]>
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 | * | |
0771dfef IM |
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 | * | |
c87e2837 IM |
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 | * | |
1da177e4 LT |
19 | * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly |
20 | * enough at me, Linus for the original (flawed) idea, Matthew | |
21 | * Kirkwood for proof-of-concept implementation. | |
22 | * | |
23 | * "The futexes are also cursed." | |
24 | * "But they come in a choice of three flavours!" | |
25 | * | |
26 | * This program is free software; you can redistribute it and/or modify | |
27 | * it under the terms of the GNU General Public License as published by | |
28 | * the Free Software Foundation; either version 2 of the License, or | |
29 | * (at your option) any later version. | |
30 | * | |
31 | * This program is distributed in the hope that it will be useful, | |
32 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
33 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
34 | * GNU General Public License for more details. | |
35 | * | |
36 | * You should have received a copy of the GNU General Public License | |
37 | * along with this program; if not, write to the Free Software | |
38 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
39 | */ | |
40 | #include <linux/slab.h> | |
41 | #include <linux/poll.h> | |
42 | #include <linux/fs.h> | |
43 | #include <linux/file.h> | |
44 | #include <linux/jhash.h> | |
45 | #include <linux/init.h> | |
46 | #include <linux/futex.h> | |
47 | #include <linux/mount.h> | |
48 | #include <linux/pagemap.h> | |
49 | #include <linux/syscalls.h> | |
7ed20e1a | 50 | #include <linux/signal.h> |
4732efbe | 51 | #include <asm/futex.h> |
1da177e4 | 52 | |
c87e2837 IM |
53 | #include "rtmutex_common.h" |
54 | ||
1da177e4 LT |
55 | #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8) |
56 | ||
57 | /* | |
58 | * Futexes are matched on equal values of this key. | |
59 | * The key type depends on whether it's a shared or private mapping. | |
60 | * Don't rearrange members without looking at hash_futex(). | |
61 | * | |
62 | * offset is aligned to a multiple of sizeof(u32) (== 4) by definition. | |
63 | * We set bit 0 to indicate if it's an inode-based key. | |
64 | */ | |
65 | union futex_key { | |
66 | struct { | |
67 | unsigned long pgoff; | |
68 | struct inode *inode; | |
69 | int offset; | |
70 | } shared; | |
71 | struct { | |
e2970f2f | 72 | unsigned long address; |
1da177e4 LT |
73 | struct mm_struct *mm; |
74 | int offset; | |
75 | } private; | |
76 | struct { | |
77 | unsigned long word; | |
78 | void *ptr; | |
79 | int offset; | |
80 | } both; | |
81 | }; | |
82 | ||
c87e2837 IM |
83 | /* |
84 | * Priority Inheritance state: | |
85 | */ | |
86 | struct futex_pi_state { | |
87 | /* | |
88 | * list of 'owned' pi_state instances - these have to be | |
89 | * cleaned up in do_exit() if the task exits prematurely: | |
90 | */ | |
91 | struct list_head list; | |
92 | ||
93 | /* | |
94 | * The PI object: | |
95 | */ | |
96 | struct rt_mutex pi_mutex; | |
97 | ||
98 | struct task_struct *owner; | |
99 | atomic_t refcount; | |
100 | ||
101 | union futex_key key; | |
102 | }; | |
103 | ||
1da177e4 LT |
104 | /* |
105 | * We use this hashed waitqueue instead of a normal wait_queue_t, so | |
106 | * we can wake only the relevant ones (hashed queues may be shared). | |
107 | * | |
108 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
109 | * It is considered woken when list_empty(&q->list) || q->lock_ptr == 0. | |
110 | * The order of wakup is always to make the first condition true, then | |
111 | * wake up q->waiters, then make the second condition true. | |
112 | */ | |
113 | struct futex_q { | |
114 | struct list_head list; | |
115 | wait_queue_head_t waiters; | |
116 | ||
e2970f2f | 117 | /* Which hash list lock to use: */ |
1da177e4 LT |
118 | spinlock_t *lock_ptr; |
119 | ||
e2970f2f | 120 | /* Key which the futex is hashed on: */ |
1da177e4 LT |
121 | union futex_key key; |
122 | ||
e2970f2f | 123 | /* For fd, sigio sent using these: */ |
1da177e4 LT |
124 | int fd; |
125 | struct file *filp; | |
c87e2837 IM |
126 | |
127 | /* Optional priority inheritance state: */ | |
128 | struct futex_pi_state *pi_state; | |
129 | struct task_struct *task; | |
1da177e4 LT |
130 | }; |
131 | ||
132 | /* | |
133 | * Split the global futex_lock into every hash list lock. | |
134 | */ | |
135 | struct futex_hash_bucket { | |
136 | spinlock_t lock; | |
137 | struct list_head chain; | |
138 | }; | |
139 | ||
140 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; | |
141 | ||
142 | /* Futex-fs vfsmount entry: */ | |
143 | static struct vfsmount *futex_mnt; | |
144 | ||
145 | /* | |
146 | * We hash on the keys returned from get_futex_key (see below). | |
147 | */ | |
148 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
149 | { | |
150 | u32 hash = jhash2((u32*)&key->both.word, | |
151 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
152 | key->both.offset); | |
153 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; | |
154 | } | |
155 | ||
156 | /* | |
157 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
158 | */ | |
159 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
160 | { | |
161 | return (key1->both.word == key2->both.word | |
162 | && key1->both.ptr == key2->both.ptr | |
163 | && key1->both.offset == key2->both.offset); | |
164 | } | |
165 | ||
166 | /* | |
167 | * Get parameters which are the keys for a futex. | |
168 | * | |
169 | * For shared mappings, it's (page->index, vma->vm_file->f_dentry->d_inode, | |
170 | * offset_within_page). For private mappings, it's (uaddr, current->mm). | |
171 | * We can usually work out the index without swapping in the page. | |
172 | * | |
173 | * Returns: 0, or negative error code. | |
174 | * The key words are stored in *key on success. | |
175 | * | |
176 | * Should be called with ¤t->mm->mmap_sem but NOT any spinlocks. | |
177 | */ | |
e2970f2f | 178 | static int get_futex_key(u32 __user *uaddr, union futex_key *key) |
1da177e4 | 179 | { |
e2970f2f | 180 | unsigned long address = (unsigned long)uaddr; |
1da177e4 LT |
181 | struct mm_struct *mm = current->mm; |
182 | struct vm_area_struct *vma; | |
183 | struct page *page; | |
184 | int err; | |
185 | ||
186 | /* | |
187 | * The futex address must be "naturally" aligned. | |
188 | */ | |
e2970f2f | 189 | key->both.offset = address % PAGE_SIZE; |
1da177e4 LT |
190 | if (unlikely((key->both.offset % sizeof(u32)) != 0)) |
191 | return -EINVAL; | |
e2970f2f | 192 | address -= key->both.offset; |
1da177e4 LT |
193 | |
194 | /* | |
195 | * The futex is hashed differently depending on whether | |
196 | * it's in a shared or private mapping. So check vma first. | |
197 | */ | |
e2970f2f | 198 | vma = find_extend_vma(mm, address); |
1da177e4 LT |
199 | if (unlikely(!vma)) |
200 | return -EFAULT; | |
201 | ||
202 | /* | |
203 | * Permissions. | |
204 | */ | |
205 | if (unlikely((vma->vm_flags & (VM_IO|VM_READ)) != VM_READ)) | |
206 | return (vma->vm_flags & VM_IO) ? -EPERM : -EACCES; | |
207 | ||
208 | /* | |
209 | * Private mappings are handled in a simple way. | |
210 | * | |
211 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
212 | * it's a read-only handle, it's expected that futexes attach to | |
213 | * the object not the particular process. Therefore we use | |
214 | * VM_MAYSHARE here, not VM_SHARED which is restricted to shared | |
215 | * mappings of _writable_ handles. | |
216 | */ | |
217 | if (likely(!(vma->vm_flags & VM_MAYSHARE))) { | |
218 | key->private.mm = mm; | |
e2970f2f | 219 | key->private.address = address; |
1da177e4 LT |
220 | return 0; |
221 | } | |
222 | ||
223 | /* | |
224 | * Linear file mappings are also simple. | |
225 | */ | |
226 | key->shared.inode = vma->vm_file->f_dentry->d_inode; | |
227 | key->both.offset++; /* Bit 0 of offset indicates inode-based key. */ | |
228 | if (likely(!(vma->vm_flags & VM_NONLINEAR))) { | |
e2970f2f | 229 | key->shared.pgoff = (((address - vma->vm_start) >> PAGE_SHIFT) |
1da177e4 LT |
230 | + vma->vm_pgoff); |
231 | return 0; | |
232 | } | |
233 | ||
234 | /* | |
235 | * We could walk the page table to read the non-linear | |
236 | * pte, and get the page index without fetching the page | |
237 | * from swap. But that's a lot of code to duplicate here | |
238 | * for a rare case, so we simply fetch the page. | |
239 | */ | |
e2970f2f | 240 | err = get_user_pages(current, mm, address, 1, 0, 0, &page, NULL); |
1da177e4 LT |
241 | if (err >= 0) { |
242 | key->shared.pgoff = | |
243 | page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
244 | put_page(page); | |
245 | return 0; | |
246 | } | |
247 | return err; | |
248 | } | |
249 | ||
250 | /* | |
251 | * Take a reference to the resource addressed by a key. | |
252 | * Can be called while holding spinlocks. | |
253 | * | |
254 | * NOTE: mmap_sem MUST be held between get_futex_key() and calling this | |
255 | * function, if it is called at all. mmap_sem keeps key->shared.inode valid. | |
256 | */ | |
257 | static inline void get_key_refs(union futex_key *key) | |
258 | { | |
259 | if (key->both.ptr != 0) { | |
260 | if (key->both.offset & 1) | |
261 | atomic_inc(&key->shared.inode->i_count); | |
262 | else | |
263 | atomic_inc(&key->private.mm->mm_count); | |
264 | } | |
265 | } | |
266 | ||
267 | /* | |
268 | * Drop a reference to the resource addressed by a key. | |
269 | * The hash bucket spinlock must not be held. | |
270 | */ | |
271 | static void drop_key_refs(union futex_key *key) | |
272 | { | |
273 | if (key->both.ptr != 0) { | |
274 | if (key->both.offset & 1) | |
275 | iput(key->shared.inode); | |
276 | else | |
277 | mmdrop(key->private.mm); | |
278 | } | |
279 | } | |
280 | ||
e2970f2f | 281 | static inline int get_futex_value_locked(u32 *dest, u32 __user *from) |
1da177e4 LT |
282 | { |
283 | int ret; | |
284 | ||
285 | inc_preempt_count(); | |
e2970f2f | 286 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
1da177e4 LT |
287 | dec_preempt_count(); |
288 | ||
289 | return ret ? -EFAULT : 0; | |
290 | } | |
291 | ||
c87e2837 IM |
292 | /* |
293 | * Fault handling. Called with current->mm->mmap_sem held. | |
294 | */ | |
295 | static int futex_handle_fault(unsigned long address, int attempt) | |
296 | { | |
297 | struct vm_area_struct * vma; | |
298 | struct mm_struct *mm = current->mm; | |
299 | ||
300 | if (attempt >= 2 || !(vma = find_vma(mm, address)) || | |
301 | vma->vm_start > address || !(vma->vm_flags & VM_WRITE)) | |
302 | return -EFAULT; | |
303 | ||
304 | switch (handle_mm_fault(mm, vma, address, 1)) { | |
305 | case VM_FAULT_MINOR: | |
306 | current->min_flt++; | |
307 | break; | |
308 | case VM_FAULT_MAJOR: | |
309 | current->maj_flt++; | |
310 | break; | |
311 | default: | |
312 | return -EFAULT; | |
313 | } | |
314 | return 0; | |
315 | } | |
316 | ||
317 | /* | |
318 | * PI code: | |
319 | */ | |
320 | static int refill_pi_state_cache(void) | |
321 | { | |
322 | struct futex_pi_state *pi_state; | |
323 | ||
324 | if (likely(current->pi_state_cache)) | |
325 | return 0; | |
326 | ||
327 | pi_state = kmalloc(sizeof(*pi_state), GFP_KERNEL); | |
328 | ||
329 | if (!pi_state) | |
330 | return -ENOMEM; | |
331 | ||
332 | memset(pi_state, 0, sizeof(*pi_state)); | |
333 | INIT_LIST_HEAD(&pi_state->list); | |
334 | /* pi_mutex gets initialized later */ | |
335 | pi_state->owner = NULL; | |
336 | atomic_set(&pi_state->refcount, 1); | |
337 | ||
338 | current->pi_state_cache = pi_state; | |
339 | ||
340 | return 0; | |
341 | } | |
342 | ||
343 | static struct futex_pi_state * alloc_pi_state(void) | |
344 | { | |
345 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
346 | ||
347 | WARN_ON(!pi_state); | |
348 | current->pi_state_cache = NULL; | |
349 | ||
350 | return pi_state; | |
351 | } | |
352 | ||
353 | static void free_pi_state(struct futex_pi_state *pi_state) | |
354 | { | |
355 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
356 | return; | |
357 | ||
358 | /* | |
359 | * If pi_state->owner is NULL, the owner is most probably dying | |
360 | * and has cleaned up the pi_state already | |
361 | */ | |
362 | if (pi_state->owner) { | |
363 | spin_lock_irq(&pi_state->owner->pi_lock); | |
364 | list_del_init(&pi_state->list); | |
365 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
366 | ||
367 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
368 | } | |
369 | ||
370 | if (current->pi_state_cache) | |
371 | kfree(pi_state); | |
372 | else { | |
373 | /* | |
374 | * pi_state->list is already empty. | |
375 | * clear pi_state->owner. | |
376 | * refcount is at 0 - put it back to 1. | |
377 | */ | |
378 | pi_state->owner = NULL; | |
379 | atomic_set(&pi_state->refcount, 1); | |
380 | current->pi_state_cache = pi_state; | |
381 | } | |
382 | } | |
383 | ||
384 | /* | |
385 | * Look up the task based on what TID userspace gave us. | |
386 | * We dont trust it. | |
387 | */ | |
388 | static struct task_struct * futex_find_get_task(pid_t pid) | |
389 | { | |
390 | struct task_struct *p; | |
391 | ||
392 | read_lock(&tasklist_lock); | |
393 | p = find_task_by_pid(pid); | |
394 | if (!p) | |
395 | goto out_unlock; | |
396 | if ((current->euid != p->euid) && (current->euid != p->uid)) { | |
397 | p = NULL; | |
398 | goto out_unlock; | |
399 | } | |
400 | if (p->state == EXIT_ZOMBIE || p->exit_state == EXIT_ZOMBIE) { | |
401 | p = NULL; | |
402 | goto out_unlock; | |
403 | } | |
404 | get_task_struct(p); | |
405 | out_unlock: | |
406 | read_unlock(&tasklist_lock); | |
407 | ||
408 | return p; | |
409 | } | |
410 | ||
411 | /* | |
412 | * This task is holding PI mutexes at exit time => bad. | |
413 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
414 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
415 | */ | |
416 | void exit_pi_state_list(struct task_struct *curr) | |
417 | { | |
c87e2837 IM |
418 | struct list_head *next, *head = &curr->pi_state_list; |
419 | struct futex_pi_state *pi_state; | |
627371d7 | 420 | struct futex_hash_bucket *hb; |
c87e2837 IM |
421 | union futex_key key; |
422 | ||
423 | /* | |
424 | * We are a ZOMBIE and nobody can enqueue itself on | |
425 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 426 | * versus waiters unqueueing themselves: |
c87e2837 IM |
427 | */ |
428 | spin_lock_irq(&curr->pi_lock); | |
429 | while (!list_empty(head)) { | |
430 | ||
431 | next = head->next; | |
432 | pi_state = list_entry(next, struct futex_pi_state, list); | |
433 | key = pi_state->key; | |
627371d7 | 434 | hb = hash_futex(&key); |
c87e2837 IM |
435 | spin_unlock_irq(&curr->pi_lock); |
436 | ||
c87e2837 IM |
437 | spin_lock(&hb->lock); |
438 | ||
439 | spin_lock_irq(&curr->pi_lock); | |
627371d7 IM |
440 | /* |
441 | * We dropped the pi-lock, so re-check whether this | |
442 | * task still owns the PI-state: | |
443 | */ | |
c87e2837 IM |
444 | if (head->next != next) { |
445 | spin_unlock(&hb->lock); | |
446 | continue; | |
447 | } | |
448 | ||
c87e2837 | 449 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
450 | WARN_ON(list_empty(&pi_state->list)); |
451 | list_del_init(&pi_state->list); | |
c87e2837 IM |
452 | pi_state->owner = NULL; |
453 | spin_unlock_irq(&curr->pi_lock); | |
454 | ||
455 | rt_mutex_unlock(&pi_state->pi_mutex); | |
456 | ||
457 | spin_unlock(&hb->lock); | |
458 | ||
459 | spin_lock_irq(&curr->pi_lock); | |
460 | } | |
461 | spin_unlock_irq(&curr->pi_lock); | |
462 | } | |
463 | ||
464 | static int | |
465 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, struct futex_q *me) | |
466 | { | |
467 | struct futex_pi_state *pi_state = NULL; | |
468 | struct futex_q *this, *next; | |
469 | struct list_head *head; | |
470 | struct task_struct *p; | |
471 | pid_t pid; | |
472 | ||
473 | head = &hb->chain; | |
474 | ||
475 | list_for_each_entry_safe(this, next, head, list) { | |
627371d7 | 476 | if (match_futex(&this->key, &me->key)) { |
c87e2837 IM |
477 | /* |
478 | * Another waiter already exists - bump up | |
479 | * the refcount and return its pi_state: | |
480 | */ | |
481 | pi_state = this->pi_state; | |
06a9ec29 TG |
482 | /* |
483 | * Userspace might have messed up non PI and PI futexes | |
484 | */ | |
485 | if (unlikely(!pi_state)) | |
486 | return -EINVAL; | |
487 | ||
627371d7 IM |
488 | WARN_ON(!atomic_read(&pi_state->refcount)); |
489 | ||
c87e2837 IM |
490 | atomic_inc(&pi_state->refcount); |
491 | me->pi_state = pi_state; | |
492 | ||
493 | return 0; | |
494 | } | |
495 | } | |
496 | ||
497 | /* | |
e3f2ddea IM |
498 | * We are the first waiter - try to look up the real owner and attach |
499 | * the new pi_state to it, but bail out when the owner died bit is set | |
500 | * and TID = 0: | |
c87e2837 IM |
501 | */ |
502 | pid = uval & FUTEX_TID_MASK; | |
e3f2ddea IM |
503 | if (!pid && (uval & FUTEX_OWNER_DIED)) |
504 | return -ESRCH; | |
c87e2837 IM |
505 | p = futex_find_get_task(pid); |
506 | if (!p) | |
507 | return -ESRCH; | |
508 | ||
509 | pi_state = alloc_pi_state(); | |
510 | ||
511 | /* | |
512 | * Initialize the pi_mutex in locked state and make 'p' | |
513 | * the owner of it: | |
514 | */ | |
515 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
516 | ||
517 | /* Store the key for possible exit cleanups: */ | |
518 | pi_state->key = me->key; | |
519 | ||
520 | spin_lock_irq(&p->pi_lock); | |
627371d7 | 521 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
522 | list_add(&pi_state->list, &p->pi_state_list); |
523 | pi_state->owner = p; | |
524 | spin_unlock_irq(&p->pi_lock); | |
525 | ||
526 | put_task_struct(p); | |
527 | ||
528 | me->pi_state = pi_state; | |
529 | ||
530 | return 0; | |
531 | } | |
532 | ||
1da177e4 LT |
533 | /* |
534 | * The hash bucket lock must be held when this is called. | |
535 | * Afterwards, the futex_q must not be accessed. | |
536 | */ | |
537 | static void wake_futex(struct futex_q *q) | |
538 | { | |
539 | list_del_init(&q->list); | |
540 | if (q->filp) | |
541 | send_sigio(&q->filp->f_owner, q->fd, POLL_IN); | |
542 | /* | |
543 | * The lock in wake_up_all() is a crucial memory barrier after the | |
544 | * list_del_init() and also before assigning to q->lock_ptr. | |
545 | */ | |
546 | wake_up_all(&q->waiters); | |
547 | /* | |
548 | * The waiting task can free the futex_q as soon as this is written, | |
549 | * without taking any locks. This must come last. | |
8e31108b AM |
550 | * |
551 | * A memory barrier is required here to prevent the following store | |
552 | * to lock_ptr from getting ahead of the wakeup. Clearing the lock | |
553 | * at the end of wake_up_all() does not prevent this store from | |
554 | * moving. | |
1da177e4 | 555 | */ |
8e31108b | 556 | wmb(); |
1da177e4 LT |
557 | q->lock_ptr = NULL; |
558 | } | |
559 | ||
c87e2837 IM |
560 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
561 | { | |
562 | struct task_struct *new_owner; | |
563 | struct futex_pi_state *pi_state = this->pi_state; | |
564 | u32 curval, newval; | |
565 | ||
566 | if (!pi_state) | |
567 | return -EINVAL; | |
568 | ||
569 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); | |
570 | ||
571 | /* | |
572 | * This happens when we have stolen the lock and the original | |
573 | * pending owner did not enqueue itself back on the rt_mutex. | |
574 | * Thats not a tragedy. We know that way, that a lock waiter | |
575 | * is on the fly. We make the futex_q waiter the pending owner. | |
576 | */ | |
577 | if (!new_owner) | |
578 | new_owner = this->task; | |
579 | ||
580 | /* | |
581 | * We pass it to the next owner. (The WAITERS bit is always | |
582 | * kept enabled while there is PI state around. We must also | |
583 | * preserve the owner died bit.) | |
584 | */ | |
e3f2ddea IM |
585 | if (!(uval & FUTEX_OWNER_DIED)) { |
586 | newval = FUTEX_WAITERS | new_owner->pid; | |
587 | ||
588 | inc_preempt_count(); | |
589 | curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval); | |
590 | dec_preempt_count(); | |
591 | if (curval == -EFAULT) | |
592 | return -EFAULT; | |
593 | if (curval != uval) | |
594 | return -EINVAL; | |
595 | } | |
c87e2837 | 596 | |
627371d7 IM |
597 | spin_lock_irq(&pi_state->owner->pi_lock); |
598 | WARN_ON(list_empty(&pi_state->list)); | |
599 | list_del_init(&pi_state->list); | |
600 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
601 | ||
602 | spin_lock_irq(&new_owner->pi_lock); | |
603 | WARN_ON(!list_empty(&pi_state->list)); | |
c87e2837 IM |
604 | list_add(&pi_state->list, &new_owner->pi_state_list); |
605 | pi_state->owner = new_owner; | |
627371d7 IM |
606 | spin_unlock_irq(&new_owner->pi_lock); |
607 | ||
c87e2837 IM |
608 | rt_mutex_unlock(&pi_state->pi_mutex); |
609 | ||
610 | return 0; | |
611 | } | |
612 | ||
613 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) | |
614 | { | |
615 | u32 oldval; | |
616 | ||
617 | /* | |
618 | * There is no waiter, so we unlock the futex. The owner died | |
619 | * bit has not to be preserved here. We are the owner: | |
620 | */ | |
621 | inc_preempt_count(); | |
622 | oldval = futex_atomic_cmpxchg_inatomic(uaddr, uval, 0); | |
623 | dec_preempt_count(); | |
624 | ||
625 | if (oldval == -EFAULT) | |
626 | return oldval; | |
627 | if (oldval != uval) | |
628 | return -EAGAIN; | |
629 | ||
630 | return 0; | |
631 | } | |
632 | ||
8b8f319f IM |
633 | /* |
634 | * Express the locking dependencies for lockdep: | |
635 | */ | |
636 | static inline void | |
637 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
638 | { | |
639 | if (hb1 <= hb2) { | |
640 | spin_lock(&hb1->lock); | |
641 | if (hb1 < hb2) | |
642 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
643 | } else { /* hb1 > hb2 */ | |
644 | spin_lock(&hb2->lock); | |
645 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
646 | } | |
647 | } | |
648 | ||
1da177e4 LT |
649 | /* |
650 | * Wake up all waiters hashed on the physical page that is mapped | |
651 | * to this virtual address: | |
652 | */ | |
e2970f2f | 653 | static int futex_wake(u32 __user *uaddr, int nr_wake) |
1da177e4 | 654 | { |
e2970f2f | 655 | struct futex_hash_bucket *hb; |
1da177e4 | 656 | struct futex_q *this, *next; |
e2970f2f IM |
657 | struct list_head *head; |
658 | union futex_key key; | |
1da177e4 LT |
659 | int ret; |
660 | ||
661 | down_read(¤t->mm->mmap_sem); | |
662 | ||
663 | ret = get_futex_key(uaddr, &key); | |
664 | if (unlikely(ret != 0)) | |
665 | goto out; | |
666 | ||
e2970f2f IM |
667 | hb = hash_futex(&key); |
668 | spin_lock(&hb->lock); | |
669 | head = &hb->chain; | |
1da177e4 LT |
670 | |
671 | list_for_each_entry_safe(this, next, head, list) { | |
672 | if (match_futex (&this->key, &key)) { | |
ed6f7b10 IM |
673 | if (this->pi_state) { |
674 | ret = -EINVAL; | |
675 | break; | |
676 | } | |
1da177e4 LT |
677 | wake_futex(this); |
678 | if (++ret >= nr_wake) | |
679 | break; | |
680 | } | |
681 | } | |
682 | ||
e2970f2f | 683 | spin_unlock(&hb->lock); |
1da177e4 LT |
684 | out: |
685 | up_read(¤t->mm->mmap_sem); | |
686 | return ret; | |
687 | } | |
688 | ||
4732efbe JJ |
689 | /* |
690 | * Wake up all waiters hashed on the physical page that is mapped | |
691 | * to this virtual address: | |
692 | */ | |
e2970f2f IM |
693 | static int |
694 | futex_wake_op(u32 __user *uaddr1, u32 __user *uaddr2, | |
695 | int nr_wake, int nr_wake2, int op) | |
4732efbe JJ |
696 | { |
697 | union futex_key key1, key2; | |
e2970f2f | 698 | struct futex_hash_bucket *hb1, *hb2; |
4732efbe JJ |
699 | struct list_head *head; |
700 | struct futex_q *this, *next; | |
701 | int ret, op_ret, attempt = 0; | |
702 | ||
703 | retryfull: | |
704 | down_read(¤t->mm->mmap_sem); | |
705 | ||
706 | ret = get_futex_key(uaddr1, &key1); | |
707 | if (unlikely(ret != 0)) | |
708 | goto out; | |
709 | ret = get_futex_key(uaddr2, &key2); | |
710 | if (unlikely(ret != 0)) | |
711 | goto out; | |
712 | ||
e2970f2f IM |
713 | hb1 = hash_futex(&key1); |
714 | hb2 = hash_futex(&key2); | |
4732efbe JJ |
715 | |
716 | retry: | |
8b8f319f | 717 | double_lock_hb(hb1, hb2); |
4732efbe | 718 | |
e2970f2f | 719 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 720 | if (unlikely(op_ret < 0)) { |
e2970f2f | 721 | u32 dummy; |
4732efbe | 722 | |
e2970f2f IM |
723 | spin_unlock(&hb1->lock); |
724 | if (hb1 != hb2) | |
725 | spin_unlock(&hb2->lock); | |
4732efbe | 726 | |
7ee1dd3f | 727 | #ifndef CONFIG_MMU |
e2970f2f IM |
728 | /* |
729 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
730 | * but we might get them from range checking | |
731 | */ | |
7ee1dd3f DH |
732 | ret = op_ret; |
733 | goto out; | |
734 | #endif | |
735 | ||
796f8d9b DG |
736 | if (unlikely(op_ret != -EFAULT)) { |
737 | ret = op_ret; | |
738 | goto out; | |
739 | } | |
740 | ||
e2970f2f IM |
741 | /* |
742 | * futex_atomic_op_inuser needs to both read and write | |
4732efbe JJ |
743 | * *(int __user *)uaddr2, but we can't modify it |
744 | * non-atomically. Therefore, if get_user below is not | |
745 | * enough, we need to handle the fault ourselves, while | |
e2970f2f IM |
746 | * still holding the mmap_sem. |
747 | */ | |
4732efbe | 748 | if (attempt++) { |
c87e2837 IM |
749 | if (futex_handle_fault((unsigned long)uaddr2, |
750 | attempt)) | |
4732efbe | 751 | goto out; |
4732efbe JJ |
752 | goto retry; |
753 | } | |
754 | ||
e2970f2f IM |
755 | /* |
756 | * If we would have faulted, release mmap_sem, | |
757 | * fault it in and start all over again. | |
758 | */ | |
4732efbe JJ |
759 | up_read(¤t->mm->mmap_sem); |
760 | ||
e2970f2f | 761 | ret = get_user(dummy, uaddr2); |
4732efbe JJ |
762 | if (ret) |
763 | return ret; | |
764 | ||
765 | goto retryfull; | |
766 | } | |
767 | ||
e2970f2f | 768 | head = &hb1->chain; |
4732efbe JJ |
769 | |
770 | list_for_each_entry_safe(this, next, head, list) { | |
771 | if (match_futex (&this->key, &key1)) { | |
772 | wake_futex(this); | |
773 | if (++ret >= nr_wake) | |
774 | break; | |
775 | } | |
776 | } | |
777 | ||
778 | if (op_ret > 0) { | |
e2970f2f | 779 | head = &hb2->chain; |
4732efbe JJ |
780 | |
781 | op_ret = 0; | |
782 | list_for_each_entry_safe(this, next, head, list) { | |
783 | if (match_futex (&this->key, &key2)) { | |
784 | wake_futex(this); | |
785 | if (++op_ret >= nr_wake2) | |
786 | break; | |
787 | } | |
788 | } | |
789 | ret += op_ret; | |
790 | } | |
791 | ||
e2970f2f IM |
792 | spin_unlock(&hb1->lock); |
793 | if (hb1 != hb2) | |
794 | spin_unlock(&hb2->lock); | |
4732efbe JJ |
795 | out: |
796 | up_read(¤t->mm->mmap_sem); | |
797 | return ret; | |
798 | } | |
799 | ||
1da177e4 LT |
800 | /* |
801 | * Requeue all waiters hashed on one physical page to another | |
802 | * physical page. | |
803 | */ | |
e2970f2f IM |
804 | static int futex_requeue(u32 __user *uaddr1, u32 __user *uaddr2, |
805 | int nr_wake, int nr_requeue, u32 *cmpval) | |
1da177e4 LT |
806 | { |
807 | union futex_key key1, key2; | |
e2970f2f | 808 | struct futex_hash_bucket *hb1, *hb2; |
1da177e4 LT |
809 | struct list_head *head1; |
810 | struct futex_q *this, *next; | |
811 | int ret, drop_count = 0; | |
812 | ||
813 | retry: | |
814 | down_read(¤t->mm->mmap_sem); | |
815 | ||
816 | ret = get_futex_key(uaddr1, &key1); | |
817 | if (unlikely(ret != 0)) | |
818 | goto out; | |
819 | ret = get_futex_key(uaddr2, &key2); | |
820 | if (unlikely(ret != 0)) | |
821 | goto out; | |
822 | ||
e2970f2f IM |
823 | hb1 = hash_futex(&key1); |
824 | hb2 = hash_futex(&key2); | |
1da177e4 | 825 | |
8b8f319f | 826 | double_lock_hb(hb1, hb2); |
1da177e4 | 827 | |
e2970f2f IM |
828 | if (likely(cmpval != NULL)) { |
829 | u32 curval; | |
1da177e4 | 830 | |
e2970f2f | 831 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
832 | |
833 | if (unlikely(ret)) { | |
e2970f2f IM |
834 | spin_unlock(&hb1->lock); |
835 | if (hb1 != hb2) | |
836 | spin_unlock(&hb2->lock); | |
1da177e4 | 837 | |
e2970f2f IM |
838 | /* |
839 | * If we would have faulted, release mmap_sem, fault | |
1da177e4 LT |
840 | * it in and start all over again. |
841 | */ | |
842 | up_read(¤t->mm->mmap_sem); | |
843 | ||
e2970f2f | 844 | ret = get_user(curval, uaddr1); |
1da177e4 LT |
845 | |
846 | if (!ret) | |
847 | goto retry; | |
848 | ||
849 | return ret; | |
850 | } | |
e2970f2f | 851 | if (curval != *cmpval) { |
1da177e4 LT |
852 | ret = -EAGAIN; |
853 | goto out_unlock; | |
854 | } | |
855 | } | |
856 | ||
e2970f2f | 857 | head1 = &hb1->chain; |
1da177e4 LT |
858 | list_for_each_entry_safe(this, next, head1, list) { |
859 | if (!match_futex (&this->key, &key1)) | |
860 | continue; | |
861 | if (++ret <= nr_wake) { | |
862 | wake_futex(this); | |
863 | } else { | |
59e0e0ac SD |
864 | /* |
865 | * If key1 and key2 hash to the same bucket, no need to | |
866 | * requeue. | |
867 | */ | |
868 | if (likely(head1 != &hb2->chain)) { | |
869 | list_move_tail(&this->list, &hb2->chain); | |
870 | this->lock_ptr = &hb2->lock; | |
871 | } | |
1da177e4 LT |
872 | this->key = key2; |
873 | get_key_refs(&key2); | |
874 | drop_count++; | |
875 | ||
876 | if (ret - nr_wake >= nr_requeue) | |
877 | break; | |
1da177e4 LT |
878 | } |
879 | } | |
880 | ||
881 | out_unlock: | |
e2970f2f IM |
882 | spin_unlock(&hb1->lock); |
883 | if (hb1 != hb2) | |
884 | spin_unlock(&hb2->lock); | |
1da177e4 LT |
885 | |
886 | /* drop_key_refs() must be called outside the spinlocks. */ | |
887 | while (--drop_count >= 0) | |
888 | drop_key_refs(&key1); | |
889 | ||
890 | out: | |
891 | up_read(¤t->mm->mmap_sem); | |
892 | return ret; | |
893 | } | |
894 | ||
895 | /* The key must be already stored in q->key. */ | |
896 | static inline struct futex_hash_bucket * | |
897 | queue_lock(struct futex_q *q, int fd, struct file *filp) | |
898 | { | |
e2970f2f | 899 | struct futex_hash_bucket *hb; |
1da177e4 LT |
900 | |
901 | q->fd = fd; | |
902 | q->filp = filp; | |
903 | ||
904 | init_waitqueue_head(&q->waiters); | |
905 | ||
906 | get_key_refs(&q->key); | |
e2970f2f IM |
907 | hb = hash_futex(&q->key); |
908 | q->lock_ptr = &hb->lock; | |
1da177e4 | 909 | |
e2970f2f IM |
910 | spin_lock(&hb->lock); |
911 | return hb; | |
1da177e4 LT |
912 | } |
913 | ||
e2970f2f | 914 | static inline void __queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 915 | { |
e2970f2f | 916 | list_add_tail(&q->list, &hb->chain); |
c87e2837 | 917 | q->task = current; |
e2970f2f | 918 | spin_unlock(&hb->lock); |
1da177e4 LT |
919 | } |
920 | ||
921 | static inline void | |
e2970f2f | 922 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 923 | { |
e2970f2f | 924 | spin_unlock(&hb->lock); |
1da177e4 LT |
925 | drop_key_refs(&q->key); |
926 | } | |
927 | ||
928 | /* | |
929 | * queue_me and unqueue_me must be called as a pair, each | |
930 | * exactly once. They are called with the hashed spinlock held. | |
931 | */ | |
932 | ||
933 | /* The key must be already stored in q->key. */ | |
934 | static void queue_me(struct futex_q *q, int fd, struct file *filp) | |
935 | { | |
e2970f2f IM |
936 | struct futex_hash_bucket *hb; |
937 | ||
938 | hb = queue_lock(q, fd, filp); | |
939 | __queue_me(q, hb); | |
1da177e4 LT |
940 | } |
941 | ||
942 | /* Return 1 if we were still queued (ie. 0 means we were woken) */ | |
943 | static int unqueue_me(struct futex_q *q) | |
944 | { | |
1da177e4 | 945 | spinlock_t *lock_ptr; |
e2970f2f | 946 | int ret = 0; |
1da177e4 LT |
947 | |
948 | /* In the common case we don't take the spinlock, which is nice. */ | |
949 | retry: | |
950 | lock_ptr = q->lock_ptr; | |
951 | if (lock_ptr != 0) { | |
952 | spin_lock(lock_ptr); | |
953 | /* | |
954 | * q->lock_ptr can change between reading it and | |
955 | * spin_lock(), causing us to take the wrong lock. This | |
956 | * corrects the race condition. | |
957 | * | |
958 | * Reasoning goes like this: if we have the wrong lock, | |
959 | * q->lock_ptr must have changed (maybe several times) | |
960 | * between reading it and the spin_lock(). It can | |
961 | * change again after the spin_lock() but only if it was | |
962 | * already changed before the spin_lock(). It cannot, | |
963 | * however, change back to the original value. Therefore | |
964 | * we can detect whether we acquired the correct lock. | |
965 | */ | |
966 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
967 | spin_unlock(lock_ptr); | |
968 | goto retry; | |
969 | } | |
970 | WARN_ON(list_empty(&q->list)); | |
971 | list_del(&q->list); | |
c87e2837 IM |
972 | |
973 | BUG_ON(q->pi_state); | |
974 | ||
1da177e4 LT |
975 | spin_unlock(lock_ptr); |
976 | ret = 1; | |
977 | } | |
978 | ||
979 | drop_key_refs(&q->key); | |
980 | return ret; | |
981 | } | |
982 | ||
c87e2837 IM |
983 | /* |
984 | * PI futexes can not be requeued and must remove themself from the | |
985 | * hash bucket. The hash bucket lock is held on entry and dropped here. | |
986 | */ | |
987 | static void unqueue_me_pi(struct futex_q *q, struct futex_hash_bucket *hb) | |
988 | { | |
989 | WARN_ON(list_empty(&q->list)); | |
990 | list_del(&q->list); | |
991 | ||
992 | BUG_ON(!q->pi_state); | |
993 | free_pi_state(q->pi_state); | |
994 | q->pi_state = NULL; | |
995 | ||
996 | spin_unlock(&hb->lock); | |
997 | ||
998 | drop_key_refs(&q->key); | |
999 | } | |
1000 | ||
e2970f2f | 1001 | static int futex_wait(u32 __user *uaddr, u32 val, unsigned long time) |
1da177e4 | 1002 | { |
c87e2837 IM |
1003 | struct task_struct *curr = current; |
1004 | DECLARE_WAITQUEUE(wait, curr); | |
e2970f2f | 1005 | struct futex_hash_bucket *hb; |
1da177e4 | 1006 | struct futex_q q; |
e2970f2f IM |
1007 | u32 uval; |
1008 | int ret; | |
1da177e4 | 1009 | |
c87e2837 | 1010 | q.pi_state = NULL; |
1da177e4 | 1011 | retry: |
c87e2837 | 1012 | down_read(&curr->mm->mmap_sem); |
1da177e4 LT |
1013 | |
1014 | ret = get_futex_key(uaddr, &q.key); | |
1015 | if (unlikely(ret != 0)) | |
1016 | goto out_release_sem; | |
1017 | ||
e2970f2f | 1018 | hb = queue_lock(&q, -1, NULL); |
1da177e4 LT |
1019 | |
1020 | /* | |
1021 | * Access the page AFTER the futex is queued. | |
1022 | * Order is important: | |
1023 | * | |
1024 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
1025 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
1026 | * | |
1027 | * The basic logical guarantee of a futex is that it blocks ONLY | |
1028 | * if cond(var) is known to be true at the time of blocking, for | |
1029 | * any cond. If we queued after testing *uaddr, that would open | |
1030 | * a race condition where we could block indefinitely with | |
1031 | * cond(var) false, which would violate the guarantee. | |
1032 | * | |
1033 | * A consequence is that futex_wait() can return zero and absorb | |
1034 | * a wakeup when *uaddr != val on entry to the syscall. This is | |
1035 | * rare, but normal. | |
1036 | * | |
1037 | * We hold the mmap semaphore, so the mapping cannot have changed | |
1038 | * since we looked it up in get_futex_key. | |
1039 | */ | |
e2970f2f | 1040 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 LT |
1041 | |
1042 | if (unlikely(ret)) { | |
e2970f2f | 1043 | queue_unlock(&q, hb); |
1da177e4 | 1044 | |
e2970f2f IM |
1045 | /* |
1046 | * If we would have faulted, release mmap_sem, fault it in and | |
1da177e4 LT |
1047 | * start all over again. |
1048 | */ | |
c87e2837 | 1049 | up_read(&curr->mm->mmap_sem); |
1da177e4 | 1050 | |
e2970f2f | 1051 | ret = get_user(uval, uaddr); |
1da177e4 LT |
1052 | |
1053 | if (!ret) | |
1054 | goto retry; | |
1055 | return ret; | |
1056 | } | |
c87e2837 IM |
1057 | ret = -EWOULDBLOCK; |
1058 | if (uval != val) | |
1059 | goto out_unlock_release_sem; | |
1da177e4 LT |
1060 | |
1061 | /* Only actually queue if *uaddr contained val. */ | |
e2970f2f | 1062 | __queue_me(&q, hb); |
1da177e4 LT |
1063 | |
1064 | /* | |
1065 | * Now the futex is queued and we have checked the data, we | |
1066 | * don't want to hold mmap_sem while we sleep. | |
c87e2837 IM |
1067 | */ |
1068 | up_read(&curr->mm->mmap_sem); | |
1da177e4 LT |
1069 | |
1070 | /* | |
1071 | * There might have been scheduling since the queue_me(), as we | |
1072 | * cannot hold a spinlock across the get_user() in case it | |
1073 | * faults, and we cannot just set TASK_INTERRUPTIBLE state when | |
1074 | * queueing ourselves into the futex hash. This code thus has to | |
1075 | * rely on the futex_wake() code removing us from hash when it | |
1076 | * wakes us up. | |
1077 | */ | |
1078 | ||
1079 | /* add_wait_queue is the barrier after __set_current_state. */ | |
1080 | __set_current_state(TASK_INTERRUPTIBLE); | |
1081 | add_wait_queue(&q.waiters, &wait); | |
1082 | /* | |
1083 | * !list_empty() is safe here without any lock. | |
1084 | * q.lock_ptr != 0 is not safe, because of ordering against wakeup. | |
1085 | */ | |
1086 | if (likely(!list_empty(&q.list))) | |
1087 | time = schedule_timeout(time); | |
1088 | __set_current_state(TASK_RUNNING); | |
1089 | ||
1090 | /* | |
1091 | * NOTE: we don't remove ourselves from the waitqueue because | |
1092 | * we are the only user of it. | |
1093 | */ | |
1094 | ||
1095 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
1096 | if (!unqueue_me(&q)) | |
1097 | return 0; | |
1098 | if (time == 0) | |
1099 | return -ETIMEDOUT; | |
e2970f2f IM |
1100 | /* |
1101 | * We expect signal_pending(current), but another thread may | |
1102 | * have handled it for us already. | |
1103 | */ | |
1da177e4 LT |
1104 | return -EINTR; |
1105 | ||
c87e2837 IM |
1106 | out_unlock_release_sem: |
1107 | queue_unlock(&q, hb); | |
1108 | ||
1da177e4 | 1109 | out_release_sem: |
c87e2837 IM |
1110 | up_read(&curr->mm->mmap_sem); |
1111 | return ret; | |
1112 | } | |
1113 | ||
1114 | /* | |
1115 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
1116 | * and failed. The kernel side here does the whole locking operation: | |
1117 | * if there are waiters then it will block, it does PI, etc. (Due to | |
1118 | * races the kernel might see a 0 value of the futex too.) | |
1119 | */ | |
1120 | static int do_futex_lock_pi(u32 __user *uaddr, int detect, int trylock, | |
1121 | struct hrtimer_sleeper *to) | |
1122 | { | |
1123 | struct task_struct *curr = current; | |
1124 | struct futex_hash_bucket *hb; | |
1125 | u32 uval, newval, curval; | |
1126 | struct futex_q q; | |
1127 | int ret, attempt = 0; | |
1128 | ||
1129 | if (refill_pi_state_cache()) | |
1130 | return -ENOMEM; | |
1131 | ||
1132 | q.pi_state = NULL; | |
1133 | retry: | |
1134 | down_read(&curr->mm->mmap_sem); | |
1135 | ||
1136 | ret = get_futex_key(uaddr, &q.key); | |
1137 | if (unlikely(ret != 0)) | |
1138 | goto out_release_sem; | |
1139 | ||
1140 | hb = queue_lock(&q, -1, NULL); | |
1141 | ||
1142 | retry_locked: | |
1143 | /* | |
1144 | * To avoid races, we attempt to take the lock here again | |
1145 | * (by doing a 0 -> TID atomic cmpxchg), while holding all | |
1146 | * the locks. It will most likely not succeed. | |
1147 | */ | |
1148 | newval = current->pid; | |
1149 | ||
1150 | inc_preempt_count(); | |
1151 | curval = futex_atomic_cmpxchg_inatomic(uaddr, 0, newval); | |
1152 | dec_preempt_count(); | |
1153 | ||
1154 | if (unlikely(curval == -EFAULT)) | |
1155 | goto uaddr_faulted; | |
1156 | ||
1157 | /* We own the lock already */ | |
1158 | if (unlikely((curval & FUTEX_TID_MASK) == current->pid)) { | |
1159 | if (!detect && 0) | |
1160 | force_sig(SIGKILL, current); | |
1161 | ret = -EDEADLK; | |
1162 | goto out_unlock_release_sem; | |
1163 | } | |
1164 | ||
1165 | /* | |
1166 | * Surprise - we got the lock. Just return | |
1167 | * to userspace: | |
1168 | */ | |
1169 | if (unlikely(!curval)) | |
1170 | goto out_unlock_release_sem; | |
1171 | ||
1172 | uval = curval; | |
1173 | newval = uval | FUTEX_WAITERS; | |
1174 | ||
1175 | inc_preempt_count(); | |
1176 | curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval); | |
1177 | dec_preempt_count(); | |
1178 | ||
1179 | if (unlikely(curval == -EFAULT)) | |
1180 | goto uaddr_faulted; | |
1181 | if (unlikely(curval != uval)) | |
1182 | goto retry_locked; | |
1183 | ||
1184 | /* | |
1185 | * We dont have the lock. Look up the PI state (or create it if | |
1186 | * we are the first waiter): | |
1187 | */ | |
1188 | ret = lookup_pi_state(uval, hb, &q); | |
1189 | ||
1190 | if (unlikely(ret)) { | |
1191 | /* | |
1192 | * There were no waiters and the owner task lookup | |
1193 | * failed. When the OWNER_DIED bit is set, then we | |
1194 | * know that this is a robust futex and we actually | |
1195 | * take the lock. This is safe as we are protected by | |
1196 | * the hash bucket lock. We also set the waiters bit | |
1197 | * unconditionally here, to simplify glibc handling of | |
1198 | * multiple tasks racing to acquire the lock and | |
1199 | * cleanup the problems which were left by the dead | |
1200 | * owner. | |
1201 | */ | |
1202 | if (curval & FUTEX_OWNER_DIED) { | |
1203 | uval = newval; | |
1204 | newval = current->pid | | |
1205 | FUTEX_OWNER_DIED | FUTEX_WAITERS; | |
1206 | ||
1207 | inc_preempt_count(); | |
1208 | curval = futex_atomic_cmpxchg_inatomic(uaddr, | |
1209 | uval, newval); | |
1210 | dec_preempt_count(); | |
1211 | ||
1212 | if (unlikely(curval == -EFAULT)) | |
1213 | goto uaddr_faulted; | |
1214 | if (unlikely(curval != uval)) | |
1215 | goto retry_locked; | |
1216 | ret = 0; | |
1217 | } | |
1218 | goto out_unlock_release_sem; | |
1219 | } | |
1220 | ||
1221 | /* | |
1222 | * Only actually queue now that the atomic ops are done: | |
1223 | */ | |
1224 | __queue_me(&q, hb); | |
1225 | ||
1226 | /* | |
1227 | * Now the futex is queued and we have checked the data, we | |
1228 | * don't want to hold mmap_sem while we sleep. | |
1229 | */ | |
1230 | up_read(&curr->mm->mmap_sem); | |
1231 | ||
1232 | WARN_ON(!q.pi_state); | |
1233 | /* | |
1234 | * Block on the PI mutex: | |
1235 | */ | |
1236 | if (!trylock) | |
1237 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); | |
1238 | else { | |
1239 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); | |
1240 | /* Fixup the trylock return value: */ | |
1241 | ret = ret ? 0 : -EWOULDBLOCK; | |
1242 | } | |
1243 | ||
1244 | down_read(&curr->mm->mmap_sem); | |
a99e4e41 | 1245 | spin_lock(q.lock_ptr); |
c87e2837 IM |
1246 | |
1247 | /* | |
1248 | * Got the lock. We might not be the anticipated owner if we | |
1249 | * did a lock-steal - fix up the PI-state in that case. | |
1250 | */ | |
1251 | if (!ret && q.pi_state->owner != curr) { | |
1252 | u32 newtid = current->pid | FUTEX_WAITERS; | |
1253 | ||
1254 | /* Owner died? */ | |
1255 | if (q.pi_state->owner != NULL) { | |
1256 | spin_lock_irq(&q.pi_state->owner->pi_lock); | |
627371d7 | 1257 | WARN_ON(list_empty(&q.pi_state->list)); |
c87e2837 IM |
1258 | list_del_init(&q.pi_state->list); |
1259 | spin_unlock_irq(&q.pi_state->owner->pi_lock); | |
1260 | } else | |
1261 | newtid |= FUTEX_OWNER_DIED; | |
1262 | ||
1263 | q.pi_state->owner = current; | |
1264 | ||
1265 | spin_lock_irq(¤t->pi_lock); | |
627371d7 | 1266 | WARN_ON(!list_empty(&q.pi_state->list)); |
c87e2837 IM |
1267 | list_add(&q.pi_state->list, ¤t->pi_state_list); |
1268 | spin_unlock_irq(¤t->pi_lock); | |
1269 | ||
1270 | /* Unqueue and drop the lock */ | |
1271 | unqueue_me_pi(&q, hb); | |
1272 | up_read(&curr->mm->mmap_sem); | |
1273 | /* | |
1274 | * We own it, so we have to replace the pending owner | |
1275 | * TID. This must be atomic as we have preserve the | |
1276 | * owner died bit here. | |
1277 | */ | |
1278 | ret = get_user(uval, uaddr); | |
1279 | while (!ret) { | |
1280 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
1281 | curval = futex_atomic_cmpxchg_inatomic(uaddr, | |
1282 | uval, newval); | |
1283 | if (curval == -EFAULT) | |
1284 | ret = -EFAULT; | |
1285 | if (curval == uval) | |
1286 | break; | |
1287 | uval = curval; | |
1288 | } | |
1289 | } else { | |
1290 | /* | |
1291 | * Catch the rare case, where the lock was released | |
1292 | * when we were on the way back before we locked | |
1293 | * the hash bucket. | |
1294 | */ | |
1295 | if (ret && q.pi_state->owner == curr) { | |
1296 | if (rt_mutex_trylock(&q.pi_state->pi_mutex)) | |
1297 | ret = 0; | |
1298 | } | |
1299 | /* Unqueue and drop the lock */ | |
1300 | unqueue_me_pi(&q, hb); | |
1301 | up_read(&curr->mm->mmap_sem); | |
1302 | } | |
1303 | ||
1304 | if (!detect && ret == -EDEADLK && 0) | |
1305 | force_sig(SIGKILL, current); | |
1306 | ||
1307 | return ret; | |
1308 | ||
1309 | out_unlock_release_sem: | |
1310 | queue_unlock(&q, hb); | |
1311 | ||
1312 | out_release_sem: | |
1313 | up_read(&curr->mm->mmap_sem); | |
1314 | return ret; | |
1315 | ||
1316 | uaddr_faulted: | |
1317 | /* | |
1318 | * We have to r/w *(int __user *)uaddr, but we can't modify it | |
1319 | * non-atomically. Therefore, if get_user below is not | |
1320 | * enough, we need to handle the fault ourselves, while | |
1321 | * still holding the mmap_sem. | |
1322 | */ | |
1323 | if (attempt++) { | |
1324 | if (futex_handle_fault((unsigned long)uaddr, attempt)) | |
1325 | goto out_unlock_release_sem; | |
1326 | ||
1327 | goto retry_locked; | |
1328 | } | |
1329 | ||
1330 | queue_unlock(&q, hb); | |
1331 | up_read(&curr->mm->mmap_sem); | |
1332 | ||
1333 | ret = get_user(uval, uaddr); | |
1334 | if (!ret && (uval != -EFAULT)) | |
1335 | goto retry; | |
1336 | ||
1337 | return ret; | |
1338 | } | |
1339 | ||
1340 | /* | |
1341 | * Restart handler | |
1342 | */ | |
1343 | static long futex_lock_pi_restart(struct restart_block *restart) | |
1344 | { | |
1345 | struct hrtimer_sleeper timeout, *to = NULL; | |
1346 | int ret; | |
1347 | ||
1348 | restart->fn = do_no_restart_syscall; | |
1349 | ||
1350 | if (restart->arg2 || restart->arg3) { | |
1351 | to = &timeout; | |
1352 | hrtimer_init(&to->timer, CLOCK_REALTIME, HRTIMER_ABS); | |
1353 | hrtimer_init_sleeper(to, current); | |
1354 | to->timer.expires.tv64 = ((u64)restart->arg1 << 32) | | |
1355 | (u64) restart->arg0; | |
1356 | } | |
1357 | ||
1358 | pr_debug("lock_pi restart: %p, %d (%d)\n", | |
1359 | (u32 __user *)restart->arg0, current->pid); | |
1360 | ||
1361 | ret = do_futex_lock_pi((u32 __user *)restart->arg0, restart->arg1, | |
1362 | 0, to); | |
1363 | ||
1364 | if (ret != -EINTR) | |
1365 | return ret; | |
1366 | ||
1367 | restart->fn = futex_lock_pi_restart; | |
1368 | ||
1369 | /* The other values are filled in */ | |
1370 | return -ERESTART_RESTARTBLOCK; | |
1371 | } | |
1372 | ||
1373 | /* | |
1374 | * Called from the syscall entry below. | |
1375 | */ | |
1376 | static int futex_lock_pi(u32 __user *uaddr, int detect, unsigned long sec, | |
1377 | long nsec, int trylock) | |
1378 | { | |
1379 | struct hrtimer_sleeper timeout, *to = NULL; | |
1380 | struct restart_block *restart; | |
1381 | int ret; | |
1382 | ||
1383 | if (sec != MAX_SCHEDULE_TIMEOUT) { | |
1384 | to = &timeout; | |
1385 | hrtimer_init(&to->timer, CLOCK_REALTIME, HRTIMER_ABS); | |
1386 | hrtimer_init_sleeper(to, current); | |
1387 | to->timer.expires = ktime_set(sec, nsec); | |
1388 | } | |
1389 | ||
1390 | ret = do_futex_lock_pi(uaddr, detect, trylock, to); | |
1391 | ||
1392 | if (ret != -EINTR) | |
1393 | return ret; | |
1394 | ||
1395 | pr_debug("lock_pi interrupted: %p, %d (%d)\n", uaddr, current->pid); | |
1396 | ||
1397 | restart = ¤t_thread_info()->restart_block; | |
1398 | restart->fn = futex_lock_pi_restart; | |
1399 | restart->arg0 = (unsigned long) uaddr; | |
1400 | restart->arg1 = detect; | |
1401 | if (to) { | |
1402 | restart->arg2 = to->timer.expires.tv64 & 0xFFFFFFFF; | |
1403 | restart->arg3 = to->timer.expires.tv64 >> 32; | |
1404 | } else | |
1405 | restart->arg2 = restart->arg3 = 0; | |
1406 | ||
1407 | return -ERESTART_RESTARTBLOCK; | |
1408 | } | |
1409 | ||
1410 | /* | |
1411 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
1412 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
1413 | * and do the rt-mutex unlock. | |
1414 | */ | |
1415 | static int futex_unlock_pi(u32 __user *uaddr) | |
1416 | { | |
1417 | struct futex_hash_bucket *hb; | |
1418 | struct futex_q *this, *next; | |
1419 | u32 uval; | |
1420 | struct list_head *head; | |
1421 | union futex_key key; | |
1422 | int ret, attempt = 0; | |
1423 | ||
1424 | retry: | |
1425 | if (get_user(uval, uaddr)) | |
1426 | return -EFAULT; | |
1427 | /* | |
1428 | * We release only a lock we actually own: | |
1429 | */ | |
1430 | if ((uval & FUTEX_TID_MASK) != current->pid) | |
1431 | return -EPERM; | |
1432 | /* | |
1433 | * First take all the futex related locks: | |
1434 | */ | |
1435 | down_read(¤t->mm->mmap_sem); | |
1436 | ||
1437 | ret = get_futex_key(uaddr, &key); | |
1438 | if (unlikely(ret != 0)) | |
1439 | goto out; | |
1440 | ||
1441 | hb = hash_futex(&key); | |
1442 | spin_lock(&hb->lock); | |
1443 | ||
1444 | retry_locked: | |
1445 | /* | |
1446 | * To avoid races, try to do the TID -> 0 atomic transition | |
1447 | * again. If it succeeds then we can return without waking | |
1448 | * anyone else up: | |
1449 | */ | |
e3f2ddea IM |
1450 | if (!(uval & FUTEX_OWNER_DIED)) { |
1451 | inc_preempt_count(); | |
1452 | uval = futex_atomic_cmpxchg_inatomic(uaddr, current->pid, 0); | |
1453 | dec_preempt_count(); | |
1454 | } | |
c87e2837 IM |
1455 | |
1456 | if (unlikely(uval == -EFAULT)) | |
1457 | goto pi_faulted; | |
1458 | /* | |
1459 | * Rare case: we managed to release the lock atomically, | |
1460 | * no need to wake anyone else up: | |
1461 | */ | |
1462 | if (unlikely(uval == current->pid)) | |
1463 | goto out_unlock; | |
1464 | ||
1465 | /* | |
1466 | * Ok, other tasks may need to be woken up - check waiters | |
1467 | * and do the wakeup if necessary: | |
1468 | */ | |
1469 | head = &hb->chain; | |
1470 | ||
1471 | list_for_each_entry_safe(this, next, head, list) { | |
1472 | if (!match_futex (&this->key, &key)) | |
1473 | continue; | |
1474 | ret = wake_futex_pi(uaddr, uval, this); | |
1475 | /* | |
1476 | * The atomic access to the futex value | |
1477 | * generated a pagefault, so retry the | |
1478 | * user-access and the wakeup: | |
1479 | */ | |
1480 | if (ret == -EFAULT) | |
1481 | goto pi_faulted; | |
1482 | goto out_unlock; | |
1483 | } | |
1484 | /* | |
1485 | * No waiters - kernel unlocks the futex: | |
1486 | */ | |
e3f2ddea IM |
1487 | if (!(uval & FUTEX_OWNER_DIED)) { |
1488 | ret = unlock_futex_pi(uaddr, uval); | |
1489 | if (ret == -EFAULT) | |
1490 | goto pi_faulted; | |
1491 | } | |
c87e2837 IM |
1492 | |
1493 | out_unlock: | |
1494 | spin_unlock(&hb->lock); | |
1495 | out: | |
1496 | up_read(¤t->mm->mmap_sem); | |
1497 | ||
1498 | return ret; | |
1499 | ||
1500 | pi_faulted: | |
1501 | /* | |
1502 | * We have to r/w *(int __user *)uaddr, but we can't modify it | |
1503 | * non-atomically. Therefore, if get_user below is not | |
1504 | * enough, we need to handle the fault ourselves, while | |
1505 | * still holding the mmap_sem. | |
1506 | */ | |
1507 | if (attempt++) { | |
1508 | if (futex_handle_fault((unsigned long)uaddr, attempt)) | |
1509 | goto out_unlock; | |
1510 | ||
1511 | goto retry_locked; | |
1512 | } | |
1513 | ||
1514 | spin_unlock(&hb->lock); | |
1da177e4 | 1515 | up_read(¤t->mm->mmap_sem); |
c87e2837 IM |
1516 | |
1517 | ret = get_user(uval, uaddr); | |
1518 | if (!ret && (uval != -EFAULT)) | |
1519 | goto retry; | |
1520 | ||
1da177e4 LT |
1521 | return ret; |
1522 | } | |
1523 | ||
1524 | static int futex_close(struct inode *inode, struct file *filp) | |
1525 | { | |
1526 | struct futex_q *q = filp->private_data; | |
1527 | ||
1528 | unqueue_me(q); | |
1529 | kfree(q); | |
e2970f2f | 1530 | |
1da177e4 LT |
1531 | return 0; |
1532 | } | |
1533 | ||
1534 | /* This is one-shot: once it's gone off you need a new fd */ | |
1535 | static unsigned int futex_poll(struct file *filp, | |
1536 | struct poll_table_struct *wait) | |
1537 | { | |
1538 | struct futex_q *q = filp->private_data; | |
1539 | int ret = 0; | |
1540 | ||
1541 | poll_wait(filp, &q->waiters, wait); | |
1542 | ||
1543 | /* | |
1544 | * list_empty() is safe here without any lock. | |
1545 | * q->lock_ptr != 0 is not safe, because of ordering against wakeup. | |
1546 | */ | |
1547 | if (list_empty(&q->list)) | |
1548 | ret = POLLIN | POLLRDNORM; | |
1549 | ||
1550 | return ret; | |
1551 | } | |
1552 | ||
1553 | static struct file_operations futex_fops = { | |
1554 | .release = futex_close, | |
1555 | .poll = futex_poll, | |
1556 | }; | |
1557 | ||
1558 | /* | |
1559 | * Signal allows caller to avoid the race which would occur if they | |
1560 | * set the sigio stuff up afterwards. | |
1561 | */ | |
e2970f2f | 1562 | static int futex_fd(u32 __user *uaddr, int signal) |
1da177e4 LT |
1563 | { |
1564 | struct futex_q *q; | |
1565 | struct file *filp; | |
1566 | int ret, err; | |
1567 | ||
1568 | ret = -EINVAL; | |
7ed20e1a | 1569 | if (!valid_signal(signal)) |
1da177e4 LT |
1570 | goto out; |
1571 | ||
1572 | ret = get_unused_fd(); | |
1573 | if (ret < 0) | |
1574 | goto out; | |
1575 | filp = get_empty_filp(); | |
1576 | if (!filp) { | |
1577 | put_unused_fd(ret); | |
1578 | ret = -ENFILE; | |
1579 | goto out; | |
1580 | } | |
1581 | filp->f_op = &futex_fops; | |
1582 | filp->f_vfsmnt = mntget(futex_mnt); | |
1583 | filp->f_dentry = dget(futex_mnt->mnt_root); | |
1584 | filp->f_mapping = filp->f_dentry->d_inode->i_mapping; | |
1585 | ||
1586 | if (signal) { | |
1da177e4 LT |
1587 | err = f_setown(filp, current->pid, 1); |
1588 | if (err < 0) { | |
39ed3fde | 1589 | goto error; |
1da177e4 LT |
1590 | } |
1591 | filp->f_owner.signum = signal; | |
1592 | } | |
1593 | ||
1594 | q = kmalloc(sizeof(*q), GFP_KERNEL); | |
1595 | if (!q) { | |
39ed3fde PE |
1596 | err = -ENOMEM; |
1597 | goto error; | |
1da177e4 | 1598 | } |
c87e2837 | 1599 | q->pi_state = NULL; |
1da177e4 LT |
1600 | |
1601 | down_read(¤t->mm->mmap_sem); | |
1602 | err = get_futex_key(uaddr, &q->key); | |
1603 | ||
1604 | if (unlikely(err != 0)) { | |
1605 | up_read(¤t->mm->mmap_sem); | |
1da177e4 | 1606 | kfree(q); |
39ed3fde | 1607 | goto error; |
1da177e4 LT |
1608 | } |
1609 | ||
1610 | /* | |
1611 | * queue_me() must be called before releasing mmap_sem, because | |
1612 | * key->shared.inode needs to be referenced while holding it. | |
1613 | */ | |
1614 | filp->private_data = q; | |
1615 | ||
1616 | queue_me(q, ret, filp); | |
1617 | up_read(¤t->mm->mmap_sem); | |
1618 | ||
1619 | /* Now we map fd to filp, so userspace can access it */ | |
1620 | fd_install(ret, filp); | |
1621 | out: | |
1622 | return ret; | |
39ed3fde PE |
1623 | error: |
1624 | put_unused_fd(ret); | |
1625 | put_filp(filp); | |
1626 | ret = err; | |
1627 | goto out; | |
1da177e4 LT |
1628 | } |
1629 | ||
0771dfef IM |
1630 | /* |
1631 | * Support for robust futexes: the kernel cleans up held futexes at | |
1632 | * thread exit time. | |
1633 | * | |
1634 | * Implementation: user-space maintains a per-thread list of locks it | |
1635 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
1636 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 1637 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
1638 | * always manipulated with the lock held, so the list is private and |
1639 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
1640 | * field, to allow the kernel to clean up if the thread dies after | |
1641 | * acquiring the lock, but just before it could have added itself to | |
1642 | * the list. There can only be one such pending lock. | |
1643 | */ | |
1644 | ||
1645 | /** | |
1646 | * sys_set_robust_list - set the robust-futex list head of a task | |
1647 | * @head: pointer to the list-head | |
1648 | * @len: length of the list-head, as userspace expects | |
1649 | */ | |
1650 | asmlinkage long | |
1651 | sys_set_robust_list(struct robust_list_head __user *head, | |
1652 | size_t len) | |
1653 | { | |
1654 | /* | |
1655 | * The kernel knows only one size for now: | |
1656 | */ | |
1657 | if (unlikely(len != sizeof(*head))) | |
1658 | return -EINVAL; | |
1659 | ||
1660 | current->robust_list = head; | |
1661 | ||
1662 | return 0; | |
1663 | } | |
1664 | ||
1665 | /** | |
1666 | * sys_get_robust_list - get the robust-futex list head of a task | |
1667 | * @pid: pid of the process [zero for current task] | |
1668 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
1669 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
1670 | */ | |
1671 | asmlinkage long | |
1672 | sys_get_robust_list(int pid, struct robust_list_head __user **head_ptr, | |
1673 | size_t __user *len_ptr) | |
1674 | { | |
1675 | struct robust_list_head *head; | |
1676 | unsigned long ret; | |
1677 | ||
1678 | if (!pid) | |
1679 | head = current->robust_list; | |
1680 | else { | |
1681 | struct task_struct *p; | |
1682 | ||
1683 | ret = -ESRCH; | |
1684 | read_lock(&tasklist_lock); | |
1685 | p = find_task_by_pid(pid); | |
1686 | if (!p) | |
1687 | goto err_unlock; | |
1688 | ret = -EPERM; | |
1689 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
1690 | !capable(CAP_SYS_PTRACE)) | |
1691 | goto err_unlock; | |
1692 | head = p->robust_list; | |
1693 | read_unlock(&tasklist_lock); | |
1694 | } | |
1695 | ||
1696 | if (put_user(sizeof(*head), len_ptr)) | |
1697 | return -EFAULT; | |
1698 | return put_user(head, head_ptr); | |
1699 | ||
1700 | err_unlock: | |
1701 | read_unlock(&tasklist_lock); | |
1702 | ||
1703 | return ret; | |
1704 | } | |
1705 | ||
1706 | /* | |
1707 | * Process a futex-list entry, check whether it's owned by the | |
1708 | * dying task, and do notification if so: | |
1709 | */ | |
e3f2ddea | 1710 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 1711 | { |
e3f2ddea | 1712 | u32 uval, nval, mval; |
0771dfef | 1713 | |
8f17d3a5 IM |
1714 | retry: |
1715 | if (get_user(uval, uaddr)) | |
0771dfef IM |
1716 | return -1; |
1717 | ||
8f17d3a5 | 1718 | if ((uval & FUTEX_TID_MASK) == curr->pid) { |
0771dfef IM |
1719 | /* |
1720 | * Ok, this dying thread is truly holding a futex | |
1721 | * of interest. Set the OWNER_DIED bit atomically | |
1722 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
1723 | * set, wake up a waiter (if any). (We have to do a | |
1724 | * futex_wake() even if OWNER_DIED is already set - | |
1725 | * to handle the rare but possible case of recursive | |
1726 | * thread-death.) The rest of the cleanup is done in | |
1727 | * userspace. | |
1728 | */ | |
e3f2ddea IM |
1729 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
1730 | nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval); | |
1731 | ||
c87e2837 IM |
1732 | if (nval == -EFAULT) |
1733 | return -1; | |
1734 | ||
1735 | if (nval != uval) | |
8f17d3a5 | 1736 | goto retry; |
0771dfef | 1737 | |
e3f2ddea IM |
1738 | /* |
1739 | * Wake robust non-PI futexes here. The wakeup of | |
1740 | * PI futexes happens in exit_pi_state(): | |
1741 | */ | |
1742 | if (!pi) { | |
1743 | if (uval & FUTEX_WAITERS) | |
1744 | futex_wake(uaddr, 1); | |
1745 | } | |
0771dfef IM |
1746 | } |
1747 | return 0; | |
1748 | } | |
1749 | ||
e3f2ddea IM |
1750 | /* |
1751 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
1752 | */ | |
1753 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
1754 | struct robust_list __user **head, int *pi) | |
1755 | { | |
1756 | unsigned long uentry; | |
1757 | ||
1758 | if (get_user(uentry, (unsigned long *)head)) | |
1759 | return -EFAULT; | |
1760 | ||
1761 | *entry = (void *)(uentry & ~1UL); | |
1762 | *pi = uentry & 1; | |
1763 | ||
1764 | return 0; | |
1765 | } | |
1766 | ||
0771dfef IM |
1767 | /* |
1768 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
1769 | * and mark any locks found there dead, and notify any waiters. | |
1770 | * | |
1771 | * We silently return on any sign of list-walking problem. | |
1772 | */ | |
1773 | void exit_robust_list(struct task_struct *curr) | |
1774 | { | |
1775 | struct robust_list_head __user *head = curr->robust_list; | |
1776 | struct robust_list __user *entry, *pending; | |
e3f2ddea | 1777 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
0771dfef IM |
1778 | unsigned long futex_offset; |
1779 | ||
1780 | /* | |
1781 | * Fetch the list head (which was registered earlier, via | |
1782 | * sys_set_robust_list()): | |
1783 | */ | |
e3f2ddea | 1784 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
1785 | return; |
1786 | /* | |
1787 | * Fetch the relative futex offset: | |
1788 | */ | |
1789 | if (get_user(futex_offset, &head->futex_offset)) | |
1790 | return; | |
1791 | /* | |
1792 | * Fetch any possibly pending lock-add first, and handle it | |
1793 | * if it exists: | |
1794 | */ | |
e3f2ddea | 1795 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 1796 | return; |
e3f2ddea | 1797 | |
0771dfef | 1798 | if (pending) |
e3f2ddea | 1799 | handle_futex_death((void *)pending + futex_offset, curr, pip); |
0771dfef IM |
1800 | |
1801 | while (entry != &head->list) { | |
1802 | /* | |
1803 | * A pending lock might already be on the list, so | |
c87e2837 | 1804 | * don't process it twice: |
0771dfef IM |
1805 | */ |
1806 | if (entry != pending) | |
1807 | if (handle_futex_death((void *)entry + futex_offset, | |
e3f2ddea | 1808 | curr, pi)) |
0771dfef | 1809 | return; |
0771dfef IM |
1810 | /* |
1811 | * Fetch the next entry in the list: | |
1812 | */ | |
e3f2ddea | 1813 | if (fetch_robust_entry(&entry, &entry->next, &pi)) |
0771dfef IM |
1814 | return; |
1815 | /* | |
1816 | * Avoid excessively long or circular lists: | |
1817 | */ | |
1818 | if (!--limit) | |
1819 | break; | |
1820 | ||
1821 | cond_resched(); | |
1822 | } | |
1823 | } | |
1824 | ||
e2970f2f IM |
1825 | long do_futex(u32 __user *uaddr, int op, u32 val, unsigned long timeout, |
1826 | u32 __user *uaddr2, u32 val2, u32 val3) | |
1da177e4 LT |
1827 | { |
1828 | int ret; | |
1829 | ||
1830 | switch (op) { | |
1831 | case FUTEX_WAIT: | |
1832 | ret = futex_wait(uaddr, val, timeout); | |
1833 | break; | |
1834 | case FUTEX_WAKE: | |
1835 | ret = futex_wake(uaddr, val); | |
1836 | break; | |
1837 | case FUTEX_FD: | |
1838 | /* non-zero val means F_SETOWN(getpid()) & F_SETSIG(val) */ | |
1839 | ret = futex_fd(uaddr, val); | |
1840 | break; | |
1841 | case FUTEX_REQUEUE: | |
1842 | ret = futex_requeue(uaddr, uaddr2, val, val2, NULL); | |
1843 | break; | |
1844 | case FUTEX_CMP_REQUEUE: | |
1845 | ret = futex_requeue(uaddr, uaddr2, val, val2, &val3); | |
1846 | break; | |
4732efbe JJ |
1847 | case FUTEX_WAKE_OP: |
1848 | ret = futex_wake_op(uaddr, uaddr2, val, val2, val3); | |
1849 | break; | |
c87e2837 IM |
1850 | case FUTEX_LOCK_PI: |
1851 | ret = futex_lock_pi(uaddr, val, timeout, val2, 0); | |
1852 | break; | |
1853 | case FUTEX_UNLOCK_PI: | |
1854 | ret = futex_unlock_pi(uaddr); | |
1855 | break; | |
1856 | case FUTEX_TRYLOCK_PI: | |
1857 | ret = futex_lock_pi(uaddr, 0, timeout, val2, 1); | |
1858 | break; | |
1da177e4 LT |
1859 | default: |
1860 | ret = -ENOSYS; | |
1861 | } | |
1862 | return ret; | |
1863 | } | |
1864 | ||
1865 | ||
e2970f2f | 1866 | asmlinkage long sys_futex(u32 __user *uaddr, int op, u32 val, |
1da177e4 | 1867 | struct timespec __user *utime, u32 __user *uaddr2, |
e2970f2f | 1868 | u32 val3) |
1da177e4 LT |
1869 | { |
1870 | struct timespec t; | |
1871 | unsigned long timeout = MAX_SCHEDULE_TIMEOUT; | |
e2970f2f | 1872 | u32 val2 = 0; |
1da177e4 | 1873 | |
c87e2837 | 1874 | if (utime && (op == FUTEX_WAIT || op == FUTEX_LOCK_PI)) { |
1da177e4 LT |
1875 | if (copy_from_user(&t, utime, sizeof(t)) != 0) |
1876 | return -EFAULT; | |
9741ef96 TG |
1877 | if (!timespec_valid(&t)) |
1878 | return -EINVAL; | |
c87e2837 IM |
1879 | if (op == FUTEX_WAIT) |
1880 | timeout = timespec_to_jiffies(&t) + 1; | |
1881 | else { | |
1882 | timeout = t.tv_sec; | |
1883 | val2 = t.tv_nsec; | |
1884 | } | |
1da177e4 LT |
1885 | } |
1886 | /* | |
1887 | * requeue parameter in 'utime' if op == FUTEX_REQUEUE. | |
1888 | */ | |
c87e2837 | 1889 | if (op == FUTEX_REQUEUE || op == FUTEX_CMP_REQUEUE) |
e2970f2f | 1890 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 1891 | |
e2970f2f | 1892 | return do_futex(uaddr, op, val, timeout, uaddr2, val2, val3); |
1da177e4 LT |
1893 | } |
1894 | ||
454e2398 DH |
1895 | static int futexfs_get_sb(struct file_system_type *fs_type, |
1896 | int flags, const char *dev_name, void *data, | |
1897 | struct vfsmount *mnt) | |
1da177e4 | 1898 | { |
454e2398 | 1899 | return get_sb_pseudo(fs_type, "futex", NULL, 0xBAD1DEA, mnt); |
1da177e4 LT |
1900 | } |
1901 | ||
1902 | static struct file_system_type futex_fs_type = { | |
1903 | .name = "futexfs", | |
1904 | .get_sb = futexfs_get_sb, | |
1905 | .kill_sb = kill_anon_super, | |
1906 | }; | |
1907 | ||
1908 | static int __init init(void) | |
1909 | { | |
1910 | unsigned int i; | |
1911 | ||
1912 | register_filesystem(&futex_fs_type); | |
1913 | futex_mnt = kern_mount(&futex_fs_type); | |
1914 | ||
1915 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { | |
1916 | INIT_LIST_HEAD(&futex_queues[i].chain); | |
1917 | spin_lock_init(&futex_queues[i].lock); | |
1918 | } | |
1919 | return 0; | |
1920 | } | |
1921 | __initcall(init); |