key:
2060 queue_unlock(&q, hb);
2061
2062out_put_key:
2063 put_futex_key(&q.key);
2064out:
2065 if (to)
2066 destroy_hrtimer_on_stack(&to->timer);
2067 return ret != -EINTR ? ret : -ERESTARTNOINTR;
2068
2069uaddr_faulted:
2070 queue_unlock(&q, hb);
2071
2072 ret = fault_in_user_writeable(uaddr);
2073 if (ret)
2074 goto out_put_key;
2075
2076 if (!(flags & FLAGS_SHARED))
2077 goto retry_private;
2078
2079 put_futex_key(&q.key);
2080 goto retry;
2081}
2082
2083/*
2084 * Userspace attempted a TID -> 0 atomic transition, and failed.
2085 * This is the in-kernel slowpath: we look up the PI state (if any),
2086 * and do the rt-mutex unlock.
2087 */
2088static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags)
2089{
2090 struct futex_hash_bucket *hb;
2091 struct futex_q *this, *next;
2092 struct plist_head *head;
2093 union futex_key key = FUTEX_KEY_INIT;
2094 u32 uval, vpid = task_pid_vnr(current);
2095 int ret;
2096
2097retry:
2098 if (get_user(uval, uaddr))
2099 return -EFAULT;
2100 /*
2101 * We release only a lock we actually own:
2102 */
2103 if ((uval & FUTEX_TID_MASK) != vpid)
2104 return -EPERM;
2105
2106 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE);
2107 if (unlikely(ret != 0))
2108 goto out;
2109
2110 hb = hash_futex(&key);
2111 spin_lock(&hb->lock);
2112
2113 /*
2114 * To avoid races, try to do the TID -> 0 atomic transition
2115 * again. If it succeeds then we can return without waking
2116 * anyone else up:
2117 */
2118 if (!(uval & FUTEX_OWNER_DIED) &&
2119 cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0))
2120 goto pi_faulted;
2121 /*
2122 * Rare case: we managed to release the lock atomically,
2123 * no need to wake anyone else up:
2124 */
2125 if (unlikely(uval == vpid))
2126 goto out_unlock;
2127
2128 /*
2129 * Ok, other tasks may need to be woken up - check waiters
2130 * and do the wakeup if necessary:
2131 */
2132 head = &hb->chain;
2133
2134 plist_for_each_entry_safe(this, next, head, list) {
2135 if (!match_futex (&this->key, &key))
2136 continue;
2137 ret = wake_futex_pi(uaddr, uval, this);
2138 /*
2139 * The atomic access to the futex value
2140 * generated a pagefault, so retry the
2141 * user-access and the wakeup:
2142 */
2143 if (ret == -EFAULT)
2144 goto pi_faulted;
2145 goto out_unlock;
2146 }
2147 /*
2148 * No waiters - kernel unlocks the futex:
2149 */
2150 if (!(uval & FUTEX_OWNER_DIED)) {
2151 ret = unlock_futex_pi(uaddr, uval);
2152 if (ret == -EFAULT)
2153 goto pi_faulted;
2154 }
2155
2156out_unlock:
2157 spin_unlock(&hb->lock);
2158 put_futex_key(&key);
2159
2160out:
2161 return ret;
2162
2163pi_faulted:
2164 spin_unlock(&hb->lock);
2165 put_futex_key(&key);
2166
2167 ret = fault_in_user_writeable(uaddr);
2168 if (!ret)
2169 goto retry;
2170
2171 return ret;
2172}
2173
2174/**
2175 * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2176 * @hb: the hash_bucket futex_q was original enqueued on
2177 * @q: the futex_q woken while waiting to be requeued
2178 * @key2: the futex_key of the requeue target futex
2179 * @timeout: the timeout associated with the wait (NULL if none)
2180 *
2181 * Detect if the task was woken on the initial futex as opposed to the requeue
2182 * target futex. If so, determine if it was a timeout or a signal that caused
2183 * the wakeup and return the appropriate error code to the caller. Must be
2184 * called with the hb lock held.
2185 *
2186 * Returns
2187 * 0 - no early wakeup detected
2188 * <0 - -ETIMEDOUT or -ERESTARTNOINTR
2189 */
2190static inline
2191int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb,
2192 struct futex_q *q, union futex_key *key2,
2193 struct hrtimer_sleeper *timeout)
2194{
2195 int ret = 0;
2196
2197 /*
2198 * With the hb lock held, we avoid races while we process the wakeup.
2199 * We only need to hold hb (and not hb2) to ensure atomicity as the
2200 * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2201 * It can't be requeued from uaddr2 to something else since we don't
2202 * support a PI aware source futex for requeue.
2203 */
2204 if (!match_futex(&q->key, key2)) {
2205 WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr));
2206 /*
2207 * We were woken prior to requeue by a timeout or a signal.
2208 * Unqueue the futex_q and determine which it was.
2209 */
2210 plist_del(&q->list, &hb->chain);
2211
2212 /* Handle spurious wakeups gracefully */
2213 ret = -EWOULDBLOCK;
2214 if (timeout && !timeout->task)
2215 ret = -ETIMEDOUT;
2216 else if (signal_pending(current))
2217 ret = -ERESTARTNOINTR;
2218 }
2219 return ret;
2220}
2221
2222/**
2223 * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
2224 * @uaddr: the futex we initially wait on (non-pi)
2225 * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be
2226 * the same type, no requeueing from private to shared, etc.
2227 * @val: the expected value of uaddr
2228 * @abs_time: absolute timeout
2229 * @bitset: 32 bit wakeup bitset set by userspace, defaults to all
2230 * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0)
2231 * @uaddr2: the pi futex we will take prior to returning to user-space
2232 *
2233 * The caller will wait on uaddr and will be requeued by futex_requeue() to
2234 * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake
2235 * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to
2236 * userspace. This ensures the rt_mutex maintains an owner when it has waiters;
2237 * without one, the pi logic would not know which task to boost/deboost, if
2238 * there was a need to.
2239 *
2240 * We call schedule in futex_wait_queue_me() when we enqueue and return there
2241 * via the following:
2242 * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
2243 * 2) wakeup on uaddr2 after a requeue
2244 * 3) signal
2245 * 4) timeout
2246 *
2247 * If 3, cleanup and return -ERESTARTNOINTR.
2248 *
2249 * If 2, we may then block on trying to take the rt_mutex and return via:
2250 * 5) successful lock
2251 * 6) signal
2252 * 7) timeout
2253 * 8) other lock acquisition failure
2254 *
2255 * If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
2256 *
2257 * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2258 *
2259 * Returns:
2260 * 0 - On success
2261 * <0 - On error
2262 */
2263static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
2264 u32 val, ktime_t *abs_time, u32 bitset,
2265 u32 __user *uaddr2)
2266{
2267 struct hrtimer_sleeper timeout, *to = NULL;
2268 struct rt_mutex_waiter rt_waiter;
2269 struct rt_mutex *pi_mutex = NULL;
2270 struct futex_hash_bucket *hb;
2271 union futex_key key2 = FUTEX_KEY_INIT;
2272 struct futex_q q = futex_q_init;
2273 int res, ret;
2274
2275 if (uaddr == uaddr2)
2276 return -EINVAL;
2277
2278 if (!bitset)
2279 return -EINVAL;
2280
2281 if (abs_time) {
2282 to = &timeout;
2283 hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
2284 CLOCK_REALTIME : CLOCK_MONOTONIC,
2285 HRTIMER_MODE_ABS);
2286 hrtimer_init_sleeper(to, current);
2287 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
2288 current->timer_slack_ns);
2289 }
2290
2291 /*
2292 * The waiter is allocated on our stack, manipulated by the requeue
2293 * code while we sleep on uaddr.
2294 */
2295 debug_rt_mutex_init_waiter(&rt_waiter);
2296 rt_waiter.task = NULL;
2297
2298 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
2299 if (unlikely(ret != 0))
2300 goto out;
2301
2302 q.bitset = bitset;
2303 q.rt_waiter = &rt_waiter;
2304 q.requeue_pi_key = &key2;
2305
2306 /*
2307 * Prepare to wait on uaddr. On success, increments q.key (key1) ref
2308 * count.
2309 */
2310 ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
2311 if (ret)
2312 goto out_key2;
2313
2314 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
2315 futex_wait_queue_me(hb, &q, to);
2316
2317 spin_lock(&hb->lock);
2318 ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
2319 spin_unlock(&hb->lock);
2320 if (ret)
2321 goto out_put_keys;
2322
2323 /*
2324 * In order for us to be here, we know our q.key == key2, and since
2325 * we took the hb->lock above, we also know that futex_requeue() has
2326 * completed and we no longer have to concern ourselves with a wakeup
2327 * race with the atomic proxy lock acquisition by the requeue code. The
2328 * futex_requeue dropped our key1 reference and incremented our key2
2329 * reference count.
2330 */
2331
2332 /* Check if the requeue code acquired the second futex for us. */
2333 if (!q.rt_waiter) {
2334 /*
2335 * Got the lock. We might not be the anticipated owner if we
2336 * did a lock-steal - fix up the PI-state in that case.
2337 */
2338 if (q.pi_state && (q.pi_state->owner != current)) {
2339 spin_lock(q.lock_ptr);
2340 ret = fixup_pi_state_owner(uaddr2, &q, current);
2341 spin_unlock(q.lock_ptr);
2342 }
2343 } else {
2344 /*
2345 * We have been woken up by futex_unlock_pi(), a timeout, or a
2346 * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2347 * the pi_state.
2348 */
2349 WARN_ON(!q.pi_state);
2350 pi_mutex = &q.pi_state->pi_mutex;
2351 ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1);
2352 debug_rt_mutex_free_waiter(&rt_waiter);
2353
2354 spin_lock(q.lock_ptr);
2355 /*
2356 * Fixup the pi_state owner and possibly acquire the lock if we
2357 * haven't already.
2358 */
2359 res = fixup_owner(uaddr2, &q, !ret);
2360 /*
2361 * If fixup_owner() returned an error, proprogate that. If it
2362 * acquired the lock, clear -ETIMEDOUT or -EINTR.
2363 */
2364 if (res)
2365 ret = (res < 0) ? res : 0;
2366
2367 /* Unqueue and drop the lock. */
2368 unqueue_me_pi(&q);
2369 }
2370
2371 /*
2372 * If fixup_pi_state_owner() faulted and was unable to handle the
2373 * fault, unlock the rt_mutex and return the fault to userspace.
2374 */
2375 if (ret == -EFAULT) {
2376 if (pi_mutex && rt_mutex_owner(pi_mutex) == current)
2377 rt_mutex_unlock(pi_mutex);
2378 } else if (ret == -EINTR) {
2379 /*
2380 * We've already been requeued, but cannot restart by calling
2381 * futex_lock_pi() directly. We could restart this syscall, but
2382 * it would detect that the user space "val" changed and return
2383 * -EWOULDBLOCK. Save the overhead of the restart and return
2384 * -EWOULDBLOCK directly.
2385 */
2386 ret = -EWOULDBLOCK;
2387 }
2388
2389out_put_keys:
2390 put_futex_key(&q.key);
2391out_key2:
2392 put_futex_key(&key2);
2393
2394out:
2395 if (to) {
2396 hrtimer_cancel(&to->timer);
2397 destroy_hrtimer_on_stack(&to->timer);
2398 }
2399 return ret;
2400}
2401
2402/*
2403 * Support for robust futexes: the kernel cleans up held futexes at
2404 * thread exit time.
2405 *
2406 * Implementation: user-space maintains a per-thread list of locks it
2407 * is holding. Upon do_exit(), the kernel carefully walks this list,
2408 * and marks all locks that are owned by this thread with the
2409 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
2410 * always manipulated with the lock held, so the list is private and
2411 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2412 * field, to allow the kernel to clean up if the thread dies after
2413 * acquiring the lock, but just before it could have added itself to
2414 * the list. There can only be one such pending lock.
2415 */
2416
2417/**
2418 * sys_set_robust_list() - Set the robust-futex list head of a task
2419 * @head: pointer to the list-head
2420 * @len: length of the list-head, as userspace expects
2421 */
2422SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head,
2423 size_t, len)
2424{
2425 if (!futex_cmpxchg_enabled)
2426 return -ENOSYS;
2427 /*
2428 * The kernel knows only one size for now:
2429 */
2430 if (unlikely(len != sizeof(*head)))
2431 return -EINVAL;
2432
2433 current->robust_list = head;
2434
2435 return 0;
2436}
2437
2438/**
2439 * sys_get_robust_list() - Get the robust-futex list head of a task
2440 * @pid: pid of the process [zero for current task]
2441 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2442 * @len_ptr: pointer to a length field, the kernel fills in the header size
2443 */
2444SYSCALL_DEFINE3(get_robust_list, int, pid,
2445 struct robust_list_head __user * __user *, head_ptr,
2446 size_t __user *, len_ptr)
2447{
2448 struct robust_list_head __user *head;
2449 unsigned long ret;
2450 struct task_struct *p;
2451
2452 if (!futex_cmpxchg_enabled)
2453 return -ENOSYS;
2454
2455 WARN_ONCE(1, "deprecated: get_robust_list will be deleted in 2013.\n");
2456
2457 rcu_read_lock();
2458
2459 ret = -ESRCH;
2460 if (!pid)
2461 p = current;
2462 else {
2463 p = find_task_by_vpid(pid);
2464 if (!p)
2465 goto err_unlock;
2466 }
2467
2468 ret = -EPERM;
2469 if (!ptrace_may_access(p, PTRACE_MODE_READ))
2470 goto err_unlock;
2471
2472 head = p->robust_list;
2473 rcu_read_unlock();
2474
2475 if (put_user(sizeof(*head), len_ptr))
2476 return -EFAULT;
2477 return put_user(head, head_ptr);
2478
2479err_unlock:
2480 rcu_read_unlock();
2481
2482 return ret;
2483}
2484
2485/*
2486 * Process a futex-list entry, check whether it's owned by the
2487 * dying task, and do notification if so:
2488 */
2489int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
2490{
2491 u32 uval, uninitialized_var(nval), mval;
2492
2493retry:
2494 if (get_user(uval, uaddr))
2495 return -1;
2496
2497 if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) {
2498 /*
2499 * Ok, this dying thread is truly holding a futex
2500 * of interest. Set the OWNER_DIED bit atomically
2501 * via cmpxchg, and if the value had FUTEX_WAITERS
2502 * set, wake up a waiter (if any). (We have to do a
2503 * futex_wake() even if OWNER_DIED is already set -
2504 * to handle the rare but possible case of recursive
2505 * thread-death.) The rest of the cleanup is done in
2506 * userspace.
2507 */
2508 mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
2509 /*
2510 * We are not holding a lock here, but we want to have
2511 * the pagefault_disable/enable() protection because
2512 * we want to handle the fault gracefully. If the
2513 * access fails we try to fault in the futex with R/W
2514 * verification via get_user_pages. get_user() above
2515 * does not guarantee R/W access. If that fails we
2516 * give up and leave the futex locked.
2517 */
2518 if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) {
2519 if (fault_in_user_writeable(uaddr))
2520 return -1;
2521 goto retry;
2522 }
2523 if (nval != uval)
2524 goto retry;
2525
2526 /*
2527 * Wake robust non-PI futexes here. The wakeup of
2528 * PI futexes happens in exit_pi_state():
2529 */
2530 if (!pi && (uval & FUTEX_WAITERS))
2531 futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
2532 }
2533 return 0;
2534}
2535
2536/*
2537 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2538 */
2539static inline int fetch_robust_entry(struct robust_list __user **entry,
2540 struct robust_list __user * __user *head,
2541 unsigned int *pi)
2542{
2543 unsigned long uentry;
2544
2545 if (get_user(uentry, (unsigned long __user *)head))
2546 return -EFAULT;
2547
2548 *entry = (void __user *)(uentry & ~1UL);
2549 *pi = uentry & 1;
2550
2551 return 0;
2552}
2553
2554/*
2555 * Walk curr->robust_list (very carefully, it's a userspace list!)
2556 * and mark any locks found there dead, and notify any waiters.
2557 *
2558 * We silently return on any sign of list-walking problem.
2559 */
2560void exit_robust_list(struct task_struct *curr)
2561{
2562 struct robust_list_head __user *head = curr->robust_list;
2563 struct robust_list __user *entry, *next_entry, *pending;
2564 unsigned int limit = ROBUST_LIST_LIMIT, pi, pip;
2565 unsigned int uninitialized_var(next_pi);
2566 unsigned long futex_offset;
2567 int rc;
2568
2569 if (!futex_cmpxchg_enabled)
2570 return;
2571
2572 /*
2573 * Fetch the list head (which was registered earlier, via
2574 * sys_set_robust_list()):
2575 */
2576 if (fetch_robust_entry(&entry, &head->list.next, &pi))
2577 return;
2578 /*
2579 * Fetch the relative futex offset:
2580 */
2581 if (get_user(futex_offset, &head->futex_offset))
2582 return;
2583 /*
2584 * Fetch any possibly pending lock-add first, and handle it
2585 * if it exists:
2586 */
2587 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
2588 return;
2589
2590 next_entry = NULL; /* avoid warning with gcc */
2591 while (entry != &head->list) {
2592 /*
2593 * Fetch the next entry in the list before calling
2594 * handle_futex_death:
2595 */
2596 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
2597 /*
2598 * A pending lock might already be on the list, so
2599 * don't process it twice:
2600 */
2601 if (entry != pending)
2602 if (handle_futex_death((void __user *)entry + futex_offset,
2603 curr, pi))
2604 return;
2605 if (rc)
2606 return;
2607 entry = next_entry;
2608 pi = next_pi;
2609 /*
2610 * Avoid excessively long or circular lists:
2611 */
2612 if (!--limit)
2613 break;
2614
2615 cond_resched();
2616 }
2617
2618 if (pending)
2619 handle_futex_death((void __user *)pending + futex_offset,
2620 curr, pip);
2621}
2622
2623long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
2624 u32 __user *uaddr2, u32 val2, u32 val3)
2625{
2626 int cmd = op & FUTEX_CMD_MASK;
2627 unsigned int flags = 0;
2628
2629 if (!(op & FUTEX_PRIVATE_FLAG))
2630 flags |= FLAGS_SHARED;
2631
2632 if (op & FUTEX_CLOCK_REALTIME) {
2633 flags |= FLAGS_CLOCKRT;
2634 if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
2635 return -ENOSYS;
2636 }
2637
2638 switch (cmd) {
2639 case FUTEX_LOCK_PI:
2640 case FUTEX_UNLOCK_PI:
2641 case FUTEX_TRYLOCK_PI:
2642 case FUTEX_WAIT_REQUEUE_PI:
2643 case FUTEX_CMP_REQUEUE_PI:
2644 if (!futex_cmpxchg_enabled)
2645 return -ENOSYS;
2646 }
2647
2648 switch (cmd) {
2649 case FUTEX_WAIT:
2650 val3 = FUTEX_BITSET_MATCH_ANY;
2651 case FUTEX_WAIT_BITSET:
2652 return futex_wait(uaddr, flags, val, timeout, val3);
2653 case FUTEX_WAKE:
2654 val3 = FUTEX_BITSET_MATCH_ANY;
2655 case FUTEX_WAKE_BITSET:
2656 return futex_wake(uaddr, flags, val, val3);
2657 case FUTEX_REQUEUE:
2658 return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0);
2659 case FUTEX_CMP_REQUEUE:
2660 return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0);
2661 case FUTEX_WAKE_OP:
2662 return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3);
2663 case FUTEX_LOCK_PI:
2664 return futex_lock_pi(uaddr, flags, val, timeout, 0);
2665 case FUTEX_UNLOCK_PI:
2666 return futex_unlock_pi(uaddr, flags);
2667 case FUTEX_TRYLOCK_PI:
2668 return futex_lock_pi(uaddr, flags, 0, timeout, 1);
2669 case FUTEX_WAIT_REQUEUE_PI:
2670 val3 = FUTEX_BITSET_MATCH_ANY;
2671 return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3,
2672 uaddr2);
2673 case FUTEX_CMP_REQUEUE_PI:
2674 return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1);
2675 }
2676 return -ENOSYS;
2677}
2678
2679
2680SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
2681 struct timespec __user *, utime, u32 __user *, uaddr2,
2682 u32, val3)
2683{
2684 struct timespec ts;
2685 ktime_t t, *tp = NULL;
2686 u32 val2 = 0;
2687 int cmd = op & FUTEX_CMD_MASK;
2688
2689 if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
2690 cmd == FUTEX_WAIT_BITSET ||
2691 cmd == FUTEX_WAIT_REQUEUE_PI)) {
2692 if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
2693 return -EFAULT;
2694 if (!timespec_valid(&ts))
2695 return -EINVAL;
2696
2697 t = timespec_to_ktime(ts);
2698 if (cmd == FUTEX_WAIT)
2699 t = ktime_add_safe(ktime_get(), t);
2700 tp = &t;
2701 }
2702 /*
2703 * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
2704 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
2705 */
2706 if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
2707 cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP)
2708 val2 = (u32) (unsigned long) utime;
2709
2710 return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
2711}
2712
2713static int __init futex_init(void)
2714{
2715 u32 curval;
2716 int i;
2717
2718 /*
2719 * This will fail and we want it. Some arch implementations do
2720 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2721 * functionality. We want to know that before we call in any
2722 * of the complex code paths. Also we want to prevent
2723 * registration of robust lists in that case. NULL is
2724 * guaranteed to fault and we get -EFAULT on functional
2725 * implementation, the non-functional ones will return
2726 * -ENOSYS.
2727 */
2728 if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT)
2729 futex_cmpxchg_enabled = 1;
2730
2731 for (i = 0; i < ARRAY_SIZE(futex_queues); i++) {
2732 plist_head_init(&futex_queues[i].chain);
2733 spin_lock_init(&futex_queues[i].lock);
2734 }
2735
2736 return 0;
2737}
2738__initcall(futex_init);