]>
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> |
4732efbe | 55 | #include <asm/futex.h> |
1da177e4 | 56 | |
c87e2837 IM |
57 | #include "rtmutex_common.h" |
58 | ||
1da177e4 LT |
59 | #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8) |
60 | ||
c87e2837 IM |
61 | /* |
62 | * Priority Inheritance state: | |
63 | */ | |
64 | struct futex_pi_state { | |
65 | /* | |
66 | * list of 'owned' pi_state instances - these have to be | |
67 | * cleaned up in do_exit() if the task exits prematurely: | |
68 | */ | |
69 | struct list_head list; | |
70 | ||
71 | /* | |
72 | * The PI object: | |
73 | */ | |
74 | struct rt_mutex pi_mutex; | |
75 | ||
76 | struct task_struct *owner; | |
77 | atomic_t refcount; | |
78 | ||
79 | union futex_key key; | |
80 | }; | |
81 | ||
1da177e4 LT |
82 | /* |
83 | * We use this hashed waitqueue instead of a normal wait_queue_t, so | |
84 | * we can wake only the relevant ones (hashed queues may be shared). | |
85 | * | |
86 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 87 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
1da177e4 LT |
88 | * The order of wakup is always to make the first condition true, then |
89 | * wake up q->waiters, then make the second condition true. | |
90 | */ | |
91 | struct futex_q { | |
ec92d082 | 92 | struct plist_node list; |
1da177e4 LT |
93 | wait_queue_head_t waiters; |
94 | ||
e2970f2f | 95 | /* Which hash list lock to use: */ |
1da177e4 LT |
96 | spinlock_t *lock_ptr; |
97 | ||
e2970f2f | 98 | /* Key which the futex is hashed on: */ |
1da177e4 LT |
99 | union futex_key key; |
100 | ||
e2970f2f | 101 | /* For fd, sigio sent using these: */ |
1da177e4 LT |
102 | int fd; |
103 | struct file *filp; | |
c87e2837 IM |
104 | |
105 | /* Optional priority inheritance state: */ | |
106 | struct futex_pi_state *pi_state; | |
107 | struct task_struct *task; | |
1da177e4 LT |
108 | }; |
109 | ||
110 | /* | |
111 | * Split the global futex_lock into every hash list lock. | |
112 | */ | |
113 | struct futex_hash_bucket { | |
ec92d082 PP |
114 | spinlock_t lock; |
115 | struct plist_head chain; | |
1da177e4 LT |
116 | }; |
117 | ||
118 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; | |
119 | ||
120 | /* Futex-fs vfsmount entry: */ | |
121 | static struct vfsmount *futex_mnt; | |
122 | ||
36cf3b5c TG |
123 | /* |
124 | * Take mm->mmap_sem, when futex is shared | |
125 | */ | |
126 | static inline void futex_lock_mm(struct rw_semaphore *fshared) | |
127 | { | |
128 | if (fshared) | |
129 | down_read(fshared); | |
130 | } | |
131 | ||
132 | /* | |
133 | * Release mm->mmap_sem, when the futex is shared | |
134 | */ | |
135 | static inline void futex_unlock_mm(struct rw_semaphore *fshared) | |
136 | { | |
137 | if (fshared) | |
138 | up_read(fshared); | |
139 | } | |
140 | ||
1da177e4 LT |
141 | /* |
142 | * We hash on the keys returned from get_futex_key (see below). | |
143 | */ | |
144 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
145 | { | |
146 | u32 hash = jhash2((u32*)&key->both.word, | |
147 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
148 | key->both.offset); | |
149 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; | |
150 | } | |
151 | ||
152 | /* | |
153 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
154 | */ | |
155 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
156 | { | |
157 | return (key1->both.word == key2->both.word | |
158 | && key1->both.ptr == key2->both.ptr | |
159 | && key1->both.offset == key2->both.offset); | |
160 | } | |
161 | ||
34f01cc1 ED |
162 | /** |
163 | * get_futex_key - Get parameters which are the keys for a futex. | |
164 | * @uaddr: virtual address of the futex | |
165 | * @shared: NULL for a PROCESS_PRIVATE futex, | |
166 | * ¤t->mm->mmap_sem for a PROCESS_SHARED futex | |
167 | * @key: address where result is stored. | |
168 | * | |
169 | * Returns a negative error code or 0 | |
170 | * The key words are stored in *key on success. | |
1da177e4 | 171 | * |
f3a43f3f | 172 | * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode, |
1da177e4 LT |
173 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
174 | * We can usually work out the index without swapping in the page. | |
175 | * | |
34f01cc1 ED |
176 | * fshared is NULL for PROCESS_PRIVATE futexes |
177 | * For other futexes, it points to ¤t->mm->mmap_sem and | |
178 | * caller must have taken the reader lock. but NOT any spinlocks. | |
1da177e4 | 179 | */ |
34f01cc1 ED |
180 | int get_futex_key(u32 __user *uaddr, struct rw_semaphore *fshared, |
181 | union futex_key *key) | |
1da177e4 | 182 | { |
e2970f2f | 183 | unsigned long address = (unsigned long)uaddr; |
1da177e4 LT |
184 | struct mm_struct *mm = current->mm; |
185 | struct vm_area_struct *vma; | |
186 | struct page *page; | |
187 | int err; | |
188 | ||
189 | /* | |
190 | * The futex address must be "naturally" aligned. | |
191 | */ | |
e2970f2f | 192 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 193 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 194 | return -EINVAL; |
e2970f2f | 195 | address -= key->both.offset; |
1da177e4 | 196 | |
34f01cc1 ED |
197 | /* |
198 | * PROCESS_PRIVATE futexes are fast. | |
199 | * As the mm cannot disappear under us and the 'key' only needs | |
200 | * virtual address, we dont even have to find the underlying vma. | |
201 | * Note : We do have to check 'uaddr' is a valid user address, | |
202 | * but access_ok() should be faster than find_vma() | |
203 | */ | |
204 | if (!fshared) { | |
205 | if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32)))) | |
206 | return -EFAULT; | |
207 | key->private.mm = mm; | |
208 | key->private.address = address; | |
209 | return 0; | |
210 | } | |
1da177e4 LT |
211 | /* |
212 | * The futex is hashed differently depending on whether | |
213 | * it's in a shared or private mapping. So check vma first. | |
214 | */ | |
e2970f2f | 215 | vma = find_extend_vma(mm, address); |
1da177e4 LT |
216 | if (unlikely(!vma)) |
217 | return -EFAULT; | |
218 | ||
219 | /* | |
220 | * Permissions. | |
221 | */ | |
222 | if (unlikely((vma->vm_flags & (VM_IO|VM_READ)) != VM_READ)) | |
223 | return (vma->vm_flags & VM_IO) ? -EPERM : -EACCES; | |
224 | ||
225 | /* | |
226 | * Private mappings are handled in a simple way. | |
227 | * | |
228 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
229 | * it's a read-only handle, it's expected that futexes attach to | |
230 | * the object not the particular process. Therefore we use | |
231 | * VM_MAYSHARE here, not VM_SHARED which is restricted to shared | |
232 | * mappings of _writable_ handles. | |
233 | */ | |
234 | if (likely(!(vma->vm_flags & VM_MAYSHARE))) { | |
34f01cc1 | 235 | key->both.offset |= FUT_OFF_MMSHARED; /* reference taken on mm */ |
1da177e4 | 236 | key->private.mm = mm; |
e2970f2f | 237 | key->private.address = address; |
1da177e4 LT |
238 | return 0; |
239 | } | |
240 | ||
241 | /* | |
242 | * Linear file mappings are also simple. | |
243 | */ | |
f3a43f3f | 244 | key->shared.inode = vma->vm_file->f_path.dentry->d_inode; |
34f01cc1 | 245 | key->both.offset |= FUT_OFF_INODE; /* inode-based key. */ |
1da177e4 | 246 | if (likely(!(vma->vm_flags & VM_NONLINEAR))) { |
e2970f2f | 247 | key->shared.pgoff = (((address - vma->vm_start) >> PAGE_SHIFT) |
1da177e4 LT |
248 | + vma->vm_pgoff); |
249 | return 0; | |
250 | } | |
251 | ||
252 | /* | |
253 | * We could walk the page table to read the non-linear | |
254 | * pte, and get the page index without fetching the page | |
255 | * from swap. But that's a lot of code to duplicate here | |
256 | * for a rare case, so we simply fetch the page. | |
257 | */ | |
e2970f2f | 258 | err = get_user_pages(current, mm, address, 1, 0, 0, &page, NULL); |
1da177e4 LT |
259 | if (err >= 0) { |
260 | key->shared.pgoff = | |
261 | page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
262 | put_page(page); | |
263 | return 0; | |
264 | } | |
265 | return err; | |
266 | } | |
9adef58b | 267 | EXPORT_SYMBOL_GPL(get_futex_key); |
1da177e4 LT |
268 | |
269 | /* | |
270 | * Take a reference to the resource addressed by a key. | |
271 | * Can be called while holding spinlocks. | |
272 | * | |
1da177e4 | 273 | */ |
9adef58b | 274 | inline void get_futex_key_refs(union futex_key *key) |
1da177e4 | 275 | { |
34f01cc1 ED |
276 | if (key->both.ptr == 0) |
277 | return; | |
278 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
279 | case FUT_OFF_INODE: | |
1da177e4 | 280 | atomic_inc(&key->shared.inode->i_count); |
34f01cc1 ED |
281 | break; |
282 | case FUT_OFF_MMSHARED: | |
1da177e4 | 283 | atomic_inc(&key->private.mm->mm_count); |
34f01cc1 | 284 | break; |
1da177e4 LT |
285 | } |
286 | } | |
9adef58b | 287 | EXPORT_SYMBOL_GPL(get_futex_key_refs); |
1da177e4 LT |
288 | |
289 | /* | |
290 | * Drop a reference to the resource addressed by a key. | |
291 | * The hash bucket spinlock must not be held. | |
292 | */ | |
9adef58b | 293 | void drop_futex_key_refs(union futex_key *key) |
1da177e4 | 294 | { |
34f01cc1 ED |
295 | if (key->both.ptr == 0) |
296 | return; | |
297 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
298 | case FUT_OFF_INODE: | |
1da177e4 | 299 | iput(key->shared.inode); |
34f01cc1 ED |
300 | break; |
301 | case FUT_OFF_MMSHARED: | |
1da177e4 | 302 | mmdrop(key->private.mm); |
34f01cc1 | 303 | break; |
1da177e4 LT |
304 | } |
305 | } | |
9adef58b | 306 | EXPORT_SYMBOL_GPL(drop_futex_key_refs); |
1da177e4 | 307 | |
36cf3b5c TG |
308 | static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval) |
309 | { | |
310 | u32 curval; | |
311 | ||
312 | pagefault_disable(); | |
313 | curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval); | |
314 | pagefault_enable(); | |
315 | ||
316 | return curval; | |
317 | } | |
318 | ||
319 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
320 | { |
321 | int ret; | |
322 | ||
a866374a | 323 | pagefault_disable(); |
e2970f2f | 324 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 325 | pagefault_enable(); |
1da177e4 LT |
326 | |
327 | return ret ? -EFAULT : 0; | |
328 | } | |
329 | ||
c87e2837 | 330 | /* |
34f01cc1 ED |
331 | * Fault handling. |
332 | * if fshared is non NULL, current->mm->mmap_sem is already held | |
c87e2837 | 333 | */ |
34f01cc1 ED |
334 | static int futex_handle_fault(unsigned long address, |
335 | struct rw_semaphore *fshared, int attempt) | |
c87e2837 IM |
336 | { |
337 | struct vm_area_struct * vma; | |
338 | struct mm_struct *mm = current->mm; | |
34f01cc1 | 339 | int ret = -EFAULT; |
c87e2837 | 340 | |
34f01cc1 ED |
341 | if (attempt > 2) |
342 | return ret; | |
c87e2837 | 343 | |
34f01cc1 ED |
344 | if (!fshared) |
345 | down_read(&mm->mmap_sem); | |
346 | vma = find_vma(mm, address); | |
347 | if (vma && address >= vma->vm_start && | |
348 | (vma->vm_flags & VM_WRITE)) { | |
349 | switch (handle_mm_fault(mm, vma, address, 1)) { | |
350 | case VM_FAULT_MINOR: | |
351 | ret = 0; | |
352 | current->min_flt++; | |
353 | break; | |
354 | case VM_FAULT_MAJOR: | |
355 | ret = 0; | |
356 | current->maj_flt++; | |
357 | break; | |
358 | } | |
c87e2837 | 359 | } |
34f01cc1 ED |
360 | if (!fshared) |
361 | up_read(&mm->mmap_sem); | |
362 | return ret; | |
c87e2837 IM |
363 | } |
364 | ||
365 | /* | |
366 | * PI code: | |
367 | */ | |
368 | static int refill_pi_state_cache(void) | |
369 | { | |
370 | struct futex_pi_state *pi_state; | |
371 | ||
372 | if (likely(current->pi_state_cache)) | |
373 | return 0; | |
374 | ||
4668edc3 | 375 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
376 | |
377 | if (!pi_state) | |
378 | return -ENOMEM; | |
379 | ||
c87e2837 IM |
380 | INIT_LIST_HEAD(&pi_state->list); |
381 | /* pi_mutex gets initialized later */ | |
382 | pi_state->owner = NULL; | |
383 | atomic_set(&pi_state->refcount, 1); | |
384 | ||
385 | current->pi_state_cache = pi_state; | |
386 | ||
387 | return 0; | |
388 | } | |
389 | ||
390 | static struct futex_pi_state * alloc_pi_state(void) | |
391 | { | |
392 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
393 | ||
394 | WARN_ON(!pi_state); | |
395 | current->pi_state_cache = NULL; | |
396 | ||
397 | return pi_state; | |
398 | } | |
399 | ||
400 | static void free_pi_state(struct futex_pi_state *pi_state) | |
401 | { | |
402 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
403 | return; | |
404 | ||
405 | /* | |
406 | * If pi_state->owner is NULL, the owner is most probably dying | |
407 | * and has cleaned up the pi_state already | |
408 | */ | |
409 | if (pi_state->owner) { | |
410 | spin_lock_irq(&pi_state->owner->pi_lock); | |
411 | list_del_init(&pi_state->list); | |
412 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
413 | ||
414 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
415 | } | |
416 | ||
417 | if (current->pi_state_cache) | |
418 | kfree(pi_state); | |
419 | else { | |
420 | /* | |
421 | * pi_state->list is already empty. | |
422 | * clear pi_state->owner. | |
423 | * refcount is at 0 - put it back to 1. | |
424 | */ | |
425 | pi_state->owner = NULL; | |
426 | atomic_set(&pi_state->refcount, 1); | |
427 | current->pi_state_cache = pi_state; | |
428 | } | |
429 | } | |
430 | ||
431 | /* | |
432 | * Look up the task based on what TID userspace gave us. | |
433 | * We dont trust it. | |
434 | */ | |
435 | static struct task_struct * futex_find_get_task(pid_t pid) | |
436 | { | |
437 | struct task_struct *p; | |
438 | ||
d359b549 | 439 | rcu_read_lock(); |
c87e2837 | 440 | p = find_task_by_pid(pid); |
a06381fe TG |
441 | |
442 | if (!p || ((current->euid != p->euid) && (current->euid != p->uid))) | |
443 | p = ERR_PTR(-ESRCH); | |
444 | else | |
445 | get_task_struct(p); | |
446 | ||
d359b549 | 447 | rcu_read_unlock(); |
c87e2837 IM |
448 | |
449 | return p; | |
450 | } | |
451 | ||
452 | /* | |
453 | * This task is holding PI mutexes at exit time => bad. | |
454 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
455 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
456 | */ | |
457 | void exit_pi_state_list(struct task_struct *curr) | |
458 | { | |
c87e2837 IM |
459 | struct list_head *next, *head = &curr->pi_state_list; |
460 | struct futex_pi_state *pi_state; | |
627371d7 | 461 | struct futex_hash_bucket *hb; |
c87e2837 IM |
462 | union futex_key key; |
463 | ||
464 | /* | |
465 | * We are a ZOMBIE and nobody can enqueue itself on | |
466 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 467 | * versus waiters unqueueing themselves: |
c87e2837 IM |
468 | */ |
469 | spin_lock_irq(&curr->pi_lock); | |
470 | while (!list_empty(head)) { | |
471 | ||
472 | next = head->next; | |
473 | pi_state = list_entry(next, struct futex_pi_state, list); | |
474 | key = pi_state->key; | |
627371d7 | 475 | hb = hash_futex(&key); |
c87e2837 IM |
476 | spin_unlock_irq(&curr->pi_lock); |
477 | ||
c87e2837 IM |
478 | spin_lock(&hb->lock); |
479 | ||
480 | spin_lock_irq(&curr->pi_lock); | |
627371d7 IM |
481 | /* |
482 | * We dropped the pi-lock, so re-check whether this | |
483 | * task still owns the PI-state: | |
484 | */ | |
c87e2837 IM |
485 | if (head->next != next) { |
486 | spin_unlock(&hb->lock); | |
487 | continue; | |
488 | } | |
489 | ||
c87e2837 | 490 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
491 | WARN_ON(list_empty(&pi_state->list)); |
492 | list_del_init(&pi_state->list); | |
c87e2837 IM |
493 | pi_state->owner = NULL; |
494 | spin_unlock_irq(&curr->pi_lock); | |
495 | ||
496 | rt_mutex_unlock(&pi_state->pi_mutex); | |
497 | ||
498 | spin_unlock(&hb->lock); | |
499 | ||
500 | spin_lock_irq(&curr->pi_lock); | |
501 | } | |
502 | spin_unlock_irq(&curr->pi_lock); | |
503 | } | |
504 | ||
505 | static int | |
d0aa7a70 PP |
506 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
507 | union futex_key *key, struct futex_pi_state **ps) | |
c87e2837 IM |
508 | { |
509 | struct futex_pi_state *pi_state = NULL; | |
510 | struct futex_q *this, *next; | |
ec92d082 | 511 | struct plist_head *head; |
c87e2837 | 512 | struct task_struct *p; |
778e9a9c | 513 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 IM |
514 | |
515 | head = &hb->chain; | |
516 | ||
ec92d082 | 517 | plist_for_each_entry_safe(this, next, head, list) { |
d0aa7a70 | 518 | if (match_futex(&this->key, key)) { |
c87e2837 IM |
519 | /* |
520 | * Another waiter already exists - bump up | |
521 | * the refcount and return its pi_state: | |
522 | */ | |
523 | pi_state = this->pi_state; | |
06a9ec29 TG |
524 | /* |
525 | * Userspace might have messed up non PI and PI futexes | |
526 | */ | |
527 | if (unlikely(!pi_state)) | |
528 | return -EINVAL; | |
529 | ||
627371d7 | 530 | WARN_ON(!atomic_read(&pi_state->refcount)); |
778e9a9c AK |
531 | WARN_ON(pid && pi_state->owner && |
532 | pi_state->owner->pid != pid); | |
627371d7 | 533 | |
c87e2837 | 534 | atomic_inc(&pi_state->refcount); |
d0aa7a70 | 535 | *ps = pi_state; |
c87e2837 IM |
536 | |
537 | return 0; | |
538 | } | |
539 | } | |
540 | ||
541 | /* | |
e3f2ddea | 542 | * We are the first waiter - try to look up the real owner and attach |
778e9a9c | 543 | * the new pi_state to it, but bail out when TID = 0 |
c87e2837 | 544 | */ |
778e9a9c | 545 | if (!pid) |
e3f2ddea | 546 | return -ESRCH; |
c87e2837 | 547 | p = futex_find_get_task(pid); |
778e9a9c AK |
548 | if (IS_ERR(p)) |
549 | return PTR_ERR(p); | |
550 | ||
551 | /* | |
552 | * We need to look at the task state flags to figure out, | |
553 | * whether the task is exiting. To protect against the do_exit | |
554 | * change of the task flags, we do this protected by | |
555 | * p->pi_lock: | |
556 | */ | |
557 | spin_lock_irq(&p->pi_lock); | |
558 | if (unlikely(p->flags & PF_EXITING)) { | |
559 | /* | |
560 | * The task is on the way out. When PF_EXITPIDONE is | |
561 | * set, we know that the task has finished the | |
562 | * cleanup: | |
563 | */ | |
564 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
565 | ||
566 | spin_unlock_irq(&p->pi_lock); | |
567 | put_task_struct(p); | |
568 | return ret; | |
569 | } | |
c87e2837 IM |
570 | |
571 | pi_state = alloc_pi_state(); | |
572 | ||
573 | /* | |
574 | * Initialize the pi_mutex in locked state and make 'p' | |
575 | * the owner of it: | |
576 | */ | |
577 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
578 | ||
579 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 580 | pi_state->key = *key; |
c87e2837 | 581 | |
627371d7 | 582 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
583 | list_add(&pi_state->list, &p->pi_state_list); |
584 | pi_state->owner = p; | |
585 | spin_unlock_irq(&p->pi_lock); | |
586 | ||
587 | put_task_struct(p); | |
588 | ||
d0aa7a70 | 589 | *ps = pi_state; |
c87e2837 IM |
590 | |
591 | return 0; | |
592 | } | |
593 | ||
1da177e4 LT |
594 | /* |
595 | * The hash bucket lock must be held when this is called. | |
596 | * Afterwards, the futex_q must not be accessed. | |
597 | */ | |
598 | static void wake_futex(struct futex_q *q) | |
599 | { | |
ec92d082 | 600 | plist_del(&q->list, &q->list.plist); |
1da177e4 LT |
601 | if (q->filp) |
602 | send_sigio(&q->filp->f_owner, q->fd, POLL_IN); | |
603 | /* | |
604 | * The lock in wake_up_all() is a crucial memory barrier after the | |
ec92d082 | 605 | * plist_del() and also before assigning to q->lock_ptr. |
1da177e4 LT |
606 | */ |
607 | wake_up_all(&q->waiters); | |
608 | /* | |
609 | * The waiting task can free the futex_q as soon as this is written, | |
610 | * without taking any locks. This must come last. | |
8e31108b AM |
611 | * |
612 | * A memory barrier is required here to prevent the following store | |
613 | * to lock_ptr from getting ahead of the wakeup. Clearing the lock | |
614 | * at the end of wake_up_all() does not prevent this store from | |
615 | * moving. | |
1da177e4 | 616 | */ |
ccdea2f8 | 617 | smp_wmb(); |
1da177e4 LT |
618 | q->lock_ptr = NULL; |
619 | } | |
620 | ||
c87e2837 IM |
621 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
622 | { | |
623 | struct task_struct *new_owner; | |
624 | struct futex_pi_state *pi_state = this->pi_state; | |
625 | u32 curval, newval; | |
626 | ||
627 | if (!pi_state) | |
628 | return -EINVAL; | |
629 | ||
21778867 | 630 | spin_lock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
631 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
632 | ||
633 | /* | |
634 | * This happens when we have stolen the lock and the original | |
635 | * pending owner did not enqueue itself back on the rt_mutex. | |
636 | * Thats not a tragedy. We know that way, that a lock waiter | |
637 | * is on the fly. We make the futex_q waiter the pending owner. | |
638 | */ | |
639 | if (!new_owner) | |
640 | new_owner = this->task; | |
641 | ||
642 | /* | |
643 | * We pass it to the next owner. (The WAITERS bit is always | |
644 | * kept enabled while there is PI state around. We must also | |
645 | * preserve the owner died bit.) | |
646 | */ | |
e3f2ddea | 647 | if (!(uval & FUTEX_OWNER_DIED)) { |
778e9a9c AK |
648 | int ret = 0; |
649 | ||
e3f2ddea IM |
650 | newval = FUTEX_WAITERS | new_owner->pid; |
651 | ||
36cf3b5c | 652 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
778e9a9c | 653 | |
e3f2ddea | 654 | if (curval == -EFAULT) |
778e9a9c | 655 | ret = -EFAULT; |
e3f2ddea | 656 | if (curval != uval) |
778e9a9c AK |
657 | ret = -EINVAL; |
658 | if (ret) { | |
659 | spin_unlock(&pi_state->pi_mutex.wait_lock); | |
660 | return ret; | |
661 | } | |
e3f2ddea | 662 | } |
c87e2837 | 663 | |
627371d7 IM |
664 | spin_lock_irq(&pi_state->owner->pi_lock); |
665 | WARN_ON(list_empty(&pi_state->list)); | |
666 | list_del_init(&pi_state->list); | |
667 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
668 | ||
669 | spin_lock_irq(&new_owner->pi_lock); | |
670 | WARN_ON(!list_empty(&pi_state->list)); | |
c87e2837 IM |
671 | list_add(&pi_state->list, &new_owner->pi_state_list); |
672 | pi_state->owner = new_owner; | |
627371d7 IM |
673 | spin_unlock_irq(&new_owner->pi_lock); |
674 | ||
21778867 | 675 | spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
676 | rt_mutex_unlock(&pi_state->pi_mutex); |
677 | ||
678 | return 0; | |
679 | } | |
680 | ||
681 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) | |
682 | { | |
683 | u32 oldval; | |
684 | ||
685 | /* | |
686 | * There is no waiter, so we unlock the futex. The owner died | |
687 | * bit has not to be preserved here. We are the owner: | |
688 | */ | |
36cf3b5c | 689 | oldval = cmpxchg_futex_value_locked(uaddr, uval, 0); |
c87e2837 IM |
690 | |
691 | if (oldval == -EFAULT) | |
692 | return oldval; | |
693 | if (oldval != uval) | |
694 | return -EAGAIN; | |
695 | ||
696 | return 0; | |
697 | } | |
698 | ||
8b8f319f IM |
699 | /* |
700 | * Express the locking dependencies for lockdep: | |
701 | */ | |
702 | static inline void | |
703 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
704 | { | |
705 | if (hb1 <= hb2) { | |
706 | spin_lock(&hb1->lock); | |
707 | if (hb1 < hb2) | |
708 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
709 | } else { /* hb1 > hb2 */ | |
710 | spin_lock(&hb2->lock); | |
711 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
712 | } | |
713 | } | |
714 | ||
1da177e4 LT |
715 | /* |
716 | * Wake up all waiters hashed on the physical page that is mapped | |
717 | * to this virtual address: | |
718 | */ | |
34f01cc1 ED |
719 | static int futex_wake(u32 __user *uaddr, struct rw_semaphore *fshared, |
720 | int nr_wake) | |
1da177e4 | 721 | { |
e2970f2f | 722 | struct futex_hash_bucket *hb; |
1da177e4 | 723 | struct futex_q *this, *next; |
ec92d082 | 724 | struct plist_head *head; |
e2970f2f | 725 | union futex_key key; |
1da177e4 LT |
726 | int ret; |
727 | ||
36cf3b5c | 728 | futex_lock_mm(fshared); |
1da177e4 | 729 | |
34f01cc1 | 730 | ret = get_futex_key(uaddr, fshared, &key); |
1da177e4 LT |
731 | if (unlikely(ret != 0)) |
732 | goto out; | |
733 | ||
e2970f2f IM |
734 | hb = hash_futex(&key); |
735 | spin_lock(&hb->lock); | |
736 | head = &hb->chain; | |
1da177e4 | 737 | |
ec92d082 | 738 | plist_for_each_entry_safe(this, next, head, list) { |
1da177e4 | 739 | if (match_futex (&this->key, &key)) { |
ed6f7b10 IM |
740 | if (this->pi_state) { |
741 | ret = -EINVAL; | |
742 | break; | |
743 | } | |
1da177e4 LT |
744 | wake_futex(this); |
745 | if (++ret >= nr_wake) | |
746 | break; | |
747 | } | |
748 | } | |
749 | ||
e2970f2f | 750 | spin_unlock(&hb->lock); |
1da177e4 | 751 | out: |
36cf3b5c | 752 | futex_unlock_mm(fshared); |
1da177e4 LT |
753 | return ret; |
754 | } | |
755 | ||
4732efbe JJ |
756 | /* |
757 | * Wake up all waiters hashed on the physical page that is mapped | |
758 | * to this virtual address: | |
759 | */ | |
e2970f2f | 760 | static int |
34f01cc1 ED |
761 | futex_wake_op(u32 __user *uaddr1, struct rw_semaphore *fshared, |
762 | u32 __user *uaddr2, | |
e2970f2f | 763 | int nr_wake, int nr_wake2, int op) |
4732efbe JJ |
764 | { |
765 | union futex_key key1, key2; | |
e2970f2f | 766 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 767 | struct plist_head *head; |
4732efbe JJ |
768 | struct futex_q *this, *next; |
769 | int ret, op_ret, attempt = 0; | |
770 | ||
771 | retryfull: | |
36cf3b5c | 772 | futex_lock_mm(fshared); |
4732efbe | 773 | |
34f01cc1 | 774 | ret = get_futex_key(uaddr1, fshared, &key1); |
4732efbe JJ |
775 | if (unlikely(ret != 0)) |
776 | goto out; | |
34f01cc1 | 777 | ret = get_futex_key(uaddr2, fshared, &key2); |
4732efbe JJ |
778 | if (unlikely(ret != 0)) |
779 | goto out; | |
780 | ||
e2970f2f IM |
781 | hb1 = hash_futex(&key1); |
782 | hb2 = hash_futex(&key2); | |
4732efbe JJ |
783 | |
784 | retry: | |
8b8f319f | 785 | double_lock_hb(hb1, hb2); |
4732efbe | 786 | |
e2970f2f | 787 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 788 | if (unlikely(op_ret < 0)) { |
e2970f2f | 789 | u32 dummy; |
4732efbe | 790 | |
e2970f2f IM |
791 | spin_unlock(&hb1->lock); |
792 | if (hb1 != hb2) | |
793 | spin_unlock(&hb2->lock); | |
4732efbe | 794 | |
7ee1dd3f | 795 | #ifndef CONFIG_MMU |
e2970f2f IM |
796 | /* |
797 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
798 | * but we might get them from range checking | |
799 | */ | |
7ee1dd3f DH |
800 | ret = op_ret; |
801 | goto out; | |
802 | #endif | |
803 | ||
796f8d9b DG |
804 | if (unlikely(op_ret != -EFAULT)) { |
805 | ret = op_ret; | |
806 | goto out; | |
807 | } | |
808 | ||
e2970f2f IM |
809 | /* |
810 | * futex_atomic_op_inuser needs to both read and write | |
4732efbe JJ |
811 | * *(int __user *)uaddr2, but we can't modify it |
812 | * non-atomically. Therefore, if get_user below is not | |
813 | * enough, we need to handle the fault ourselves, while | |
e2970f2f IM |
814 | * still holding the mmap_sem. |
815 | */ | |
4732efbe | 816 | if (attempt++) { |
34f01cc1 | 817 | ret = futex_handle_fault((unsigned long)uaddr2, |
36cf3b5c | 818 | fshared, attempt); |
34f01cc1 | 819 | if (ret) |
4732efbe | 820 | goto out; |
4732efbe JJ |
821 | goto retry; |
822 | } | |
823 | ||
e2970f2f IM |
824 | /* |
825 | * If we would have faulted, release mmap_sem, | |
826 | * fault it in and start all over again. | |
827 | */ | |
36cf3b5c | 828 | futex_unlock_mm(fshared); |
4732efbe | 829 | |
e2970f2f | 830 | ret = get_user(dummy, uaddr2); |
4732efbe JJ |
831 | if (ret) |
832 | return ret; | |
833 | ||
834 | goto retryfull; | |
835 | } | |
836 | ||
e2970f2f | 837 | head = &hb1->chain; |
4732efbe | 838 | |
ec92d082 | 839 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
840 | if (match_futex (&this->key, &key1)) { |
841 | wake_futex(this); | |
842 | if (++ret >= nr_wake) | |
843 | break; | |
844 | } | |
845 | } | |
846 | ||
847 | if (op_ret > 0) { | |
e2970f2f | 848 | head = &hb2->chain; |
4732efbe JJ |
849 | |
850 | op_ret = 0; | |
ec92d082 | 851 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
852 | if (match_futex (&this->key, &key2)) { |
853 | wake_futex(this); | |
854 | if (++op_ret >= nr_wake2) | |
855 | break; | |
856 | } | |
857 | } | |
858 | ret += op_ret; | |
859 | } | |
860 | ||
e2970f2f IM |
861 | spin_unlock(&hb1->lock); |
862 | if (hb1 != hb2) | |
863 | spin_unlock(&hb2->lock); | |
4732efbe | 864 | out: |
36cf3b5c TG |
865 | futex_unlock_mm(fshared); |
866 | ||
4732efbe JJ |
867 | return ret; |
868 | } | |
869 | ||
1da177e4 LT |
870 | /* |
871 | * Requeue all waiters hashed on one physical page to another | |
872 | * physical page. | |
873 | */ | |
34f01cc1 ED |
874 | static int futex_requeue(u32 __user *uaddr1, struct rw_semaphore *fshared, |
875 | u32 __user *uaddr2, | |
e2970f2f | 876 | int nr_wake, int nr_requeue, u32 *cmpval) |
1da177e4 LT |
877 | { |
878 | union futex_key key1, key2; | |
e2970f2f | 879 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 880 | struct plist_head *head1; |
1da177e4 LT |
881 | struct futex_q *this, *next; |
882 | int ret, drop_count = 0; | |
883 | ||
884 | retry: | |
36cf3b5c | 885 | futex_lock_mm(fshared); |
1da177e4 | 886 | |
34f01cc1 | 887 | ret = get_futex_key(uaddr1, fshared, &key1); |
1da177e4 LT |
888 | if (unlikely(ret != 0)) |
889 | goto out; | |
34f01cc1 | 890 | ret = get_futex_key(uaddr2, fshared, &key2); |
1da177e4 LT |
891 | if (unlikely(ret != 0)) |
892 | goto out; | |
893 | ||
e2970f2f IM |
894 | hb1 = hash_futex(&key1); |
895 | hb2 = hash_futex(&key2); | |
1da177e4 | 896 | |
8b8f319f | 897 | double_lock_hb(hb1, hb2); |
1da177e4 | 898 | |
e2970f2f IM |
899 | if (likely(cmpval != NULL)) { |
900 | u32 curval; | |
1da177e4 | 901 | |
e2970f2f | 902 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
903 | |
904 | if (unlikely(ret)) { | |
e2970f2f IM |
905 | spin_unlock(&hb1->lock); |
906 | if (hb1 != hb2) | |
907 | spin_unlock(&hb2->lock); | |
1da177e4 | 908 | |
e2970f2f IM |
909 | /* |
910 | * If we would have faulted, release mmap_sem, fault | |
1da177e4 LT |
911 | * it in and start all over again. |
912 | */ | |
36cf3b5c | 913 | futex_unlock_mm(fshared); |
1da177e4 | 914 | |
e2970f2f | 915 | ret = get_user(curval, uaddr1); |
1da177e4 LT |
916 | |
917 | if (!ret) | |
918 | goto retry; | |
919 | ||
920 | return ret; | |
921 | } | |
e2970f2f | 922 | if (curval != *cmpval) { |
1da177e4 LT |
923 | ret = -EAGAIN; |
924 | goto out_unlock; | |
925 | } | |
926 | } | |
927 | ||
e2970f2f | 928 | head1 = &hb1->chain; |
ec92d082 | 929 | plist_for_each_entry_safe(this, next, head1, list) { |
1da177e4 LT |
930 | if (!match_futex (&this->key, &key1)) |
931 | continue; | |
932 | if (++ret <= nr_wake) { | |
933 | wake_futex(this); | |
934 | } else { | |
59e0e0ac SD |
935 | /* |
936 | * If key1 and key2 hash to the same bucket, no need to | |
937 | * requeue. | |
938 | */ | |
939 | if (likely(head1 != &hb2->chain)) { | |
ec92d082 PP |
940 | plist_del(&this->list, &hb1->chain); |
941 | plist_add(&this->list, &hb2->chain); | |
59e0e0ac | 942 | this->lock_ptr = &hb2->lock; |
ec92d082 PP |
943 | #ifdef CONFIG_DEBUG_PI_LIST |
944 | this->list.plist.lock = &hb2->lock; | |
945 | #endif | |
778e9a9c | 946 | } |
1da177e4 | 947 | this->key = key2; |
9adef58b | 948 | get_futex_key_refs(&key2); |
1da177e4 LT |
949 | drop_count++; |
950 | ||
951 | if (ret - nr_wake >= nr_requeue) | |
952 | break; | |
1da177e4 LT |
953 | } |
954 | } | |
955 | ||
956 | out_unlock: | |
e2970f2f IM |
957 | spin_unlock(&hb1->lock); |
958 | if (hb1 != hb2) | |
959 | spin_unlock(&hb2->lock); | |
1da177e4 | 960 | |
9adef58b | 961 | /* drop_futex_key_refs() must be called outside the spinlocks. */ |
1da177e4 | 962 | while (--drop_count >= 0) |
9adef58b | 963 | drop_futex_key_refs(&key1); |
1da177e4 LT |
964 | |
965 | out: | |
36cf3b5c | 966 | futex_unlock_mm(fshared); |
1da177e4 LT |
967 | return ret; |
968 | } | |
969 | ||
970 | /* The key must be already stored in q->key. */ | |
971 | static inline struct futex_hash_bucket * | |
972 | queue_lock(struct futex_q *q, int fd, struct file *filp) | |
973 | { | |
e2970f2f | 974 | struct futex_hash_bucket *hb; |
1da177e4 LT |
975 | |
976 | q->fd = fd; | |
977 | q->filp = filp; | |
978 | ||
979 | init_waitqueue_head(&q->waiters); | |
980 | ||
9adef58b | 981 | get_futex_key_refs(&q->key); |
e2970f2f IM |
982 | hb = hash_futex(&q->key); |
983 | q->lock_ptr = &hb->lock; | |
1da177e4 | 984 | |
e2970f2f IM |
985 | spin_lock(&hb->lock); |
986 | return hb; | |
1da177e4 LT |
987 | } |
988 | ||
e2970f2f | 989 | static inline void __queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 990 | { |
ec92d082 PP |
991 | int prio; |
992 | ||
993 | /* | |
994 | * The priority used to register this element is | |
995 | * - either the real thread-priority for the real-time threads | |
996 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
997 | * - or MAX_RT_PRIO for non-RT threads. | |
998 | * Thus, all RT-threads are woken first in priority order, and | |
999 | * the others are woken last, in FIFO order. | |
1000 | */ | |
1001 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1002 | ||
1003 | plist_node_init(&q->list, prio); | |
1004 | #ifdef CONFIG_DEBUG_PI_LIST | |
1005 | q->list.plist.lock = &hb->lock; | |
1006 | #endif | |
1007 | plist_add(&q->list, &hb->chain); | |
c87e2837 | 1008 | q->task = current; |
e2970f2f | 1009 | spin_unlock(&hb->lock); |
1da177e4 LT |
1010 | } |
1011 | ||
1012 | static inline void | |
e2970f2f | 1013 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 1014 | { |
e2970f2f | 1015 | spin_unlock(&hb->lock); |
9adef58b | 1016 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1017 | } |
1018 | ||
1019 | /* | |
1020 | * queue_me and unqueue_me must be called as a pair, each | |
1021 | * exactly once. They are called with the hashed spinlock held. | |
1022 | */ | |
1023 | ||
1024 | /* The key must be already stored in q->key. */ | |
1025 | static void queue_me(struct futex_q *q, int fd, struct file *filp) | |
1026 | { | |
e2970f2f IM |
1027 | struct futex_hash_bucket *hb; |
1028 | ||
1029 | hb = queue_lock(q, fd, filp); | |
1030 | __queue_me(q, hb); | |
1da177e4 LT |
1031 | } |
1032 | ||
1033 | /* Return 1 if we were still queued (ie. 0 means we were woken) */ | |
1034 | static int unqueue_me(struct futex_q *q) | |
1035 | { | |
1da177e4 | 1036 | spinlock_t *lock_ptr; |
e2970f2f | 1037 | int ret = 0; |
1da177e4 LT |
1038 | |
1039 | /* In the common case we don't take the spinlock, which is nice. */ | |
1040 | retry: | |
1041 | lock_ptr = q->lock_ptr; | |
e91467ec | 1042 | barrier(); |
1da177e4 LT |
1043 | if (lock_ptr != 0) { |
1044 | spin_lock(lock_ptr); | |
1045 | /* | |
1046 | * q->lock_ptr can change between reading it and | |
1047 | * spin_lock(), causing us to take the wrong lock. This | |
1048 | * corrects the race condition. | |
1049 | * | |
1050 | * Reasoning goes like this: if we have the wrong lock, | |
1051 | * q->lock_ptr must have changed (maybe several times) | |
1052 | * between reading it and the spin_lock(). It can | |
1053 | * change again after the spin_lock() but only if it was | |
1054 | * already changed before the spin_lock(). It cannot, | |
1055 | * however, change back to the original value. Therefore | |
1056 | * we can detect whether we acquired the correct lock. | |
1057 | */ | |
1058 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
1059 | spin_unlock(lock_ptr); | |
1060 | goto retry; | |
1061 | } | |
ec92d082 PP |
1062 | WARN_ON(plist_node_empty(&q->list)); |
1063 | plist_del(&q->list, &q->list.plist); | |
c87e2837 IM |
1064 | |
1065 | BUG_ON(q->pi_state); | |
1066 | ||
1da177e4 LT |
1067 | spin_unlock(lock_ptr); |
1068 | ret = 1; | |
1069 | } | |
1070 | ||
9adef58b | 1071 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1072 | return ret; |
1073 | } | |
1074 | ||
c87e2837 IM |
1075 | /* |
1076 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
1077 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
1078 | * and dropped here. | |
c87e2837 | 1079 | */ |
d0aa7a70 | 1080 | static void unqueue_me_pi(struct futex_q *q) |
c87e2837 | 1081 | { |
ec92d082 PP |
1082 | WARN_ON(plist_node_empty(&q->list)); |
1083 | plist_del(&q->list, &q->list.plist); | |
c87e2837 IM |
1084 | |
1085 | BUG_ON(!q->pi_state); | |
1086 | free_pi_state(q->pi_state); | |
1087 | q->pi_state = NULL; | |
1088 | ||
d0aa7a70 | 1089 | spin_unlock(q->lock_ptr); |
c87e2837 | 1090 | |
9adef58b | 1091 | drop_futex_key_refs(&q->key); |
c87e2837 IM |
1092 | } |
1093 | ||
d0aa7a70 PP |
1094 | /* |
1095 | * Fixup the pi_state owner with current. | |
1096 | * | |
778e9a9c AK |
1097 | * Must be called with hash bucket lock held and mm->sem held for non |
1098 | * private futexes. | |
d0aa7a70 | 1099 | */ |
778e9a9c | 1100 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
d0aa7a70 PP |
1101 | struct task_struct *curr) |
1102 | { | |
1103 | u32 newtid = curr->pid | FUTEX_WAITERS; | |
1104 | struct futex_pi_state *pi_state = q->pi_state; | |
1105 | u32 uval, curval, newval; | |
1106 | int ret; | |
1107 | ||
1108 | /* Owner died? */ | |
1109 | if (pi_state->owner != NULL) { | |
1110 | spin_lock_irq(&pi_state->owner->pi_lock); | |
1111 | WARN_ON(list_empty(&pi_state->list)); | |
1112 | list_del_init(&pi_state->list); | |
1113 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
1114 | } else | |
1115 | newtid |= FUTEX_OWNER_DIED; | |
1116 | ||
1117 | pi_state->owner = curr; | |
1118 | ||
1119 | spin_lock_irq(&curr->pi_lock); | |
1120 | WARN_ON(!list_empty(&pi_state->list)); | |
1121 | list_add(&pi_state->list, &curr->pi_state_list); | |
1122 | spin_unlock_irq(&curr->pi_lock); | |
1123 | ||
d0aa7a70 PP |
1124 | /* |
1125 | * We own it, so we have to replace the pending owner | |
1126 | * TID. This must be atomic as we have preserve the | |
1127 | * owner died bit here. | |
1128 | */ | |
778e9a9c AK |
1129 | ret = get_futex_value_locked(&uval, uaddr); |
1130 | ||
d0aa7a70 PP |
1131 | while (!ret) { |
1132 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
778e9a9c | 1133 | |
36cf3b5c | 1134 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
778e9a9c | 1135 | |
d0aa7a70 | 1136 | if (curval == -EFAULT) |
778e9a9c | 1137 | ret = -EFAULT; |
d0aa7a70 PP |
1138 | if (curval == uval) |
1139 | break; | |
1140 | uval = curval; | |
1141 | } | |
1142 | return ret; | |
1143 | } | |
1144 | ||
34f01cc1 ED |
1145 | /* |
1146 | * In case we must use restart_block to restart a futex_wait, | |
1147 | * we encode in the 'arg3' shared capability | |
1148 | */ | |
1149 | #define ARG3_SHARED 1 | |
1150 | ||
72c1bbf3 | 1151 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 1152 | |
34f01cc1 ED |
1153 | static int futex_wait(u32 __user *uaddr, struct rw_semaphore *fshared, |
1154 | u32 val, ktime_t *abs_time) | |
1da177e4 | 1155 | { |
c87e2837 IM |
1156 | struct task_struct *curr = current; |
1157 | DECLARE_WAITQUEUE(wait, curr); | |
e2970f2f | 1158 | struct futex_hash_bucket *hb; |
1da177e4 | 1159 | struct futex_q q; |
e2970f2f IM |
1160 | u32 uval; |
1161 | int ret; | |
bd197234 | 1162 | struct hrtimer_sleeper t; |
c19384b5 | 1163 | int rem = 0; |
1da177e4 | 1164 | |
c87e2837 | 1165 | q.pi_state = NULL; |
1da177e4 | 1166 | retry: |
36cf3b5c | 1167 | futex_lock_mm(fshared); |
1da177e4 | 1168 | |
34f01cc1 | 1169 | ret = get_futex_key(uaddr, fshared, &q.key); |
1da177e4 LT |
1170 | if (unlikely(ret != 0)) |
1171 | goto out_release_sem; | |
1172 | ||
e2970f2f | 1173 | hb = queue_lock(&q, -1, NULL); |
1da177e4 LT |
1174 | |
1175 | /* | |
1176 | * Access the page AFTER the futex is queued. | |
1177 | * Order is important: | |
1178 | * | |
1179 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
1180 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
1181 | * | |
1182 | * The basic logical guarantee of a futex is that it blocks ONLY | |
1183 | * if cond(var) is known to be true at the time of blocking, for | |
1184 | * any cond. If we queued after testing *uaddr, that would open | |
1185 | * a race condition where we could block indefinitely with | |
1186 | * cond(var) false, which would violate the guarantee. | |
1187 | * | |
1188 | * A consequence is that futex_wait() can return zero and absorb | |
1189 | * a wakeup when *uaddr != val on entry to the syscall. This is | |
1190 | * rare, but normal. | |
1191 | * | |
34f01cc1 ED |
1192 | * for shared futexes, we hold the mmap semaphore, so the mapping |
1193 | * cannot have changed since we looked it up in get_futex_key. | |
1da177e4 | 1194 | */ |
e2970f2f | 1195 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 LT |
1196 | |
1197 | if (unlikely(ret)) { | |
e2970f2f | 1198 | queue_unlock(&q, hb); |
1da177e4 | 1199 | |
e2970f2f IM |
1200 | /* |
1201 | * If we would have faulted, release mmap_sem, fault it in and | |
1da177e4 LT |
1202 | * start all over again. |
1203 | */ | |
36cf3b5c | 1204 | futex_unlock_mm(fshared); |
1da177e4 | 1205 | |
e2970f2f | 1206 | ret = get_user(uval, uaddr); |
1da177e4 LT |
1207 | |
1208 | if (!ret) | |
1209 | goto retry; | |
1210 | return ret; | |
1211 | } | |
c87e2837 IM |
1212 | ret = -EWOULDBLOCK; |
1213 | if (uval != val) | |
1214 | goto out_unlock_release_sem; | |
1da177e4 LT |
1215 | |
1216 | /* Only actually queue if *uaddr contained val. */ | |
e2970f2f | 1217 | __queue_me(&q, hb); |
1da177e4 LT |
1218 | |
1219 | /* | |
1220 | * Now the futex is queued and we have checked the data, we | |
1221 | * don't want to hold mmap_sem while we sleep. | |
c87e2837 | 1222 | */ |
36cf3b5c | 1223 | futex_unlock_mm(fshared); |
1da177e4 LT |
1224 | |
1225 | /* | |
1226 | * There might have been scheduling since the queue_me(), as we | |
1227 | * cannot hold a spinlock across the get_user() in case it | |
1228 | * faults, and we cannot just set TASK_INTERRUPTIBLE state when | |
1229 | * queueing ourselves into the futex hash. This code thus has to | |
1230 | * rely on the futex_wake() code removing us from hash when it | |
1231 | * wakes us up. | |
1232 | */ | |
1233 | ||
1234 | /* add_wait_queue is the barrier after __set_current_state. */ | |
1235 | __set_current_state(TASK_INTERRUPTIBLE); | |
1236 | add_wait_queue(&q.waiters, &wait); | |
1237 | /* | |
ec92d082 | 1238 | * !plist_node_empty() is safe here without any lock. |
1da177e4 LT |
1239 | * q.lock_ptr != 0 is not safe, because of ordering against wakeup. |
1240 | */ | |
ec92d082 | 1241 | if (likely(!plist_node_empty(&q.list))) { |
c19384b5 PP |
1242 | if (!abs_time) |
1243 | schedule(); | |
1244 | else { | |
1245 | hrtimer_init(&t.timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1246 | hrtimer_init_sleeper(&t, current); | |
1247 | t.timer.expires = *abs_time; | |
1248 | ||
1249 | hrtimer_start(&t.timer, t.timer.expires, HRTIMER_MODE_ABS); | |
1250 | ||
1251 | /* | |
1252 | * the timer could have already expired, in which | |
1253 | * case current would be flagged for rescheduling. | |
1254 | * Don't bother calling schedule. | |
1255 | */ | |
1256 | if (likely(t.task)) | |
1257 | schedule(); | |
1258 | ||
1259 | hrtimer_cancel(&t.timer); | |
72c1bbf3 | 1260 | |
c19384b5 PP |
1261 | /* Flag if a timeout occured */ |
1262 | rem = (t.task == NULL); | |
1263 | } | |
72c1bbf3 | 1264 | } |
1da177e4 LT |
1265 | __set_current_state(TASK_RUNNING); |
1266 | ||
1267 | /* | |
1268 | * NOTE: we don't remove ourselves from the waitqueue because | |
1269 | * we are the only user of it. | |
1270 | */ | |
1271 | ||
1272 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
1273 | if (!unqueue_me(&q)) | |
1274 | return 0; | |
c19384b5 | 1275 | if (rem) |
1da177e4 | 1276 | return -ETIMEDOUT; |
72c1bbf3 | 1277 | |
e2970f2f IM |
1278 | /* |
1279 | * We expect signal_pending(current), but another thread may | |
1280 | * have handled it for us already. | |
1281 | */ | |
c19384b5 | 1282 | if (!abs_time) |
72c1bbf3 NP |
1283 | return -ERESTARTSYS; |
1284 | else { | |
1285 | struct restart_block *restart; | |
1286 | restart = ¤t_thread_info()->restart_block; | |
1287 | restart->fn = futex_wait_restart; | |
1288 | restart->arg0 = (unsigned long)uaddr; | |
1289 | restart->arg1 = (unsigned long)val; | |
c19384b5 | 1290 | restart->arg2 = (unsigned long)abs_time; |
34f01cc1 ED |
1291 | restart->arg3 = 0; |
1292 | if (fshared) | |
1293 | restart->arg3 |= ARG3_SHARED; | |
72c1bbf3 NP |
1294 | return -ERESTART_RESTARTBLOCK; |
1295 | } | |
1da177e4 | 1296 | |
c87e2837 IM |
1297 | out_unlock_release_sem: |
1298 | queue_unlock(&q, hb); | |
1299 | ||
1da177e4 | 1300 | out_release_sem: |
36cf3b5c | 1301 | futex_unlock_mm(fshared); |
c87e2837 IM |
1302 | return ret; |
1303 | } | |
1304 | ||
72c1bbf3 NP |
1305 | |
1306 | static long futex_wait_restart(struct restart_block *restart) | |
1307 | { | |
1308 | u32 __user *uaddr = (u32 __user *)restart->arg0; | |
1309 | u32 val = (u32)restart->arg1; | |
c19384b5 | 1310 | ktime_t *abs_time = (ktime_t *)restart->arg2; |
34f01cc1 | 1311 | struct rw_semaphore *fshared = NULL; |
72c1bbf3 NP |
1312 | |
1313 | restart->fn = do_no_restart_syscall; | |
34f01cc1 ED |
1314 | if (restart->arg3 & ARG3_SHARED) |
1315 | fshared = ¤t->mm->mmap_sem; | |
1316 | return (long)futex_wait(uaddr, fshared, val, abs_time); | |
72c1bbf3 NP |
1317 | } |
1318 | ||
1319 | ||
c87e2837 IM |
1320 | /* |
1321 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
1322 | * and failed. The kernel side here does the whole locking operation: | |
1323 | * if there are waiters then it will block, it does PI, etc. (Due to | |
1324 | * races the kernel might see a 0 value of the futex too.) | |
1325 | */ | |
34f01cc1 ED |
1326 | static int futex_lock_pi(u32 __user *uaddr, struct rw_semaphore *fshared, |
1327 | int detect, ktime_t *time, int trylock) | |
c87e2837 | 1328 | { |
c5780e97 | 1329 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 IM |
1330 | struct task_struct *curr = current; |
1331 | struct futex_hash_bucket *hb; | |
1332 | u32 uval, newval, curval; | |
1333 | struct futex_q q; | |
778e9a9c | 1334 | int ret, lock_taken, ownerdied = 0, attempt = 0; |
c87e2837 IM |
1335 | |
1336 | if (refill_pi_state_cache()) | |
1337 | return -ENOMEM; | |
1338 | ||
c19384b5 | 1339 | if (time) { |
c5780e97 | 1340 | to = &timeout; |
c9cb2e3d | 1341 | hrtimer_init(&to->timer, CLOCK_REALTIME, HRTIMER_MODE_ABS); |
c5780e97 | 1342 | hrtimer_init_sleeper(to, current); |
c19384b5 | 1343 | to->timer.expires = *time; |
c5780e97 TG |
1344 | } |
1345 | ||
c87e2837 IM |
1346 | q.pi_state = NULL; |
1347 | retry: | |
36cf3b5c | 1348 | futex_lock_mm(fshared); |
c87e2837 | 1349 | |
34f01cc1 | 1350 | ret = get_futex_key(uaddr, fshared, &q.key); |
c87e2837 IM |
1351 | if (unlikely(ret != 0)) |
1352 | goto out_release_sem; | |
1353 | ||
778e9a9c | 1354 | retry_unlocked: |
c87e2837 IM |
1355 | hb = queue_lock(&q, -1, NULL); |
1356 | ||
1357 | retry_locked: | |
778e9a9c | 1358 | ret = lock_taken = 0; |
d0aa7a70 | 1359 | |
c87e2837 IM |
1360 | /* |
1361 | * To avoid races, we attempt to take the lock here again | |
1362 | * (by doing a 0 -> TID atomic cmpxchg), while holding all | |
1363 | * the locks. It will most likely not succeed. | |
1364 | */ | |
1365 | newval = current->pid; | |
1366 | ||
36cf3b5c | 1367 | curval = cmpxchg_futex_value_locked(uaddr, 0, newval); |
c87e2837 IM |
1368 | |
1369 | if (unlikely(curval == -EFAULT)) | |
1370 | goto uaddr_faulted; | |
1371 | ||
778e9a9c AK |
1372 | /* |
1373 | * Detect deadlocks. In case of REQUEUE_PI this is a valid | |
1374 | * situation and we return success to user space. | |
1375 | */ | |
c87e2837 | 1376 | if (unlikely((curval & FUTEX_TID_MASK) == current->pid)) { |
bd197234 | 1377 | ret = -EDEADLK; |
c87e2837 IM |
1378 | goto out_unlock_release_sem; |
1379 | } | |
1380 | ||
1381 | /* | |
778e9a9c | 1382 | * Surprise - we got the lock. Just return to userspace: |
c87e2837 IM |
1383 | */ |
1384 | if (unlikely(!curval)) | |
1385 | goto out_unlock_release_sem; | |
1386 | ||
1387 | uval = curval; | |
778e9a9c | 1388 | |
d0aa7a70 | 1389 | /* |
778e9a9c AK |
1390 | * Set the WAITERS flag, so the owner will know it has someone |
1391 | * to wake at next unlock | |
d0aa7a70 | 1392 | */ |
778e9a9c AK |
1393 | newval = curval | FUTEX_WAITERS; |
1394 | ||
1395 | /* | |
1396 | * There are two cases, where a futex might have no owner (the | |
bd197234 TG |
1397 | * owner TID is 0): OWNER_DIED. We take over the futex in this |
1398 | * case. We also do an unconditional take over, when the owner | |
1399 | * of the futex died. | |
778e9a9c AK |
1400 | * |
1401 | * This is safe as we are protected by the hash bucket lock ! | |
1402 | */ | |
1403 | if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { | |
bd197234 | 1404 | /* Keep the OWNER_DIED bit */ |
778e9a9c AK |
1405 | newval = (curval & ~FUTEX_TID_MASK) | current->pid; |
1406 | ownerdied = 0; | |
1407 | lock_taken = 1; | |
1408 | } | |
c87e2837 | 1409 | |
36cf3b5c | 1410 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
c87e2837 IM |
1411 | |
1412 | if (unlikely(curval == -EFAULT)) | |
1413 | goto uaddr_faulted; | |
1414 | if (unlikely(curval != uval)) | |
1415 | goto retry_locked; | |
1416 | ||
778e9a9c | 1417 | /* |
bd197234 | 1418 | * We took the lock due to owner died take over. |
778e9a9c | 1419 | */ |
bd197234 | 1420 | if (unlikely(lock_taken)) |
d0aa7a70 | 1421 | goto out_unlock_release_sem; |
d0aa7a70 | 1422 | |
c87e2837 IM |
1423 | /* |
1424 | * We dont have the lock. Look up the PI state (or create it if | |
1425 | * we are the first waiter): | |
1426 | */ | |
d0aa7a70 | 1427 | ret = lookup_pi_state(uval, hb, &q.key, &q.pi_state); |
c87e2837 IM |
1428 | |
1429 | if (unlikely(ret)) { | |
778e9a9c | 1430 | switch (ret) { |
c87e2837 | 1431 | |
778e9a9c AK |
1432 | case -EAGAIN: |
1433 | /* | |
1434 | * Task is exiting and we just wait for the | |
1435 | * exit to complete. | |
1436 | */ | |
1437 | queue_unlock(&q, hb); | |
36cf3b5c | 1438 | futex_unlock_mm(fshared); |
778e9a9c AK |
1439 | cond_resched(); |
1440 | goto retry; | |
c87e2837 | 1441 | |
778e9a9c AK |
1442 | case -ESRCH: |
1443 | /* | |
1444 | * No owner found for this futex. Check if the | |
1445 | * OWNER_DIED bit is set to figure out whether | |
1446 | * this is a robust futex or not. | |
1447 | */ | |
1448 | if (get_futex_value_locked(&curval, uaddr)) | |
c87e2837 | 1449 | goto uaddr_faulted; |
778e9a9c AK |
1450 | |
1451 | /* | |
1452 | * We simply start over in case of a robust | |
1453 | * futex. The code above will take the futex | |
1454 | * and return happy. | |
1455 | */ | |
1456 | if (curval & FUTEX_OWNER_DIED) { | |
1457 | ownerdied = 1; | |
c87e2837 | 1458 | goto retry_locked; |
778e9a9c AK |
1459 | } |
1460 | default: | |
1461 | goto out_unlock_release_sem; | |
c87e2837 | 1462 | } |
c87e2837 IM |
1463 | } |
1464 | ||
1465 | /* | |
1466 | * Only actually queue now that the atomic ops are done: | |
1467 | */ | |
1468 | __queue_me(&q, hb); | |
1469 | ||
1470 | /* | |
1471 | * Now the futex is queued and we have checked the data, we | |
1472 | * don't want to hold mmap_sem while we sleep. | |
1473 | */ | |
36cf3b5c | 1474 | futex_unlock_mm(fshared); |
c87e2837 IM |
1475 | |
1476 | WARN_ON(!q.pi_state); | |
1477 | /* | |
1478 | * Block on the PI mutex: | |
1479 | */ | |
1480 | if (!trylock) | |
1481 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); | |
1482 | else { | |
1483 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); | |
1484 | /* Fixup the trylock return value: */ | |
1485 | ret = ret ? 0 : -EWOULDBLOCK; | |
1486 | } | |
1487 | ||
36cf3b5c | 1488 | futex_lock_mm(fshared); |
a99e4e41 | 1489 | spin_lock(q.lock_ptr); |
c87e2837 | 1490 | |
778e9a9c AK |
1491 | if (!ret) { |
1492 | /* | |
1493 | * Got the lock. We might not be the anticipated owner | |
1494 | * if we did a lock-steal - fix up the PI-state in | |
1495 | * that case: | |
1496 | */ | |
1497 | if (q.pi_state->owner != curr) | |
1498 | ret = fixup_pi_state_owner(uaddr, &q, curr); | |
1499 | } else { | |
c87e2837 IM |
1500 | /* |
1501 | * Catch the rare case, where the lock was released | |
778e9a9c AK |
1502 | * when we were on the way back before we locked the |
1503 | * hash bucket. | |
c87e2837 | 1504 | */ |
778e9a9c AK |
1505 | if (q.pi_state->owner == curr && |
1506 | rt_mutex_trylock(&q.pi_state->pi_mutex)) { | |
1507 | ret = 0; | |
1508 | } else { | |
1509 | /* | |
1510 | * Paranoia check. If we did not take the lock | |
1511 | * in the trylock above, then we should not be | |
1512 | * the owner of the rtmutex, neither the real | |
1513 | * nor the pending one: | |
1514 | */ | |
1515 | if (rt_mutex_owner(&q.pi_state->pi_mutex) == curr) | |
1516 | printk(KERN_ERR "futex_lock_pi: ret = %d " | |
1517 | "pi-mutex: %p pi-state %p\n", ret, | |
1518 | q.pi_state->pi_mutex.owner, | |
1519 | q.pi_state->owner); | |
c87e2837 | 1520 | } |
c87e2837 IM |
1521 | } |
1522 | ||
778e9a9c AK |
1523 | /* Unqueue and drop the lock */ |
1524 | unqueue_me_pi(&q); | |
36cf3b5c | 1525 | futex_unlock_mm(fshared); |
c87e2837 | 1526 | |
c5780e97 | 1527 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 IM |
1528 | |
1529 | out_unlock_release_sem: | |
1530 | queue_unlock(&q, hb); | |
1531 | ||
1532 | out_release_sem: | |
36cf3b5c | 1533 | futex_unlock_mm(fshared); |
c87e2837 IM |
1534 | return ret; |
1535 | ||
1536 | uaddr_faulted: | |
1537 | /* | |
1538 | * We have to r/w *(int __user *)uaddr, but we can't modify it | |
1539 | * non-atomically. Therefore, if get_user below is not | |
1540 | * enough, we need to handle the fault ourselves, while | |
1541 | * still holding the mmap_sem. | |
778e9a9c AK |
1542 | * |
1543 | * ... and hb->lock. :-) --ANK | |
c87e2837 | 1544 | */ |
778e9a9c AK |
1545 | queue_unlock(&q, hb); |
1546 | ||
c87e2837 | 1547 | if (attempt++) { |
34f01cc1 ED |
1548 | ret = futex_handle_fault((unsigned long)uaddr, fshared, |
1549 | attempt); | |
1550 | if (ret) | |
778e9a9c AK |
1551 | goto out_release_sem; |
1552 | goto retry_unlocked; | |
c87e2837 IM |
1553 | } |
1554 | ||
36cf3b5c | 1555 | futex_unlock_mm(fshared); |
c87e2837 IM |
1556 | |
1557 | ret = get_user(uval, uaddr); | |
1558 | if (!ret && (uval != -EFAULT)) | |
1559 | goto retry; | |
1560 | ||
1561 | return ret; | |
1562 | } | |
1563 | ||
c87e2837 IM |
1564 | /* |
1565 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
1566 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
1567 | * and do the rt-mutex unlock. | |
1568 | */ | |
34f01cc1 | 1569 | static int futex_unlock_pi(u32 __user *uaddr, struct rw_semaphore *fshared) |
c87e2837 IM |
1570 | { |
1571 | struct futex_hash_bucket *hb; | |
1572 | struct futex_q *this, *next; | |
1573 | u32 uval; | |
ec92d082 | 1574 | struct plist_head *head; |
c87e2837 IM |
1575 | union futex_key key; |
1576 | int ret, attempt = 0; | |
1577 | ||
1578 | retry: | |
1579 | if (get_user(uval, uaddr)) | |
1580 | return -EFAULT; | |
1581 | /* | |
1582 | * We release only a lock we actually own: | |
1583 | */ | |
1584 | if ((uval & FUTEX_TID_MASK) != current->pid) | |
1585 | return -EPERM; | |
1586 | /* | |
1587 | * First take all the futex related locks: | |
1588 | */ | |
36cf3b5c | 1589 | futex_lock_mm(fshared); |
c87e2837 | 1590 | |
34f01cc1 | 1591 | ret = get_futex_key(uaddr, fshared, &key); |
c87e2837 IM |
1592 | if (unlikely(ret != 0)) |
1593 | goto out; | |
1594 | ||
1595 | hb = hash_futex(&key); | |
778e9a9c | 1596 | retry_unlocked: |
c87e2837 IM |
1597 | spin_lock(&hb->lock); |
1598 | ||
c87e2837 IM |
1599 | /* |
1600 | * To avoid races, try to do the TID -> 0 atomic transition | |
1601 | * again. If it succeeds then we can return without waking | |
1602 | * anyone else up: | |
1603 | */ | |
36cf3b5c TG |
1604 | if (!(uval & FUTEX_OWNER_DIED)) |
1605 | uval = cmpxchg_futex_value_locked(uaddr, current->pid, 0); | |
1606 | ||
c87e2837 IM |
1607 | |
1608 | if (unlikely(uval == -EFAULT)) | |
1609 | goto pi_faulted; | |
1610 | /* | |
1611 | * Rare case: we managed to release the lock atomically, | |
1612 | * no need to wake anyone else up: | |
1613 | */ | |
1614 | if (unlikely(uval == current->pid)) | |
1615 | goto out_unlock; | |
1616 | ||
1617 | /* | |
1618 | * Ok, other tasks may need to be woken up - check waiters | |
1619 | * and do the wakeup if necessary: | |
1620 | */ | |
1621 | head = &hb->chain; | |
1622 | ||
ec92d082 | 1623 | plist_for_each_entry_safe(this, next, head, list) { |
c87e2837 IM |
1624 | if (!match_futex (&this->key, &key)) |
1625 | continue; | |
1626 | ret = wake_futex_pi(uaddr, uval, this); | |
1627 | /* | |
1628 | * The atomic access to the futex value | |
1629 | * generated a pagefault, so retry the | |
1630 | * user-access and the wakeup: | |
1631 | */ | |
1632 | if (ret == -EFAULT) | |
1633 | goto pi_faulted; | |
1634 | goto out_unlock; | |
1635 | } | |
1636 | /* | |
1637 | * No waiters - kernel unlocks the futex: | |
1638 | */ | |
e3f2ddea IM |
1639 | if (!(uval & FUTEX_OWNER_DIED)) { |
1640 | ret = unlock_futex_pi(uaddr, uval); | |
1641 | if (ret == -EFAULT) | |
1642 | goto pi_faulted; | |
1643 | } | |
c87e2837 IM |
1644 | |
1645 | out_unlock: | |
1646 | spin_unlock(&hb->lock); | |
1647 | out: | |
36cf3b5c | 1648 | futex_unlock_mm(fshared); |
c87e2837 IM |
1649 | |
1650 | return ret; | |
1651 | ||
1652 | pi_faulted: | |
1653 | /* | |
1654 | * We have to r/w *(int __user *)uaddr, but we can't modify it | |
1655 | * non-atomically. Therefore, if get_user below is not | |
1656 | * enough, we need to handle the fault ourselves, while | |
1657 | * still holding the mmap_sem. | |
778e9a9c AK |
1658 | * |
1659 | * ... and hb->lock. --ANK | |
c87e2837 | 1660 | */ |
778e9a9c AK |
1661 | spin_unlock(&hb->lock); |
1662 | ||
c87e2837 | 1663 | if (attempt++) { |
34f01cc1 ED |
1664 | ret = futex_handle_fault((unsigned long)uaddr, fshared, |
1665 | attempt); | |
1666 | if (ret) | |
778e9a9c AK |
1667 | goto out; |
1668 | goto retry_unlocked; | |
c87e2837 IM |
1669 | } |
1670 | ||
36cf3b5c | 1671 | futex_unlock_mm(fshared); |
c87e2837 IM |
1672 | |
1673 | ret = get_user(uval, uaddr); | |
1674 | if (!ret && (uval != -EFAULT)) | |
1675 | goto retry; | |
1676 | ||
1da177e4 LT |
1677 | return ret; |
1678 | } | |
1679 | ||
1680 | static int futex_close(struct inode *inode, struct file *filp) | |
1681 | { | |
1682 | struct futex_q *q = filp->private_data; | |
1683 | ||
1684 | unqueue_me(q); | |
1685 | kfree(q); | |
e2970f2f | 1686 | |
1da177e4 LT |
1687 | return 0; |
1688 | } | |
1689 | ||
1690 | /* This is one-shot: once it's gone off you need a new fd */ | |
1691 | static unsigned int futex_poll(struct file *filp, | |
1692 | struct poll_table_struct *wait) | |
1693 | { | |
1694 | struct futex_q *q = filp->private_data; | |
1695 | int ret = 0; | |
1696 | ||
1697 | poll_wait(filp, &q->waiters, wait); | |
1698 | ||
1699 | /* | |
ec92d082 | 1700 | * plist_node_empty() is safe here without any lock. |
1da177e4 LT |
1701 | * q->lock_ptr != 0 is not safe, because of ordering against wakeup. |
1702 | */ | |
ec92d082 | 1703 | if (plist_node_empty(&q->list)) |
1da177e4 LT |
1704 | ret = POLLIN | POLLRDNORM; |
1705 | ||
1706 | return ret; | |
1707 | } | |
1708 | ||
15ad7cdc | 1709 | static const struct file_operations futex_fops = { |
1da177e4 LT |
1710 | .release = futex_close, |
1711 | .poll = futex_poll, | |
1712 | }; | |
1713 | ||
1714 | /* | |
1715 | * Signal allows caller to avoid the race which would occur if they | |
1716 | * set the sigio stuff up afterwards. | |
1717 | */ | |
e2970f2f | 1718 | static int futex_fd(u32 __user *uaddr, int signal) |
1da177e4 LT |
1719 | { |
1720 | struct futex_q *q; | |
1721 | struct file *filp; | |
1722 | int ret, err; | |
34f01cc1 | 1723 | struct rw_semaphore *fshared; |
19c6b6ed AM |
1724 | static unsigned long printk_interval; |
1725 | ||
1726 | if (printk_timed_ratelimit(&printk_interval, 60 * 60 * 1000)) { | |
1727 | printk(KERN_WARNING "Process `%s' used FUTEX_FD, which " | |
36cf3b5c TG |
1728 | "will be removed from the kernel in June 2007\n", |
1729 | current->comm); | |
19c6b6ed | 1730 | } |
1da177e4 LT |
1731 | |
1732 | ret = -EINVAL; | |
7ed20e1a | 1733 | if (!valid_signal(signal)) |
1da177e4 LT |
1734 | goto out; |
1735 | ||
1736 | ret = get_unused_fd(); | |
1737 | if (ret < 0) | |
1738 | goto out; | |
1739 | filp = get_empty_filp(); | |
1740 | if (!filp) { | |
1741 | put_unused_fd(ret); | |
1742 | ret = -ENFILE; | |
1743 | goto out; | |
1744 | } | |
1745 | filp->f_op = &futex_fops; | |
f3a43f3f JJS |
1746 | filp->f_path.mnt = mntget(futex_mnt); |
1747 | filp->f_path.dentry = dget(futex_mnt->mnt_root); | |
1748 | filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping; | |
1da177e4 LT |
1749 | |
1750 | if (signal) { | |
609d7fa9 | 1751 | err = __f_setown(filp, task_pid(current), PIDTYPE_PID, 1); |
1da177e4 | 1752 | if (err < 0) { |
39ed3fde | 1753 | goto error; |
1da177e4 LT |
1754 | } |
1755 | filp->f_owner.signum = signal; | |
1756 | } | |
1757 | ||
1758 | q = kmalloc(sizeof(*q), GFP_KERNEL); | |
1759 | if (!q) { | |
39ed3fde PE |
1760 | err = -ENOMEM; |
1761 | goto error; | |
1da177e4 | 1762 | } |
c87e2837 | 1763 | q->pi_state = NULL; |
1da177e4 | 1764 | |
34f01cc1 ED |
1765 | fshared = ¤t->mm->mmap_sem; |
1766 | down_read(fshared); | |
1767 | err = get_futex_key(uaddr, fshared, &q->key); | |
1da177e4 LT |
1768 | |
1769 | if (unlikely(err != 0)) { | |
34f01cc1 | 1770 | up_read(fshared); |
1da177e4 | 1771 | kfree(q); |
39ed3fde | 1772 | goto error; |
1da177e4 LT |
1773 | } |
1774 | ||
1775 | /* | |
1776 | * queue_me() must be called before releasing mmap_sem, because | |
1777 | * key->shared.inode needs to be referenced while holding it. | |
1778 | */ | |
1779 | filp->private_data = q; | |
1780 | ||
1781 | queue_me(q, ret, filp); | |
34f01cc1 | 1782 | up_read(fshared); |
1da177e4 LT |
1783 | |
1784 | /* Now we map fd to filp, so userspace can access it */ | |
1785 | fd_install(ret, filp); | |
1786 | out: | |
1787 | return ret; | |
39ed3fde PE |
1788 | error: |
1789 | put_unused_fd(ret); | |
1790 | put_filp(filp); | |
1791 | ret = err; | |
1792 | goto out; | |
1da177e4 LT |
1793 | } |
1794 | ||
0771dfef IM |
1795 | /* |
1796 | * Support for robust futexes: the kernel cleans up held futexes at | |
1797 | * thread exit time. | |
1798 | * | |
1799 | * Implementation: user-space maintains a per-thread list of locks it | |
1800 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
1801 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 1802 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
1803 | * always manipulated with the lock held, so the list is private and |
1804 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
1805 | * field, to allow the kernel to clean up if the thread dies after | |
1806 | * acquiring the lock, but just before it could have added itself to | |
1807 | * the list. There can only be one such pending lock. | |
1808 | */ | |
1809 | ||
1810 | /** | |
1811 | * sys_set_robust_list - set the robust-futex list head of a task | |
1812 | * @head: pointer to the list-head | |
1813 | * @len: length of the list-head, as userspace expects | |
1814 | */ | |
1815 | asmlinkage long | |
1816 | sys_set_robust_list(struct robust_list_head __user *head, | |
1817 | size_t len) | |
1818 | { | |
1819 | /* | |
1820 | * The kernel knows only one size for now: | |
1821 | */ | |
1822 | if (unlikely(len != sizeof(*head))) | |
1823 | return -EINVAL; | |
1824 | ||
1825 | current->robust_list = head; | |
1826 | ||
1827 | return 0; | |
1828 | } | |
1829 | ||
1830 | /** | |
1831 | * sys_get_robust_list - get the robust-futex list head of a task | |
1832 | * @pid: pid of the process [zero for current task] | |
1833 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
1834 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
1835 | */ | |
1836 | asmlinkage long | |
ba46df98 | 1837 | sys_get_robust_list(int pid, struct robust_list_head __user * __user *head_ptr, |
0771dfef IM |
1838 | size_t __user *len_ptr) |
1839 | { | |
ba46df98 | 1840 | struct robust_list_head __user *head; |
0771dfef IM |
1841 | unsigned long ret; |
1842 | ||
1843 | if (!pid) | |
1844 | head = current->robust_list; | |
1845 | else { | |
1846 | struct task_struct *p; | |
1847 | ||
1848 | ret = -ESRCH; | |
aaa2a97e | 1849 | rcu_read_lock(); |
0771dfef IM |
1850 | p = find_task_by_pid(pid); |
1851 | if (!p) | |
1852 | goto err_unlock; | |
1853 | ret = -EPERM; | |
1854 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
1855 | !capable(CAP_SYS_PTRACE)) | |
1856 | goto err_unlock; | |
1857 | head = p->robust_list; | |
aaa2a97e | 1858 | rcu_read_unlock(); |
0771dfef IM |
1859 | } |
1860 | ||
1861 | if (put_user(sizeof(*head), len_ptr)) | |
1862 | return -EFAULT; | |
1863 | return put_user(head, head_ptr); | |
1864 | ||
1865 | err_unlock: | |
aaa2a97e | 1866 | rcu_read_unlock(); |
0771dfef IM |
1867 | |
1868 | return ret; | |
1869 | } | |
1870 | ||
1871 | /* | |
1872 | * Process a futex-list entry, check whether it's owned by the | |
1873 | * dying task, and do notification if so: | |
1874 | */ | |
e3f2ddea | 1875 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 1876 | { |
e3f2ddea | 1877 | u32 uval, nval, mval; |
0771dfef | 1878 | |
8f17d3a5 IM |
1879 | retry: |
1880 | if (get_user(uval, uaddr)) | |
0771dfef IM |
1881 | return -1; |
1882 | ||
8f17d3a5 | 1883 | if ((uval & FUTEX_TID_MASK) == curr->pid) { |
0771dfef IM |
1884 | /* |
1885 | * Ok, this dying thread is truly holding a futex | |
1886 | * of interest. Set the OWNER_DIED bit atomically | |
1887 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
1888 | * set, wake up a waiter (if any). (We have to do a | |
1889 | * futex_wake() even if OWNER_DIED is already set - | |
1890 | * to handle the rare but possible case of recursive | |
1891 | * thread-death.) The rest of the cleanup is done in | |
1892 | * userspace. | |
1893 | */ | |
e3f2ddea IM |
1894 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
1895 | nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval); | |
1896 | ||
c87e2837 IM |
1897 | if (nval == -EFAULT) |
1898 | return -1; | |
1899 | ||
1900 | if (nval != uval) | |
8f17d3a5 | 1901 | goto retry; |
0771dfef | 1902 | |
e3f2ddea IM |
1903 | /* |
1904 | * Wake robust non-PI futexes here. The wakeup of | |
1905 | * PI futexes happens in exit_pi_state(): | |
1906 | */ | |
36cf3b5c | 1907 | if (!pi && (uval & FUTEX_WAITERS)) |
34f01cc1 | 1908 | futex_wake(uaddr, &curr->mm->mmap_sem, 1); |
0771dfef IM |
1909 | } |
1910 | return 0; | |
1911 | } | |
1912 | ||
e3f2ddea IM |
1913 | /* |
1914 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
1915 | */ | |
1916 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 AV |
1917 | struct robust_list __user * __user *head, |
1918 | int *pi) | |
e3f2ddea IM |
1919 | { |
1920 | unsigned long uentry; | |
1921 | ||
ba46df98 | 1922 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
1923 | return -EFAULT; |
1924 | ||
ba46df98 | 1925 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
1926 | *pi = uentry & 1; |
1927 | ||
1928 | return 0; | |
1929 | } | |
1930 | ||
0771dfef IM |
1931 | /* |
1932 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
1933 | * and mark any locks found there dead, and notify any waiters. | |
1934 | * | |
1935 | * We silently return on any sign of list-walking problem. | |
1936 | */ | |
1937 | void exit_robust_list(struct task_struct *curr) | |
1938 | { | |
1939 | struct robust_list_head __user *head = curr->robust_list; | |
1940 | struct robust_list __user *entry, *pending; | |
e3f2ddea | 1941 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
0771dfef IM |
1942 | unsigned long futex_offset; |
1943 | ||
1944 | /* | |
1945 | * Fetch the list head (which was registered earlier, via | |
1946 | * sys_set_robust_list()): | |
1947 | */ | |
e3f2ddea | 1948 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
1949 | return; |
1950 | /* | |
1951 | * Fetch the relative futex offset: | |
1952 | */ | |
1953 | if (get_user(futex_offset, &head->futex_offset)) | |
1954 | return; | |
1955 | /* | |
1956 | * Fetch any possibly pending lock-add first, and handle it | |
1957 | * if it exists: | |
1958 | */ | |
e3f2ddea | 1959 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 1960 | return; |
e3f2ddea | 1961 | |
0771dfef | 1962 | if (pending) |
34f01cc1 ED |
1963 | handle_futex_death((void __user *)pending + futex_offset, |
1964 | curr, pip); | |
0771dfef IM |
1965 | |
1966 | while (entry != &head->list) { | |
1967 | /* | |
1968 | * A pending lock might already be on the list, so | |
c87e2837 | 1969 | * don't process it twice: |
0771dfef IM |
1970 | */ |
1971 | if (entry != pending) | |
ba46df98 | 1972 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 1973 | curr, pi)) |
0771dfef | 1974 | return; |
0771dfef IM |
1975 | /* |
1976 | * Fetch the next entry in the list: | |
1977 | */ | |
e3f2ddea | 1978 | if (fetch_robust_entry(&entry, &entry->next, &pi)) |
0771dfef IM |
1979 | return; |
1980 | /* | |
1981 | * Avoid excessively long or circular lists: | |
1982 | */ | |
1983 | if (!--limit) | |
1984 | break; | |
1985 | ||
1986 | cond_resched(); | |
1987 | } | |
1988 | } | |
1989 | ||
c19384b5 | 1990 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 1991 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 LT |
1992 | { |
1993 | int ret; | |
34f01cc1 ED |
1994 | int cmd = op & FUTEX_CMD_MASK; |
1995 | struct rw_semaphore *fshared = NULL; | |
1996 | ||
1997 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
1998 | fshared = ¤t->mm->mmap_sem; | |
1da177e4 | 1999 | |
34f01cc1 | 2000 | switch (cmd) { |
1da177e4 | 2001 | case FUTEX_WAIT: |
34f01cc1 | 2002 | ret = futex_wait(uaddr, fshared, val, timeout); |
1da177e4 LT |
2003 | break; |
2004 | case FUTEX_WAKE: | |
34f01cc1 | 2005 | ret = futex_wake(uaddr, fshared, val); |
1da177e4 LT |
2006 | break; |
2007 | case FUTEX_FD: | |
2008 | /* non-zero val means F_SETOWN(getpid()) & F_SETSIG(val) */ | |
2009 | ret = futex_fd(uaddr, val); | |
2010 | break; | |
2011 | case FUTEX_REQUEUE: | |
34f01cc1 | 2012 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL); |
1da177e4 LT |
2013 | break; |
2014 | case FUTEX_CMP_REQUEUE: | |
34f01cc1 | 2015 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3); |
1da177e4 | 2016 | break; |
4732efbe | 2017 | case FUTEX_WAKE_OP: |
34f01cc1 | 2018 | ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3); |
4732efbe | 2019 | break; |
c87e2837 | 2020 | case FUTEX_LOCK_PI: |
34f01cc1 | 2021 | ret = futex_lock_pi(uaddr, fshared, val, timeout, 0); |
c87e2837 IM |
2022 | break; |
2023 | case FUTEX_UNLOCK_PI: | |
34f01cc1 | 2024 | ret = futex_unlock_pi(uaddr, fshared); |
c87e2837 IM |
2025 | break; |
2026 | case FUTEX_TRYLOCK_PI: | |
34f01cc1 | 2027 | ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1); |
c87e2837 | 2028 | break; |
1da177e4 LT |
2029 | default: |
2030 | ret = -ENOSYS; | |
2031 | } | |
2032 | return ret; | |
2033 | } | |
2034 | ||
2035 | ||
e2970f2f | 2036 | asmlinkage long sys_futex(u32 __user *uaddr, int op, u32 val, |
1da177e4 | 2037 | struct timespec __user *utime, u32 __user *uaddr2, |
e2970f2f | 2038 | u32 val3) |
1da177e4 | 2039 | { |
c19384b5 PP |
2040 | struct timespec ts; |
2041 | ktime_t t, *tp = NULL; | |
e2970f2f | 2042 | u32 val2 = 0; |
34f01cc1 | 2043 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 2044 | |
34f01cc1 | 2045 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI)) { |
c19384b5 | 2046 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 2047 | return -EFAULT; |
c19384b5 | 2048 | if (!timespec_valid(&ts)) |
9741ef96 | 2049 | return -EINVAL; |
c19384b5 PP |
2050 | |
2051 | t = timespec_to_ktime(ts); | |
34f01cc1 | 2052 | if (cmd == FUTEX_WAIT) |
c19384b5 PP |
2053 | t = ktime_add(ktime_get(), t); |
2054 | tp = &t; | |
1da177e4 LT |
2055 | } |
2056 | /* | |
34f01cc1 | 2057 | * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE. |
1da177e4 | 2058 | */ |
bd197234 | 2059 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE) |
e2970f2f | 2060 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 2061 | |
c19384b5 | 2062 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
2063 | } |
2064 | ||
454e2398 DH |
2065 | static int futexfs_get_sb(struct file_system_type *fs_type, |
2066 | int flags, const char *dev_name, void *data, | |
2067 | struct vfsmount *mnt) | |
1da177e4 | 2068 | { |
454e2398 | 2069 | return get_sb_pseudo(fs_type, "futex", NULL, 0xBAD1DEA, mnt); |
1da177e4 LT |
2070 | } |
2071 | ||
2072 | static struct file_system_type futex_fs_type = { | |
2073 | .name = "futexfs", | |
2074 | .get_sb = futexfs_get_sb, | |
2075 | .kill_sb = kill_anon_super, | |
2076 | }; | |
2077 | ||
2078 | static int __init init(void) | |
2079 | { | |
95362fa9 AM |
2080 | int i = register_filesystem(&futex_fs_type); |
2081 | ||
2082 | if (i) | |
2083 | return i; | |
1da177e4 | 2084 | |
1da177e4 | 2085 | futex_mnt = kern_mount(&futex_fs_type); |
95362fa9 AM |
2086 | if (IS_ERR(futex_mnt)) { |
2087 | unregister_filesystem(&futex_fs_type); | |
2088 | return PTR_ERR(futex_mnt); | |
2089 | } | |
1da177e4 LT |
2090 | |
2091 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { | |
ec92d082 | 2092 | plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock); |
1da177e4 LT |
2093 | spin_lock_init(&futex_queues[i].lock); |
2094 | } | |
2095 | return 0; | |
2096 | } | |
2097 | __initcall(init); |