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 | * | |
34f01cc1 ED |
19 | * PRIVATE futexes by Eric Dumazet |
20 | * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com> | |
21 | * | |
1da177e4 LT |
22 | * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly |
23 | * enough at me, Linus for the original (flawed) idea, Matthew | |
24 | * Kirkwood for proof-of-concept implementation. | |
25 | * | |
26 | * "The futexes are also cursed." | |
27 | * "But they come in a choice of three flavours!" | |
28 | * | |
29 | * This program is free software; you can redistribute it and/or modify | |
30 | * it under the terms of the GNU General Public License as published by | |
31 | * the Free Software Foundation; either version 2 of the License, or | |
32 | * (at your option) any later version. | |
33 | * | |
34 | * This program is distributed in the hope that it will be useful, | |
35 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
36 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
37 | * GNU General Public License for more details. | |
38 | * | |
39 | * You should have received a copy of the GNU General Public License | |
40 | * along with this program; if not, write to the Free Software | |
41 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
42 | */ | |
43 | #include <linux/slab.h> | |
44 | #include <linux/poll.h> | |
45 | #include <linux/fs.h> | |
46 | #include <linux/file.h> | |
47 | #include <linux/jhash.h> | |
48 | #include <linux/init.h> | |
49 | #include <linux/futex.h> | |
50 | #include <linux/mount.h> | |
51 | #include <linux/pagemap.h> | |
52 | #include <linux/syscalls.h> | |
7ed20e1a | 53 | #include <linux/signal.h> |
9adef58b | 54 | #include <linux/module.h> |
fd5eea42 | 55 | #include <linux/magic.h> |
b488893a PE |
56 | #include <linux/pid.h> |
57 | #include <linux/nsproxy.h> | |
58 | ||
4732efbe | 59 | #include <asm/futex.h> |
1da177e4 | 60 | |
c87e2837 IM |
61 | #include "rtmutex_common.h" |
62 | ||
a0c1e907 TG |
63 | int __read_mostly futex_cmpxchg_enabled; |
64 | ||
1da177e4 LT |
65 | #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8) |
66 | ||
c87e2837 IM |
67 | /* |
68 | * Priority Inheritance state: | |
69 | */ | |
70 | struct futex_pi_state { | |
71 | /* | |
72 | * list of 'owned' pi_state instances - these have to be | |
73 | * cleaned up in do_exit() if the task exits prematurely: | |
74 | */ | |
75 | struct list_head list; | |
76 | ||
77 | /* | |
78 | * The PI object: | |
79 | */ | |
80 | struct rt_mutex pi_mutex; | |
81 | ||
82 | struct task_struct *owner; | |
83 | atomic_t refcount; | |
84 | ||
85 | union futex_key key; | |
86 | }; | |
87 | ||
1da177e4 LT |
88 | /* |
89 | * We use this hashed waitqueue instead of a normal wait_queue_t, so | |
90 | * we can wake only the relevant ones (hashed queues may be shared). | |
91 | * | |
92 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 93 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
1da177e4 | 94 | * The order of wakup is always to make the first condition true, then |
73500ac5 | 95 | * wake up q->waiter, then make the second condition true. |
1da177e4 LT |
96 | */ |
97 | struct futex_q { | |
ec92d082 | 98 | struct plist_node list; |
73500ac5 DH |
99 | /* There can only be a single waiter */ |
100 | wait_queue_head_t waiter; | |
1da177e4 | 101 | |
e2970f2f | 102 | /* Which hash list lock to use: */ |
1da177e4 LT |
103 | spinlock_t *lock_ptr; |
104 | ||
e2970f2f | 105 | /* Key which the futex is hashed on: */ |
1da177e4 LT |
106 | union futex_key key; |
107 | ||
c87e2837 IM |
108 | /* Optional priority inheritance state: */ |
109 | struct futex_pi_state *pi_state; | |
110 | struct task_struct *task; | |
cd689985 TG |
111 | |
112 | /* Bitset for the optional bitmasked wakeup */ | |
113 | u32 bitset; | |
1da177e4 LT |
114 | }; |
115 | ||
116 | /* | |
117 | * Split the global futex_lock into every hash list lock. | |
118 | */ | |
119 | struct futex_hash_bucket { | |
ec92d082 PP |
120 | spinlock_t lock; |
121 | struct plist_head chain; | |
1da177e4 LT |
122 | }; |
123 | ||
124 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; | |
125 | ||
1da177e4 LT |
126 | /* |
127 | * We hash on the keys returned from get_futex_key (see below). | |
128 | */ | |
129 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
130 | { | |
131 | u32 hash = jhash2((u32*)&key->both.word, | |
132 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
133 | key->both.offset); | |
134 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; | |
135 | } | |
136 | ||
137 | /* | |
138 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
139 | */ | |
140 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
141 | { | |
142 | return (key1->both.word == key2->both.word | |
143 | && key1->both.ptr == key2->both.ptr | |
144 | && key1->both.offset == key2->both.offset); | |
145 | } | |
146 | ||
38d47c1b PZ |
147 | /* |
148 | * Take a reference to the resource addressed by a key. | |
149 | * Can be called while holding spinlocks. | |
150 | * | |
151 | */ | |
152 | static void get_futex_key_refs(union futex_key *key) | |
153 | { | |
154 | if (!key->both.ptr) | |
155 | return; | |
156 | ||
157 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
158 | case FUT_OFF_INODE: | |
159 | atomic_inc(&key->shared.inode->i_count); | |
160 | break; | |
161 | case FUT_OFF_MMSHARED: | |
162 | atomic_inc(&key->private.mm->mm_count); | |
163 | break; | |
164 | } | |
165 | } | |
166 | ||
167 | /* | |
168 | * Drop a reference to the resource addressed by a key. | |
169 | * The hash bucket spinlock must not be held. | |
170 | */ | |
171 | static void drop_futex_key_refs(union futex_key *key) | |
172 | { | |
90621c40 DH |
173 | if (!key->both.ptr) { |
174 | /* If we're here then we tried to put a key we failed to get */ | |
175 | WARN_ON_ONCE(1); | |
38d47c1b | 176 | return; |
90621c40 | 177 | } |
38d47c1b PZ |
178 | |
179 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
180 | case FUT_OFF_INODE: | |
181 | iput(key->shared.inode); | |
182 | break; | |
183 | case FUT_OFF_MMSHARED: | |
184 | mmdrop(key->private.mm); | |
185 | break; | |
186 | } | |
187 | } | |
188 | ||
34f01cc1 ED |
189 | /** |
190 | * get_futex_key - Get parameters which are the keys for a futex. | |
191 | * @uaddr: virtual address of the futex | |
192 | * @shared: NULL for a PROCESS_PRIVATE futex, | |
193 | * ¤t->mm->mmap_sem for a PROCESS_SHARED futex | |
194 | * @key: address where result is stored. | |
195 | * | |
196 | * Returns a negative error code or 0 | |
197 | * The key words are stored in *key on success. | |
1da177e4 | 198 | * |
f3a43f3f | 199 | * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode, |
1da177e4 LT |
200 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
201 | * We can usually work out the index without swapping in the page. | |
202 | * | |
34f01cc1 ED |
203 | * fshared is NULL for PROCESS_PRIVATE futexes |
204 | * For other futexes, it points to ¤t->mm->mmap_sem and | |
205 | * caller must have taken the reader lock. but NOT any spinlocks. | |
1da177e4 | 206 | */ |
c2f9f201 | 207 | static int get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key) |
1da177e4 | 208 | { |
e2970f2f | 209 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 210 | struct mm_struct *mm = current->mm; |
1da177e4 LT |
211 | struct page *page; |
212 | int err; | |
213 | ||
214 | /* | |
215 | * The futex address must be "naturally" aligned. | |
216 | */ | |
e2970f2f | 217 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 218 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 219 | return -EINVAL; |
e2970f2f | 220 | address -= key->both.offset; |
1da177e4 | 221 | |
34f01cc1 ED |
222 | /* |
223 | * PROCESS_PRIVATE futexes are fast. | |
224 | * As the mm cannot disappear under us and the 'key' only needs | |
225 | * virtual address, we dont even have to find the underlying vma. | |
226 | * Note : We do have to check 'uaddr' is a valid user address, | |
227 | * but access_ok() should be faster than find_vma() | |
228 | */ | |
229 | if (!fshared) { | |
230 | if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32)))) | |
231 | return -EFAULT; | |
232 | key->private.mm = mm; | |
233 | key->private.address = address; | |
42569c39 | 234 | get_futex_key_refs(key); |
34f01cc1 ED |
235 | return 0; |
236 | } | |
1da177e4 | 237 | |
38d47c1b | 238 | again: |
734b05b1 | 239 | err = get_user_pages_fast(address, 1, 0, &page); |
38d47c1b PZ |
240 | if (err < 0) |
241 | return err; | |
242 | ||
243 | lock_page(page); | |
244 | if (!page->mapping) { | |
245 | unlock_page(page); | |
246 | put_page(page); | |
247 | goto again; | |
248 | } | |
1da177e4 LT |
249 | |
250 | /* | |
251 | * Private mappings are handled in a simple way. | |
252 | * | |
253 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
254 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 255 | * the object not the particular process. |
1da177e4 | 256 | */ |
38d47c1b PZ |
257 | if (PageAnon(page)) { |
258 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ | |
1da177e4 | 259 | key->private.mm = mm; |
e2970f2f | 260 | key->private.address = address; |
38d47c1b PZ |
261 | } else { |
262 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ | |
263 | key->shared.inode = page->mapping->host; | |
264 | key->shared.pgoff = page->index; | |
1da177e4 LT |
265 | } |
266 | ||
38d47c1b | 267 | get_futex_key_refs(key); |
1da177e4 | 268 | |
38d47c1b PZ |
269 | unlock_page(page); |
270 | put_page(page); | |
271 | return 0; | |
1da177e4 LT |
272 | } |
273 | ||
38d47c1b | 274 | static inline |
c2f9f201 | 275 | void put_futex_key(int fshared, union futex_key *key) |
1da177e4 | 276 | { |
38d47c1b | 277 | drop_futex_key_refs(key); |
1da177e4 LT |
278 | } |
279 | ||
36cf3b5c TG |
280 | static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval) |
281 | { | |
282 | u32 curval; | |
283 | ||
284 | pagefault_disable(); | |
285 | curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval); | |
286 | pagefault_enable(); | |
287 | ||
288 | return curval; | |
289 | } | |
290 | ||
291 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
292 | { |
293 | int ret; | |
294 | ||
a866374a | 295 | pagefault_disable(); |
e2970f2f | 296 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 297 | pagefault_enable(); |
1da177e4 LT |
298 | |
299 | return ret ? -EFAULT : 0; | |
300 | } | |
301 | ||
c87e2837 | 302 | /* |
34f01cc1 | 303 | * Fault handling. |
c87e2837 | 304 | */ |
c2f9f201 | 305 | static int futex_handle_fault(unsigned long address, int attempt) |
c87e2837 IM |
306 | { |
307 | struct vm_area_struct * vma; | |
308 | struct mm_struct *mm = current->mm; | |
34f01cc1 | 309 | int ret = -EFAULT; |
c87e2837 | 310 | |
34f01cc1 ED |
311 | if (attempt > 2) |
312 | return ret; | |
c87e2837 | 313 | |
61270708 | 314 | down_read(&mm->mmap_sem); |
34f01cc1 ED |
315 | vma = find_vma(mm, address); |
316 | if (vma && address >= vma->vm_start && | |
317 | (vma->vm_flags & VM_WRITE)) { | |
83c54070 NP |
318 | int fault; |
319 | fault = handle_mm_fault(mm, vma, address, 1); | |
320 | if (unlikely((fault & VM_FAULT_ERROR))) { | |
321 | #if 0 | |
322 | /* XXX: let's do this when we verify it is OK */ | |
323 | if (ret & VM_FAULT_OOM) | |
324 | ret = -ENOMEM; | |
325 | #endif | |
326 | } else { | |
34f01cc1 | 327 | ret = 0; |
83c54070 NP |
328 | if (fault & VM_FAULT_MAJOR) |
329 | current->maj_flt++; | |
330 | else | |
331 | current->min_flt++; | |
34f01cc1 | 332 | } |
c87e2837 | 333 | } |
61270708 | 334 | up_read(&mm->mmap_sem); |
34f01cc1 | 335 | return ret; |
c87e2837 IM |
336 | } |
337 | ||
338 | /* | |
339 | * PI code: | |
340 | */ | |
341 | static int refill_pi_state_cache(void) | |
342 | { | |
343 | struct futex_pi_state *pi_state; | |
344 | ||
345 | if (likely(current->pi_state_cache)) | |
346 | return 0; | |
347 | ||
4668edc3 | 348 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
349 | |
350 | if (!pi_state) | |
351 | return -ENOMEM; | |
352 | ||
c87e2837 IM |
353 | INIT_LIST_HEAD(&pi_state->list); |
354 | /* pi_mutex gets initialized later */ | |
355 | pi_state->owner = NULL; | |
356 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 357 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
358 | |
359 | current->pi_state_cache = pi_state; | |
360 | ||
361 | return 0; | |
362 | } | |
363 | ||
364 | static struct futex_pi_state * alloc_pi_state(void) | |
365 | { | |
366 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
367 | ||
368 | WARN_ON(!pi_state); | |
369 | current->pi_state_cache = NULL; | |
370 | ||
371 | return pi_state; | |
372 | } | |
373 | ||
374 | static void free_pi_state(struct futex_pi_state *pi_state) | |
375 | { | |
376 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
377 | return; | |
378 | ||
379 | /* | |
380 | * If pi_state->owner is NULL, the owner is most probably dying | |
381 | * and has cleaned up the pi_state already | |
382 | */ | |
383 | if (pi_state->owner) { | |
384 | spin_lock_irq(&pi_state->owner->pi_lock); | |
385 | list_del_init(&pi_state->list); | |
386 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
387 | ||
388 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
389 | } | |
390 | ||
391 | if (current->pi_state_cache) | |
392 | kfree(pi_state); | |
393 | else { | |
394 | /* | |
395 | * pi_state->list is already empty. | |
396 | * clear pi_state->owner. | |
397 | * refcount is at 0 - put it back to 1. | |
398 | */ | |
399 | pi_state->owner = NULL; | |
400 | atomic_set(&pi_state->refcount, 1); | |
401 | current->pi_state_cache = pi_state; | |
402 | } | |
403 | } | |
404 | ||
405 | /* | |
406 | * Look up the task based on what TID userspace gave us. | |
407 | * We dont trust it. | |
408 | */ | |
409 | static struct task_struct * futex_find_get_task(pid_t pid) | |
410 | { | |
411 | struct task_struct *p; | |
c69e8d9c | 412 | const struct cred *cred = current_cred(), *pcred; |
c87e2837 | 413 | |
d359b549 | 414 | rcu_read_lock(); |
228ebcbe | 415 | p = find_task_by_vpid(pid); |
c69e8d9c | 416 | if (!p) { |
a06381fe | 417 | p = ERR_PTR(-ESRCH); |
c69e8d9c DH |
418 | } else { |
419 | pcred = __task_cred(p); | |
420 | if (cred->euid != pcred->euid && | |
421 | cred->euid != pcred->uid) | |
422 | p = ERR_PTR(-ESRCH); | |
423 | else | |
424 | get_task_struct(p); | |
425 | } | |
a06381fe | 426 | |
d359b549 | 427 | rcu_read_unlock(); |
c87e2837 IM |
428 | |
429 | return p; | |
430 | } | |
431 | ||
432 | /* | |
433 | * This task is holding PI mutexes at exit time => bad. | |
434 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
435 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
436 | */ | |
437 | void exit_pi_state_list(struct task_struct *curr) | |
438 | { | |
c87e2837 IM |
439 | struct list_head *next, *head = &curr->pi_state_list; |
440 | struct futex_pi_state *pi_state; | |
627371d7 | 441 | struct futex_hash_bucket *hb; |
38d47c1b | 442 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 443 | |
a0c1e907 TG |
444 | if (!futex_cmpxchg_enabled) |
445 | return; | |
c87e2837 IM |
446 | /* |
447 | * We are a ZOMBIE and nobody can enqueue itself on | |
448 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 449 | * versus waiters unqueueing themselves: |
c87e2837 IM |
450 | */ |
451 | spin_lock_irq(&curr->pi_lock); | |
452 | while (!list_empty(head)) { | |
453 | ||
454 | next = head->next; | |
455 | pi_state = list_entry(next, struct futex_pi_state, list); | |
456 | key = pi_state->key; | |
627371d7 | 457 | hb = hash_futex(&key); |
c87e2837 IM |
458 | spin_unlock_irq(&curr->pi_lock); |
459 | ||
c87e2837 IM |
460 | spin_lock(&hb->lock); |
461 | ||
462 | spin_lock_irq(&curr->pi_lock); | |
627371d7 IM |
463 | /* |
464 | * We dropped the pi-lock, so re-check whether this | |
465 | * task still owns the PI-state: | |
466 | */ | |
c87e2837 IM |
467 | if (head->next != next) { |
468 | spin_unlock(&hb->lock); | |
469 | continue; | |
470 | } | |
471 | ||
c87e2837 | 472 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
473 | WARN_ON(list_empty(&pi_state->list)); |
474 | list_del_init(&pi_state->list); | |
c87e2837 IM |
475 | pi_state->owner = NULL; |
476 | spin_unlock_irq(&curr->pi_lock); | |
477 | ||
478 | rt_mutex_unlock(&pi_state->pi_mutex); | |
479 | ||
480 | spin_unlock(&hb->lock); | |
481 | ||
482 | spin_lock_irq(&curr->pi_lock); | |
483 | } | |
484 | spin_unlock_irq(&curr->pi_lock); | |
485 | } | |
486 | ||
487 | static int | |
d0aa7a70 PP |
488 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
489 | union futex_key *key, struct futex_pi_state **ps) | |
c87e2837 IM |
490 | { |
491 | struct futex_pi_state *pi_state = NULL; | |
492 | struct futex_q *this, *next; | |
ec92d082 | 493 | struct plist_head *head; |
c87e2837 | 494 | struct task_struct *p; |
778e9a9c | 495 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 IM |
496 | |
497 | head = &hb->chain; | |
498 | ||
ec92d082 | 499 | plist_for_each_entry_safe(this, next, head, list) { |
d0aa7a70 | 500 | if (match_futex(&this->key, key)) { |
c87e2837 IM |
501 | /* |
502 | * Another waiter already exists - bump up | |
503 | * the refcount and return its pi_state: | |
504 | */ | |
505 | pi_state = this->pi_state; | |
06a9ec29 TG |
506 | /* |
507 | * Userspace might have messed up non PI and PI futexes | |
508 | */ | |
509 | if (unlikely(!pi_state)) | |
510 | return -EINVAL; | |
511 | ||
627371d7 | 512 | WARN_ON(!atomic_read(&pi_state->refcount)); |
778e9a9c AK |
513 | WARN_ON(pid && pi_state->owner && |
514 | pi_state->owner->pid != pid); | |
627371d7 | 515 | |
c87e2837 | 516 | atomic_inc(&pi_state->refcount); |
d0aa7a70 | 517 | *ps = pi_state; |
c87e2837 IM |
518 | |
519 | return 0; | |
520 | } | |
521 | } | |
522 | ||
523 | /* | |
e3f2ddea | 524 | * We are the first waiter - try to look up the real owner and attach |
778e9a9c | 525 | * the new pi_state to it, but bail out when TID = 0 |
c87e2837 | 526 | */ |
778e9a9c | 527 | if (!pid) |
e3f2ddea | 528 | return -ESRCH; |
c87e2837 | 529 | p = futex_find_get_task(pid); |
778e9a9c AK |
530 | if (IS_ERR(p)) |
531 | return PTR_ERR(p); | |
532 | ||
533 | /* | |
534 | * We need to look at the task state flags to figure out, | |
535 | * whether the task is exiting. To protect against the do_exit | |
536 | * change of the task flags, we do this protected by | |
537 | * p->pi_lock: | |
538 | */ | |
539 | spin_lock_irq(&p->pi_lock); | |
540 | if (unlikely(p->flags & PF_EXITING)) { | |
541 | /* | |
542 | * The task is on the way out. When PF_EXITPIDONE is | |
543 | * set, we know that the task has finished the | |
544 | * cleanup: | |
545 | */ | |
546 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
547 | ||
548 | spin_unlock_irq(&p->pi_lock); | |
549 | put_task_struct(p); | |
550 | return ret; | |
551 | } | |
c87e2837 IM |
552 | |
553 | pi_state = alloc_pi_state(); | |
554 | ||
555 | /* | |
556 | * Initialize the pi_mutex in locked state and make 'p' | |
557 | * the owner of it: | |
558 | */ | |
559 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
560 | ||
561 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 562 | pi_state->key = *key; |
c87e2837 | 563 | |
627371d7 | 564 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
565 | list_add(&pi_state->list, &p->pi_state_list); |
566 | pi_state->owner = p; | |
567 | spin_unlock_irq(&p->pi_lock); | |
568 | ||
569 | put_task_struct(p); | |
570 | ||
d0aa7a70 | 571 | *ps = pi_state; |
c87e2837 IM |
572 | |
573 | return 0; | |
574 | } | |
575 | ||
1da177e4 LT |
576 | /* |
577 | * The hash bucket lock must be held when this is called. | |
578 | * Afterwards, the futex_q must not be accessed. | |
579 | */ | |
580 | static void wake_futex(struct futex_q *q) | |
581 | { | |
ec92d082 | 582 | plist_del(&q->list, &q->list.plist); |
1da177e4 LT |
583 | /* |
584 | * The lock in wake_up_all() is a crucial memory barrier after the | |
ec92d082 | 585 | * plist_del() and also before assigning to q->lock_ptr. |
1da177e4 | 586 | */ |
73500ac5 | 587 | wake_up(&q->waiter); |
1da177e4 LT |
588 | /* |
589 | * The waiting task can free the futex_q as soon as this is written, | |
590 | * without taking any locks. This must come last. | |
8e31108b AM |
591 | * |
592 | * A memory barrier is required here to prevent the following store | |
593 | * to lock_ptr from getting ahead of the wakeup. Clearing the lock | |
594 | * at the end of wake_up_all() does not prevent this store from | |
595 | * moving. | |
1da177e4 | 596 | */ |
ccdea2f8 | 597 | smp_wmb(); |
1da177e4 LT |
598 | q->lock_ptr = NULL; |
599 | } | |
600 | ||
c87e2837 IM |
601 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
602 | { | |
603 | struct task_struct *new_owner; | |
604 | struct futex_pi_state *pi_state = this->pi_state; | |
605 | u32 curval, newval; | |
606 | ||
607 | if (!pi_state) | |
608 | return -EINVAL; | |
609 | ||
21778867 | 610 | spin_lock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
611 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
612 | ||
613 | /* | |
614 | * This happens when we have stolen the lock and the original | |
615 | * pending owner did not enqueue itself back on the rt_mutex. | |
616 | * Thats not a tragedy. We know that way, that a lock waiter | |
617 | * is on the fly. We make the futex_q waiter the pending owner. | |
618 | */ | |
619 | if (!new_owner) | |
620 | new_owner = this->task; | |
621 | ||
622 | /* | |
623 | * We pass it to the next owner. (The WAITERS bit is always | |
624 | * kept enabled while there is PI state around. We must also | |
625 | * preserve the owner died bit.) | |
626 | */ | |
e3f2ddea | 627 | if (!(uval & FUTEX_OWNER_DIED)) { |
778e9a9c AK |
628 | int ret = 0; |
629 | ||
b488893a | 630 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 631 | |
36cf3b5c | 632 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
778e9a9c | 633 | |
e3f2ddea | 634 | if (curval == -EFAULT) |
778e9a9c | 635 | ret = -EFAULT; |
cde898fa | 636 | else if (curval != uval) |
778e9a9c AK |
637 | ret = -EINVAL; |
638 | if (ret) { | |
639 | spin_unlock(&pi_state->pi_mutex.wait_lock); | |
640 | return ret; | |
641 | } | |
e3f2ddea | 642 | } |
c87e2837 | 643 | |
627371d7 IM |
644 | spin_lock_irq(&pi_state->owner->pi_lock); |
645 | WARN_ON(list_empty(&pi_state->list)); | |
646 | list_del_init(&pi_state->list); | |
647 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
648 | ||
649 | spin_lock_irq(&new_owner->pi_lock); | |
650 | WARN_ON(!list_empty(&pi_state->list)); | |
c87e2837 IM |
651 | list_add(&pi_state->list, &new_owner->pi_state_list); |
652 | pi_state->owner = new_owner; | |
627371d7 IM |
653 | spin_unlock_irq(&new_owner->pi_lock); |
654 | ||
21778867 | 655 | spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
656 | rt_mutex_unlock(&pi_state->pi_mutex); |
657 | ||
658 | return 0; | |
659 | } | |
660 | ||
661 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) | |
662 | { | |
663 | u32 oldval; | |
664 | ||
665 | /* | |
666 | * There is no waiter, so we unlock the futex. The owner died | |
667 | * bit has not to be preserved here. We are the owner: | |
668 | */ | |
36cf3b5c | 669 | oldval = cmpxchg_futex_value_locked(uaddr, uval, 0); |
c87e2837 IM |
670 | |
671 | if (oldval == -EFAULT) | |
672 | return oldval; | |
673 | if (oldval != uval) | |
674 | return -EAGAIN; | |
675 | ||
676 | return 0; | |
677 | } | |
678 | ||
8b8f319f IM |
679 | /* |
680 | * Express the locking dependencies for lockdep: | |
681 | */ | |
682 | static inline void | |
683 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
684 | { | |
685 | if (hb1 <= hb2) { | |
686 | spin_lock(&hb1->lock); | |
687 | if (hb1 < hb2) | |
688 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
689 | } else { /* hb1 > hb2 */ | |
690 | spin_lock(&hb2->lock); | |
691 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
692 | } | |
693 | } | |
694 | ||
1da177e4 LT |
695 | /* |
696 | * Wake up all waiters hashed on the physical page that is mapped | |
697 | * to this virtual address: | |
698 | */ | |
c2f9f201 | 699 | static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset) |
1da177e4 | 700 | { |
e2970f2f | 701 | struct futex_hash_bucket *hb; |
1da177e4 | 702 | struct futex_q *this, *next; |
ec92d082 | 703 | struct plist_head *head; |
38d47c1b | 704 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 LT |
705 | int ret; |
706 | ||
cd689985 TG |
707 | if (!bitset) |
708 | return -EINVAL; | |
709 | ||
34f01cc1 | 710 | ret = get_futex_key(uaddr, fshared, &key); |
1da177e4 LT |
711 | if (unlikely(ret != 0)) |
712 | goto out; | |
713 | ||
e2970f2f IM |
714 | hb = hash_futex(&key); |
715 | spin_lock(&hb->lock); | |
716 | head = &hb->chain; | |
1da177e4 | 717 | |
ec92d082 | 718 | plist_for_each_entry_safe(this, next, head, list) { |
1da177e4 | 719 | if (match_futex (&this->key, &key)) { |
ed6f7b10 IM |
720 | if (this->pi_state) { |
721 | ret = -EINVAL; | |
722 | break; | |
723 | } | |
cd689985 TG |
724 | |
725 | /* Check if one of the bits is set in both bitsets */ | |
726 | if (!(this->bitset & bitset)) | |
727 | continue; | |
728 | ||
1da177e4 LT |
729 | wake_futex(this); |
730 | if (++ret >= nr_wake) | |
731 | break; | |
732 | } | |
733 | } | |
734 | ||
e2970f2f | 735 | spin_unlock(&hb->lock); |
38d47c1b | 736 | put_futex_key(fshared, &key); |
42d35d48 | 737 | out: |
1da177e4 LT |
738 | return ret; |
739 | } | |
740 | ||
4732efbe JJ |
741 | /* |
742 | * Wake up all waiters hashed on the physical page that is mapped | |
743 | * to this virtual address: | |
744 | */ | |
e2970f2f | 745 | static int |
c2f9f201 | 746 | futex_wake_op(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, |
e2970f2f | 747 | int nr_wake, int nr_wake2, int op) |
4732efbe | 748 | { |
38d47c1b | 749 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 750 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 751 | struct plist_head *head; |
4732efbe JJ |
752 | struct futex_q *this, *next; |
753 | int ret, op_ret, attempt = 0; | |
754 | ||
755 | retryfull: | |
34f01cc1 | 756 | ret = get_futex_key(uaddr1, fshared, &key1); |
4732efbe JJ |
757 | if (unlikely(ret != 0)) |
758 | goto out; | |
34f01cc1 | 759 | ret = get_futex_key(uaddr2, fshared, &key2); |
4732efbe | 760 | if (unlikely(ret != 0)) |
42d35d48 | 761 | goto out_put_key1; |
4732efbe | 762 | |
e2970f2f IM |
763 | hb1 = hash_futex(&key1); |
764 | hb2 = hash_futex(&key2); | |
4732efbe JJ |
765 | |
766 | retry: | |
8b8f319f | 767 | double_lock_hb(hb1, hb2); |
4732efbe | 768 | |
e2970f2f | 769 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 770 | if (unlikely(op_ret < 0)) { |
e2970f2f | 771 | u32 dummy; |
4732efbe | 772 | |
e2970f2f IM |
773 | spin_unlock(&hb1->lock); |
774 | if (hb1 != hb2) | |
775 | spin_unlock(&hb2->lock); | |
4732efbe | 776 | |
7ee1dd3f | 777 | #ifndef CONFIG_MMU |
e2970f2f IM |
778 | /* |
779 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
780 | * but we might get them from range checking | |
781 | */ | |
7ee1dd3f | 782 | ret = op_ret; |
42d35d48 | 783 | goto out_put_keys; |
7ee1dd3f DH |
784 | #endif |
785 | ||
796f8d9b DG |
786 | if (unlikely(op_ret != -EFAULT)) { |
787 | ret = op_ret; | |
42d35d48 | 788 | goto out_put_keys; |
796f8d9b DG |
789 | } |
790 | ||
e2970f2f IM |
791 | /* |
792 | * futex_atomic_op_inuser needs to both read and write | |
4732efbe JJ |
793 | * *(int __user *)uaddr2, but we can't modify it |
794 | * non-atomically. Therefore, if get_user below is not | |
795 | * enough, we need to handle the fault ourselves, while | |
e2970f2f IM |
796 | * still holding the mmap_sem. |
797 | */ | |
4732efbe | 798 | if (attempt++) { |
34f01cc1 | 799 | ret = futex_handle_fault((unsigned long)uaddr2, |
c2f9f201 | 800 | attempt); |
34f01cc1 | 801 | if (ret) |
42d35d48 | 802 | goto out_put_keys; |
4732efbe JJ |
803 | goto retry; |
804 | } | |
805 | ||
e2970f2f | 806 | ret = get_user(dummy, uaddr2); |
4732efbe JJ |
807 | if (ret) |
808 | return ret; | |
809 | ||
810 | goto retryfull; | |
811 | } | |
812 | ||
e2970f2f | 813 | head = &hb1->chain; |
4732efbe | 814 | |
ec92d082 | 815 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
816 | if (match_futex (&this->key, &key1)) { |
817 | wake_futex(this); | |
818 | if (++ret >= nr_wake) | |
819 | break; | |
820 | } | |
821 | } | |
822 | ||
823 | if (op_ret > 0) { | |
e2970f2f | 824 | head = &hb2->chain; |
4732efbe JJ |
825 | |
826 | op_ret = 0; | |
ec92d082 | 827 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
828 | if (match_futex (&this->key, &key2)) { |
829 | wake_futex(this); | |
830 | if (++op_ret >= nr_wake2) | |
831 | break; | |
832 | } | |
833 | } | |
834 | ret += op_ret; | |
835 | } | |
836 | ||
e2970f2f IM |
837 | spin_unlock(&hb1->lock); |
838 | if (hb1 != hb2) | |
839 | spin_unlock(&hb2->lock); | |
42d35d48 | 840 | out_put_keys: |
38d47c1b | 841 | put_futex_key(fshared, &key2); |
42d35d48 | 842 | out_put_key1: |
38d47c1b | 843 | put_futex_key(fshared, &key1); |
42d35d48 | 844 | out: |
4732efbe JJ |
845 | return ret; |
846 | } | |
847 | ||
1da177e4 LT |
848 | /* |
849 | * Requeue all waiters hashed on one physical page to another | |
850 | * physical page. | |
851 | */ | |
c2f9f201 | 852 | static int futex_requeue(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, |
e2970f2f | 853 | int nr_wake, int nr_requeue, u32 *cmpval) |
1da177e4 | 854 | { |
38d47c1b | 855 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 856 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 857 | struct plist_head *head1; |
1da177e4 LT |
858 | struct futex_q *this, *next; |
859 | int ret, drop_count = 0; | |
860 | ||
42d35d48 | 861 | retry: |
34f01cc1 | 862 | ret = get_futex_key(uaddr1, fshared, &key1); |
1da177e4 LT |
863 | if (unlikely(ret != 0)) |
864 | goto out; | |
34f01cc1 | 865 | ret = get_futex_key(uaddr2, fshared, &key2); |
1da177e4 | 866 | if (unlikely(ret != 0)) |
42d35d48 | 867 | goto out_put_key1; |
1da177e4 | 868 | |
e2970f2f IM |
869 | hb1 = hash_futex(&key1); |
870 | hb2 = hash_futex(&key2); | |
1da177e4 | 871 | |
8b8f319f | 872 | double_lock_hb(hb1, hb2); |
1da177e4 | 873 | |
e2970f2f IM |
874 | if (likely(cmpval != NULL)) { |
875 | u32 curval; | |
1da177e4 | 876 | |
e2970f2f | 877 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
878 | |
879 | if (unlikely(ret)) { | |
e2970f2f IM |
880 | spin_unlock(&hb1->lock); |
881 | if (hb1 != hb2) | |
882 | spin_unlock(&hb2->lock); | |
1da177e4 | 883 | |
e2970f2f | 884 | ret = get_user(curval, uaddr1); |
1da177e4 LT |
885 | |
886 | if (!ret) | |
887 | goto retry; | |
888 | ||
42d35d48 | 889 | goto out_put_keys; |
1da177e4 | 890 | } |
e2970f2f | 891 | if (curval != *cmpval) { |
1da177e4 LT |
892 | ret = -EAGAIN; |
893 | goto out_unlock; | |
894 | } | |
895 | } | |
896 | ||
e2970f2f | 897 | head1 = &hb1->chain; |
ec92d082 | 898 | plist_for_each_entry_safe(this, next, head1, list) { |
1da177e4 LT |
899 | if (!match_futex (&this->key, &key1)) |
900 | continue; | |
901 | if (++ret <= nr_wake) { | |
902 | wake_futex(this); | |
903 | } else { | |
59e0e0ac SD |
904 | /* |
905 | * If key1 and key2 hash to the same bucket, no need to | |
906 | * requeue. | |
907 | */ | |
908 | if (likely(head1 != &hb2->chain)) { | |
ec92d082 PP |
909 | plist_del(&this->list, &hb1->chain); |
910 | plist_add(&this->list, &hb2->chain); | |
59e0e0ac | 911 | this->lock_ptr = &hb2->lock; |
ec92d082 PP |
912 | #ifdef CONFIG_DEBUG_PI_LIST |
913 | this->list.plist.lock = &hb2->lock; | |
914 | #endif | |
778e9a9c | 915 | } |
1da177e4 | 916 | this->key = key2; |
9adef58b | 917 | get_futex_key_refs(&key2); |
1da177e4 LT |
918 | drop_count++; |
919 | ||
920 | if (ret - nr_wake >= nr_requeue) | |
921 | break; | |
1da177e4 LT |
922 | } |
923 | } | |
924 | ||
925 | out_unlock: | |
e2970f2f IM |
926 | spin_unlock(&hb1->lock); |
927 | if (hb1 != hb2) | |
928 | spin_unlock(&hb2->lock); | |
1da177e4 | 929 | |
9adef58b | 930 | /* drop_futex_key_refs() must be called outside the spinlocks. */ |
1da177e4 | 931 | while (--drop_count >= 0) |
9adef58b | 932 | drop_futex_key_refs(&key1); |
1da177e4 | 933 | |
42d35d48 | 934 | out_put_keys: |
38d47c1b | 935 | put_futex_key(fshared, &key2); |
42d35d48 | 936 | out_put_key1: |
38d47c1b | 937 | put_futex_key(fshared, &key1); |
42d35d48 | 938 | out: |
1da177e4 LT |
939 | return ret; |
940 | } | |
941 | ||
942 | /* The key must be already stored in q->key. */ | |
82af7aca | 943 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
1da177e4 | 944 | { |
e2970f2f | 945 | struct futex_hash_bucket *hb; |
1da177e4 | 946 | |
73500ac5 | 947 | init_waitqueue_head(&q->waiter); |
1da177e4 | 948 | |
9adef58b | 949 | get_futex_key_refs(&q->key); |
e2970f2f IM |
950 | hb = hash_futex(&q->key); |
951 | q->lock_ptr = &hb->lock; | |
1da177e4 | 952 | |
e2970f2f IM |
953 | spin_lock(&hb->lock); |
954 | return hb; | |
1da177e4 LT |
955 | } |
956 | ||
82af7aca | 957 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 958 | { |
ec92d082 PP |
959 | int prio; |
960 | ||
961 | /* | |
962 | * The priority used to register this element is | |
963 | * - either the real thread-priority for the real-time threads | |
964 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
965 | * - or MAX_RT_PRIO for non-RT threads. | |
966 | * Thus, all RT-threads are woken first in priority order, and | |
967 | * the others are woken last, in FIFO order. | |
968 | */ | |
969 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
970 | ||
971 | plist_node_init(&q->list, prio); | |
972 | #ifdef CONFIG_DEBUG_PI_LIST | |
973 | q->list.plist.lock = &hb->lock; | |
974 | #endif | |
975 | plist_add(&q->list, &hb->chain); | |
c87e2837 | 976 | q->task = current; |
e2970f2f | 977 | spin_unlock(&hb->lock); |
1da177e4 LT |
978 | } |
979 | ||
980 | static inline void | |
e2970f2f | 981 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 982 | { |
e2970f2f | 983 | spin_unlock(&hb->lock); |
9adef58b | 984 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
985 | } |
986 | ||
987 | /* | |
988 | * queue_me and unqueue_me must be called as a pair, each | |
989 | * exactly once. They are called with the hashed spinlock held. | |
990 | */ | |
991 | ||
1da177e4 LT |
992 | /* Return 1 if we were still queued (ie. 0 means we were woken) */ |
993 | static int unqueue_me(struct futex_q *q) | |
994 | { | |
1da177e4 | 995 | spinlock_t *lock_ptr; |
e2970f2f | 996 | int ret = 0; |
1da177e4 LT |
997 | |
998 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 999 | retry: |
1da177e4 | 1000 | lock_ptr = q->lock_ptr; |
e91467ec | 1001 | barrier(); |
c80544dc | 1002 | if (lock_ptr != NULL) { |
1da177e4 LT |
1003 | spin_lock(lock_ptr); |
1004 | /* | |
1005 | * q->lock_ptr can change between reading it and | |
1006 | * spin_lock(), causing us to take the wrong lock. This | |
1007 | * corrects the race condition. | |
1008 | * | |
1009 | * Reasoning goes like this: if we have the wrong lock, | |
1010 | * q->lock_ptr must have changed (maybe several times) | |
1011 | * between reading it and the spin_lock(). It can | |
1012 | * change again after the spin_lock() but only if it was | |
1013 | * already changed before the spin_lock(). It cannot, | |
1014 | * however, change back to the original value. Therefore | |
1015 | * we can detect whether we acquired the correct lock. | |
1016 | */ | |
1017 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
1018 | spin_unlock(lock_ptr); | |
1019 | goto retry; | |
1020 | } | |
ec92d082 PP |
1021 | WARN_ON(plist_node_empty(&q->list)); |
1022 | plist_del(&q->list, &q->list.plist); | |
c87e2837 IM |
1023 | |
1024 | BUG_ON(q->pi_state); | |
1025 | ||
1da177e4 LT |
1026 | spin_unlock(lock_ptr); |
1027 | ret = 1; | |
1028 | } | |
1029 | ||
9adef58b | 1030 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1031 | return ret; |
1032 | } | |
1033 | ||
c87e2837 IM |
1034 | /* |
1035 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
1036 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
1037 | * and dropped here. | |
c87e2837 | 1038 | */ |
d0aa7a70 | 1039 | static void unqueue_me_pi(struct futex_q *q) |
c87e2837 | 1040 | { |
ec92d082 PP |
1041 | WARN_ON(plist_node_empty(&q->list)); |
1042 | plist_del(&q->list, &q->list.plist); | |
c87e2837 IM |
1043 | |
1044 | BUG_ON(!q->pi_state); | |
1045 | free_pi_state(q->pi_state); | |
1046 | q->pi_state = NULL; | |
1047 | ||
d0aa7a70 | 1048 | spin_unlock(q->lock_ptr); |
c87e2837 | 1049 | |
9adef58b | 1050 | drop_futex_key_refs(&q->key); |
c87e2837 IM |
1051 | } |
1052 | ||
d0aa7a70 | 1053 | /* |
cdf71a10 | 1054 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 1055 | * |
778e9a9c AK |
1056 | * Must be called with hash bucket lock held and mm->sem held for non |
1057 | * private futexes. | |
d0aa7a70 | 1058 | */ |
778e9a9c | 1059 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
c2f9f201 | 1060 | struct task_struct *newowner, int fshared) |
d0aa7a70 | 1061 | { |
cdf71a10 | 1062 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 1063 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 1064 | struct task_struct *oldowner = pi_state->owner; |
d0aa7a70 | 1065 | u32 uval, curval, newval; |
1b7558e4 | 1066 | int ret, attempt = 0; |
d0aa7a70 PP |
1067 | |
1068 | /* Owner died? */ | |
1b7558e4 TG |
1069 | if (!pi_state->owner) |
1070 | newtid |= FUTEX_OWNER_DIED; | |
1071 | ||
1072 | /* | |
1073 | * We are here either because we stole the rtmutex from the | |
1074 | * pending owner or we are the pending owner which failed to | |
1075 | * get the rtmutex. We have to replace the pending owner TID | |
1076 | * in the user space variable. This must be atomic as we have | |
1077 | * to preserve the owner died bit here. | |
1078 | * | |
1079 | * Note: We write the user space value _before_ changing the | |
1080 | * pi_state because we can fault here. Imagine swapped out | |
1081 | * pages or a fork, which was running right before we acquired | |
1082 | * mmap_sem, that marked all the anonymous memory readonly for | |
1083 | * cow. | |
1084 | * | |
1085 | * Modifying pi_state _before_ the user space value would | |
1086 | * leave the pi_state in an inconsistent state when we fault | |
1087 | * here, because we need to drop the hash bucket lock to | |
1088 | * handle the fault. This might be observed in the PID check | |
1089 | * in lookup_pi_state. | |
1090 | */ | |
1091 | retry: | |
1092 | if (get_futex_value_locked(&uval, uaddr)) | |
1093 | goto handle_fault; | |
1094 | ||
1095 | while (1) { | |
1096 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
1097 | ||
1098 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); | |
1099 | ||
1100 | if (curval == -EFAULT) | |
1101 | goto handle_fault; | |
1102 | if (curval == uval) | |
1103 | break; | |
1104 | uval = curval; | |
1105 | } | |
1106 | ||
1107 | /* | |
1108 | * We fixed up user space. Now we need to fix the pi_state | |
1109 | * itself. | |
1110 | */ | |
d0aa7a70 PP |
1111 | if (pi_state->owner != NULL) { |
1112 | spin_lock_irq(&pi_state->owner->pi_lock); | |
1113 | WARN_ON(list_empty(&pi_state->list)); | |
1114 | list_del_init(&pi_state->list); | |
1115 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
1b7558e4 | 1116 | } |
d0aa7a70 | 1117 | |
cdf71a10 | 1118 | pi_state->owner = newowner; |
d0aa7a70 | 1119 | |
cdf71a10 | 1120 | spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 1121 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 TG |
1122 | list_add(&pi_state->list, &newowner->pi_state_list); |
1123 | spin_unlock_irq(&newowner->pi_lock); | |
1b7558e4 | 1124 | return 0; |
d0aa7a70 | 1125 | |
d0aa7a70 | 1126 | /* |
1b7558e4 TG |
1127 | * To handle the page fault we need to drop the hash bucket |
1128 | * lock here. That gives the other task (either the pending | |
1129 | * owner itself or the task which stole the rtmutex) the | |
1130 | * chance to try the fixup of the pi_state. So once we are | |
1131 | * back from handling the fault we need to check the pi_state | |
1132 | * after reacquiring the hash bucket lock and before trying to | |
1133 | * do another fixup. When the fixup has been done already we | |
1134 | * simply return. | |
d0aa7a70 | 1135 | */ |
1b7558e4 TG |
1136 | handle_fault: |
1137 | spin_unlock(q->lock_ptr); | |
778e9a9c | 1138 | |
c2f9f201 | 1139 | ret = futex_handle_fault((unsigned long)uaddr, attempt++); |
778e9a9c | 1140 | |
1b7558e4 | 1141 | spin_lock(q->lock_ptr); |
778e9a9c | 1142 | |
1b7558e4 TG |
1143 | /* |
1144 | * Check if someone else fixed it for us: | |
1145 | */ | |
1146 | if (pi_state->owner != oldowner) | |
1147 | return 0; | |
1148 | ||
1149 | if (ret) | |
1150 | return ret; | |
1151 | ||
1152 | goto retry; | |
d0aa7a70 PP |
1153 | } |
1154 | ||
34f01cc1 ED |
1155 | /* |
1156 | * In case we must use restart_block to restart a futex_wait, | |
ce6bd420 | 1157 | * we encode in the 'flags' shared capability |
34f01cc1 | 1158 | */ |
1acdac10 TG |
1159 | #define FLAGS_SHARED 0x01 |
1160 | #define FLAGS_CLOCKRT 0x02 | |
34f01cc1 | 1161 | |
72c1bbf3 | 1162 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 1163 | |
c2f9f201 | 1164 | static int futex_wait(u32 __user *uaddr, int fshared, |
1acdac10 | 1165 | u32 val, ktime_t *abs_time, u32 bitset, int clockrt) |
1da177e4 | 1166 | { |
c87e2837 IM |
1167 | struct task_struct *curr = current; |
1168 | DECLARE_WAITQUEUE(wait, curr); | |
e2970f2f | 1169 | struct futex_hash_bucket *hb; |
1da177e4 | 1170 | struct futex_q q; |
e2970f2f IM |
1171 | u32 uval; |
1172 | int ret; | |
bd197234 | 1173 | struct hrtimer_sleeper t; |
c19384b5 | 1174 | int rem = 0; |
1da177e4 | 1175 | |
cd689985 TG |
1176 | if (!bitset) |
1177 | return -EINVAL; | |
1178 | ||
c87e2837 | 1179 | q.pi_state = NULL; |
cd689985 | 1180 | q.bitset = bitset; |
42d35d48 | 1181 | retry: |
38d47c1b | 1182 | q.key = FUTEX_KEY_INIT; |
34f01cc1 | 1183 | ret = get_futex_key(uaddr, fshared, &q.key); |
1da177e4 | 1184 | if (unlikely(ret != 0)) |
42d35d48 | 1185 | goto out; |
1da177e4 | 1186 | |
82af7aca | 1187 | hb = queue_lock(&q); |
1da177e4 LT |
1188 | |
1189 | /* | |
1190 | * Access the page AFTER the futex is queued. | |
1191 | * Order is important: | |
1192 | * | |
1193 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
1194 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
1195 | * | |
1196 | * The basic logical guarantee of a futex is that it blocks ONLY | |
1197 | * if cond(var) is known to be true at the time of blocking, for | |
1198 | * any cond. If we queued after testing *uaddr, that would open | |
1199 | * a race condition where we could block indefinitely with | |
1200 | * cond(var) false, which would violate the guarantee. | |
1201 | * | |
1202 | * A consequence is that futex_wait() can return zero and absorb | |
1203 | * a wakeup when *uaddr != val on entry to the syscall. This is | |
1204 | * rare, but normal. | |
1205 | * | |
34f01cc1 ED |
1206 | * for shared futexes, we hold the mmap semaphore, so the mapping |
1207 | * cannot have changed since we looked it up in get_futex_key. | |
1da177e4 | 1208 | */ |
e2970f2f | 1209 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 LT |
1210 | |
1211 | if (unlikely(ret)) { | |
e2970f2f | 1212 | queue_unlock(&q, hb); |
42d35d48 | 1213 | put_futex_key(fshared, &q.key); |
1da177e4 | 1214 | |
e2970f2f | 1215 | ret = get_user(uval, uaddr); |
1da177e4 LT |
1216 | |
1217 | if (!ret) | |
1218 | goto retry; | |
1219 | return ret; | |
1220 | } | |
c87e2837 IM |
1221 | ret = -EWOULDBLOCK; |
1222 | if (uval != val) | |
42d35d48 | 1223 | goto out_unlock_put_key; |
1da177e4 LT |
1224 | |
1225 | /* Only actually queue if *uaddr contained val. */ | |
82af7aca | 1226 | queue_me(&q, hb); |
1da177e4 | 1227 | |
1da177e4 LT |
1228 | /* |
1229 | * There might have been scheduling since the queue_me(), as we | |
1230 | * cannot hold a spinlock across the get_user() in case it | |
1231 | * faults, and we cannot just set TASK_INTERRUPTIBLE state when | |
1232 | * queueing ourselves into the futex hash. This code thus has to | |
1233 | * rely on the futex_wake() code removing us from hash when it | |
1234 | * wakes us up. | |
1235 | */ | |
1236 | ||
1237 | /* add_wait_queue is the barrier after __set_current_state. */ | |
1238 | __set_current_state(TASK_INTERRUPTIBLE); | |
73500ac5 | 1239 | add_wait_queue(&q.waiter, &wait); |
1da177e4 | 1240 | /* |
ec92d082 | 1241 | * !plist_node_empty() is safe here without any lock. |
1da177e4 LT |
1242 | * q.lock_ptr != 0 is not safe, because of ordering against wakeup. |
1243 | */ | |
ec92d082 | 1244 | if (likely(!plist_node_empty(&q.list))) { |
c19384b5 PP |
1245 | if (!abs_time) |
1246 | schedule(); | |
1247 | else { | |
ae4b748e AV |
1248 | unsigned long slack; |
1249 | slack = current->timer_slack_ns; | |
1250 | if (rt_task(current)) | |
1251 | slack = 0; | |
1acdac10 TG |
1252 | hrtimer_init_on_stack(&t.timer, |
1253 | clockrt ? CLOCK_REALTIME : | |
1254 | CLOCK_MONOTONIC, | |
1255 | HRTIMER_MODE_ABS); | |
c19384b5 | 1256 | hrtimer_init_sleeper(&t, current); |
ae4b748e | 1257 | hrtimer_set_expires_range_ns(&t.timer, *abs_time, slack); |
c19384b5 | 1258 | |
cc584b21 | 1259 | hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS); |
3588a085 PZ |
1260 | if (!hrtimer_active(&t.timer)) |
1261 | t.task = NULL; | |
c19384b5 PP |
1262 | |
1263 | /* | |
1264 | * the timer could have already expired, in which | |
1265 | * case current would be flagged for rescheduling. | |
1266 | * Don't bother calling schedule. | |
1267 | */ | |
1268 | if (likely(t.task)) | |
1269 | schedule(); | |
1270 | ||
1271 | hrtimer_cancel(&t.timer); | |
72c1bbf3 | 1272 | |
c19384b5 PP |
1273 | /* Flag if a timeout occured */ |
1274 | rem = (t.task == NULL); | |
237fc6e7 TG |
1275 | |
1276 | destroy_hrtimer_on_stack(&t.timer); | |
c19384b5 | 1277 | } |
72c1bbf3 | 1278 | } |
1da177e4 LT |
1279 | __set_current_state(TASK_RUNNING); |
1280 | ||
1281 | /* | |
1282 | * NOTE: we don't remove ourselves from the waitqueue because | |
1283 | * we are the only user of it. | |
1284 | */ | |
1285 | ||
1286 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
1287 | if (!unqueue_me(&q)) | |
1288 | return 0; | |
c19384b5 | 1289 | if (rem) |
1da177e4 | 1290 | return -ETIMEDOUT; |
72c1bbf3 | 1291 | |
e2970f2f IM |
1292 | /* |
1293 | * We expect signal_pending(current), but another thread may | |
1294 | * have handled it for us already. | |
1295 | */ | |
c19384b5 | 1296 | if (!abs_time) |
72c1bbf3 NP |
1297 | return -ERESTARTSYS; |
1298 | else { | |
1299 | struct restart_block *restart; | |
1300 | restart = ¤t_thread_info()->restart_block; | |
1301 | restart->fn = futex_wait_restart; | |
ce6bd420 SR |
1302 | restart->futex.uaddr = (u32 *)uaddr; |
1303 | restart->futex.val = val; | |
1304 | restart->futex.time = abs_time->tv64; | |
cd689985 | 1305 | restart->futex.bitset = bitset; |
ce6bd420 SR |
1306 | restart->futex.flags = 0; |
1307 | ||
34f01cc1 | 1308 | if (fshared) |
ce6bd420 | 1309 | restart->futex.flags |= FLAGS_SHARED; |
1acdac10 TG |
1310 | if (clockrt) |
1311 | restart->futex.flags |= FLAGS_CLOCKRT; | |
72c1bbf3 NP |
1312 | return -ERESTART_RESTARTBLOCK; |
1313 | } | |
1da177e4 | 1314 | |
42d35d48 | 1315 | out_unlock_put_key: |
c87e2837 | 1316 | queue_unlock(&q, hb); |
38d47c1b | 1317 | put_futex_key(fshared, &q.key); |
42d35d48 DH |
1318 | |
1319 | out: | |
c87e2837 IM |
1320 | return ret; |
1321 | } | |
1322 | ||
72c1bbf3 NP |
1323 | |
1324 | static long futex_wait_restart(struct restart_block *restart) | |
1325 | { | |
ce6bd420 | 1326 | u32 __user *uaddr = (u32 __user *)restart->futex.uaddr; |
c2f9f201 | 1327 | int fshared = 0; |
ce6bd420 | 1328 | ktime_t t; |
72c1bbf3 | 1329 | |
ce6bd420 | 1330 | t.tv64 = restart->futex.time; |
72c1bbf3 | 1331 | restart->fn = do_no_restart_syscall; |
ce6bd420 | 1332 | if (restart->futex.flags & FLAGS_SHARED) |
c2f9f201 | 1333 | fshared = 1; |
cd689985 | 1334 | return (long)futex_wait(uaddr, fshared, restart->futex.val, &t, |
1acdac10 TG |
1335 | restart->futex.bitset, |
1336 | restart->futex.flags & FLAGS_CLOCKRT); | |
72c1bbf3 NP |
1337 | } |
1338 | ||
1339 | ||
c87e2837 IM |
1340 | /* |
1341 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
1342 | * and failed. The kernel side here does the whole locking operation: | |
1343 | * if there are waiters then it will block, it does PI, etc. (Due to | |
1344 | * races the kernel might see a 0 value of the futex too.) | |
1345 | */ | |
c2f9f201 | 1346 | static int futex_lock_pi(u32 __user *uaddr, int fshared, |
34f01cc1 | 1347 | int detect, ktime_t *time, int trylock) |
c87e2837 | 1348 | { |
c5780e97 | 1349 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 IM |
1350 | struct task_struct *curr = current; |
1351 | struct futex_hash_bucket *hb; | |
1352 | u32 uval, newval, curval; | |
1353 | struct futex_q q; | |
778e9a9c | 1354 | int ret, lock_taken, ownerdied = 0, attempt = 0; |
c87e2837 IM |
1355 | |
1356 | if (refill_pi_state_cache()) | |
1357 | return -ENOMEM; | |
1358 | ||
c19384b5 | 1359 | if (time) { |
c5780e97 | 1360 | to = &timeout; |
237fc6e7 TG |
1361 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
1362 | HRTIMER_MODE_ABS); | |
c5780e97 | 1363 | hrtimer_init_sleeper(to, current); |
cc584b21 | 1364 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
1365 | } |
1366 | ||
c87e2837 | 1367 | q.pi_state = NULL; |
42d35d48 | 1368 | retry: |
38d47c1b | 1369 | q.key = FUTEX_KEY_INIT; |
34f01cc1 | 1370 | ret = get_futex_key(uaddr, fshared, &q.key); |
c87e2837 | 1371 | if (unlikely(ret != 0)) |
42d35d48 | 1372 | goto out; |
c87e2837 | 1373 | |
42d35d48 | 1374 | retry_unlocked: |
82af7aca | 1375 | hb = queue_lock(&q); |
c87e2837 | 1376 | |
42d35d48 | 1377 | retry_locked: |
778e9a9c | 1378 | ret = lock_taken = 0; |
d0aa7a70 | 1379 | |
c87e2837 IM |
1380 | /* |
1381 | * To avoid races, we attempt to take the lock here again | |
1382 | * (by doing a 0 -> TID atomic cmpxchg), while holding all | |
1383 | * the locks. It will most likely not succeed. | |
1384 | */ | |
b488893a | 1385 | newval = task_pid_vnr(current); |
c87e2837 | 1386 | |
36cf3b5c | 1387 | curval = cmpxchg_futex_value_locked(uaddr, 0, newval); |
c87e2837 IM |
1388 | |
1389 | if (unlikely(curval == -EFAULT)) | |
1390 | goto uaddr_faulted; | |
1391 | ||
778e9a9c AK |
1392 | /* |
1393 | * Detect deadlocks. In case of REQUEUE_PI this is a valid | |
1394 | * situation and we return success to user space. | |
1395 | */ | |
b488893a | 1396 | if (unlikely((curval & FUTEX_TID_MASK) == task_pid_vnr(current))) { |
bd197234 | 1397 | ret = -EDEADLK; |
42d35d48 | 1398 | goto out_unlock_put_key; |
c87e2837 IM |
1399 | } |
1400 | ||
1401 | /* | |
778e9a9c | 1402 | * Surprise - we got the lock. Just return to userspace: |
c87e2837 IM |
1403 | */ |
1404 | if (unlikely(!curval)) | |
42d35d48 | 1405 | goto out_unlock_put_key; |
c87e2837 IM |
1406 | |
1407 | uval = curval; | |
778e9a9c | 1408 | |
d0aa7a70 | 1409 | /* |
778e9a9c AK |
1410 | * Set the WAITERS flag, so the owner will know it has someone |
1411 | * to wake at next unlock | |
d0aa7a70 | 1412 | */ |
778e9a9c AK |
1413 | newval = curval | FUTEX_WAITERS; |
1414 | ||
1415 | /* | |
1416 | * There are two cases, where a futex might have no owner (the | |
bd197234 TG |
1417 | * owner TID is 0): OWNER_DIED. We take over the futex in this |
1418 | * case. We also do an unconditional take over, when the owner | |
1419 | * of the futex died. | |
778e9a9c AK |
1420 | * |
1421 | * This is safe as we are protected by the hash bucket lock ! | |
1422 | */ | |
1423 | if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { | |
bd197234 | 1424 | /* Keep the OWNER_DIED bit */ |
b488893a | 1425 | newval = (curval & ~FUTEX_TID_MASK) | task_pid_vnr(current); |
778e9a9c AK |
1426 | ownerdied = 0; |
1427 | lock_taken = 1; | |
1428 | } | |
c87e2837 | 1429 | |
36cf3b5c | 1430 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
c87e2837 IM |
1431 | |
1432 | if (unlikely(curval == -EFAULT)) | |
1433 | goto uaddr_faulted; | |
1434 | if (unlikely(curval != uval)) | |
1435 | goto retry_locked; | |
1436 | ||
778e9a9c | 1437 | /* |
bd197234 | 1438 | * We took the lock due to owner died take over. |
778e9a9c | 1439 | */ |
bd197234 | 1440 | if (unlikely(lock_taken)) |
42d35d48 | 1441 | goto out_unlock_put_key; |
d0aa7a70 | 1442 | |
c87e2837 IM |
1443 | /* |
1444 | * We dont have the lock. Look up the PI state (or create it if | |
1445 | * we are the first waiter): | |
1446 | */ | |
d0aa7a70 | 1447 | ret = lookup_pi_state(uval, hb, &q.key, &q.pi_state); |
c87e2837 IM |
1448 | |
1449 | if (unlikely(ret)) { | |
778e9a9c | 1450 | switch (ret) { |
c87e2837 | 1451 | |
778e9a9c AK |
1452 | case -EAGAIN: |
1453 | /* | |
1454 | * Task is exiting and we just wait for the | |
1455 | * exit to complete. | |
1456 | */ | |
1457 | queue_unlock(&q, hb); | |
778e9a9c AK |
1458 | cond_resched(); |
1459 | goto retry; | |
c87e2837 | 1460 | |
778e9a9c AK |
1461 | case -ESRCH: |
1462 | /* | |
1463 | * No owner found for this futex. Check if the | |
1464 | * OWNER_DIED bit is set to figure out whether | |
1465 | * this is a robust futex or not. | |
1466 | */ | |
1467 | if (get_futex_value_locked(&curval, uaddr)) | |
c87e2837 | 1468 | goto uaddr_faulted; |
778e9a9c AK |
1469 | |
1470 | /* | |
1471 | * We simply start over in case of a robust | |
1472 | * futex. The code above will take the futex | |
1473 | * and return happy. | |
1474 | */ | |
1475 | if (curval & FUTEX_OWNER_DIED) { | |
1476 | ownerdied = 1; | |
c87e2837 | 1477 | goto retry_locked; |
778e9a9c AK |
1478 | } |
1479 | default: | |
42d35d48 | 1480 | goto out_unlock_put_key; |
c87e2837 | 1481 | } |
c87e2837 IM |
1482 | } |
1483 | ||
1484 | /* | |
1485 | * Only actually queue now that the atomic ops are done: | |
1486 | */ | |
82af7aca | 1487 | queue_me(&q, hb); |
c87e2837 | 1488 | |
c87e2837 IM |
1489 | WARN_ON(!q.pi_state); |
1490 | /* | |
1491 | * Block on the PI mutex: | |
1492 | */ | |
1493 | if (!trylock) | |
1494 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); | |
1495 | else { | |
1496 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); | |
1497 | /* Fixup the trylock return value: */ | |
1498 | ret = ret ? 0 : -EWOULDBLOCK; | |
1499 | } | |
1500 | ||
a99e4e41 | 1501 | spin_lock(q.lock_ptr); |
c87e2837 | 1502 | |
778e9a9c AK |
1503 | if (!ret) { |
1504 | /* | |
1505 | * Got the lock. We might not be the anticipated owner | |
1506 | * if we did a lock-steal - fix up the PI-state in | |
1507 | * that case: | |
1508 | */ | |
1509 | if (q.pi_state->owner != curr) | |
1b7558e4 | 1510 | ret = fixup_pi_state_owner(uaddr, &q, curr, fshared); |
778e9a9c | 1511 | } else { |
c87e2837 IM |
1512 | /* |
1513 | * Catch the rare case, where the lock was released | |
778e9a9c AK |
1514 | * when we were on the way back before we locked the |
1515 | * hash bucket. | |
c87e2837 | 1516 | */ |
cdf71a10 TG |
1517 | if (q.pi_state->owner == curr) { |
1518 | /* | |
1519 | * Try to get the rt_mutex now. This might | |
1520 | * fail as some other task acquired the | |
1521 | * rt_mutex after we removed ourself from the | |
1522 | * rt_mutex waiters list. | |
1523 | */ | |
1524 | if (rt_mutex_trylock(&q.pi_state->pi_mutex)) | |
1525 | ret = 0; | |
1526 | else { | |
1527 | /* | |
1528 | * pi_state is incorrect, some other | |
1529 | * task did a lock steal and we | |
1530 | * returned due to timeout or signal | |
1531 | * without taking the rt_mutex. Too | |
1532 | * late. We can access the | |
1533 | * rt_mutex_owner without locking, as | |
1534 | * the other task is now blocked on | |
1535 | * the hash bucket lock. Fix the state | |
1536 | * up. | |
1537 | */ | |
1538 | struct task_struct *owner; | |
1539 | int res; | |
1540 | ||
1541 | owner = rt_mutex_owner(&q.pi_state->pi_mutex); | |
1b7558e4 TG |
1542 | res = fixup_pi_state_owner(uaddr, &q, owner, |
1543 | fshared); | |
cdf71a10 | 1544 | |
cdf71a10 TG |
1545 | /* propagate -EFAULT, if the fixup failed */ |
1546 | if (res) | |
1547 | ret = res; | |
1548 | } | |
778e9a9c AK |
1549 | } else { |
1550 | /* | |
1551 | * Paranoia check. If we did not take the lock | |
1552 | * in the trylock above, then we should not be | |
1553 | * the owner of the rtmutex, neither the real | |
1554 | * nor the pending one: | |
1555 | */ | |
1556 | if (rt_mutex_owner(&q.pi_state->pi_mutex) == curr) | |
1557 | printk(KERN_ERR "futex_lock_pi: ret = %d " | |
1558 | "pi-mutex: %p pi-state %p\n", ret, | |
1559 | q.pi_state->pi_mutex.owner, | |
1560 | q.pi_state->owner); | |
c87e2837 | 1561 | } |
c87e2837 IM |
1562 | } |
1563 | ||
778e9a9c AK |
1564 | /* Unqueue and drop the lock */ |
1565 | unqueue_me_pi(&q); | |
c87e2837 | 1566 | |
237fc6e7 TG |
1567 | if (to) |
1568 | destroy_hrtimer_on_stack(&to->timer); | |
c5780e97 | 1569 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 1570 | |
42d35d48 | 1571 | out_unlock_put_key: |
c87e2837 IM |
1572 | queue_unlock(&q, hb); |
1573 | ||
42d35d48 | 1574 | out_put_key: |
38d47c1b | 1575 | put_futex_key(fshared, &q.key); |
42d35d48 | 1576 | out: |
237fc6e7 TG |
1577 | if (to) |
1578 | destroy_hrtimer_on_stack(&to->timer); | |
c87e2837 IM |
1579 | return ret; |
1580 | ||
42d35d48 | 1581 | uaddr_faulted: |
c87e2837 | 1582 | /* |
b5686363 DH |
1583 | * We have to r/w *(int __user *)uaddr, and we have to modify it |
1584 | * atomically. Therefore, if we continue to fault after get_user() | |
1585 | * below, we need to handle the fault ourselves, while still holding | |
1586 | * the mmap_sem. This can occur if the uaddr is under contention as | |
1587 | * we have to drop the mmap_sem in order to call get_user(). | |
c87e2837 | 1588 | */ |
778e9a9c AK |
1589 | queue_unlock(&q, hb); |
1590 | ||
c87e2837 | 1591 | if (attempt++) { |
c2f9f201 | 1592 | ret = futex_handle_fault((unsigned long)uaddr, attempt); |
34f01cc1 | 1593 | if (ret) |
42d35d48 | 1594 | goto out_put_key; |
778e9a9c | 1595 | goto retry_unlocked; |
c87e2837 IM |
1596 | } |
1597 | ||
c87e2837 | 1598 | ret = get_user(uval, uaddr); |
b5686363 | 1599 | if (!ret) |
c87e2837 IM |
1600 | goto retry; |
1601 | ||
237fc6e7 TG |
1602 | if (to) |
1603 | destroy_hrtimer_on_stack(&to->timer); | |
c87e2837 IM |
1604 | return ret; |
1605 | } | |
1606 | ||
c87e2837 IM |
1607 | /* |
1608 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
1609 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
1610 | * and do the rt-mutex unlock. | |
1611 | */ | |
c2f9f201 | 1612 | static int futex_unlock_pi(u32 __user *uaddr, int fshared) |
c87e2837 IM |
1613 | { |
1614 | struct futex_hash_bucket *hb; | |
1615 | struct futex_q *this, *next; | |
1616 | u32 uval; | |
ec92d082 | 1617 | struct plist_head *head; |
38d47c1b | 1618 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 IM |
1619 | int ret, attempt = 0; |
1620 | ||
1621 | retry: | |
1622 | if (get_user(uval, uaddr)) | |
1623 | return -EFAULT; | |
1624 | /* | |
1625 | * We release only a lock we actually own: | |
1626 | */ | |
b488893a | 1627 | if ((uval & FUTEX_TID_MASK) != task_pid_vnr(current)) |
c87e2837 | 1628 | return -EPERM; |
c87e2837 | 1629 | |
34f01cc1 | 1630 | ret = get_futex_key(uaddr, fshared, &key); |
c87e2837 IM |
1631 | if (unlikely(ret != 0)) |
1632 | goto out; | |
1633 | ||
1634 | hb = hash_futex(&key); | |
778e9a9c | 1635 | retry_unlocked: |
c87e2837 IM |
1636 | spin_lock(&hb->lock); |
1637 | ||
c87e2837 IM |
1638 | /* |
1639 | * To avoid races, try to do the TID -> 0 atomic transition | |
1640 | * again. If it succeeds then we can return without waking | |
1641 | * anyone else up: | |
1642 | */ | |
36cf3b5c | 1643 | if (!(uval & FUTEX_OWNER_DIED)) |
b488893a | 1644 | uval = cmpxchg_futex_value_locked(uaddr, task_pid_vnr(current), 0); |
36cf3b5c | 1645 | |
c87e2837 IM |
1646 | |
1647 | if (unlikely(uval == -EFAULT)) | |
1648 | goto pi_faulted; | |
1649 | /* | |
1650 | * Rare case: we managed to release the lock atomically, | |
1651 | * no need to wake anyone else up: | |
1652 | */ | |
b488893a | 1653 | if (unlikely(uval == task_pid_vnr(current))) |
c87e2837 IM |
1654 | goto out_unlock; |
1655 | ||
1656 | /* | |
1657 | * Ok, other tasks may need to be woken up - check waiters | |
1658 | * and do the wakeup if necessary: | |
1659 | */ | |
1660 | head = &hb->chain; | |
1661 | ||
ec92d082 | 1662 | plist_for_each_entry_safe(this, next, head, list) { |
c87e2837 IM |
1663 | if (!match_futex (&this->key, &key)) |
1664 | continue; | |
1665 | ret = wake_futex_pi(uaddr, uval, this); | |
1666 | /* | |
1667 | * The atomic access to the futex value | |
1668 | * generated a pagefault, so retry the | |
1669 | * user-access and the wakeup: | |
1670 | */ | |
1671 | if (ret == -EFAULT) | |
1672 | goto pi_faulted; | |
1673 | goto out_unlock; | |
1674 | } | |
1675 | /* | |
1676 | * No waiters - kernel unlocks the futex: | |
1677 | */ | |
e3f2ddea IM |
1678 | if (!(uval & FUTEX_OWNER_DIED)) { |
1679 | ret = unlock_futex_pi(uaddr, uval); | |
1680 | if (ret == -EFAULT) | |
1681 | goto pi_faulted; | |
1682 | } | |
c87e2837 IM |
1683 | |
1684 | out_unlock: | |
1685 | spin_unlock(&hb->lock); | |
38d47c1b | 1686 | put_futex_key(fshared, &key); |
c87e2837 | 1687 | |
42d35d48 | 1688 | out: |
c87e2837 IM |
1689 | return ret; |
1690 | ||
1691 | pi_faulted: | |
1692 | /* | |
b5686363 DH |
1693 | * We have to r/w *(int __user *)uaddr, and we have to modify it |
1694 | * atomically. Therefore, if we continue to fault after get_user() | |
1695 | * below, we need to handle the fault ourselves, while still holding | |
1696 | * the mmap_sem. This can occur if the uaddr is under contention as | |
1697 | * we have to drop the mmap_sem in order to call get_user(). | |
c87e2837 | 1698 | */ |
778e9a9c AK |
1699 | spin_unlock(&hb->lock); |
1700 | ||
c87e2837 | 1701 | if (attempt++) { |
c2f9f201 | 1702 | ret = futex_handle_fault((unsigned long)uaddr, attempt); |
34f01cc1 | 1703 | if (ret) |
778e9a9c | 1704 | goto out; |
187226f5 | 1705 | uval = 0; |
778e9a9c | 1706 | goto retry_unlocked; |
c87e2837 IM |
1707 | } |
1708 | ||
c87e2837 | 1709 | ret = get_user(uval, uaddr); |
b5686363 | 1710 | if (!ret) |
c87e2837 IM |
1711 | goto retry; |
1712 | ||
1da177e4 LT |
1713 | return ret; |
1714 | } | |
1715 | ||
0771dfef IM |
1716 | /* |
1717 | * Support for robust futexes: the kernel cleans up held futexes at | |
1718 | * thread exit time. | |
1719 | * | |
1720 | * Implementation: user-space maintains a per-thread list of locks it | |
1721 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
1722 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 1723 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
1724 | * always manipulated with the lock held, so the list is private and |
1725 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
1726 | * field, to allow the kernel to clean up if the thread dies after | |
1727 | * acquiring the lock, but just before it could have added itself to | |
1728 | * the list. There can only be one such pending lock. | |
1729 | */ | |
1730 | ||
1731 | /** | |
1732 | * sys_set_robust_list - set the robust-futex list head of a task | |
1733 | * @head: pointer to the list-head | |
1734 | * @len: length of the list-head, as userspace expects | |
1735 | */ | |
836f92ad HC |
1736 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
1737 | size_t, len) | |
0771dfef | 1738 | { |
a0c1e907 TG |
1739 | if (!futex_cmpxchg_enabled) |
1740 | return -ENOSYS; | |
0771dfef IM |
1741 | /* |
1742 | * The kernel knows only one size for now: | |
1743 | */ | |
1744 | if (unlikely(len != sizeof(*head))) | |
1745 | return -EINVAL; | |
1746 | ||
1747 | current->robust_list = head; | |
1748 | ||
1749 | return 0; | |
1750 | } | |
1751 | ||
1752 | /** | |
1753 | * sys_get_robust_list - get the robust-futex list head of a task | |
1754 | * @pid: pid of the process [zero for current task] | |
1755 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
1756 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
1757 | */ | |
836f92ad HC |
1758 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
1759 | struct robust_list_head __user * __user *, head_ptr, | |
1760 | size_t __user *, len_ptr) | |
0771dfef | 1761 | { |
ba46df98 | 1762 | struct robust_list_head __user *head; |
0771dfef | 1763 | unsigned long ret; |
c69e8d9c | 1764 | const struct cred *cred = current_cred(), *pcred; |
0771dfef | 1765 | |
a0c1e907 TG |
1766 | if (!futex_cmpxchg_enabled) |
1767 | return -ENOSYS; | |
1768 | ||
0771dfef IM |
1769 | if (!pid) |
1770 | head = current->robust_list; | |
1771 | else { | |
1772 | struct task_struct *p; | |
1773 | ||
1774 | ret = -ESRCH; | |
aaa2a97e | 1775 | rcu_read_lock(); |
228ebcbe | 1776 | p = find_task_by_vpid(pid); |
0771dfef IM |
1777 | if (!p) |
1778 | goto err_unlock; | |
1779 | ret = -EPERM; | |
c69e8d9c DH |
1780 | pcred = __task_cred(p); |
1781 | if (cred->euid != pcred->euid && | |
1782 | cred->euid != pcred->uid && | |
76aac0e9 | 1783 | !capable(CAP_SYS_PTRACE)) |
0771dfef IM |
1784 | goto err_unlock; |
1785 | head = p->robust_list; | |
aaa2a97e | 1786 | rcu_read_unlock(); |
0771dfef IM |
1787 | } |
1788 | ||
1789 | if (put_user(sizeof(*head), len_ptr)) | |
1790 | return -EFAULT; | |
1791 | return put_user(head, head_ptr); | |
1792 | ||
1793 | err_unlock: | |
aaa2a97e | 1794 | rcu_read_unlock(); |
0771dfef IM |
1795 | |
1796 | return ret; | |
1797 | } | |
1798 | ||
1799 | /* | |
1800 | * Process a futex-list entry, check whether it's owned by the | |
1801 | * dying task, and do notification if so: | |
1802 | */ | |
e3f2ddea | 1803 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 1804 | { |
e3f2ddea | 1805 | u32 uval, nval, mval; |
0771dfef | 1806 | |
8f17d3a5 IM |
1807 | retry: |
1808 | if (get_user(uval, uaddr)) | |
0771dfef IM |
1809 | return -1; |
1810 | ||
b488893a | 1811 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
1812 | /* |
1813 | * Ok, this dying thread is truly holding a futex | |
1814 | * of interest. Set the OWNER_DIED bit atomically | |
1815 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
1816 | * set, wake up a waiter (if any). (We have to do a | |
1817 | * futex_wake() even if OWNER_DIED is already set - | |
1818 | * to handle the rare but possible case of recursive | |
1819 | * thread-death.) The rest of the cleanup is done in | |
1820 | * userspace. | |
1821 | */ | |
e3f2ddea IM |
1822 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
1823 | nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval); | |
1824 | ||
c87e2837 IM |
1825 | if (nval == -EFAULT) |
1826 | return -1; | |
1827 | ||
1828 | if (nval != uval) | |
8f17d3a5 | 1829 | goto retry; |
0771dfef | 1830 | |
e3f2ddea IM |
1831 | /* |
1832 | * Wake robust non-PI futexes here. The wakeup of | |
1833 | * PI futexes happens in exit_pi_state(): | |
1834 | */ | |
36cf3b5c | 1835 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 1836 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
1837 | } |
1838 | return 0; | |
1839 | } | |
1840 | ||
e3f2ddea IM |
1841 | /* |
1842 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
1843 | */ | |
1844 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 AV |
1845 | struct robust_list __user * __user *head, |
1846 | int *pi) | |
e3f2ddea IM |
1847 | { |
1848 | unsigned long uentry; | |
1849 | ||
ba46df98 | 1850 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
1851 | return -EFAULT; |
1852 | ||
ba46df98 | 1853 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
1854 | *pi = uentry & 1; |
1855 | ||
1856 | return 0; | |
1857 | } | |
1858 | ||
0771dfef IM |
1859 | /* |
1860 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
1861 | * and mark any locks found there dead, and notify any waiters. | |
1862 | * | |
1863 | * We silently return on any sign of list-walking problem. | |
1864 | */ | |
1865 | void exit_robust_list(struct task_struct *curr) | |
1866 | { | |
1867 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e MS |
1868 | struct robust_list __user *entry, *next_entry, *pending; |
1869 | unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip; | |
0771dfef | 1870 | unsigned long futex_offset; |
9f96cb1e | 1871 | int rc; |
0771dfef | 1872 | |
a0c1e907 TG |
1873 | if (!futex_cmpxchg_enabled) |
1874 | return; | |
1875 | ||
0771dfef IM |
1876 | /* |
1877 | * Fetch the list head (which was registered earlier, via | |
1878 | * sys_set_robust_list()): | |
1879 | */ | |
e3f2ddea | 1880 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
1881 | return; |
1882 | /* | |
1883 | * Fetch the relative futex offset: | |
1884 | */ | |
1885 | if (get_user(futex_offset, &head->futex_offset)) | |
1886 | return; | |
1887 | /* | |
1888 | * Fetch any possibly pending lock-add first, and handle it | |
1889 | * if it exists: | |
1890 | */ | |
e3f2ddea | 1891 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 1892 | return; |
e3f2ddea | 1893 | |
9f96cb1e | 1894 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 1895 | while (entry != &head->list) { |
9f96cb1e MS |
1896 | /* |
1897 | * Fetch the next entry in the list before calling | |
1898 | * handle_futex_death: | |
1899 | */ | |
1900 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
1901 | /* |
1902 | * A pending lock might already be on the list, so | |
c87e2837 | 1903 | * don't process it twice: |
0771dfef IM |
1904 | */ |
1905 | if (entry != pending) | |
ba46df98 | 1906 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 1907 | curr, pi)) |
0771dfef | 1908 | return; |
9f96cb1e | 1909 | if (rc) |
0771dfef | 1910 | return; |
9f96cb1e MS |
1911 | entry = next_entry; |
1912 | pi = next_pi; | |
0771dfef IM |
1913 | /* |
1914 | * Avoid excessively long or circular lists: | |
1915 | */ | |
1916 | if (!--limit) | |
1917 | break; | |
1918 | ||
1919 | cond_resched(); | |
1920 | } | |
9f96cb1e MS |
1921 | |
1922 | if (pending) | |
1923 | handle_futex_death((void __user *)pending + futex_offset, | |
1924 | curr, pip); | |
0771dfef IM |
1925 | } |
1926 | ||
c19384b5 | 1927 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 1928 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 1929 | { |
1acdac10 | 1930 | int clockrt, ret = -ENOSYS; |
34f01cc1 | 1931 | int cmd = op & FUTEX_CMD_MASK; |
c2f9f201 | 1932 | int fshared = 0; |
34f01cc1 ED |
1933 | |
1934 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
c2f9f201 | 1935 | fshared = 1; |
1da177e4 | 1936 | |
1acdac10 TG |
1937 | clockrt = op & FUTEX_CLOCK_REALTIME; |
1938 | if (clockrt && cmd != FUTEX_WAIT_BITSET) | |
1939 | return -ENOSYS; | |
1da177e4 | 1940 | |
34f01cc1 | 1941 | switch (cmd) { |
1da177e4 | 1942 | case FUTEX_WAIT: |
cd689985 TG |
1943 | val3 = FUTEX_BITSET_MATCH_ANY; |
1944 | case FUTEX_WAIT_BITSET: | |
1acdac10 | 1945 | ret = futex_wait(uaddr, fshared, val, timeout, val3, clockrt); |
1da177e4 LT |
1946 | break; |
1947 | case FUTEX_WAKE: | |
cd689985 TG |
1948 | val3 = FUTEX_BITSET_MATCH_ANY; |
1949 | case FUTEX_WAKE_BITSET: | |
1950 | ret = futex_wake(uaddr, fshared, val, val3); | |
1da177e4 | 1951 | break; |
1da177e4 | 1952 | case FUTEX_REQUEUE: |
34f01cc1 | 1953 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL); |
1da177e4 LT |
1954 | break; |
1955 | case FUTEX_CMP_REQUEUE: | |
34f01cc1 | 1956 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3); |
1da177e4 | 1957 | break; |
4732efbe | 1958 | case FUTEX_WAKE_OP: |
34f01cc1 | 1959 | ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3); |
4732efbe | 1960 | break; |
c87e2837 | 1961 | case FUTEX_LOCK_PI: |
a0c1e907 TG |
1962 | if (futex_cmpxchg_enabled) |
1963 | ret = futex_lock_pi(uaddr, fshared, val, timeout, 0); | |
c87e2837 IM |
1964 | break; |
1965 | case FUTEX_UNLOCK_PI: | |
a0c1e907 TG |
1966 | if (futex_cmpxchg_enabled) |
1967 | ret = futex_unlock_pi(uaddr, fshared); | |
c87e2837 IM |
1968 | break; |
1969 | case FUTEX_TRYLOCK_PI: | |
a0c1e907 TG |
1970 | if (futex_cmpxchg_enabled) |
1971 | ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1); | |
c87e2837 | 1972 | break; |
1da177e4 LT |
1973 | default: |
1974 | ret = -ENOSYS; | |
1975 | } | |
1976 | return ret; | |
1977 | } | |
1978 | ||
1979 | ||
17da2bd9 HC |
1980 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
1981 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
1982 | u32, val3) | |
1da177e4 | 1983 | { |
c19384b5 PP |
1984 | struct timespec ts; |
1985 | ktime_t t, *tp = NULL; | |
e2970f2f | 1986 | u32 val2 = 0; |
34f01cc1 | 1987 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 1988 | |
cd689985 TG |
1989 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
1990 | cmd == FUTEX_WAIT_BITSET)) { | |
c19384b5 | 1991 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 1992 | return -EFAULT; |
c19384b5 | 1993 | if (!timespec_valid(&ts)) |
9741ef96 | 1994 | return -EINVAL; |
c19384b5 PP |
1995 | |
1996 | t = timespec_to_ktime(ts); | |
34f01cc1 | 1997 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 1998 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 1999 | tp = &t; |
1da177e4 LT |
2000 | } |
2001 | /* | |
34f01cc1 | 2002 | * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE. |
f54f0986 | 2003 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 2004 | */ |
f54f0986 AS |
2005 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
2006 | cmd == FUTEX_WAKE_OP) | |
e2970f2f | 2007 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 2008 | |
c19384b5 | 2009 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
2010 | } |
2011 | ||
f6d107fb | 2012 | static int __init futex_init(void) |
1da177e4 | 2013 | { |
a0c1e907 | 2014 | u32 curval; |
3e4ab747 | 2015 | int i; |
95362fa9 | 2016 | |
a0c1e907 TG |
2017 | /* |
2018 | * This will fail and we want it. Some arch implementations do | |
2019 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
2020 | * functionality. We want to know that before we call in any | |
2021 | * of the complex code paths. Also we want to prevent | |
2022 | * registration of robust lists in that case. NULL is | |
2023 | * guaranteed to fault and we get -EFAULT on functional | |
2024 | * implementation, the non functional ones will return | |
2025 | * -ENOSYS. | |
2026 | */ | |
2027 | curval = cmpxchg_futex_value_locked(NULL, 0, 0); | |
2028 | if (curval == -EFAULT) | |
2029 | futex_cmpxchg_enabled = 1; | |
2030 | ||
3e4ab747 TG |
2031 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { |
2032 | plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock); | |
2033 | spin_lock_init(&futex_queues[i].lock); | |
2034 | } | |
2035 | ||
1da177e4 LT |
2036 | return 0; |
2037 | } | |
f6d107fb | 2038 | __initcall(futex_init); |