Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly | |
12 | * enough at me, Linus for the original (flawed) idea, Matthew | |
13 | * Kirkwood for proof-of-concept implementation. | |
14 | * | |
15 | * "The futexes are also cursed." | |
16 | * "But they come in a choice of three flavours!" | |
17 | * | |
18 | * This program is free software; you can redistribute it and/or modify | |
19 | * it under the terms of the GNU General Public License as published by | |
20 | * the Free Software Foundation; either version 2 of the License, or | |
21 | * (at your option) any later version. | |
22 | * | |
23 | * This program is distributed in the hope that it will be useful, | |
24 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
25 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
26 | * GNU General Public License for more details. | |
27 | * | |
28 | * You should have received a copy of the GNU General Public License | |
29 | * along with this program; if not, write to the Free Software | |
30 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
31 | */ | |
32 | #include <linux/slab.h> | |
33 | #include <linux/poll.h> | |
34 | #include <linux/fs.h> | |
35 | #include <linux/file.h> | |
36 | #include <linux/jhash.h> | |
37 | #include <linux/init.h> | |
38 | #include <linux/futex.h> | |
39 | #include <linux/mount.h> | |
40 | #include <linux/pagemap.h> | |
41 | #include <linux/syscalls.h> | |
7ed20e1a | 42 | #include <linux/signal.h> |
1da177e4 LT |
43 | |
44 | #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8) | |
45 | ||
46 | /* | |
47 | * Futexes are matched on equal values of this key. | |
48 | * The key type depends on whether it's a shared or private mapping. | |
49 | * Don't rearrange members without looking at hash_futex(). | |
50 | * | |
51 | * offset is aligned to a multiple of sizeof(u32) (== 4) by definition. | |
52 | * We set bit 0 to indicate if it's an inode-based key. | |
53 | */ | |
54 | union futex_key { | |
55 | struct { | |
56 | unsigned long pgoff; | |
57 | struct inode *inode; | |
58 | int offset; | |
59 | } shared; | |
60 | struct { | |
61 | unsigned long uaddr; | |
62 | struct mm_struct *mm; | |
63 | int offset; | |
64 | } private; | |
65 | struct { | |
66 | unsigned long word; | |
67 | void *ptr; | |
68 | int offset; | |
69 | } both; | |
70 | }; | |
71 | ||
72 | /* | |
73 | * We use this hashed waitqueue instead of a normal wait_queue_t, so | |
74 | * we can wake only the relevant ones (hashed queues may be shared). | |
75 | * | |
76 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
77 | * It is considered woken when list_empty(&q->list) || q->lock_ptr == 0. | |
78 | * The order of wakup is always to make the first condition true, then | |
79 | * wake up q->waiters, then make the second condition true. | |
80 | */ | |
81 | struct futex_q { | |
82 | struct list_head list; | |
83 | wait_queue_head_t waiters; | |
84 | ||
85 | /* Which hash list lock to use. */ | |
86 | spinlock_t *lock_ptr; | |
87 | ||
88 | /* Key which the futex is hashed on. */ | |
89 | union futex_key key; | |
90 | ||
91 | /* For fd, sigio sent using these. */ | |
92 | int fd; | |
93 | struct file *filp; | |
94 | }; | |
95 | ||
96 | /* | |
97 | * Split the global futex_lock into every hash list lock. | |
98 | */ | |
99 | struct futex_hash_bucket { | |
100 | spinlock_t lock; | |
101 | struct list_head chain; | |
102 | }; | |
103 | ||
104 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; | |
105 | ||
106 | /* Futex-fs vfsmount entry: */ | |
107 | static struct vfsmount *futex_mnt; | |
108 | ||
109 | /* | |
110 | * We hash on the keys returned from get_futex_key (see below). | |
111 | */ | |
112 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
113 | { | |
114 | u32 hash = jhash2((u32*)&key->both.word, | |
115 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
116 | key->both.offset); | |
117 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; | |
118 | } | |
119 | ||
120 | /* | |
121 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
122 | */ | |
123 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
124 | { | |
125 | return (key1->both.word == key2->both.word | |
126 | && key1->both.ptr == key2->both.ptr | |
127 | && key1->both.offset == key2->both.offset); | |
128 | } | |
129 | ||
130 | /* | |
131 | * Get parameters which are the keys for a futex. | |
132 | * | |
133 | * For shared mappings, it's (page->index, vma->vm_file->f_dentry->d_inode, | |
134 | * offset_within_page). For private mappings, it's (uaddr, current->mm). | |
135 | * We can usually work out the index without swapping in the page. | |
136 | * | |
137 | * Returns: 0, or negative error code. | |
138 | * The key words are stored in *key on success. | |
139 | * | |
140 | * Should be called with ¤t->mm->mmap_sem but NOT any spinlocks. | |
141 | */ | |
142 | static int get_futex_key(unsigned long uaddr, union futex_key *key) | |
143 | { | |
144 | struct mm_struct *mm = current->mm; | |
145 | struct vm_area_struct *vma; | |
146 | struct page *page; | |
147 | int err; | |
148 | ||
149 | /* | |
150 | * The futex address must be "naturally" aligned. | |
151 | */ | |
152 | key->both.offset = uaddr % PAGE_SIZE; | |
153 | if (unlikely((key->both.offset % sizeof(u32)) != 0)) | |
154 | return -EINVAL; | |
155 | uaddr -= key->both.offset; | |
156 | ||
157 | /* | |
158 | * The futex is hashed differently depending on whether | |
159 | * it's in a shared or private mapping. So check vma first. | |
160 | */ | |
161 | vma = find_extend_vma(mm, uaddr); | |
162 | if (unlikely(!vma)) | |
163 | return -EFAULT; | |
164 | ||
165 | /* | |
166 | * Permissions. | |
167 | */ | |
168 | if (unlikely((vma->vm_flags & (VM_IO|VM_READ)) != VM_READ)) | |
169 | return (vma->vm_flags & VM_IO) ? -EPERM : -EACCES; | |
170 | ||
171 | /* | |
172 | * Private mappings are handled in a simple way. | |
173 | * | |
174 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
175 | * it's a read-only handle, it's expected that futexes attach to | |
176 | * the object not the particular process. Therefore we use | |
177 | * VM_MAYSHARE here, not VM_SHARED which is restricted to shared | |
178 | * mappings of _writable_ handles. | |
179 | */ | |
180 | if (likely(!(vma->vm_flags & VM_MAYSHARE))) { | |
181 | key->private.mm = mm; | |
182 | key->private.uaddr = uaddr; | |
183 | return 0; | |
184 | } | |
185 | ||
186 | /* | |
187 | * Linear file mappings are also simple. | |
188 | */ | |
189 | key->shared.inode = vma->vm_file->f_dentry->d_inode; | |
190 | key->both.offset++; /* Bit 0 of offset indicates inode-based key. */ | |
191 | if (likely(!(vma->vm_flags & VM_NONLINEAR))) { | |
192 | key->shared.pgoff = (((uaddr - vma->vm_start) >> PAGE_SHIFT) | |
193 | + vma->vm_pgoff); | |
194 | return 0; | |
195 | } | |
196 | ||
197 | /* | |
198 | * We could walk the page table to read the non-linear | |
199 | * pte, and get the page index without fetching the page | |
200 | * from swap. But that's a lot of code to duplicate here | |
201 | * for a rare case, so we simply fetch the page. | |
202 | */ | |
203 | ||
204 | /* | |
205 | * Do a quick atomic lookup first - this is the fastpath. | |
206 | */ | |
207 | spin_lock(¤t->mm->page_table_lock); | |
208 | page = follow_page(mm, uaddr, 0); | |
209 | if (likely(page != NULL)) { | |
210 | key->shared.pgoff = | |
211 | page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
212 | spin_unlock(¤t->mm->page_table_lock); | |
213 | return 0; | |
214 | } | |
215 | spin_unlock(¤t->mm->page_table_lock); | |
216 | ||
217 | /* | |
218 | * Do it the general way. | |
219 | */ | |
220 | err = get_user_pages(current, mm, uaddr, 1, 0, 0, &page, NULL); | |
221 | if (err >= 0) { | |
222 | key->shared.pgoff = | |
223 | page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
224 | put_page(page); | |
225 | return 0; | |
226 | } | |
227 | return err; | |
228 | } | |
229 | ||
230 | /* | |
231 | * Take a reference to the resource addressed by a key. | |
232 | * Can be called while holding spinlocks. | |
233 | * | |
234 | * NOTE: mmap_sem MUST be held between get_futex_key() and calling this | |
235 | * function, if it is called at all. mmap_sem keeps key->shared.inode valid. | |
236 | */ | |
237 | static inline void get_key_refs(union futex_key *key) | |
238 | { | |
239 | if (key->both.ptr != 0) { | |
240 | if (key->both.offset & 1) | |
241 | atomic_inc(&key->shared.inode->i_count); | |
242 | else | |
243 | atomic_inc(&key->private.mm->mm_count); | |
244 | } | |
245 | } | |
246 | ||
247 | /* | |
248 | * Drop a reference to the resource addressed by a key. | |
249 | * The hash bucket spinlock must not be held. | |
250 | */ | |
251 | static void drop_key_refs(union futex_key *key) | |
252 | { | |
253 | if (key->both.ptr != 0) { | |
254 | if (key->both.offset & 1) | |
255 | iput(key->shared.inode); | |
256 | else | |
257 | mmdrop(key->private.mm); | |
258 | } | |
259 | } | |
260 | ||
261 | static inline int get_futex_value_locked(int *dest, int __user *from) | |
262 | { | |
263 | int ret; | |
264 | ||
265 | inc_preempt_count(); | |
266 | ret = __copy_from_user_inatomic(dest, from, sizeof(int)); | |
267 | dec_preempt_count(); | |
268 | ||
269 | return ret ? -EFAULT : 0; | |
270 | } | |
271 | ||
272 | /* | |
273 | * The hash bucket lock must be held when this is called. | |
274 | * Afterwards, the futex_q must not be accessed. | |
275 | */ | |
276 | static void wake_futex(struct futex_q *q) | |
277 | { | |
278 | list_del_init(&q->list); | |
279 | if (q->filp) | |
280 | send_sigio(&q->filp->f_owner, q->fd, POLL_IN); | |
281 | /* | |
282 | * The lock in wake_up_all() is a crucial memory barrier after the | |
283 | * list_del_init() and also before assigning to q->lock_ptr. | |
284 | */ | |
285 | wake_up_all(&q->waiters); | |
286 | /* | |
287 | * The waiting task can free the futex_q as soon as this is written, | |
288 | * without taking any locks. This must come last. | |
289 | */ | |
290 | q->lock_ptr = NULL; | |
291 | } | |
292 | ||
293 | /* | |
294 | * Wake up all waiters hashed on the physical page that is mapped | |
295 | * to this virtual address: | |
296 | */ | |
297 | static int futex_wake(unsigned long uaddr, int nr_wake) | |
298 | { | |
299 | union futex_key key; | |
300 | struct futex_hash_bucket *bh; | |
301 | struct list_head *head; | |
302 | struct futex_q *this, *next; | |
303 | int ret; | |
304 | ||
305 | down_read(¤t->mm->mmap_sem); | |
306 | ||
307 | ret = get_futex_key(uaddr, &key); | |
308 | if (unlikely(ret != 0)) | |
309 | goto out; | |
310 | ||
311 | bh = hash_futex(&key); | |
312 | spin_lock(&bh->lock); | |
313 | head = &bh->chain; | |
314 | ||
315 | list_for_each_entry_safe(this, next, head, list) { | |
316 | if (match_futex (&this->key, &key)) { | |
317 | wake_futex(this); | |
318 | if (++ret >= nr_wake) | |
319 | break; | |
320 | } | |
321 | } | |
322 | ||
323 | spin_unlock(&bh->lock); | |
324 | out: | |
325 | up_read(¤t->mm->mmap_sem); | |
326 | return ret; | |
327 | } | |
328 | ||
329 | /* | |
330 | * Requeue all waiters hashed on one physical page to another | |
331 | * physical page. | |
332 | */ | |
333 | static int futex_requeue(unsigned long uaddr1, unsigned long uaddr2, | |
334 | int nr_wake, int nr_requeue, int *valp) | |
335 | { | |
336 | union futex_key key1, key2; | |
337 | struct futex_hash_bucket *bh1, *bh2; | |
338 | struct list_head *head1; | |
339 | struct futex_q *this, *next; | |
340 | int ret, drop_count = 0; | |
341 | ||
342 | retry: | |
343 | down_read(¤t->mm->mmap_sem); | |
344 | ||
345 | ret = get_futex_key(uaddr1, &key1); | |
346 | if (unlikely(ret != 0)) | |
347 | goto out; | |
348 | ret = get_futex_key(uaddr2, &key2); | |
349 | if (unlikely(ret != 0)) | |
350 | goto out; | |
351 | ||
352 | bh1 = hash_futex(&key1); | |
353 | bh2 = hash_futex(&key2); | |
354 | ||
355 | if (bh1 < bh2) | |
356 | spin_lock(&bh1->lock); | |
357 | spin_lock(&bh2->lock); | |
358 | if (bh1 > bh2) | |
359 | spin_lock(&bh1->lock); | |
360 | ||
361 | if (likely(valp != NULL)) { | |
362 | int curval; | |
363 | ||
364 | ret = get_futex_value_locked(&curval, (int __user *)uaddr1); | |
365 | ||
366 | if (unlikely(ret)) { | |
367 | spin_unlock(&bh1->lock); | |
368 | if (bh1 != bh2) | |
369 | spin_unlock(&bh2->lock); | |
370 | ||
371 | /* If we would have faulted, release mmap_sem, fault | |
372 | * it in and start all over again. | |
373 | */ | |
374 | up_read(¤t->mm->mmap_sem); | |
375 | ||
376 | ret = get_user(curval, (int __user *)uaddr1); | |
377 | ||
378 | if (!ret) | |
379 | goto retry; | |
380 | ||
381 | return ret; | |
382 | } | |
383 | if (curval != *valp) { | |
384 | ret = -EAGAIN; | |
385 | goto out_unlock; | |
386 | } | |
387 | } | |
388 | ||
389 | head1 = &bh1->chain; | |
390 | list_for_each_entry_safe(this, next, head1, list) { | |
391 | if (!match_futex (&this->key, &key1)) | |
392 | continue; | |
393 | if (++ret <= nr_wake) { | |
394 | wake_futex(this); | |
395 | } else { | |
396 | list_move_tail(&this->list, &bh2->chain); | |
397 | this->lock_ptr = &bh2->lock; | |
398 | this->key = key2; | |
399 | get_key_refs(&key2); | |
400 | drop_count++; | |
401 | ||
402 | if (ret - nr_wake >= nr_requeue) | |
403 | break; | |
404 | /* Make sure to stop if key1 == key2 */ | |
405 | if (head1 == &bh2->chain && head1 != &next->list) | |
406 | head1 = &this->list; | |
407 | } | |
408 | } | |
409 | ||
410 | out_unlock: | |
411 | spin_unlock(&bh1->lock); | |
412 | if (bh1 != bh2) | |
413 | spin_unlock(&bh2->lock); | |
414 | ||
415 | /* drop_key_refs() must be called outside the spinlocks. */ | |
416 | while (--drop_count >= 0) | |
417 | drop_key_refs(&key1); | |
418 | ||
419 | out: | |
420 | up_read(¤t->mm->mmap_sem); | |
421 | return ret; | |
422 | } | |
423 | ||
424 | /* The key must be already stored in q->key. */ | |
425 | static inline struct futex_hash_bucket * | |
426 | queue_lock(struct futex_q *q, int fd, struct file *filp) | |
427 | { | |
428 | struct futex_hash_bucket *bh; | |
429 | ||
430 | q->fd = fd; | |
431 | q->filp = filp; | |
432 | ||
433 | init_waitqueue_head(&q->waiters); | |
434 | ||
435 | get_key_refs(&q->key); | |
436 | bh = hash_futex(&q->key); | |
437 | q->lock_ptr = &bh->lock; | |
438 | ||
439 | spin_lock(&bh->lock); | |
440 | return bh; | |
441 | } | |
442 | ||
443 | static inline void __queue_me(struct futex_q *q, struct futex_hash_bucket *bh) | |
444 | { | |
445 | list_add_tail(&q->list, &bh->chain); | |
446 | spin_unlock(&bh->lock); | |
447 | } | |
448 | ||
449 | static inline void | |
450 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *bh) | |
451 | { | |
452 | spin_unlock(&bh->lock); | |
453 | drop_key_refs(&q->key); | |
454 | } | |
455 | ||
456 | /* | |
457 | * queue_me and unqueue_me must be called as a pair, each | |
458 | * exactly once. They are called with the hashed spinlock held. | |
459 | */ | |
460 | ||
461 | /* The key must be already stored in q->key. */ | |
462 | static void queue_me(struct futex_q *q, int fd, struct file *filp) | |
463 | { | |
464 | struct futex_hash_bucket *bh; | |
465 | bh = queue_lock(q, fd, filp); | |
466 | __queue_me(q, bh); | |
467 | } | |
468 | ||
469 | /* Return 1 if we were still queued (ie. 0 means we were woken) */ | |
470 | static int unqueue_me(struct futex_q *q) | |
471 | { | |
472 | int ret = 0; | |
473 | spinlock_t *lock_ptr; | |
474 | ||
475 | /* In the common case we don't take the spinlock, which is nice. */ | |
476 | retry: | |
477 | lock_ptr = q->lock_ptr; | |
478 | if (lock_ptr != 0) { | |
479 | spin_lock(lock_ptr); | |
480 | /* | |
481 | * q->lock_ptr can change between reading it and | |
482 | * spin_lock(), causing us to take the wrong lock. This | |
483 | * corrects the race condition. | |
484 | * | |
485 | * Reasoning goes like this: if we have the wrong lock, | |
486 | * q->lock_ptr must have changed (maybe several times) | |
487 | * between reading it and the spin_lock(). It can | |
488 | * change again after the spin_lock() but only if it was | |
489 | * already changed before the spin_lock(). It cannot, | |
490 | * however, change back to the original value. Therefore | |
491 | * we can detect whether we acquired the correct lock. | |
492 | */ | |
493 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
494 | spin_unlock(lock_ptr); | |
495 | goto retry; | |
496 | } | |
497 | WARN_ON(list_empty(&q->list)); | |
498 | list_del(&q->list); | |
499 | spin_unlock(lock_ptr); | |
500 | ret = 1; | |
501 | } | |
502 | ||
503 | drop_key_refs(&q->key); | |
504 | return ret; | |
505 | } | |
506 | ||
507 | static int futex_wait(unsigned long uaddr, int val, unsigned long time) | |
508 | { | |
509 | DECLARE_WAITQUEUE(wait, current); | |
510 | int ret, curval; | |
511 | struct futex_q q; | |
512 | struct futex_hash_bucket *bh; | |
513 | ||
514 | retry: | |
515 | down_read(¤t->mm->mmap_sem); | |
516 | ||
517 | ret = get_futex_key(uaddr, &q.key); | |
518 | if (unlikely(ret != 0)) | |
519 | goto out_release_sem; | |
520 | ||
521 | bh = queue_lock(&q, -1, NULL); | |
522 | ||
523 | /* | |
524 | * Access the page AFTER the futex is queued. | |
525 | * Order is important: | |
526 | * | |
527 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
528 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
529 | * | |
530 | * The basic logical guarantee of a futex is that it blocks ONLY | |
531 | * if cond(var) is known to be true at the time of blocking, for | |
532 | * any cond. If we queued after testing *uaddr, that would open | |
533 | * a race condition where we could block indefinitely with | |
534 | * cond(var) false, which would violate the guarantee. | |
535 | * | |
536 | * A consequence is that futex_wait() can return zero and absorb | |
537 | * a wakeup when *uaddr != val on entry to the syscall. This is | |
538 | * rare, but normal. | |
539 | * | |
540 | * We hold the mmap semaphore, so the mapping cannot have changed | |
541 | * since we looked it up in get_futex_key. | |
542 | */ | |
543 | ||
544 | ret = get_futex_value_locked(&curval, (int __user *)uaddr); | |
545 | ||
546 | if (unlikely(ret)) { | |
547 | queue_unlock(&q, bh); | |
548 | ||
549 | /* If we would have faulted, release mmap_sem, fault it in and | |
550 | * start all over again. | |
551 | */ | |
552 | up_read(¤t->mm->mmap_sem); | |
553 | ||
554 | ret = get_user(curval, (int __user *)uaddr); | |
555 | ||
556 | if (!ret) | |
557 | goto retry; | |
558 | return ret; | |
559 | } | |
560 | if (curval != val) { | |
561 | ret = -EWOULDBLOCK; | |
562 | queue_unlock(&q, bh); | |
563 | goto out_release_sem; | |
564 | } | |
565 | ||
566 | /* Only actually queue if *uaddr contained val. */ | |
567 | __queue_me(&q, bh); | |
568 | ||
569 | /* | |
570 | * Now the futex is queued and we have checked the data, we | |
571 | * don't want to hold mmap_sem while we sleep. | |
572 | */ | |
573 | up_read(¤t->mm->mmap_sem); | |
574 | ||
575 | /* | |
576 | * There might have been scheduling since the queue_me(), as we | |
577 | * cannot hold a spinlock across the get_user() in case it | |
578 | * faults, and we cannot just set TASK_INTERRUPTIBLE state when | |
579 | * queueing ourselves into the futex hash. This code thus has to | |
580 | * rely on the futex_wake() code removing us from hash when it | |
581 | * wakes us up. | |
582 | */ | |
583 | ||
584 | /* add_wait_queue is the barrier after __set_current_state. */ | |
585 | __set_current_state(TASK_INTERRUPTIBLE); | |
586 | add_wait_queue(&q.waiters, &wait); | |
587 | /* | |
588 | * !list_empty() is safe here without any lock. | |
589 | * q.lock_ptr != 0 is not safe, because of ordering against wakeup. | |
590 | */ | |
591 | if (likely(!list_empty(&q.list))) | |
592 | time = schedule_timeout(time); | |
593 | __set_current_state(TASK_RUNNING); | |
594 | ||
595 | /* | |
596 | * NOTE: we don't remove ourselves from the waitqueue because | |
597 | * we are the only user of it. | |
598 | */ | |
599 | ||
600 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
601 | if (!unqueue_me(&q)) | |
602 | return 0; | |
603 | if (time == 0) | |
604 | return -ETIMEDOUT; | |
605 | /* We expect signal_pending(current), but another thread may | |
606 | * have handled it for us already. */ | |
607 | return -EINTR; | |
608 | ||
609 | out_release_sem: | |
610 | up_read(¤t->mm->mmap_sem); | |
611 | return ret; | |
612 | } | |
613 | ||
614 | static int futex_close(struct inode *inode, struct file *filp) | |
615 | { | |
616 | struct futex_q *q = filp->private_data; | |
617 | ||
618 | unqueue_me(q); | |
619 | kfree(q); | |
620 | return 0; | |
621 | } | |
622 | ||
623 | /* This is one-shot: once it's gone off you need a new fd */ | |
624 | static unsigned int futex_poll(struct file *filp, | |
625 | struct poll_table_struct *wait) | |
626 | { | |
627 | struct futex_q *q = filp->private_data; | |
628 | int ret = 0; | |
629 | ||
630 | poll_wait(filp, &q->waiters, wait); | |
631 | ||
632 | /* | |
633 | * list_empty() is safe here without any lock. | |
634 | * q->lock_ptr != 0 is not safe, because of ordering against wakeup. | |
635 | */ | |
636 | if (list_empty(&q->list)) | |
637 | ret = POLLIN | POLLRDNORM; | |
638 | ||
639 | return ret; | |
640 | } | |
641 | ||
642 | static struct file_operations futex_fops = { | |
643 | .release = futex_close, | |
644 | .poll = futex_poll, | |
645 | }; | |
646 | ||
647 | /* | |
648 | * Signal allows caller to avoid the race which would occur if they | |
649 | * set the sigio stuff up afterwards. | |
650 | */ | |
651 | static int futex_fd(unsigned long uaddr, int signal) | |
652 | { | |
653 | struct futex_q *q; | |
654 | struct file *filp; | |
655 | int ret, err; | |
656 | ||
657 | ret = -EINVAL; | |
7ed20e1a | 658 | if (!valid_signal(signal)) |
1da177e4 LT |
659 | goto out; |
660 | ||
661 | ret = get_unused_fd(); | |
662 | if (ret < 0) | |
663 | goto out; | |
664 | filp = get_empty_filp(); | |
665 | if (!filp) { | |
666 | put_unused_fd(ret); | |
667 | ret = -ENFILE; | |
668 | goto out; | |
669 | } | |
670 | filp->f_op = &futex_fops; | |
671 | filp->f_vfsmnt = mntget(futex_mnt); | |
672 | filp->f_dentry = dget(futex_mnt->mnt_root); | |
673 | filp->f_mapping = filp->f_dentry->d_inode->i_mapping; | |
674 | ||
675 | if (signal) { | |
676 | int err; | |
677 | err = f_setown(filp, current->pid, 1); | |
678 | if (err < 0) { | |
679 | put_unused_fd(ret); | |
680 | put_filp(filp); | |
681 | ret = err; | |
682 | goto out; | |
683 | } | |
684 | filp->f_owner.signum = signal; | |
685 | } | |
686 | ||
687 | q = kmalloc(sizeof(*q), GFP_KERNEL); | |
688 | if (!q) { | |
689 | put_unused_fd(ret); | |
690 | put_filp(filp); | |
691 | ret = -ENOMEM; | |
692 | goto out; | |
693 | } | |
694 | ||
695 | down_read(¤t->mm->mmap_sem); | |
696 | err = get_futex_key(uaddr, &q->key); | |
697 | ||
698 | if (unlikely(err != 0)) { | |
699 | up_read(¤t->mm->mmap_sem); | |
700 | put_unused_fd(ret); | |
701 | put_filp(filp); | |
702 | kfree(q); | |
703 | return err; | |
704 | } | |
705 | ||
706 | /* | |
707 | * queue_me() must be called before releasing mmap_sem, because | |
708 | * key->shared.inode needs to be referenced while holding it. | |
709 | */ | |
710 | filp->private_data = q; | |
711 | ||
712 | queue_me(q, ret, filp); | |
713 | up_read(¤t->mm->mmap_sem); | |
714 | ||
715 | /* Now we map fd to filp, so userspace can access it */ | |
716 | fd_install(ret, filp); | |
717 | out: | |
718 | return ret; | |
719 | } | |
720 | ||
721 | long do_futex(unsigned long uaddr, int op, int val, unsigned long timeout, | |
722 | unsigned long uaddr2, int val2, int val3) | |
723 | { | |
724 | int ret; | |
725 | ||
726 | switch (op) { | |
727 | case FUTEX_WAIT: | |
728 | ret = futex_wait(uaddr, val, timeout); | |
729 | break; | |
730 | case FUTEX_WAKE: | |
731 | ret = futex_wake(uaddr, val); | |
732 | break; | |
733 | case FUTEX_FD: | |
734 | /* non-zero val means F_SETOWN(getpid()) & F_SETSIG(val) */ | |
735 | ret = futex_fd(uaddr, val); | |
736 | break; | |
737 | case FUTEX_REQUEUE: | |
738 | ret = futex_requeue(uaddr, uaddr2, val, val2, NULL); | |
739 | break; | |
740 | case FUTEX_CMP_REQUEUE: | |
741 | ret = futex_requeue(uaddr, uaddr2, val, val2, &val3); | |
742 | break; | |
743 | default: | |
744 | ret = -ENOSYS; | |
745 | } | |
746 | return ret; | |
747 | } | |
748 | ||
749 | ||
750 | asmlinkage long sys_futex(u32 __user *uaddr, int op, int val, | |
751 | struct timespec __user *utime, u32 __user *uaddr2, | |
752 | int val3) | |
753 | { | |
754 | struct timespec t; | |
755 | unsigned long timeout = MAX_SCHEDULE_TIMEOUT; | |
756 | int val2 = 0; | |
757 | ||
758 | if ((op == FUTEX_WAIT) && utime) { | |
759 | if (copy_from_user(&t, utime, sizeof(t)) != 0) | |
760 | return -EFAULT; | |
761 | timeout = timespec_to_jiffies(&t) + 1; | |
762 | } | |
763 | /* | |
764 | * requeue parameter in 'utime' if op == FUTEX_REQUEUE. | |
765 | */ | |
766 | if (op >= FUTEX_REQUEUE) | |
767 | val2 = (int) (unsigned long) utime; | |
768 | ||
769 | return do_futex((unsigned long)uaddr, op, val, timeout, | |
770 | (unsigned long)uaddr2, val2, val3); | |
771 | } | |
772 | ||
773 | static struct super_block * | |
774 | futexfs_get_sb(struct file_system_type *fs_type, | |
775 | int flags, const char *dev_name, void *data) | |
776 | { | |
777 | return get_sb_pseudo(fs_type, "futex", NULL, 0xBAD1DEA); | |
778 | } | |
779 | ||
780 | static struct file_system_type futex_fs_type = { | |
781 | .name = "futexfs", | |
782 | .get_sb = futexfs_get_sb, | |
783 | .kill_sb = kill_anon_super, | |
784 | }; | |
785 | ||
786 | static int __init init(void) | |
787 | { | |
788 | unsigned int i; | |
789 | ||
790 | register_filesystem(&futex_fs_type); | |
791 | futex_mnt = kern_mount(&futex_fs_type); | |
792 | ||
793 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { | |
794 | INIT_LIST_HEAD(&futex_queues[i].chain); | |
795 | spin_lock_init(&futex_queues[i].lock); | |
796 | } | |
797 | return 0; | |
798 | } | |
799 | __initcall(init); |