]>
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> |
4732efbe | 43 | #include <asm/futex.h> |
1da177e4 LT |
44 | |
45 | #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8) | |
46 | ||
47 | /* | |
48 | * Futexes are matched on equal values of this key. | |
49 | * The key type depends on whether it's a shared or private mapping. | |
50 | * Don't rearrange members without looking at hash_futex(). | |
51 | * | |
52 | * offset is aligned to a multiple of sizeof(u32) (== 4) by definition. | |
53 | * We set bit 0 to indicate if it's an inode-based key. | |
54 | */ | |
55 | union futex_key { | |
56 | struct { | |
57 | unsigned long pgoff; | |
58 | struct inode *inode; | |
59 | int offset; | |
60 | } shared; | |
61 | struct { | |
62 | unsigned long uaddr; | |
63 | struct mm_struct *mm; | |
64 | int offset; | |
65 | } private; | |
66 | struct { | |
67 | unsigned long word; | |
68 | void *ptr; | |
69 | int offset; | |
70 | } both; | |
71 | }; | |
72 | ||
73 | /* | |
74 | * We use this hashed waitqueue instead of a normal wait_queue_t, so | |
75 | * we can wake only the relevant ones (hashed queues may be shared). | |
76 | * | |
77 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
78 | * It is considered woken when list_empty(&q->list) || q->lock_ptr == 0. | |
79 | * The order of wakup is always to make the first condition true, then | |
80 | * wake up q->waiters, then make the second condition true. | |
81 | */ | |
82 | struct futex_q { | |
83 | struct list_head list; | |
84 | wait_queue_head_t waiters; | |
85 | ||
86 | /* Which hash list lock to use. */ | |
87 | spinlock_t *lock_ptr; | |
88 | ||
89 | /* Key which the futex is hashed on. */ | |
90 | union futex_key key; | |
91 | ||
92 | /* For fd, sigio sent using these. */ | |
93 | int fd; | |
94 | struct file *filp; | |
95 | }; | |
96 | ||
97 | /* | |
98 | * Split the global futex_lock into every hash list lock. | |
99 | */ | |
100 | struct futex_hash_bucket { | |
101 | spinlock_t lock; | |
102 | struct list_head chain; | |
103 | }; | |
104 | ||
105 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; | |
106 | ||
107 | /* Futex-fs vfsmount entry: */ | |
108 | static struct vfsmount *futex_mnt; | |
109 | ||
110 | /* | |
111 | * We hash on the keys returned from get_futex_key (see below). | |
112 | */ | |
113 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
114 | { | |
115 | u32 hash = jhash2((u32*)&key->both.word, | |
116 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
117 | key->both.offset); | |
118 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; | |
119 | } | |
120 | ||
121 | /* | |
122 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
123 | */ | |
124 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
125 | { | |
126 | return (key1->both.word == key2->both.word | |
127 | && key1->both.ptr == key2->both.ptr | |
128 | && key1->both.offset == key2->both.offset); | |
129 | } | |
130 | ||
131 | /* | |
132 | * Get parameters which are the keys for a futex. | |
133 | * | |
134 | * For shared mappings, it's (page->index, vma->vm_file->f_dentry->d_inode, | |
135 | * offset_within_page). For private mappings, it's (uaddr, current->mm). | |
136 | * We can usually work out the index without swapping in the page. | |
137 | * | |
138 | * Returns: 0, or negative error code. | |
139 | * The key words are stored in *key on success. | |
140 | * | |
141 | * Should be called with ¤t->mm->mmap_sem but NOT any spinlocks. | |
142 | */ | |
143 | static int get_futex_key(unsigned long uaddr, union futex_key *key) | |
144 | { | |
145 | struct mm_struct *mm = current->mm; | |
146 | struct vm_area_struct *vma; | |
147 | struct page *page; | |
148 | int err; | |
149 | ||
150 | /* | |
151 | * The futex address must be "naturally" aligned. | |
152 | */ | |
153 | key->both.offset = uaddr % PAGE_SIZE; | |
154 | if (unlikely((key->both.offset % sizeof(u32)) != 0)) | |
155 | return -EINVAL; | |
156 | uaddr -= key->both.offset; | |
157 | ||
158 | /* | |
159 | * The futex is hashed differently depending on whether | |
160 | * it's in a shared or private mapping. So check vma first. | |
161 | */ | |
162 | vma = find_extend_vma(mm, uaddr); | |
163 | if (unlikely(!vma)) | |
164 | return -EFAULT; | |
165 | ||
166 | /* | |
167 | * Permissions. | |
168 | */ | |
169 | if (unlikely((vma->vm_flags & (VM_IO|VM_READ)) != VM_READ)) | |
170 | return (vma->vm_flags & VM_IO) ? -EPERM : -EACCES; | |
171 | ||
172 | /* | |
173 | * Private mappings are handled in a simple way. | |
174 | * | |
175 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
176 | * it's a read-only handle, it's expected that futexes attach to | |
177 | * the object not the particular process. Therefore we use | |
178 | * VM_MAYSHARE here, not VM_SHARED which is restricted to shared | |
179 | * mappings of _writable_ handles. | |
180 | */ | |
181 | if (likely(!(vma->vm_flags & VM_MAYSHARE))) { | |
182 | key->private.mm = mm; | |
183 | key->private.uaddr = uaddr; | |
184 | return 0; | |
185 | } | |
186 | ||
187 | /* | |
188 | * Linear file mappings are also simple. | |
189 | */ | |
190 | key->shared.inode = vma->vm_file->f_dentry->d_inode; | |
191 | key->both.offset++; /* Bit 0 of offset indicates inode-based key. */ | |
192 | if (likely(!(vma->vm_flags & VM_NONLINEAR))) { | |
193 | key->shared.pgoff = (((uaddr - vma->vm_start) >> PAGE_SHIFT) | |
194 | + vma->vm_pgoff); | |
195 | return 0; | |
196 | } | |
197 | ||
198 | /* | |
199 | * We could walk the page table to read the non-linear | |
200 | * pte, and get the page index without fetching the page | |
201 | * from swap. But that's a lot of code to duplicate here | |
202 | * for a rare case, so we simply fetch the page. | |
203 | */ | |
1da177e4 LT |
204 | err = get_user_pages(current, mm, uaddr, 1, 0, 0, &page, NULL); |
205 | if (err >= 0) { | |
206 | key->shared.pgoff = | |
207 | page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
208 | put_page(page); | |
209 | return 0; | |
210 | } | |
211 | return err; | |
212 | } | |
213 | ||
214 | /* | |
215 | * Take a reference to the resource addressed by a key. | |
216 | * Can be called while holding spinlocks. | |
217 | * | |
218 | * NOTE: mmap_sem MUST be held between get_futex_key() and calling this | |
219 | * function, if it is called at all. mmap_sem keeps key->shared.inode valid. | |
220 | */ | |
221 | static inline void get_key_refs(union futex_key *key) | |
222 | { | |
223 | if (key->both.ptr != 0) { | |
224 | if (key->both.offset & 1) | |
225 | atomic_inc(&key->shared.inode->i_count); | |
226 | else | |
227 | atomic_inc(&key->private.mm->mm_count); | |
228 | } | |
229 | } | |
230 | ||
231 | /* | |
232 | * Drop a reference to the resource addressed by a key. | |
233 | * The hash bucket spinlock must not be held. | |
234 | */ | |
235 | static void drop_key_refs(union futex_key *key) | |
236 | { | |
237 | if (key->both.ptr != 0) { | |
238 | if (key->both.offset & 1) | |
239 | iput(key->shared.inode); | |
240 | else | |
241 | mmdrop(key->private.mm); | |
242 | } | |
243 | } | |
244 | ||
245 | static inline int get_futex_value_locked(int *dest, int __user *from) | |
246 | { | |
247 | int ret; | |
248 | ||
249 | inc_preempt_count(); | |
250 | ret = __copy_from_user_inatomic(dest, from, sizeof(int)); | |
251 | dec_preempt_count(); | |
252 | ||
253 | return ret ? -EFAULT : 0; | |
254 | } | |
255 | ||
256 | /* | |
257 | * The hash bucket lock must be held when this is called. | |
258 | * Afterwards, the futex_q must not be accessed. | |
259 | */ | |
260 | static void wake_futex(struct futex_q *q) | |
261 | { | |
262 | list_del_init(&q->list); | |
263 | if (q->filp) | |
264 | send_sigio(&q->filp->f_owner, q->fd, POLL_IN); | |
265 | /* | |
266 | * The lock in wake_up_all() is a crucial memory barrier after the | |
267 | * list_del_init() and also before assigning to q->lock_ptr. | |
268 | */ | |
269 | wake_up_all(&q->waiters); | |
270 | /* | |
271 | * The waiting task can free the futex_q as soon as this is written, | |
272 | * without taking any locks. This must come last. | |
8e31108b AM |
273 | * |
274 | * A memory barrier is required here to prevent the following store | |
275 | * to lock_ptr from getting ahead of the wakeup. Clearing the lock | |
276 | * at the end of wake_up_all() does not prevent this store from | |
277 | * moving. | |
1da177e4 | 278 | */ |
8e31108b | 279 | wmb(); |
1da177e4 LT |
280 | q->lock_ptr = NULL; |
281 | } | |
282 | ||
283 | /* | |
284 | * Wake up all waiters hashed on the physical page that is mapped | |
285 | * to this virtual address: | |
286 | */ | |
287 | static int futex_wake(unsigned long uaddr, int nr_wake) | |
288 | { | |
289 | union futex_key key; | |
290 | struct futex_hash_bucket *bh; | |
291 | struct list_head *head; | |
292 | struct futex_q *this, *next; | |
293 | int ret; | |
294 | ||
295 | down_read(¤t->mm->mmap_sem); | |
296 | ||
297 | ret = get_futex_key(uaddr, &key); | |
298 | if (unlikely(ret != 0)) | |
299 | goto out; | |
300 | ||
301 | bh = hash_futex(&key); | |
302 | spin_lock(&bh->lock); | |
303 | head = &bh->chain; | |
304 | ||
305 | list_for_each_entry_safe(this, next, head, list) { | |
306 | if (match_futex (&this->key, &key)) { | |
307 | wake_futex(this); | |
308 | if (++ret >= nr_wake) | |
309 | break; | |
310 | } | |
311 | } | |
312 | ||
313 | spin_unlock(&bh->lock); | |
314 | out: | |
315 | up_read(¤t->mm->mmap_sem); | |
316 | return ret; | |
317 | } | |
318 | ||
4732efbe JJ |
319 | /* |
320 | * Wake up all waiters hashed on the physical page that is mapped | |
321 | * to this virtual address: | |
322 | */ | |
323 | static int futex_wake_op(unsigned long uaddr1, unsigned long uaddr2, int nr_wake, int nr_wake2, int op) | |
324 | { | |
325 | union futex_key key1, key2; | |
326 | struct futex_hash_bucket *bh1, *bh2; | |
327 | struct list_head *head; | |
328 | struct futex_q *this, *next; | |
329 | int ret, op_ret, attempt = 0; | |
330 | ||
331 | retryfull: | |
332 | down_read(¤t->mm->mmap_sem); | |
333 | ||
334 | ret = get_futex_key(uaddr1, &key1); | |
335 | if (unlikely(ret != 0)) | |
336 | goto out; | |
337 | ret = get_futex_key(uaddr2, &key2); | |
338 | if (unlikely(ret != 0)) | |
339 | goto out; | |
340 | ||
341 | bh1 = hash_futex(&key1); | |
342 | bh2 = hash_futex(&key2); | |
343 | ||
344 | retry: | |
345 | if (bh1 < bh2) | |
346 | spin_lock(&bh1->lock); | |
347 | spin_lock(&bh2->lock); | |
348 | if (bh1 > bh2) | |
349 | spin_lock(&bh1->lock); | |
350 | ||
351 | op_ret = futex_atomic_op_inuser(op, (int __user *)uaddr2); | |
352 | if (unlikely(op_ret < 0)) { | |
353 | int dummy; | |
354 | ||
355 | spin_unlock(&bh1->lock); | |
356 | if (bh1 != bh2) | |
357 | spin_unlock(&bh2->lock); | |
358 | ||
7ee1dd3f DH |
359 | #ifndef CONFIG_MMU |
360 | /* we don't get EFAULT from MMU faults if we don't have an MMU, | |
361 | * but we might get them from range checking */ | |
362 | ret = op_ret; | |
363 | goto out; | |
364 | #endif | |
365 | ||
796f8d9b DG |
366 | if (unlikely(op_ret != -EFAULT)) { |
367 | ret = op_ret; | |
368 | goto out; | |
369 | } | |
370 | ||
4732efbe JJ |
371 | /* futex_atomic_op_inuser needs to both read and write |
372 | * *(int __user *)uaddr2, but we can't modify it | |
373 | * non-atomically. Therefore, if get_user below is not | |
374 | * enough, we need to handle the fault ourselves, while | |
375 | * still holding the mmap_sem. */ | |
376 | if (attempt++) { | |
377 | struct vm_area_struct * vma; | |
378 | struct mm_struct *mm = current->mm; | |
379 | ||
380 | ret = -EFAULT; | |
381 | if (attempt >= 2 || | |
382 | !(vma = find_vma(mm, uaddr2)) || | |
383 | vma->vm_start > uaddr2 || | |
384 | !(vma->vm_flags & VM_WRITE)) | |
385 | goto out; | |
386 | ||
387 | switch (handle_mm_fault(mm, vma, uaddr2, 1)) { | |
388 | case VM_FAULT_MINOR: | |
389 | current->min_flt++; | |
390 | break; | |
391 | case VM_FAULT_MAJOR: | |
392 | current->maj_flt++; | |
393 | break; | |
394 | default: | |
395 | goto out; | |
396 | } | |
397 | goto retry; | |
398 | } | |
399 | ||
400 | /* If we would have faulted, release mmap_sem, | |
401 | * fault it in and start all over again. */ | |
402 | up_read(¤t->mm->mmap_sem); | |
403 | ||
404 | ret = get_user(dummy, (int __user *)uaddr2); | |
405 | if (ret) | |
406 | return ret; | |
407 | ||
408 | goto retryfull; | |
409 | } | |
410 | ||
411 | head = &bh1->chain; | |
412 | ||
413 | list_for_each_entry_safe(this, next, head, list) { | |
414 | if (match_futex (&this->key, &key1)) { | |
415 | wake_futex(this); | |
416 | if (++ret >= nr_wake) | |
417 | break; | |
418 | } | |
419 | } | |
420 | ||
421 | if (op_ret > 0) { | |
422 | head = &bh2->chain; | |
423 | ||
424 | op_ret = 0; | |
425 | list_for_each_entry_safe(this, next, head, list) { | |
426 | if (match_futex (&this->key, &key2)) { | |
427 | wake_futex(this); | |
428 | if (++op_ret >= nr_wake2) | |
429 | break; | |
430 | } | |
431 | } | |
432 | ret += op_ret; | |
433 | } | |
434 | ||
435 | spin_unlock(&bh1->lock); | |
436 | if (bh1 != bh2) | |
437 | spin_unlock(&bh2->lock); | |
438 | out: | |
439 | up_read(¤t->mm->mmap_sem); | |
440 | return ret; | |
441 | } | |
442 | ||
1da177e4 LT |
443 | /* |
444 | * Requeue all waiters hashed on one physical page to another | |
445 | * physical page. | |
446 | */ | |
447 | static int futex_requeue(unsigned long uaddr1, unsigned long uaddr2, | |
448 | int nr_wake, int nr_requeue, int *valp) | |
449 | { | |
450 | union futex_key key1, key2; | |
451 | struct futex_hash_bucket *bh1, *bh2; | |
452 | struct list_head *head1; | |
453 | struct futex_q *this, *next; | |
454 | int ret, drop_count = 0; | |
455 | ||
456 | retry: | |
457 | down_read(¤t->mm->mmap_sem); | |
458 | ||
459 | ret = get_futex_key(uaddr1, &key1); | |
460 | if (unlikely(ret != 0)) | |
461 | goto out; | |
462 | ret = get_futex_key(uaddr2, &key2); | |
463 | if (unlikely(ret != 0)) | |
464 | goto out; | |
465 | ||
466 | bh1 = hash_futex(&key1); | |
467 | bh2 = hash_futex(&key2); | |
468 | ||
469 | if (bh1 < bh2) | |
470 | spin_lock(&bh1->lock); | |
471 | spin_lock(&bh2->lock); | |
472 | if (bh1 > bh2) | |
473 | spin_lock(&bh1->lock); | |
474 | ||
475 | if (likely(valp != NULL)) { | |
476 | int curval; | |
477 | ||
478 | ret = get_futex_value_locked(&curval, (int __user *)uaddr1); | |
479 | ||
480 | if (unlikely(ret)) { | |
481 | spin_unlock(&bh1->lock); | |
482 | if (bh1 != bh2) | |
483 | spin_unlock(&bh2->lock); | |
484 | ||
485 | /* If we would have faulted, release mmap_sem, fault | |
486 | * it in and start all over again. | |
487 | */ | |
488 | up_read(¤t->mm->mmap_sem); | |
489 | ||
490 | ret = get_user(curval, (int __user *)uaddr1); | |
491 | ||
492 | if (!ret) | |
493 | goto retry; | |
494 | ||
495 | return ret; | |
496 | } | |
497 | if (curval != *valp) { | |
498 | ret = -EAGAIN; | |
499 | goto out_unlock; | |
500 | } | |
501 | } | |
502 | ||
503 | head1 = &bh1->chain; | |
504 | list_for_each_entry_safe(this, next, head1, list) { | |
505 | if (!match_futex (&this->key, &key1)) | |
506 | continue; | |
507 | if (++ret <= nr_wake) { | |
508 | wake_futex(this); | |
509 | } else { | |
510 | list_move_tail(&this->list, &bh2->chain); | |
511 | this->lock_ptr = &bh2->lock; | |
512 | this->key = key2; | |
513 | get_key_refs(&key2); | |
514 | drop_count++; | |
515 | ||
516 | if (ret - nr_wake >= nr_requeue) | |
517 | break; | |
518 | /* Make sure to stop if key1 == key2 */ | |
519 | if (head1 == &bh2->chain && head1 != &next->list) | |
520 | head1 = &this->list; | |
521 | } | |
522 | } | |
523 | ||
524 | out_unlock: | |
525 | spin_unlock(&bh1->lock); | |
526 | if (bh1 != bh2) | |
527 | spin_unlock(&bh2->lock); | |
528 | ||
529 | /* drop_key_refs() must be called outside the spinlocks. */ | |
530 | while (--drop_count >= 0) | |
531 | drop_key_refs(&key1); | |
532 | ||
533 | out: | |
534 | up_read(¤t->mm->mmap_sem); | |
535 | return ret; | |
536 | } | |
537 | ||
538 | /* The key must be already stored in q->key. */ | |
539 | static inline struct futex_hash_bucket * | |
540 | queue_lock(struct futex_q *q, int fd, struct file *filp) | |
541 | { | |
542 | struct futex_hash_bucket *bh; | |
543 | ||
544 | q->fd = fd; | |
545 | q->filp = filp; | |
546 | ||
547 | init_waitqueue_head(&q->waiters); | |
548 | ||
549 | get_key_refs(&q->key); | |
550 | bh = hash_futex(&q->key); | |
551 | q->lock_ptr = &bh->lock; | |
552 | ||
553 | spin_lock(&bh->lock); | |
554 | return bh; | |
555 | } | |
556 | ||
557 | static inline void __queue_me(struct futex_q *q, struct futex_hash_bucket *bh) | |
558 | { | |
559 | list_add_tail(&q->list, &bh->chain); | |
560 | spin_unlock(&bh->lock); | |
561 | } | |
562 | ||
563 | static inline void | |
564 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *bh) | |
565 | { | |
566 | spin_unlock(&bh->lock); | |
567 | drop_key_refs(&q->key); | |
568 | } | |
569 | ||
570 | /* | |
571 | * queue_me and unqueue_me must be called as a pair, each | |
572 | * exactly once. They are called with the hashed spinlock held. | |
573 | */ | |
574 | ||
575 | /* The key must be already stored in q->key. */ | |
576 | static void queue_me(struct futex_q *q, int fd, struct file *filp) | |
577 | { | |
578 | struct futex_hash_bucket *bh; | |
579 | bh = queue_lock(q, fd, filp); | |
580 | __queue_me(q, bh); | |
581 | } | |
582 | ||
583 | /* Return 1 if we were still queued (ie. 0 means we were woken) */ | |
584 | static int unqueue_me(struct futex_q *q) | |
585 | { | |
586 | int ret = 0; | |
587 | spinlock_t *lock_ptr; | |
588 | ||
589 | /* In the common case we don't take the spinlock, which is nice. */ | |
590 | retry: | |
591 | lock_ptr = q->lock_ptr; | |
592 | if (lock_ptr != 0) { | |
593 | spin_lock(lock_ptr); | |
594 | /* | |
595 | * q->lock_ptr can change between reading it and | |
596 | * spin_lock(), causing us to take the wrong lock. This | |
597 | * corrects the race condition. | |
598 | * | |
599 | * Reasoning goes like this: if we have the wrong lock, | |
600 | * q->lock_ptr must have changed (maybe several times) | |
601 | * between reading it and the spin_lock(). It can | |
602 | * change again after the spin_lock() but only if it was | |
603 | * already changed before the spin_lock(). It cannot, | |
604 | * however, change back to the original value. Therefore | |
605 | * we can detect whether we acquired the correct lock. | |
606 | */ | |
607 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
608 | spin_unlock(lock_ptr); | |
609 | goto retry; | |
610 | } | |
611 | WARN_ON(list_empty(&q->list)); | |
612 | list_del(&q->list); | |
613 | spin_unlock(lock_ptr); | |
614 | ret = 1; | |
615 | } | |
616 | ||
617 | drop_key_refs(&q->key); | |
618 | return ret; | |
619 | } | |
620 | ||
621 | static int futex_wait(unsigned long uaddr, int val, unsigned long time) | |
622 | { | |
623 | DECLARE_WAITQUEUE(wait, current); | |
624 | int ret, curval; | |
625 | struct futex_q q; | |
626 | struct futex_hash_bucket *bh; | |
627 | ||
628 | retry: | |
629 | down_read(¤t->mm->mmap_sem); | |
630 | ||
631 | ret = get_futex_key(uaddr, &q.key); | |
632 | if (unlikely(ret != 0)) | |
633 | goto out_release_sem; | |
634 | ||
635 | bh = queue_lock(&q, -1, NULL); | |
636 | ||
637 | /* | |
638 | * Access the page AFTER the futex is queued. | |
639 | * Order is important: | |
640 | * | |
641 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
642 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
643 | * | |
644 | * The basic logical guarantee of a futex is that it blocks ONLY | |
645 | * if cond(var) is known to be true at the time of blocking, for | |
646 | * any cond. If we queued after testing *uaddr, that would open | |
647 | * a race condition where we could block indefinitely with | |
648 | * cond(var) false, which would violate the guarantee. | |
649 | * | |
650 | * A consequence is that futex_wait() can return zero and absorb | |
651 | * a wakeup when *uaddr != val on entry to the syscall. This is | |
652 | * rare, but normal. | |
653 | * | |
654 | * We hold the mmap semaphore, so the mapping cannot have changed | |
655 | * since we looked it up in get_futex_key. | |
656 | */ | |
657 | ||
658 | ret = get_futex_value_locked(&curval, (int __user *)uaddr); | |
659 | ||
660 | if (unlikely(ret)) { | |
661 | queue_unlock(&q, bh); | |
662 | ||
663 | /* If we would have faulted, release mmap_sem, fault it in and | |
664 | * start all over again. | |
665 | */ | |
666 | up_read(¤t->mm->mmap_sem); | |
667 | ||
668 | ret = get_user(curval, (int __user *)uaddr); | |
669 | ||
670 | if (!ret) | |
671 | goto retry; | |
672 | return ret; | |
673 | } | |
674 | if (curval != val) { | |
675 | ret = -EWOULDBLOCK; | |
676 | queue_unlock(&q, bh); | |
677 | goto out_release_sem; | |
678 | } | |
679 | ||
680 | /* Only actually queue if *uaddr contained val. */ | |
681 | __queue_me(&q, bh); | |
682 | ||
683 | /* | |
684 | * Now the futex is queued and we have checked the data, we | |
685 | * don't want to hold mmap_sem while we sleep. | |
686 | */ | |
687 | up_read(¤t->mm->mmap_sem); | |
688 | ||
689 | /* | |
690 | * There might have been scheduling since the queue_me(), as we | |
691 | * cannot hold a spinlock across the get_user() in case it | |
692 | * faults, and we cannot just set TASK_INTERRUPTIBLE state when | |
693 | * queueing ourselves into the futex hash. This code thus has to | |
694 | * rely on the futex_wake() code removing us from hash when it | |
695 | * wakes us up. | |
696 | */ | |
697 | ||
698 | /* add_wait_queue is the barrier after __set_current_state. */ | |
699 | __set_current_state(TASK_INTERRUPTIBLE); | |
700 | add_wait_queue(&q.waiters, &wait); | |
701 | /* | |
702 | * !list_empty() is safe here without any lock. | |
703 | * q.lock_ptr != 0 is not safe, because of ordering against wakeup. | |
704 | */ | |
705 | if (likely(!list_empty(&q.list))) | |
706 | time = schedule_timeout(time); | |
707 | __set_current_state(TASK_RUNNING); | |
708 | ||
709 | /* | |
710 | * NOTE: we don't remove ourselves from the waitqueue because | |
711 | * we are the only user of it. | |
712 | */ | |
713 | ||
714 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
715 | if (!unqueue_me(&q)) | |
716 | return 0; | |
717 | if (time == 0) | |
718 | return -ETIMEDOUT; | |
719 | /* We expect signal_pending(current), but another thread may | |
720 | * have handled it for us already. */ | |
721 | return -EINTR; | |
722 | ||
723 | out_release_sem: | |
724 | up_read(¤t->mm->mmap_sem); | |
725 | return ret; | |
726 | } | |
727 | ||
728 | static int futex_close(struct inode *inode, struct file *filp) | |
729 | { | |
730 | struct futex_q *q = filp->private_data; | |
731 | ||
732 | unqueue_me(q); | |
733 | kfree(q); | |
734 | return 0; | |
735 | } | |
736 | ||
737 | /* This is one-shot: once it's gone off you need a new fd */ | |
738 | static unsigned int futex_poll(struct file *filp, | |
739 | struct poll_table_struct *wait) | |
740 | { | |
741 | struct futex_q *q = filp->private_data; | |
742 | int ret = 0; | |
743 | ||
744 | poll_wait(filp, &q->waiters, wait); | |
745 | ||
746 | /* | |
747 | * list_empty() is safe here without any lock. | |
748 | * q->lock_ptr != 0 is not safe, because of ordering against wakeup. | |
749 | */ | |
750 | if (list_empty(&q->list)) | |
751 | ret = POLLIN | POLLRDNORM; | |
752 | ||
753 | return ret; | |
754 | } | |
755 | ||
756 | static struct file_operations futex_fops = { | |
757 | .release = futex_close, | |
758 | .poll = futex_poll, | |
759 | }; | |
760 | ||
761 | /* | |
762 | * Signal allows caller to avoid the race which would occur if they | |
763 | * set the sigio stuff up afterwards. | |
764 | */ | |
765 | static int futex_fd(unsigned long uaddr, int signal) | |
766 | { | |
767 | struct futex_q *q; | |
768 | struct file *filp; | |
769 | int ret, err; | |
770 | ||
771 | ret = -EINVAL; | |
7ed20e1a | 772 | if (!valid_signal(signal)) |
1da177e4 LT |
773 | goto out; |
774 | ||
775 | ret = get_unused_fd(); | |
776 | if (ret < 0) | |
777 | goto out; | |
778 | filp = get_empty_filp(); | |
779 | if (!filp) { | |
780 | put_unused_fd(ret); | |
781 | ret = -ENFILE; | |
782 | goto out; | |
783 | } | |
784 | filp->f_op = &futex_fops; | |
785 | filp->f_vfsmnt = mntget(futex_mnt); | |
786 | filp->f_dentry = dget(futex_mnt->mnt_root); | |
787 | filp->f_mapping = filp->f_dentry->d_inode->i_mapping; | |
788 | ||
789 | if (signal) { | |
1da177e4 LT |
790 | err = f_setown(filp, current->pid, 1); |
791 | if (err < 0) { | |
39ed3fde | 792 | goto error; |
1da177e4 LT |
793 | } |
794 | filp->f_owner.signum = signal; | |
795 | } | |
796 | ||
797 | q = kmalloc(sizeof(*q), GFP_KERNEL); | |
798 | if (!q) { | |
39ed3fde PE |
799 | err = -ENOMEM; |
800 | goto error; | |
1da177e4 LT |
801 | } |
802 | ||
803 | down_read(¤t->mm->mmap_sem); | |
804 | err = get_futex_key(uaddr, &q->key); | |
805 | ||
806 | if (unlikely(err != 0)) { | |
807 | up_read(¤t->mm->mmap_sem); | |
1da177e4 | 808 | kfree(q); |
39ed3fde | 809 | goto error; |
1da177e4 LT |
810 | } |
811 | ||
812 | /* | |
813 | * queue_me() must be called before releasing mmap_sem, because | |
814 | * key->shared.inode needs to be referenced while holding it. | |
815 | */ | |
816 | filp->private_data = q; | |
817 | ||
818 | queue_me(q, ret, filp); | |
819 | up_read(¤t->mm->mmap_sem); | |
820 | ||
821 | /* Now we map fd to filp, so userspace can access it */ | |
822 | fd_install(ret, filp); | |
823 | out: | |
824 | return ret; | |
39ed3fde PE |
825 | error: |
826 | put_unused_fd(ret); | |
827 | put_filp(filp); | |
828 | ret = err; | |
829 | goto out; | |
1da177e4 LT |
830 | } |
831 | ||
832 | long do_futex(unsigned long uaddr, int op, int val, unsigned long timeout, | |
833 | unsigned long uaddr2, int val2, int val3) | |
834 | { | |
835 | int ret; | |
836 | ||
837 | switch (op) { | |
838 | case FUTEX_WAIT: | |
839 | ret = futex_wait(uaddr, val, timeout); | |
840 | break; | |
841 | case FUTEX_WAKE: | |
842 | ret = futex_wake(uaddr, val); | |
843 | break; | |
844 | case FUTEX_FD: | |
845 | /* non-zero val means F_SETOWN(getpid()) & F_SETSIG(val) */ | |
846 | ret = futex_fd(uaddr, val); | |
847 | break; | |
848 | case FUTEX_REQUEUE: | |
849 | ret = futex_requeue(uaddr, uaddr2, val, val2, NULL); | |
850 | break; | |
851 | case FUTEX_CMP_REQUEUE: | |
852 | ret = futex_requeue(uaddr, uaddr2, val, val2, &val3); | |
853 | break; | |
4732efbe JJ |
854 | case FUTEX_WAKE_OP: |
855 | ret = futex_wake_op(uaddr, uaddr2, val, val2, val3); | |
856 | break; | |
1da177e4 LT |
857 | default: |
858 | ret = -ENOSYS; | |
859 | } | |
860 | return ret; | |
861 | } | |
862 | ||
863 | ||
864 | asmlinkage long sys_futex(u32 __user *uaddr, int op, int val, | |
865 | struct timespec __user *utime, u32 __user *uaddr2, | |
866 | int val3) | |
867 | { | |
868 | struct timespec t; | |
869 | unsigned long timeout = MAX_SCHEDULE_TIMEOUT; | |
870 | int val2 = 0; | |
871 | ||
872 | if ((op == FUTEX_WAIT) && utime) { | |
873 | if (copy_from_user(&t, utime, sizeof(t)) != 0) | |
874 | return -EFAULT; | |
875 | timeout = timespec_to_jiffies(&t) + 1; | |
876 | } | |
877 | /* | |
878 | * requeue parameter in 'utime' if op == FUTEX_REQUEUE. | |
879 | */ | |
880 | if (op >= FUTEX_REQUEUE) | |
881 | val2 = (int) (unsigned long) utime; | |
882 | ||
883 | return do_futex((unsigned long)uaddr, op, val, timeout, | |
884 | (unsigned long)uaddr2, val2, val3); | |
885 | } | |
886 | ||
887 | static struct super_block * | |
888 | futexfs_get_sb(struct file_system_type *fs_type, | |
889 | int flags, const char *dev_name, void *data) | |
890 | { | |
891 | return get_sb_pseudo(fs_type, "futex", NULL, 0xBAD1DEA); | |
892 | } | |
893 | ||
894 | static struct file_system_type futex_fs_type = { | |
895 | .name = "futexfs", | |
896 | .get_sb = futexfs_get_sb, | |
897 | .kill_sb = kill_anon_super, | |
898 | }; | |
899 | ||
900 | static int __init init(void) | |
901 | { | |
902 | unsigned int i; | |
903 | ||
904 | register_filesystem(&futex_fs_type); | |
905 | futex_mnt = kern_mount(&futex_fs_type); | |
906 | ||
907 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { | |
908 | INIT_LIST_HEAD(&futex_queues[i].chain); | |
909 | spin_lock_init(&futex_queues[i].lock); | |
910 | } | |
911 | return 0; | |
912 | } | |
913 | __initcall(init); |