]>
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 | * | |
52400ba9 DH |
22 | * Requeue-PI support by Darren Hart <dvhltc@us.ibm.com> |
23 | * Copyright (C) IBM Corporation, 2009 | |
24 | * Thanks to Thomas Gleixner for conceptual design and careful reviews. | |
25 | * | |
1da177e4 LT |
26 | * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly |
27 | * enough at me, Linus for the original (flawed) idea, Matthew | |
28 | * Kirkwood for proof-of-concept implementation. | |
29 | * | |
30 | * "The futexes are also cursed." | |
31 | * "But they come in a choice of three flavours!" | |
32 | * | |
33 | * This program is free software; you can redistribute it and/or modify | |
34 | * it under the terms of the GNU General Public License as published by | |
35 | * the Free Software Foundation; either version 2 of the License, or | |
36 | * (at your option) any later version. | |
37 | * | |
38 | * This program is distributed in the hope that it will be useful, | |
39 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
40 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
41 | * GNU General Public License for more details. | |
42 | * | |
43 | * You should have received a copy of the GNU General Public License | |
44 | * along with this program; if not, write to the Free Software | |
45 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
46 | */ | |
47 | #include <linux/slab.h> | |
48 | #include <linux/poll.h> | |
49 | #include <linux/fs.h> | |
50 | #include <linux/file.h> | |
51 | #include <linux/jhash.h> | |
52 | #include <linux/init.h> | |
53 | #include <linux/futex.h> | |
54 | #include <linux/mount.h> | |
55 | #include <linux/pagemap.h> | |
56 | #include <linux/syscalls.h> | |
7ed20e1a | 57 | #include <linux/signal.h> |
9adef58b | 58 | #include <linux/module.h> |
fd5eea42 | 59 | #include <linux/magic.h> |
b488893a PE |
60 | #include <linux/pid.h> |
61 | #include <linux/nsproxy.h> | |
62 | ||
4732efbe | 63 | #include <asm/futex.h> |
1da177e4 | 64 | |
c87e2837 IM |
65 | #include "rtmutex_common.h" |
66 | ||
a0c1e907 TG |
67 | int __read_mostly futex_cmpxchg_enabled; |
68 | ||
1da177e4 LT |
69 | #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8) |
70 | ||
c87e2837 IM |
71 | /* |
72 | * Priority Inheritance state: | |
73 | */ | |
74 | struct futex_pi_state { | |
75 | /* | |
76 | * list of 'owned' pi_state instances - these have to be | |
77 | * cleaned up in do_exit() if the task exits prematurely: | |
78 | */ | |
79 | struct list_head list; | |
80 | ||
81 | /* | |
82 | * The PI object: | |
83 | */ | |
84 | struct rt_mutex pi_mutex; | |
85 | ||
86 | struct task_struct *owner; | |
87 | atomic_t refcount; | |
88 | ||
89 | union futex_key key; | |
90 | }; | |
91 | ||
1da177e4 LT |
92 | /* |
93 | * We use this hashed waitqueue instead of a normal wait_queue_t, so | |
94 | * we can wake only the relevant ones (hashed queues may be shared). | |
95 | * | |
96 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 97 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
1da177e4 | 98 | * The order of wakup is always to make the first condition true, then |
73500ac5 | 99 | * wake up q->waiter, then make the second condition true. |
1da177e4 LT |
100 | */ |
101 | struct futex_q { | |
ec92d082 | 102 | struct plist_node list; |
f1a11e05 TG |
103 | /* Waiter reference */ |
104 | struct task_struct *task; | |
1da177e4 | 105 | |
e2970f2f | 106 | /* Which hash list lock to use: */ |
1da177e4 LT |
107 | spinlock_t *lock_ptr; |
108 | ||
e2970f2f | 109 | /* Key which the futex is hashed on: */ |
1da177e4 LT |
110 | union futex_key key; |
111 | ||
c87e2837 IM |
112 | /* Optional priority inheritance state: */ |
113 | struct futex_pi_state *pi_state; | |
cd689985 | 114 | |
52400ba9 DH |
115 | /* rt_waiter storage for requeue_pi: */ |
116 | struct rt_mutex_waiter *rt_waiter; | |
117 | ||
cd689985 TG |
118 | /* Bitset for the optional bitmasked wakeup */ |
119 | u32 bitset; | |
1da177e4 LT |
120 | }; |
121 | ||
122 | /* | |
b2d0994b DH |
123 | * Hash buckets are shared by all the futex_keys that hash to the same |
124 | * location. Each key may have multiple futex_q structures, one for each task | |
125 | * waiting on a futex. | |
1da177e4 LT |
126 | */ |
127 | struct futex_hash_bucket { | |
ec92d082 PP |
128 | spinlock_t lock; |
129 | struct plist_head chain; | |
1da177e4 LT |
130 | }; |
131 | ||
132 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; | |
133 | ||
1da177e4 LT |
134 | /* |
135 | * We hash on the keys returned from get_futex_key (see below). | |
136 | */ | |
137 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
138 | { | |
139 | u32 hash = jhash2((u32*)&key->both.word, | |
140 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
141 | key->both.offset); | |
142 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; | |
143 | } | |
144 | ||
145 | /* | |
146 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
147 | */ | |
148 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
149 | { | |
150 | return (key1->both.word == key2->both.word | |
151 | && key1->both.ptr == key2->both.ptr | |
152 | && key1->both.offset == key2->both.offset); | |
153 | } | |
154 | ||
38d47c1b PZ |
155 | /* |
156 | * Take a reference to the resource addressed by a key. | |
157 | * Can be called while holding spinlocks. | |
158 | * | |
159 | */ | |
160 | static void get_futex_key_refs(union futex_key *key) | |
161 | { | |
162 | if (!key->both.ptr) | |
163 | return; | |
164 | ||
165 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
166 | case FUT_OFF_INODE: | |
167 | atomic_inc(&key->shared.inode->i_count); | |
168 | break; | |
169 | case FUT_OFF_MMSHARED: | |
170 | atomic_inc(&key->private.mm->mm_count); | |
171 | break; | |
172 | } | |
173 | } | |
174 | ||
175 | /* | |
176 | * Drop a reference to the resource addressed by a key. | |
177 | * The hash bucket spinlock must not be held. | |
178 | */ | |
179 | static void drop_futex_key_refs(union futex_key *key) | |
180 | { | |
90621c40 DH |
181 | if (!key->both.ptr) { |
182 | /* If we're here then we tried to put a key we failed to get */ | |
183 | WARN_ON_ONCE(1); | |
38d47c1b | 184 | return; |
90621c40 | 185 | } |
38d47c1b PZ |
186 | |
187 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
188 | case FUT_OFF_INODE: | |
189 | iput(key->shared.inode); | |
190 | break; | |
191 | case FUT_OFF_MMSHARED: | |
192 | mmdrop(key->private.mm); | |
193 | break; | |
194 | } | |
195 | } | |
196 | ||
34f01cc1 ED |
197 | /** |
198 | * get_futex_key - Get parameters which are the keys for a futex. | |
199 | * @uaddr: virtual address of the futex | |
b2d0994b | 200 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED |
34f01cc1 | 201 | * @key: address where result is stored. |
64d1304a | 202 | * @rw: mapping needs to be read/write (values: VERIFY_READ, VERIFY_WRITE) |
34f01cc1 ED |
203 | * |
204 | * Returns a negative error code or 0 | |
205 | * The key words are stored in *key on success. | |
1da177e4 | 206 | * |
f3a43f3f | 207 | * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode, |
1da177e4 LT |
208 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
209 | * We can usually work out the index without swapping in the page. | |
210 | * | |
b2d0994b | 211 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 212 | */ |
64d1304a TG |
213 | static int |
214 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) | |
1da177e4 | 215 | { |
e2970f2f | 216 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 217 | struct mm_struct *mm = current->mm; |
1da177e4 LT |
218 | struct page *page; |
219 | int err; | |
220 | ||
221 | /* | |
222 | * The futex address must be "naturally" aligned. | |
223 | */ | |
e2970f2f | 224 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 225 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 226 | return -EINVAL; |
e2970f2f | 227 | address -= key->both.offset; |
1da177e4 | 228 | |
34f01cc1 ED |
229 | /* |
230 | * PROCESS_PRIVATE futexes are fast. | |
231 | * As the mm cannot disappear under us and the 'key' only needs | |
232 | * virtual address, we dont even have to find the underlying vma. | |
233 | * Note : We do have to check 'uaddr' is a valid user address, | |
234 | * but access_ok() should be faster than find_vma() | |
235 | */ | |
236 | if (!fshared) { | |
64d1304a | 237 | if (unlikely(!access_ok(rw, uaddr, sizeof(u32)))) |
34f01cc1 ED |
238 | return -EFAULT; |
239 | key->private.mm = mm; | |
240 | key->private.address = address; | |
42569c39 | 241 | get_futex_key_refs(key); |
34f01cc1 ED |
242 | return 0; |
243 | } | |
1da177e4 | 244 | |
38d47c1b | 245 | again: |
64d1304a | 246 | err = get_user_pages_fast(address, 1, rw == VERIFY_WRITE, &page); |
38d47c1b PZ |
247 | if (err < 0) |
248 | return err; | |
249 | ||
250 | lock_page(page); | |
251 | if (!page->mapping) { | |
252 | unlock_page(page); | |
253 | put_page(page); | |
254 | goto again; | |
255 | } | |
1da177e4 LT |
256 | |
257 | /* | |
258 | * Private mappings are handled in a simple way. | |
259 | * | |
260 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
261 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 262 | * the object not the particular process. |
1da177e4 | 263 | */ |
38d47c1b PZ |
264 | if (PageAnon(page)) { |
265 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ | |
1da177e4 | 266 | key->private.mm = mm; |
e2970f2f | 267 | key->private.address = address; |
38d47c1b PZ |
268 | } else { |
269 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ | |
270 | key->shared.inode = page->mapping->host; | |
271 | key->shared.pgoff = page->index; | |
1da177e4 LT |
272 | } |
273 | ||
38d47c1b | 274 | get_futex_key_refs(key); |
1da177e4 | 275 | |
38d47c1b PZ |
276 | unlock_page(page); |
277 | put_page(page); | |
278 | return 0; | |
1da177e4 LT |
279 | } |
280 | ||
38d47c1b | 281 | static inline |
c2f9f201 | 282 | void put_futex_key(int fshared, union futex_key *key) |
1da177e4 | 283 | { |
38d47c1b | 284 | drop_futex_key_refs(key); |
1da177e4 LT |
285 | } |
286 | ||
4b1c486b DH |
287 | /** |
288 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
289 | * @hb: the hash bucket the futex_q's reside in | |
290 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
291 | * | |
292 | * Must be called with the hb lock held. | |
293 | */ | |
294 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
295 | union futex_key *key) | |
296 | { | |
297 | struct futex_q *this; | |
298 | ||
299 | plist_for_each_entry(this, &hb->chain, list) { | |
300 | if (match_futex(&this->key, key)) | |
301 | return this; | |
302 | } | |
303 | return NULL; | |
304 | } | |
305 | ||
36cf3b5c TG |
306 | static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval) |
307 | { | |
308 | u32 curval; | |
309 | ||
310 | pagefault_disable(); | |
311 | curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval); | |
312 | pagefault_enable(); | |
313 | ||
314 | return curval; | |
315 | } | |
316 | ||
317 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
318 | { |
319 | int ret; | |
320 | ||
a866374a | 321 | pagefault_disable(); |
e2970f2f | 322 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 323 | pagefault_enable(); |
1da177e4 LT |
324 | |
325 | return ret ? -EFAULT : 0; | |
326 | } | |
327 | ||
c87e2837 IM |
328 | |
329 | /* | |
330 | * PI code: | |
331 | */ | |
332 | static int refill_pi_state_cache(void) | |
333 | { | |
334 | struct futex_pi_state *pi_state; | |
335 | ||
336 | if (likely(current->pi_state_cache)) | |
337 | return 0; | |
338 | ||
4668edc3 | 339 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
340 | |
341 | if (!pi_state) | |
342 | return -ENOMEM; | |
343 | ||
c87e2837 IM |
344 | INIT_LIST_HEAD(&pi_state->list); |
345 | /* pi_mutex gets initialized later */ | |
346 | pi_state->owner = NULL; | |
347 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 348 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
349 | |
350 | current->pi_state_cache = pi_state; | |
351 | ||
352 | return 0; | |
353 | } | |
354 | ||
355 | static struct futex_pi_state * alloc_pi_state(void) | |
356 | { | |
357 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
358 | ||
359 | WARN_ON(!pi_state); | |
360 | current->pi_state_cache = NULL; | |
361 | ||
362 | return pi_state; | |
363 | } | |
364 | ||
365 | static void free_pi_state(struct futex_pi_state *pi_state) | |
366 | { | |
367 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
368 | return; | |
369 | ||
370 | /* | |
371 | * If pi_state->owner is NULL, the owner is most probably dying | |
372 | * and has cleaned up the pi_state already | |
373 | */ | |
374 | if (pi_state->owner) { | |
375 | spin_lock_irq(&pi_state->owner->pi_lock); | |
376 | list_del_init(&pi_state->list); | |
377 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
378 | ||
379 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
380 | } | |
381 | ||
382 | if (current->pi_state_cache) | |
383 | kfree(pi_state); | |
384 | else { | |
385 | /* | |
386 | * pi_state->list is already empty. | |
387 | * clear pi_state->owner. | |
388 | * refcount is at 0 - put it back to 1. | |
389 | */ | |
390 | pi_state->owner = NULL; | |
391 | atomic_set(&pi_state->refcount, 1); | |
392 | current->pi_state_cache = pi_state; | |
393 | } | |
394 | } | |
395 | ||
396 | /* | |
397 | * Look up the task based on what TID userspace gave us. | |
398 | * We dont trust it. | |
399 | */ | |
400 | static struct task_struct * futex_find_get_task(pid_t pid) | |
401 | { | |
402 | struct task_struct *p; | |
c69e8d9c | 403 | const struct cred *cred = current_cred(), *pcred; |
c87e2837 | 404 | |
d359b549 | 405 | rcu_read_lock(); |
228ebcbe | 406 | p = find_task_by_vpid(pid); |
c69e8d9c | 407 | if (!p) { |
a06381fe | 408 | p = ERR_PTR(-ESRCH); |
c69e8d9c DH |
409 | } else { |
410 | pcred = __task_cred(p); | |
411 | if (cred->euid != pcred->euid && | |
412 | cred->euid != pcred->uid) | |
413 | p = ERR_PTR(-ESRCH); | |
414 | else | |
415 | get_task_struct(p); | |
416 | } | |
a06381fe | 417 | |
d359b549 | 418 | rcu_read_unlock(); |
c87e2837 IM |
419 | |
420 | return p; | |
421 | } | |
422 | ||
423 | /* | |
424 | * This task is holding PI mutexes at exit time => bad. | |
425 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
426 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
427 | */ | |
428 | void exit_pi_state_list(struct task_struct *curr) | |
429 | { | |
c87e2837 IM |
430 | struct list_head *next, *head = &curr->pi_state_list; |
431 | struct futex_pi_state *pi_state; | |
627371d7 | 432 | struct futex_hash_bucket *hb; |
38d47c1b | 433 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 434 | |
a0c1e907 TG |
435 | if (!futex_cmpxchg_enabled) |
436 | return; | |
c87e2837 IM |
437 | /* |
438 | * We are a ZOMBIE and nobody can enqueue itself on | |
439 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 440 | * versus waiters unqueueing themselves: |
c87e2837 IM |
441 | */ |
442 | spin_lock_irq(&curr->pi_lock); | |
443 | while (!list_empty(head)) { | |
444 | ||
445 | next = head->next; | |
446 | pi_state = list_entry(next, struct futex_pi_state, list); | |
447 | key = pi_state->key; | |
627371d7 | 448 | hb = hash_futex(&key); |
c87e2837 IM |
449 | spin_unlock_irq(&curr->pi_lock); |
450 | ||
c87e2837 IM |
451 | spin_lock(&hb->lock); |
452 | ||
453 | spin_lock_irq(&curr->pi_lock); | |
627371d7 IM |
454 | /* |
455 | * We dropped the pi-lock, so re-check whether this | |
456 | * task still owns the PI-state: | |
457 | */ | |
c87e2837 IM |
458 | if (head->next != next) { |
459 | spin_unlock(&hb->lock); | |
460 | continue; | |
461 | } | |
462 | ||
c87e2837 | 463 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
464 | WARN_ON(list_empty(&pi_state->list)); |
465 | list_del_init(&pi_state->list); | |
c87e2837 IM |
466 | pi_state->owner = NULL; |
467 | spin_unlock_irq(&curr->pi_lock); | |
468 | ||
469 | rt_mutex_unlock(&pi_state->pi_mutex); | |
470 | ||
471 | spin_unlock(&hb->lock); | |
472 | ||
473 | spin_lock_irq(&curr->pi_lock); | |
474 | } | |
475 | spin_unlock_irq(&curr->pi_lock); | |
476 | } | |
477 | ||
478 | static int | |
d0aa7a70 PP |
479 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
480 | union futex_key *key, struct futex_pi_state **ps) | |
c87e2837 IM |
481 | { |
482 | struct futex_pi_state *pi_state = NULL; | |
483 | struct futex_q *this, *next; | |
ec92d082 | 484 | struct plist_head *head; |
c87e2837 | 485 | struct task_struct *p; |
778e9a9c | 486 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 IM |
487 | |
488 | head = &hb->chain; | |
489 | ||
ec92d082 | 490 | plist_for_each_entry_safe(this, next, head, list) { |
d0aa7a70 | 491 | if (match_futex(&this->key, key)) { |
c87e2837 IM |
492 | /* |
493 | * Another waiter already exists - bump up | |
494 | * the refcount and return its pi_state: | |
495 | */ | |
496 | pi_state = this->pi_state; | |
06a9ec29 TG |
497 | /* |
498 | * Userspace might have messed up non PI and PI futexes | |
499 | */ | |
500 | if (unlikely(!pi_state)) | |
501 | return -EINVAL; | |
502 | ||
627371d7 | 503 | WARN_ON(!atomic_read(&pi_state->refcount)); |
778e9a9c AK |
504 | WARN_ON(pid && pi_state->owner && |
505 | pi_state->owner->pid != pid); | |
627371d7 | 506 | |
c87e2837 | 507 | atomic_inc(&pi_state->refcount); |
d0aa7a70 | 508 | *ps = pi_state; |
c87e2837 IM |
509 | |
510 | return 0; | |
511 | } | |
512 | } | |
513 | ||
514 | /* | |
e3f2ddea | 515 | * We are the first waiter - try to look up the real owner and attach |
778e9a9c | 516 | * the new pi_state to it, but bail out when TID = 0 |
c87e2837 | 517 | */ |
778e9a9c | 518 | if (!pid) |
e3f2ddea | 519 | return -ESRCH; |
c87e2837 | 520 | p = futex_find_get_task(pid); |
778e9a9c AK |
521 | if (IS_ERR(p)) |
522 | return PTR_ERR(p); | |
523 | ||
524 | /* | |
525 | * We need to look at the task state flags to figure out, | |
526 | * whether the task is exiting. To protect against the do_exit | |
527 | * change of the task flags, we do this protected by | |
528 | * p->pi_lock: | |
529 | */ | |
530 | spin_lock_irq(&p->pi_lock); | |
531 | if (unlikely(p->flags & PF_EXITING)) { | |
532 | /* | |
533 | * The task is on the way out. When PF_EXITPIDONE is | |
534 | * set, we know that the task has finished the | |
535 | * cleanup: | |
536 | */ | |
537 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
538 | ||
539 | spin_unlock_irq(&p->pi_lock); | |
540 | put_task_struct(p); | |
541 | return ret; | |
542 | } | |
c87e2837 IM |
543 | |
544 | pi_state = alloc_pi_state(); | |
545 | ||
546 | /* | |
547 | * Initialize the pi_mutex in locked state and make 'p' | |
548 | * the owner of it: | |
549 | */ | |
550 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
551 | ||
552 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 553 | pi_state->key = *key; |
c87e2837 | 554 | |
627371d7 | 555 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
556 | list_add(&pi_state->list, &p->pi_state_list); |
557 | pi_state->owner = p; | |
558 | spin_unlock_irq(&p->pi_lock); | |
559 | ||
560 | put_task_struct(p); | |
561 | ||
d0aa7a70 | 562 | *ps = pi_state; |
c87e2837 IM |
563 | |
564 | return 0; | |
565 | } | |
566 | ||
1a52084d DH |
567 | /** |
568 | * futex_lock_pi_atomic() - atomic work required to acquire a pi aware futex | |
bab5bc9e DH |
569 | * @uaddr: the pi futex user address |
570 | * @hb: the pi futex hash bucket | |
571 | * @key: the futex key associated with uaddr and hb | |
572 | * @ps: the pi_state pointer where we store the result of the | |
573 | * lookup | |
574 | * @task: the task to perform the atomic lock work for. This will | |
575 | * be "current" except in the case of requeue pi. | |
576 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d DH |
577 | * |
578 | * Returns: | |
579 | * 0 - ready to wait | |
580 | * 1 - acquired the lock | |
581 | * <0 - error | |
582 | * | |
583 | * The hb->lock and futex_key refs shall be held by the caller. | |
584 | */ | |
585 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
586 | union futex_key *key, | |
587 | struct futex_pi_state **ps, | |
bab5bc9e | 588 | struct task_struct *task, int set_waiters) |
1a52084d DH |
589 | { |
590 | int lock_taken, ret, ownerdied = 0; | |
591 | u32 uval, newval, curval; | |
592 | ||
593 | retry: | |
594 | ret = lock_taken = 0; | |
595 | ||
596 | /* | |
597 | * To avoid races, we attempt to take the lock here again | |
598 | * (by doing a 0 -> TID atomic cmpxchg), while holding all | |
599 | * the locks. It will most likely not succeed. | |
600 | */ | |
601 | newval = task_pid_vnr(task); | |
bab5bc9e DH |
602 | if (set_waiters) |
603 | newval |= FUTEX_WAITERS; | |
1a52084d DH |
604 | |
605 | curval = cmpxchg_futex_value_locked(uaddr, 0, newval); | |
606 | ||
607 | if (unlikely(curval == -EFAULT)) | |
608 | return -EFAULT; | |
609 | ||
610 | /* | |
611 | * Detect deadlocks. | |
612 | */ | |
613 | if ((unlikely((curval & FUTEX_TID_MASK) == task_pid_vnr(task)))) | |
614 | return -EDEADLK; | |
615 | ||
616 | /* | |
617 | * Surprise - we got the lock. Just return to userspace: | |
618 | */ | |
619 | if (unlikely(!curval)) | |
620 | return 1; | |
621 | ||
622 | uval = curval; | |
623 | ||
624 | /* | |
625 | * Set the FUTEX_WAITERS flag, so the owner will know it has someone | |
626 | * to wake at the next unlock. | |
627 | */ | |
628 | newval = curval | FUTEX_WAITERS; | |
629 | ||
630 | /* | |
631 | * There are two cases, where a futex might have no owner (the | |
632 | * owner TID is 0): OWNER_DIED. We take over the futex in this | |
633 | * case. We also do an unconditional take over, when the owner | |
634 | * of the futex died. | |
635 | * | |
636 | * This is safe as we are protected by the hash bucket lock ! | |
637 | */ | |
638 | if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { | |
639 | /* Keep the OWNER_DIED bit */ | |
640 | newval = (curval & ~FUTEX_TID_MASK) | task_pid_vnr(task); | |
641 | ownerdied = 0; | |
642 | lock_taken = 1; | |
643 | } | |
644 | ||
645 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); | |
646 | ||
647 | if (unlikely(curval == -EFAULT)) | |
648 | return -EFAULT; | |
649 | if (unlikely(curval != uval)) | |
650 | goto retry; | |
651 | ||
652 | /* | |
653 | * We took the lock due to owner died take over. | |
654 | */ | |
655 | if (unlikely(lock_taken)) | |
656 | return 1; | |
657 | ||
658 | /* | |
659 | * We dont have the lock. Look up the PI state (or create it if | |
660 | * we are the first waiter): | |
661 | */ | |
662 | ret = lookup_pi_state(uval, hb, key, ps); | |
663 | ||
664 | if (unlikely(ret)) { | |
665 | switch (ret) { | |
666 | case -ESRCH: | |
667 | /* | |
668 | * No owner found for this futex. Check if the | |
669 | * OWNER_DIED bit is set to figure out whether | |
670 | * this is a robust futex or not. | |
671 | */ | |
672 | if (get_futex_value_locked(&curval, uaddr)) | |
673 | return -EFAULT; | |
674 | ||
675 | /* | |
676 | * We simply start over in case of a robust | |
677 | * futex. The code above will take the futex | |
678 | * and return happy. | |
679 | */ | |
680 | if (curval & FUTEX_OWNER_DIED) { | |
681 | ownerdied = 1; | |
682 | goto retry; | |
683 | } | |
684 | default: | |
685 | break; | |
686 | } | |
687 | } | |
688 | ||
689 | return ret; | |
690 | } | |
691 | ||
1da177e4 LT |
692 | /* |
693 | * The hash bucket lock must be held when this is called. | |
694 | * Afterwards, the futex_q must not be accessed. | |
695 | */ | |
696 | static void wake_futex(struct futex_q *q) | |
697 | { | |
f1a11e05 TG |
698 | struct task_struct *p = q->task; |
699 | ||
1da177e4 | 700 | /* |
f1a11e05 TG |
701 | * We set q->lock_ptr = NULL _before_ we wake up the task. If |
702 | * a non futex wake up happens on another CPU then the task | |
703 | * might exit and p would dereference a non existing task | |
704 | * struct. Prevent this by holding a reference on p across the | |
705 | * wake up. | |
1da177e4 | 706 | */ |
f1a11e05 TG |
707 | get_task_struct(p); |
708 | ||
709 | plist_del(&q->list, &q->list.plist); | |
1da177e4 | 710 | /* |
f1a11e05 TG |
711 | * The waiting task can free the futex_q as soon as |
712 | * q->lock_ptr = NULL is written, without taking any locks. A | |
713 | * memory barrier is required here to prevent the following | |
714 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 715 | */ |
ccdea2f8 | 716 | smp_wmb(); |
1da177e4 | 717 | q->lock_ptr = NULL; |
f1a11e05 TG |
718 | |
719 | wake_up_state(p, TASK_NORMAL); | |
720 | put_task_struct(p); | |
1da177e4 LT |
721 | } |
722 | ||
c87e2837 IM |
723 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
724 | { | |
725 | struct task_struct *new_owner; | |
726 | struct futex_pi_state *pi_state = this->pi_state; | |
727 | u32 curval, newval; | |
728 | ||
729 | if (!pi_state) | |
730 | return -EINVAL; | |
731 | ||
21778867 | 732 | spin_lock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
733 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
734 | ||
735 | /* | |
736 | * This happens when we have stolen the lock and the original | |
737 | * pending owner did not enqueue itself back on the rt_mutex. | |
738 | * Thats not a tragedy. We know that way, that a lock waiter | |
739 | * is on the fly. We make the futex_q waiter the pending owner. | |
740 | */ | |
741 | if (!new_owner) | |
742 | new_owner = this->task; | |
743 | ||
744 | /* | |
745 | * We pass it to the next owner. (The WAITERS bit is always | |
746 | * kept enabled while there is PI state around. We must also | |
747 | * preserve the owner died bit.) | |
748 | */ | |
e3f2ddea | 749 | if (!(uval & FUTEX_OWNER_DIED)) { |
778e9a9c AK |
750 | int ret = 0; |
751 | ||
b488893a | 752 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 753 | |
36cf3b5c | 754 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
778e9a9c | 755 | |
e3f2ddea | 756 | if (curval == -EFAULT) |
778e9a9c | 757 | ret = -EFAULT; |
cde898fa | 758 | else if (curval != uval) |
778e9a9c AK |
759 | ret = -EINVAL; |
760 | if (ret) { | |
761 | spin_unlock(&pi_state->pi_mutex.wait_lock); | |
762 | return ret; | |
763 | } | |
e3f2ddea | 764 | } |
c87e2837 | 765 | |
627371d7 IM |
766 | spin_lock_irq(&pi_state->owner->pi_lock); |
767 | WARN_ON(list_empty(&pi_state->list)); | |
768 | list_del_init(&pi_state->list); | |
769 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
770 | ||
771 | spin_lock_irq(&new_owner->pi_lock); | |
772 | WARN_ON(!list_empty(&pi_state->list)); | |
c87e2837 IM |
773 | list_add(&pi_state->list, &new_owner->pi_state_list); |
774 | pi_state->owner = new_owner; | |
627371d7 IM |
775 | spin_unlock_irq(&new_owner->pi_lock); |
776 | ||
21778867 | 777 | spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
778 | rt_mutex_unlock(&pi_state->pi_mutex); |
779 | ||
780 | return 0; | |
781 | } | |
782 | ||
783 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) | |
784 | { | |
785 | u32 oldval; | |
786 | ||
787 | /* | |
788 | * There is no waiter, so we unlock the futex. The owner died | |
789 | * bit has not to be preserved here. We are the owner: | |
790 | */ | |
36cf3b5c | 791 | oldval = cmpxchg_futex_value_locked(uaddr, uval, 0); |
c87e2837 IM |
792 | |
793 | if (oldval == -EFAULT) | |
794 | return oldval; | |
795 | if (oldval != uval) | |
796 | return -EAGAIN; | |
797 | ||
798 | return 0; | |
799 | } | |
800 | ||
8b8f319f IM |
801 | /* |
802 | * Express the locking dependencies for lockdep: | |
803 | */ | |
804 | static inline void | |
805 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
806 | { | |
807 | if (hb1 <= hb2) { | |
808 | spin_lock(&hb1->lock); | |
809 | if (hb1 < hb2) | |
810 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
811 | } else { /* hb1 > hb2 */ | |
812 | spin_lock(&hb2->lock); | |
813 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
814 | } | |
815 | } | |
816 | ||
5eb3dc62 DH |
817 | static inline void |
818 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
819 | { | |
f061d351 | 820 | spin_unlock(&hb1->lock); |
88f502fe IM |
821 | if (hb1 != hb2) |
822 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
823 | } |
824 | ||
1da177e4 | 825 | /* |
b2d0994b | 826 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 827 | */ |
c2f9f201 | 828 | static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset) |
1da177e4 | 829 | { |
e2970f2f | 830 | struct futex_hash_bucket *hb; |
1da177e4 | 831 | struct futex_q *this, *next; |
ec92d082 | 832 | struct plist_head *head; |
38d47c1b | 833 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 LT |
834 | int ret; |
835 | ||
cd689985 TG |
836 | if (!bitset) |
837 | return -EINVAL; | |
838 | ||
64d1304a | 839 | ret = get_futex_key(uaddr, fshared, &key, VERIFY_READ); |
1da177e4 LT |
840 | if (unlikely(ret != 0)) |
841 | goto out; | |
842 | ||
e2970f2f IM |
843 | hb = hash_futex(&key); |
844 | spin_lock(&hb->lock); | |
845 | head = &hb->chain; | |
1da177e4 | 846 | |
ec92d082 | 847 | plist_for_each_entry_safe(this, next, head, list) { |
1da177e4 | 848 | if (match_futex (&this->key, &key)) { |
52400ba9 | 849 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
850 | ret = -EINVAL; |
851 | break; | |
852 | } | |
cd689985 TG |
853 | |
854 | /* Check if one of the bits is set in both bitsets */ | |
855 | if (!(this->bitset & bitset)) | |
856 | continue; | |
857 | ||
1da177e4 LT |
858 | wake_futex(this); |
859 | if (++ret >= nr_wake) | |
860 | break; | |
861 | } | |
862 | } | |
863 | ||
e2970f2f | 864 | spin_unlock(&hb->lock); |
38d47c1b | 865 | put_futex_key(fshared, &key); |
42d35d48 | 866 | out: |
1da177e4 LT |
867 | return ret; |
868 | } | |
869 | ||
4732efbe JJ |
870 | /* |
871 | * Wake up all waiters hashed on the physical page that is mapped | |
872 | * to this virtual address: | |
873 | */ | |
e2970f2f | 874 | static int |
c2f9f201 | 875 | futex_wake_op(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, |
e2970f2f | 876 | int nr_wake, int nr_wake2, int op) |
4732efbe | 877 | { |
38d47c1b | 878 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 879 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 880 | struct plist_head *head; |
4732efbe | 881 | struct futex_q *this, *next; |
e4dc5b7a | 882 | int ret, op_ret; |
4732efbe | 883 | |
e4dc5b7a | 884 | retry: |
64d1304a | 885 | ret = get_futex_key(uaddr1, fshared, &key1, VERIFY_READ); |
4732efbe JJ |
886 | if (unlikely(ret != 0)) |
887 | goto out; | |
64d1304a | 888 | ret = get_futex_key(uaddr2, fshared, &key2, VERIFY_WRITE); |
4732efbe | 889 | if (unlikely(ret != 0)) |
42d35d48 | 890 | goto out_put_key1; |
4732efbe | 891 | |
e2970f2f IM |
892 | hb1 = hash_futex(&key1); |
893 | hb2 = hash_futex(&key2); | |
4732efbe | 894 | |
8b8f319f | 895 | double_lock_hb(hb1, hb2); |
e4dc5b7a | 896 | retry_private: |
e2970f2f | 897 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 898 | if (unlikely(op_ret < 0)) { |
e2970f2f | 899 | u32 dummy; |
4732efbe | 900 | |
5eb3dc62 | 901 | double_unlock_hb(hb1, hb2); |
4732efbe | 902 | |
7ee1dd3f | 903 | #ifndef CONFIG_MMU |
e2970f2f IM |
904 | /* |
905 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
906 | * but we might get them from range checking | |
907 | */ | |
7ee1dd3f | 908 | ret = op_ret; |
42d35d48 | 909 | goto out_put_keys; |
7ee1dd3f DH |
910 | #endif |
911 | ||
796f8d9b DG |
912 | if (unlikely(op_ret != -EFAULT)) { |
913 | ret = op_ret; | |
42d35d48 | 914 | goto out_put_keys; |
796f8d9b DG |
915 | } |
916 | ||
e2970f2f | 917 | ret = get_user(dummy, uaddr2); |
4732efbe | 918 | if (ret) |
de87fcc1 | 919 | goto out_put_keys; |
4732efbe | 920 | |
e4dc5b7a DH |
921 | if (!fshared) |
922 | goto retry_private; | |
923 | ||
de87fcc1 DH |
924 | put_futex_key(fshared, &key2); |
925 | put_futex_key(fshared, &key1); | |
e4dc5b7a | 926 | goto retry; |
4732efbe JJ |
927 | } |
928 | ||
e2970f2f | 929 | head = &hb1->chain; |
4732efbe | 930 | |
ec92d082 | 931 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
932 | if (match_futex (&this->key, &key1)) { |
933 | wake_futex(this); | |
934 | if (++ret >= nr_wake) | |
935 | break; | |
936 | } | |
937 | } | |
938 | ||
939 | if (op_ret > 0) { | |
e2970f2f | 940 | head = &hb2->chain; |
4732efbe JJ |
941 | |
942 | op_ret = 0; | |
ec92d082 | 943 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
944 | if (match_futex (&this->key, &key2)) { |
945 | wake_futex(this); | |
946 | if (++op_ret >= nr_wake2) | |
947 | break; | |
948 | } | |
949 | } | |
950 | ret += op_ret; | |
951 | } | |
952 | ||
5eb3dc62 | 953 | double_unlock_hb(hb1, hb2); |
42d35d48 | 954 | out_put_keys: |
38d47c1b | 955 | put_futex_key(fshared, &key2); |
42d35d48 | 956 | out_put_key1: |
38d47c1b | 957 | put_futex_key(fshared, &key1); |
42d35d48 | 958 | out: |
4732efbe JJ |
959 | return ret; |
960 | } | |
961 | ||
9121e478 DH |
962 | /** |
963 | * requeue_futex() - Requeue a futex_q from one hb to another | |
964 | * @q: the futex_q to requeue | |
965 | * @hb1: the source hash_bucket | |
966 | * @hb2: the target hash_bucket | |
967 | * @key2: the new key for the requeued futex_q | |
968 | */ | |
969 | static inline | |
970 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
971 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
972 | { | |
973 | ||
974 | /* | |
975 | * If key1 and key2 hash to the same bucket, no need to | |
976 | * requeue. | |
977 | */ | |
978 | if (likely(&hb1->chain != &hb2->chain)) { | |
979 | plist_del(&q->list, &hb1->chain); | |
980 | plist_add(&q->list, &hb2->chain); | |
981 | q->lock_ptr = &hb2->lock; | |
982 | #ifdef CONFIG_DEBUG_PI_LIST | |
983 | q->list.plist.lock = &hb2->lock; | |
984 | #endif | |
985 | } | |
986 | get_futex_key_refs(key2); | |
987 | q->key = *key2; | |
988 | } | |
989 | ||
52400ba9 DH |
990 | /** |
991 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
992 | * q: the futex_q | |
993 | * key: the key of the requeue target futex | |
994 | * | |
995 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
996 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
997 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
998 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
999 | * atomic lock acquisition. Must be called with the q->lock_ptr held. | |
1000 | */ | |
1001 | static inline | |
1002 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key) | |
1003 | { | |
1004 | drop_futex_key_refs(&q->key); | |
1005 | get_futex_key_refs(key); | |
1006 | q->key = *key; | |
1007 | ||
1008 | WARN_ON(plist_node_empty(&q->list)); | |
1009 | plist_del(&q->list, &q->list.plist); | |
1010 | ||
1011 | WARN_ON(!q->rt_waiter); | |
1012 | q->rt_waiter = NULL; | |
1013 | ||
f1a11e05 | 1014 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1015 | } |
1016 | ||
1017 | /** | |
1018 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1019 | * @pifutex: the user address of the to futex |
1020 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1021 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1022 | * @key1: the from futex key | |
1023 | * @key2: the to futex key | |
1024 | * @ps: address to store the pi_state pointer | |
1025 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1026 | * |
1027 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1028 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1029 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1030 | * hb1 and hb2 must be held by the caller. | |
52400ba9 DH |
1031 | * |
1032 | * Returns: | |
1033 | * 0 - failed to acquire the lock atomicly | |
1034 | * 1 - acquired the lock | |
1035 | * <0 - error | |
1036 | */ | |
1037 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1038 | struct futex_hash_bucket *hb1, | |
1039 | struct futex_hash_bucket *hb2, | |
1040 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1041 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1042 | { |
bab5bc9e | 1043 | struct futex_q *top_waiter = NULL; |
52400ba9 DH |
1044 | u32 curval; |
1045 | int ret; | |
1046 | ||
1047 | if (get_futex_value_locked(&curval, pifutex)) | |
1048 | return -EFAULT; | |
1049 | ||
bab5bc9e DH |
1050 | /* |
1051 | * Find the top_waiter and determine if there are additional waiters. | |
1052 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1053 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1054 | * as we have means to handle the possible fault. If not, don't set | |
1055 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1056 | * the kernel. | |
1057 | */ | |
52400ba9 DH |
1058 | top_waiter = futex_top_waiter(hb1, key1); |
1059 | ||
1060 | /* There are no waiters, nothing for us to do. */ | |
1061 | if (!top_waiter) | |
1062 | return 0; | |
1063 | ||
1064 | /* | |
bab5bc9e DH |
1065 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1066 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1067 | * in ps in contended cases. | |
52400ba9 | 1068 | */ |
bab5bc9e DH |
1069 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1070 | set_waiters); | |
52400ba9 DH |
1071 | if (ret == 1) |
1072 | requeue_pi_wake_futex(top_waiter, key2); | |
1073 | ||
1074 | return ret; | |
1075 | } | |
1076 | ||
1077 | /** | |
1078 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
1079 | * uaddr1: source futex user address | |
1080 | * uaddr2: target futex user address | |
1081 | * nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1082 | * nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1083 | * requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
1084 | * pi futex (pi to pi requeue is not supported) | |
1085 | * | |
1086 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1087 | * uaddr2 atomically on behalf of the top waiter. | |
1088 | * | |
1089 | * Returns: | |
1090 | * >=0 - on success, the number of tasks requeued or woken | |
1091 | * <0 - on error | |
1da177e4 | 1092 | */ |
c2f9f201 | 1093 | static int futex_requeue(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, |
52400ba9 DH |
1094 | int nr_wake, int nr_requeue, u32 *cmpval, |
1095 | int requeue_pi) | |
1da177e4 | 1096 | { |
38d47c1b | 1097 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1098 | int drop_count = 0, task_count = 0, ret; |
1099 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1100 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 1101 | struct plist_head *head1; |
1da177e4 | 1102 | struct futex_q *this, *next; |
52400ba9 DH |
1103 | u32 curval2; |
1104 | ||
1105 | if (requeue_pi) { | |
1106 | /* | |
1107 | * requeue_pi requires a pi_state, try to allocate it now | |
1108 | * without any locks in case it fails. | |
1109 | */ | |
1110 | if (refill_pi_state_cache()) | |
1111 | return -ENOMEM; | |
1112 | /* | |
1113 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1114 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1115 | * returning to userspace, so as to not leave the rt_mutex with | |
1116 | * waiters and no owner. However, second and third wake-ups | |
1117 | * cannot be predicted as they involve race conditions with the | |
1118 | * first wake and a fault while looking up the pi_state. Both | |
1119 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1120 | * use nr_wake=1. | |
1121 | */ | |
1122 | if (nr_wake != 1) | |
1123 | return -EINVAL; | |
1124 | } | |
1da177e4 | 1125 | |
42d35d48 | 1126 | retry: |
52400ba9 DH |
1127 | if (pi_state != NULL) { |
1128 | /* | |
1129 | * We will have to lookup the pi_state again, so free this one | |
1130 | * to keep the accounting correct. | |
1131 | */ | |
1132 | free_pi_state(pi_state); | |
1133 | pi_state = NULL; | |
1134 | } | |
1135 | ||
64d1304a | 1136 | ret = get_futex_key(uaddr1, fshared, &key1, VERIFY_READ); |
1da177e4 LT |
1137 | if (unlikely(ret != 0)) |
1138 | goto out; | |
521c1808 TG |
1139 | ret = get_futex_key(uaddr2, fshared, &key2, |
1140 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1141 | if (unlikely(ret != 0)) |
42d35d48 | 1142 | goto out_put_key1; |
1da177e4 | 1143 | |
e2970f2f IM |
1144 | hb1 = hash_futex(&key1); |
1145 | hb2 = hash_futex(&key2); | |
1da177e4 | 1146 | |
e4dc5b7a | 1147 | retry_private: |
8b8f319f | 1148 | double_lock_hb(hb1, hb2); |
1da177e4 | 1149 | |
e2970f2f IM |
1150 | if (likely(cmpval != NULL)) { |
1151 | u32 curval; | |
1da177e4 | 1152 | |
e2970f2f | 1153 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1154 | |
1155 | if (unlikely(ret)) { | |
5eb3dc62 | 1156 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1157 | |
e2970f2f | 1158 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1159 | if (ret) |
1160 | goto out_put_keys; | |
1da177e4 | 1161 | |
e4dc5b7a DH |
1162 | if (!fshared) |
1163 | goto retry_private; | |
1da177e4 | 1164 | |
e4dc5b7a DH |
1165 | put_futex_key(fshared, &key2); |
1166 | put_futex_key(fshared, &key1); | |
1167 | goto retry; | |
1da177e4 | 1168 | } |
e2970f2f | 1169 | if (curval != *cmpval) { |
1da177e4 LT |
1170 | ret = -EAGAIN; |
1171 | goto out_unlock; | |
1172 | } | |
1173 | } | |
1174 | ||
52400ba9 | 1175 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1176 | /* |
1177 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1178 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1179 | * bit. We force this here where we are able to easily handle | |
1180 | * faults rather in the requeue loop below. | |
1181 | */ | |
52400ba9 | 1182 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1183 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1184 | |
1185 | /* | |
1186 | * At this point the top_waiter has either taken uaddr2 or is | |
1187 | * waiting on it. If the former, then the pi_state will not | |
1188 | * exist yet, look it up one more time to ensure we have a | |
1189 | * reference to it. | |
1190 | */ | |
1191 | if (ret == 1) { | |
1192 | WARN_ON(pi_state); | |
1193 | task_count++; | |
1194 | ret = get_futex_value_locked(&curval2, uaddr2); | |
1195 | if (!ret) | |
1196 | ret = lookup_pi_state(curval2, hb2, &key2, | |
1197 | &pi_state); | |
1198 | } | |
1199 | ||
1200 | switch (ret) { | |
1201 | case 0: | |
1202 | break; | |
1203 | case -EFAULT: | |
1204 | double_unlock_hb(hb1, hb2); | |
1205 | put_futex_key(fshared, &key2); | |
1206 | put_futex_key(fshared, &key1); | |
1207 | ret = get_user(curval2, uaddr2); | |
1208 | if (!ret) | |
1209 | goto retry; | |
1210 | goto out; | |
1211 | case -EAGAIN: | |
1212 | /* The owner was exiting, try again. */ | |
1213 | double_unlock_hb(hb1, hb2); | |
1214 | put_futex_key(fshared, &key2); | |
1215 | put_futex_key(fshared, &key1); | |
1216 | cond_resched(); | |
1217 | goto retry; | |
1218 | default: | |
1219 | goto out_unlock; | |
1220 | } | |
1221 | } | |
1222 | ||
e2970f2f | 1223 | head1 = &hb1->chain; |
ec92d082 | 1224 | plist_for_each_entry_safe(this, next, head1, list) { |
52400ba9 DH |
1225 | if (task_count - nr_wake >= nr_requeue) |
1226 | break; | |
1227 | ||
1228 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1229 | continue; |
52400ba9 DH |
1230 | |
1231 | WARN_ON(!requeue_pi && this->rt_waiter); | |
1232 | WARN_ON(requeue_pi && !this->rt_waiter); | |
1233 | ||
1234 | /* | |
1235 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1236 | * lock, we already woke the top_waiter. If not, it will be | |
1237 | * woken by futex_unlock_pi(). | |
1238 | */ | |
1239 | if (++task_count <= nr_wake && !requeue_pi) { | |
1da177e4 | 1240 | wake_futex(this); |
52400ba9 DH |
1241 | continue; |
1242 | } | |
1da177e4 | 1243 | |
52400ba9 DH |
1244 | /* |
1245 | * Requeue nr_requeue waiters and possibly one more in the case | |
1246 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1247 | */ | |
1248 | if (requeue_pi) { | |
1249 | /* Prepare the waiter to take the rt_mutex. */ | |
1250 | atomic_inc(&pi_state->refcount); | |
1251 | this->pi_state = pi_state; | |
1252 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1253 | this->rt_waiter, | |
1254 | this->task, 1); | |
1255 | if (ret == 1) { | |
1256 | /* We got the lock. */ | |
1257 | requeue_pi_wake_futex(this, &key2); | |
1258 | continue; | |
1259 | } else if (ret) { | |
1260 | /* -EDEADLK */ | |
1261 | this->pi_state = NULL; | |
1262 | free_pi_state(pi_state); | |
1263 | goto out_unlock; | |
1264 | } | |
1da177e4 | 1265 | } |
52400ba9 DH |
1266 | requeue_futex(this, hb1, hb2, &key2); |
1267 | drop_count++; | |
1da177e4 LT |
1268 | } |
1269 | ||
1270 | out_unlock: | |
5eb3dc62 | 1271 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1272 | |
cd84a42f DH |
1273 | /* |
1274 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1275 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1276 | * one at key2 and updated their key pointer. We no longer need to | |
1277 | * hold the references to key1. | |
1278 | */ | |
1da177e4 | 1279 | while (--drop_count >= 0) |
9adef58b | 1280 | drop_futex_key_refs(&key1); |
1da177e4 | 1281 | |
42d35d48 | 1282 | out_put_keys: |
38d47c1b | 1283 | put_futex_key(fshared, &key2); |
42d35d48 | 1284 | out_put_key1: |
38d47c1b | 1285 | put_futex_key(fshared, &key1); |
42d35d48 | 1286 | out: |
52400ba9 DH |
1287 | if (pi_state != NULL) |
1288 | free_pi_state(pi_state); | |
1289 | return ret ? ret : task_count; | |
1da177e4 LT |
1290 | } |
1291 | ||
1292 | /* The key must be already stored in q->key. */ | |
82af7aca | 1293 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
1da177e4 | 1294 | { |
e2970f2f | 1295 | struct futex_hash_bucket *hb; |
1da177e4 | 1296 | |
9adef58b | 1297 | get_futex_key_refs(&q->key); |
e2970f2f IM |
1298 | hb = hash_futex(&q->key); |
1299 | q->lock_ptr = &hb->lock; | |
1da177e4 | 1300 | |
e2970f2f IM |
1301 | spin_lock(&hb->lock); |
1302 | return hb; | |
1da177e4 LT |
1303 | } |
1304 | ||
82af7aca | 1305 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 1306 | { |
ec92d082 PP |
1307 | int prio; |
1308 | ||
1309 | /* | |
1310 | * The priority used to register this element is | |
1311 | * - either the real thread-priority for the real-time threads | |
1312 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
1313 | * - or MAX_RT_PRIO for non-RT threads. | |
1314 | * Thus, all RT-threads are woken first in priority order, and | |
1315 | * the others are woken last, in FIFO order. | |
1316 | */ | |
1317 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1318 | ||
1319 | plist_node_init(&q->list, prio); | |
1320 | #ifdef CONFIG_DEBUG_PI_LIST | |
1321 | q->list.plist.lock = &hb->lock; | |
1322 | #endif | |
1323 | plist_add(&q->list, &hb->chain); | |
c87e2837 | 1324 | q->task = current; |
e2970f2f | 1325 | spin_unlock(&hb->lock); |
1da177e4 LT |
1326 | } |
1327 | ||
1328 | static inline void | |
e2970f2f | 1329 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 1330 | { |
e2970f2f | 1331 | spin_unlock(&hb->lock); |
9adef58b | 1332 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1333 | } |
1334 | ||
1335 | /* | |
1336 | * queue_me and unqueue_me must be called as a pair, each | |
1337 | * exactly once. They are called with the hashed spinlock held. | |
1338 | */ | |
1339 | ||
1da177e4 LT |
1340 | /* Return 1 if we were still queued (ie. 0 means we were woken) */ |
1341 | static int unqueue_me(struct futex_q *q) | |
1342 | { | |
1da177e4 | 1343 | spinlock_t *lock_ptr; |
e2970f2f | 1344 | int ret = 0; |
1da177e4 LT |
1345 | |
1346 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 1347 | retry: |
1da177e4 | 1348 | lock_ptr = q->lock_ptr; |
e91467ec | 1349 | barrier(); |
c80544dc | 1350 | if (lock_ptr != NULL) { |
1da177e4 LT |
1351 | spin_lock(lock_ptr); |
1352 | /* | |
1353 | * q->lock_ptr can change between reading it and | |
1354 | * spin_lock(), causing us to take the wrong lock. This | |
1355 | * corrects the race condition. | |
1356 | * | |
1357 | * Reasoning goes like this: if we have the wrong lock, | |
1358 | * q->lock_ptr must have changed (maybe several times) | |
1359 | * between reading it and the spin_lock(). It can | |
1360 | * change again after the spin_lock() but only if it was | |
1361 | * already changed before the spin_lock(). It cannot, | |
1362 | * however, change back to the original value. Therefore | |
1363 | * we can detect whether we acquired the correct lock. | |
1364 | */ | |
1365 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
1366 | spin_unlock(lock_ptr); | |
1367 | goto retry; | |
1368 | } | |
ec92d082 PP |
1369 | WARN_ON(plist_node_empty(&q->list)); |
1370 | plist_del(&q->list, &q->list.plist); | |
c87e2837 IM |
1371 | |
1372 | BUG_ON(q->pi_state); | |
1373 | ||
1da177e4 LT |
1374 | spin_unlock(lock_ptr); |
1375 | ret = 1; | |
1376 | } | |
1377 | ||
9adef58b | 1378 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1379 | return ret; |
1380 | } | |
1381 | ||
c87e2837 IM |
1382 | /* |
1383 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
1384 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
1385 | * and dropped here. | |
c87e2837 | 1386 | */ |
d0aa7a70 | 1387 | static void unqueue_me_pi(struct futex_q *q) |
c87e2837 | 1388 | { |
ec92d082 PP |
1389 | WARN_ON(plist_node_empty(&q->list)); |
1390 | plist_del(&q->list, &q->list.plist); | |
c87e2837 IM |
1391 | |
1392 | BUG_ON(!q->pi_state); | |
1393 | free_pi_state(q->pi_state); | |
1394 | q->pi_state = NULL; | |
1395 | ||
d0aa7a70 | 1396 | spin_unlock(q->lock_ptr); |
c87e2837 | 1397 | |
9adef58b | 1398 | drop_futex_key_refs(&q->key); |
c87e2837 IM |
1399 | } |
1400 | ||
d0aa7a70 | 1401 | /* |
cdf71a10 | 1402 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 1403 | * |
778e9a9c AK |
1404 | * Must be called with hash bucket lock held and mm->sem held for non |
1405 | * private futexes. | |
d0aa7a70 | 1406 | */ |
778e9a9c | 1407 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
c2f9f201 | 1408 | struct task_struct *newowner, int fshared) |
d0aa7a70 | 1409 | { |
cdf71a10 | 1410 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 1411 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 1412 | struct task_struct *oldowner = pi_state->owner; |
d0aa7a70 | 1413 | u32 uval, curval, newval; |
e4dc5b7a | 1414 | int ret; |
d0aa7a70 PP |
1415 | |
1416 | /* Owner died? */ | |
1b7558e4 TG |
1417 | if (!pi_state->owner) |
1418 | newtid |= FUTEX_OWNER_DIED; | |
1419 | ||
1420 | /* | |
1421 | * We are here either because we stole the rtmutex from the | |
1422 | * pending owner or we are the pending owner which failed to | |
1423 | * get the rtmutex. We have to replace the pending owner TID | |
1424 | * in the user space variable. This must be atomic as we have | |
1425 | * to preserve the owner died bit here. | |
1426 | * | |
b2d0994b DH |
1427 | * Note: We write the user space value _before_ changing the pi_state |
1428 | * because we can fault here. Imagine swapped out pages or a fork | |
1429 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
1430 | * |
1431 | * Modifying pi_state _before_ the user space value would | |
1432 | * leave the pi_state in an inconsistent state when we fault | |
1433 | * here, because we need to drop the hash bucket lock to | |
1434 | * handle the fault. This might be observed in the PID check | |
1435 | * in lookup_pi_state. | |
1436 | */ | |
1437 | retry: | |
1438 | if (get_futex_value_locked(&uval, uaddr)) | |
1439 | goto handle_fault; | |
1440 | ||
1441 | while (1) { | |
1442 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
1443 | ||
1444 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); | |
1445 | ||
1446 | if (curval == -EFAULT) | |
1447 | goto handle_fault; | |
1448 | if (curval == uval) | |
1449 | break; | |
1450 | uval = curval; | |
1451 | } | |
1452 | ||
1453 | /* | |
1454 | * We fixed up user space. Now we need to fix the pi_state | |
1455 | * itself. | |
1456 | */ | |
d0aa7a70 PP |
1457 | if (pi_state->owner != NULL) { |
1458 | spin_lock_irq(&pi_state->owner->pi_lock); | |
1459 | WARN_ON(list_empty(&pi_state->list)); | |
1460 | list_del_init(&pi_state->list); | |
1461 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
1b7558e4 | 1462 | } |
d0aa7a70 | 1463 | |
cdf71a10 | 1464 | pi_state->owner = newowner; |
d0aa7a70 | 1465 | |
cdf71a10 | 1466 | spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 1467 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 TG |
1468 | list_add(&pi_state->list, &newowner->pi_state_list); |
1469 | spin_unlock_irq(&newowner->pi_lock); | |
1b7558e4 | 1470 | return 0; |
d0aa7a70 | 1471 | |
d0aa7a70 | 1472 | /* |
1b7558e4 TG |
1473 | * To handle the page fault we need to drop the hash bucket |
1474 | * lock here. That gives the other task (either the pending | |
1475 | * owner itself or the task which stole the rtmutex) the | |
1476 | * chance to try the fixup of the pi_state. So once we are | |
1477 | * back from handling the fault we need to check the pi_state | |
1478 | * after reacquiring the hash bucket lock and before trying to | |
1479 | * do another fixup. When the fixup has been done already we | |
1480 | * simply return. | |
d0aa7a70 | 1481 | */ |
1b7558e4 TG |
1482 | handle_fault: |
1483 | spin_unlock(q->lock_ptr); | |
778e9a9c | 1484 | |
e4dc5b7a | 1485 | ret = get_user(uval, uaddr); |
778e9a9c | 1486 | |
1b7558e4 | 1487 | spin_lock(q->lock_ptr); |
778e9a9c | 1488 | |
1b7558e4 TG |
1489 | /* |
1490 | * Check if someone else fixed it for us: | |
1491 | */ | |
1492 | if (pi_state->owner != oldowner) | |
1493 | return 0; | |
1494 | ||
1495 | if (ret) | |
1496 | return ret; | |
1497 | ||
1498 | goto retry; | |
d0aa7a70 PP |
1499 | } |
1500 | ||
34f01cc1 ED |
1501 | /* |
1502 | * In case we must use restart_block to restart a futex_wait, | |
ce6bd420 | 1503 | * we encode in the 'flags' shared capability |
34f01cc1 | 1504 | */ |
1acdac10 TG |
1505 | #define FLAGS_SHARED 0x01 |
1506 | #define FLAGS_CLOCKRT 0x02 | |
a72188d8 | 1507 | #define FLAGS_HAS_TIMEOUT 0x04 |
34f01cc1 | 1508 | |
72c1bbf3 | 1509 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 1510 | |
dd973998 DH |
1511 | /** |
1512 | * fixup_owner() - Post lock pi_state and corner case management | |
1513 | * @uaddr: user address of the futex | |
1514 | * @fshared: whether the futex is shared (1) or not (0) | |
1515 | * @q: futex_q (contains pi_state and access to the rt_mutex) | |
1516 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
1517 | * | |
1518 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
1519 | * the pi_state owner as well as handle race conditions that may allow us to | |
1520 | * acquire the lock. Must be called with the hb lock held. | |
1521 | * | |
1522 | * Returns: | |
1523 | * 1 - success, lock taken | |
1524 | * 0 - success, lock not taken | |
1525 | * <0 - on error (-EFAULT) | |
1526 | */ | |
1527 | static int fixup_owner(u32 __user *uaddr, int fshared, struct futex_q *q, | |
1528 | int locked) | |
1529 | { | |
1530 | struct task_struct *owner; | |
1531 | int ret = 0; | |
1532 | ||
1533 | if (locked) { | |
1534 | /* | |
1535 | * Got the lock. We might not be the anticipated owner if we | |
1536 | * did a lock-steal - fix up the PI-state in that case: | |
1537 | */ | |
1538 | if (q->pi_state->owner != current) | |
1539 | ret = fixup_pi_state_owner(uaddr, q, current, fshared); | |
1540 | goto out; | |
1541 | } | |
1542 | ||
1543 | /* | |
1544 | * Catch the rare case, where the lock was released when we were on the | |
1545 | * way back before we locked the hash bucket. | |
1546 | */ | |
1547 | if (q->pi_state->owner == current) { | |
1548 | /* | |
1549 | * Try to get the rt_mutex now. This might fail as some other | |
1550 | * task acquired the rt_mutex after we removed ourself from the | |
1551 | * rt_mutex waiters list. | |
1552 | */ | |
1553 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
1554 | locked = 1; | |
1555 | goto out; | |
1556 | } | |
1557 | ||
1558 | /* | |
1559 | * pi_state is incorrect, some other task did a lock steal and | |
1560 | * we returned due to timeout or signal without taking the | |
1561 | * rt_mutex. Too late. We can access the rt_mutex_owner without | |
1562 | * locking, as the other task is now blocked on the hash bucket | |
1563 | * lock. Fix the state up. | |
1564 | */ | |
1565 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); | |
1566 | ret = fixup_pi_state_owner(uaddr, q, owner, fshared); | |
1567 | goto out; | |
1568 | } | |
1569 | ||
1570 | /* | |
1571 | * Paranoia check. If we did not take the lock, then we should not be | |
1572 | * the owner, nor the pending owner, of the rt_mutex. | |
1573 | */ | |
1574 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
1575 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
1576 | "pi-state %p\n", ret, | |
1577 | q->pi_state->pi_mutex.owner, | |
1578 | q->pi_state->owner); | |
1579 | ||
1580 | out: | |
1581 | return ret ? ret : locked; | |
1582 | } | |
1583 | ||
ca5f9524 DH |
1584 | /** |
1585 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
1586 | * @hb: the futex hash bucket, must be locked by the caller | |
1587 | * @q: the futex_q to queue up on | |
1588 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
1589 | */ |
1590 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 1591 | struct hrtimer_sleeper *timeout) |
ca5f9524 DH |
1592 | { |
1593 | queue_me(q, hb); | |
1594 | ||
1595 | /* | |
1596 | * There might have been scheduling since the queue_me(), as we | |
1597 | * cannot hold a spinlock across the get_user() in case it | |
1598 | * faults, and we cannot just set TASK_INTERRUPTIBLE state when | |
f1a11e05 | 1599 | * queueing ourselves into the futex hash. This code thus has to |
ca5f9524 DH |
1600 | * rely on the futex_wake() code removing us from hash when it |
1601 | * wakes us up. | |
1602 | */ | |
f1a11e05 | 1603 | set_current_state(TASK_INTERRUPTIBLE); |
ca5f9524 DH |
1604 | |
1605 | /* Arm the timer */ | |
1606 | if (timeout) { | |
1607 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); | |
1608 | if (!hrtimer_active(&timeout->timer)) | |
1609 | timeout->task = NULL; | |
1610 | } | |
1611 | ||
1612 | /* | |
1613 | * !plist_node_empty() is safe here without any lock. | |
1614 | * q.lock_ptr != 0 is not safe, because of ordering against wakeup. | |
1615 | */ | |
1616 | if (likely(!plist_node_empty(&q->list))) { | |
1617 | /* | |
1618 | * If the timer has already expired, current will already be | |
1619 | * flagged for rescheduling. Only call schedule if there | |
1620 | * is no timeout, or if it has yet to expire. | |
1621 | */ | |
1622 | if (!timeout || timeout->task) | |
1623 | schedule(); | |
1624 | } | |
1625 | __set_current_state(TASK_RUNNING); | |
1626 | } | |
1627 | ||
f801073f DH |
1628 | /** |
1629 | * futex_wait_setup() - Prepare to wait on a futex | |
1630 | * @uaddr: the futex userspace address | |
1631 | * @val: the expected value | |
1632 | * @fshared: whether the futex is shared (1) or not (0) | |
1633 | * @q: the associated futex_q | |
1634 | * @hb: storage for hash_bucket pointer to be returned to caller | |
1635 | * | |
1636 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
1637 | * compare it with the expected value. Handle atomic faults internally. | |
1638 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
1639 | * with no q.key reference on failure. | |
1640 | * | |
1641 | * Returns: | |
1642 | * 0 - uaddr contains val and hb has been locked | |
1643 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlcoked | |
1644 | */ | |
1645 | static int futex_wait_setup(u32 __user *uaddr, u32 val, int fshared, | |
1646 | struct futex_q *q, struct futex_hash_bucket **hb) | |
1da177e4 | 1647 | { |
e2970f2f IM |
1648 | u32 uval; |
1649 | int ret; | |
1da177e4 | 1650 | |
1da177e4 | 1651 | /* |
b2d0994b | 1652 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
1653 | * Order is important: |
1654 | * | |
1655 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
1656 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
1657 | * | |
1658 | * The basic logical guarantee of a futex is that it blocks ONLY | |
1659 | * if cond(var) is known to be true at the time of blocking, for | |
1660 | * any cond. If we queued after testing *uaddr, that would open | |
1661 | * a race condition where we could block indefinitely with | |
1662 | * cond(var) false, which would violate the guarantee. | |
1663 | * | |
1664 | * A consequence is that futex_wait() can return zero and absorb | |
1665 | * a wakeup when *uaddr != val on entry to the syscall. This is | |
1666 | * rare, but normal. | |
1da177e4 | 1667 | */ |
f801073f DH |
1668 | retry: |
1669 | q->key = FUTEX_KEY_INIT; | |
521c1808 | 1670 | ret = get_futex_key(uaddr, fshared, &q->key, VERIFY_READ); |
f801073f | 1671 | if (unlikely(ret != 0)) |
a5a2a0c7 | 1672 | return ret; |
f801073f DH |
1673 | |
1674 | retry_private: | |
1675 | *hb = queue_lock(q); | |
1676 | ||
e2970f2f | 1677 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 1678 | |
f801073f DH |
1679 | if (ret) { |
1680 | queue_unlock(q, *hb); | |
1da177e4 | 1681 | |
e2970f2f | 1682 | ret = get_user(uval, uaddr); |
e4dc5b7a | 1683 | if (ret) |
f801073f | 1684 | goto out; |
1da177e4 | 1685 | |
e4dc5b7a DH |
1686 | if (!fshared) |
1687 | goto retry_private; | |
1688 | ||
f801073f | 1689 | put_futex_key(fshared, &q->key); |
e4dc5b7a | 1690 | goto retry; |
1da177e4 | 1691 | } |
ca5f9524 | 1692 | |
f801073f DH |
1693 | if (uval != val) { |
1694 | queue_unlock(q, *hb); | |
1695 | ret = -EWOULDBLOCK; | |
2fff78c7 | 1696 | } |
1da177e4 | 1697 | |
f801073f DH |
1698 | out: |
1699 | if (ret) | |
1700 | put_futex_key(fshared, &q->key); | |
1701 | return ret; | |
1702 | } | |
1703 | ||
1704 | static int futex_wait(u32 __user *uaddr, int fshared, | |
1705 | u32 val, ktime_t *abs_time, u32 bitset, int clockrt) | |
1706 | { | |
1707 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
1708 | struct restart_block *restart; |
1709 | struct futex_hash_bucket *hb; | |
1710 | struct futex_q q; | |
1711 | int ret; | |
1712 | ||
1713 | if (!bitset) | |
1714 | return -EINVAL; | |
1715 | ||
1716 | q.pi_state = NULL; | |
1717 | q.bitset = bitset; | |
52400ba9 | 1718 | q.rt_waiter = NULL; |
f801073f DH |
1719 | |
1720 | if (abs_time) { | |
1721 | to = &timeout; | |
1722 | ||
1723 | hrtimer_init_on_stack(&to->timer, clockrt ? CLOCK_REALTIME : | |
1724 | CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1725 | hrtimer_init_sleeper(to, current); | |
1726 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
1727 | current->timer_slack_ns); | |
1728 | } | |
1729 | ||
1730 | /* Prepare to wait on uaddr. */ | |
1731 | ret = futex_wait_setup(uaddr, val, fshared, &q, &hb); | |
1732 | if (ret) | |
1733 | goto out; | |
1734 | ||
ca5f9524 | 1735 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 1736 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
1737 | |
1738 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 1739 | ret = 0; |
1da177e4 | 1740 | if (!unqueue_me(&q)) |
2fff78c7 PZ |
1741 | goto out_put_key; |
1742 | ret = -ETIMEDOUT; | |
ca5f9524 | 1743 | if (to && !to->task) |
2fff78c7 | 1744 | goto out_put_key; |
72c1bbf3 | 1745 | |
e2970f2f IM |
1746 | /* |
1747 | * We expect signal_pending(current), but another thread may | |
1748 | * have handled it for us already. | |
1749 | */ | |
2fff78c7 | 1750 | ret = -ERESTARTSYS; |
c19384b5 | 1751 | if (!abs_time) |
2fff78c7 | 1752 | goto out_put_key; |
1da177e4 | 1753 | |
2fff78c7 PZ |
1754 | restart = ¤t_thread_info()->restart_block; |
1755 | restart->fn = futex_wait_restart; | |
1756 | restart->futex.uaddr = (u32 *)uaddr; | |
1757 | restart->futex.val = val; | |
1758 | restart->futex.time = abs_time->tv64; | |
1759 | restart->futex.bitset = bitset; | |
a72188d8 | 1760 | restart->futex.flags = FLAGS_HAS_TIMEOUT; |
2fff78c7 PZ |
1761 | |
1762 | if (fshared) | |
1763 | restart->futex.flags |= FLAGS_SHARED; | |
1764 | if (clockrt) | |
1765 | restart->futex.flags |= FLAGS_CLOCKRT; | |
42d35d48 | 1766 | |
2fff78c7 PZ |
1767 | ret = -ERESTART_RESTARTBLOCK; |
1768 | ||
1769 | out_put_key: | |
1770 | put_futex_key(fshared, &q.key); | |
42d35d48 | 1771 | out: |
ca5f9524 DH |
1772 | if (to) { |
1773 | hrtimer_cancel(&to->timer); | |
1774 | destroy_hrtimer_on_stack(&to->timer); | |
1775 | } | |
c87e2837 IM |
1776 | return ret; |
1777 | } | |
1778 | ||
72c1bbf3 NP |
1779 | |
1780 | static long futex_wait_restart(struct restart_block *restart) | |
1781 | { | |
ce6bd420 | 1782 | u32 __user *uaddr = (u32 __user *)restart->futex.uaddr; |
c2f9f201 | 1783 | int fshared = 0; |
a72188d8 | 1784 | ktime_t t, *tp = NULL; |
72c1bbf3 | 1785 | |
a72188d8 DH |
1786 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
1787 | t.tv64 = restart->futex.time; | |
1788 | tp = &t; | |
1789 | } | |
72c1bbf3 | 1790 | restart->fn = do_no_restart_syscall; |
ce6bd420 | 1791 | if (restart->futex.flags & FLAGS_SHARED) |
c2f9f201 | 1792 | fshared = 1; |
a72188d8 | 1793 | return (long)futex_wait(uaddr, fshared, restart->futex.val, tp, |
1acdac10 TG |
1794 | restart->futex.bitset, |
1795 | restart->futex.flags & FLAGS_CLOCKRT); | |
72c1bbf3 NP |
1796 | } |
1797 | ||
1798 | ||
c87e2837 IM |
1799 | /* |
1800 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
1801 | * and failed. The kernel side here does the whole locking operation: | |
1802 | * if there are waiters then it will block, it does PI, etc. (Due to | |
1803 | * races the kernel might see a 0 value of the futex too.) | |
1804 | */ | |
c2f9f201 | 1805 | static int futex_lock_pi(u32 __user *uaddr, int fshared, |
34f01cc1 | 1806 | int detect, ktime_t *time, int trylock) |
c87e2837 | 1807 | { |
c5780e97 | 1808 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 1809 | struct futex_hash_bucket *hb; |
1a52084d | 1810 | u32 uval; |
c87e2837 | 1811 | struct futex_q q; |
dd973998 | 1812 | int res, ret; |
c87e2837 IM |
1813 | |
1814 | if (refill_pi_state_cache()) | |
1815 | return -ENOMEM; | |
1816 | ||
c19384b5 | 1817 | if (time) { |
c5780e97 | 1818 | to = &timeout; |
237fc6e7 TG |
1819 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
1820 | HRTIMER_MODE_ABS); | |
c5780e97 | 1821 | hrtimer_init_sleeper(to, current); |
cc584b21 | 1822 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
1823 | } |
1824 | ||
c87e2837 | 1825 | q.pi_state = NULL; |
52400ba9 | 1826 | q.rt_waiter = NULL; |
42d35d48 | 1827 | retry: |
38d47c1b | 1828 | q.key = FUTEX_KEY_INIT; |
64d1304a | 1829 | ret = get_futex_key(uaddr, fshared, &q.key, VERIFY_WRITE); |
c87e2837 | 1830 | if (unlikely(ret != 0)) |
42d35d48 | 1831 | goto out; |
c87e2837 | 1832 | |
e4dc5b7a | 1833 | retry_private: |
82af7aca | 1834 | hb = queue_lock(&q); |
c87e2837 | 1835 | |
bab5bc9e | 1836 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 1837 | if (unlikely(ret)) { |
778e9a9c | 1838 | switch (ret) { |
1a52084d DH |
1839 | case 1: |
1840 | /* We got the lock. */ | |
1841 | ret = 0; | |
1842 | goto out_unlock_put_key; | |
1843 | case -EFAULT: | |
1844 | goto uaddr_faulted; | |
778e9a9c AK |
1845 | case -EAGAIN: |
1846 | /* | |
1847 | * Task is exiting and we just wait for the | |
1848 | * exit to complete. | |
1849 | */ | |
1850 | queue_unlock(&q, hb); | |
de87fcc1 | 1851 | put_futex_key(fshared, &q.key); |
778e9a9c AK |
1852 | cond_resched(); |
1853 | goto retry; | |
778e9a9c | 1854 | default: |
42d35d48 | 1855 | goto out_unlock_put_key; |
c87e2837 | 1856 | } |
c87e2837 IM |
1857 | } |
1858 | ||
1859 | /* | |
1860 | * Only actually queue now that the atomic ops are done: | |
1861 | */ | |
82af7aca | 1862 | queue_me(&q, hb); |
c87e2837 | 1863 | |
c87e2837 IM |
1864 | WARN_ON(!q.pi_state); |
1865 | /* | |
1866 | * Block on the PI mutex: | |
1867 | */ | |
1868 | if (!trylock) | |
1869 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); | |
1870 | else { | |
1871 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); | |
1872 | /* Fixup the trylock return value: */ | |
1873 | ret = ret ? 0 : -EWOULDBLOCK; | |
1874 | } | |
1875 | ||
a99e4e41 | 1876 | spin_lock(q.lock_ptr); |
dd973998 DH |
1877 | /* |
1878 | * Fixup the pi_state owner and possibly acquire the lock if we | |
1879 | * haven't already. | |
1880 | */ | |
1881 | res = fixup_owner(uaddr, fshared, &q, !ret); | |
1882 | /* | |
1883 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
1884 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
1885 | */ | |
1886 | if (res) | |
1887 | ret = (res < 0) ? res : 0; | |
c87e2837 | 1888 | |
e8f6386c | 1889 | /* |
dd973998 DH |
1890 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
1891 | * it and return the fault to userspace. | |
e8f6386c DH |
1892 | */ |
1893 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
1894 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
1895 | ||
778e9a9c AK |
1896 | /* Unqueue and drop the lock */ |
1897 | unqueue_me_pi(&q); | |
c87e2837 | 1898 | |
dd973998 | 1899 | goto out; |
c87e2837 | 1900 | |
42d35d48 | 1901 | out_unlock_put_key: |
c87e2837 IM |
1902 | queue_unlock(&q, hb); |
1903 | ||
42d35d48 | 1904 | out_put_key: |
38d47c1b | 1905 | put_futex_key(fshared, &q.key); |
42d35d48 | 1906 | out: |
237fc6e7 TG |
1907 | if (to) |
1908 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 1909 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 1910 | |
42d35d48 | 1911 | uaddr_faulted: |
c87e2837 | 1912 | /* |
b5686363 DH |
1913 | * We have to r/w *(int __user *)uaddr, and we have to modify it |
1914 | * atomically. Therefore, if we continue to fault after get_user() | |
1915 | * below, we need to handle the fault ourselves, while still holding | |
1916 | * the mmap_sem. This can occur if the uaddr is under contention as | |
1917 | * we have to drop the mmap_sem in order to call get_user(). | |
c87e2837 | 1918 | */ |
778e9a9c AK |
1919 | queue_unlock(&q, hb); |
1920 | ||
c87e2837 | 1921 | ret = get_user(uval, uaddr); |
e4dc5b7a DH |
1922 | if (ret) |
1923 | goto out_put_key; | |
c87e2837 | 1924 | |
e4dc5b7a DH |
1925 | if (!fshared) |
1926 | goto retry_private; | |
1927 | ||
1928 | put_futex_key(fshared, &q.key); | |
1929 | goto retry; | |
c87e2837 IM |
1930 | } |
1931 | ||
c87e2837 IM |
1932 | /* |
1933 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
1934 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
1935 | * and do the rt-mutex unlock. | |
1936 | */ | |
c2f9f201 | 1937 | static int futex_unlock_pi(u32 __user *uaddr, int fshared) |
c87e2837 IM |
1938 | { |
1939 | struct futex_hash_bucket *hb; | |
1940 | struct futex_q *this, *next; | |
1941 | u32 uval; | |
ec92d082 | 1942 | struct plist_head *head; |
38d47c1b | 1943 | union futex_key key = FUTEX_KEY_INIT; |
e4dc5b7a | 1944 | int ret; |
c87e2837 IM |
1945 | |
1946 | retry: | |
1947 | if (get_user(uval, uaddr)) | |
1948 | return -EFAULT; | |
1949 | /* | |
1950 | * We release only a lock we actually own: | |
1951 | */ | |
b488893a | 1952 | if ((uval & FUTEX_TID_MASK) != task_pid_vnr(current)) |
c87e2837 | 1953 | return -EPERM; |
c87e2837 | 1954 | |
64d1304a | 1955 | ret = get_futex_key(uaddr, fshared, &key, VERIFY_WRITE); |
c87e2837 IM |
1956 | if (unlikely(ret != 0)) |
1957 | goto out; | |
1958 | ||
1959 | hb = hash_futex(&key); | |
1960 | spin_lock(&hb->lock); | |
1961 | ||
c87e2837 IM |
1962 | /* |
1963 | * To avoid races, try to do the TID -> 0 atomic transition | |
1964 | * again. If it succeeds then we can return without waking | |
1965 | * anyone else up: | |
1966 | */ | |
36cf3b5c | 1967 | if (!(uval & FUTEX_OWNER_DIED)) |
b488893a | 1968 | uval = cmpxchg_futex_value_locked(uaddr, task_pid_vnr(current), 0); |
36cf3b5c | 1969 | |
c87e2837 IM |
1970 | |
1971 | if (unlikely(uval == -EFAULT)) | |
1972 | goto pi_faulted; | |
1973 | /* | |
1974 | * Rare case: we managed to release the lock atomically, | |
1975 | * no need to wake anyone else up: | |
1976 | */ | |
b488893a | 1977 | if (unlikely(uval == task_pid_vnr(current))) |
c87e2837 IM |
1978 | goto out_unlock; |
1979 | ||
1980 | /* | |
1981 | * Ok, other tasks may need to be woken up - check waiters | |
1982 | * and do the wakeup if necessary: | |
1983 | */ | |
1984 | head = &hb->chain; | |
1985 | ||
ec92d082 | 1986 | plist_for_each_entry_safe(this, next, head, list) { |
c87e2837 IM |
1987 | if (!match_futex (&this->key, &key)) |
1988 | continue; | |
1989 | ret = wake_futex_pi(uaddr, uval, this); | |
1990 | /* | |
1991 | * The atomic access to the futex value | |
1992 | * generated a pagefault, so retry the | |
1993 | * user-access and the wakeup: | |
1994 | */ | |
1995 | if (ret == -EFAULT) | |
1996 | goto pi_faulted; | |
1997 | goto out_unlock; | |
1998 | } | |
1999 | /* | |
2000 | * No waiters - kernel unlocks the futex: | |
2001 | */ | |
e3f2ddea IM |
2002 | if (!(uval & FUTEX_OWNER_DIED)) { |
2003 | ret = unlock_futex_pi(uaddr, uval); | |
2004 | if (ret == -EFAULT) | |
2005 | goto pi_faulted; | |
2006 | } | |
c87e2837 IM |
2007 | |
2008 | out_unlock: | |
2009 | spin_unlock(&hb->lock); | |
38d47c1b | 2010 | put_futex_key(fshared, &key); |
c87e2837 | 2011 | |
42d35d48 | 2012 | out: |
c87e2837 IM |
2013 | return ret; |
2014 | ||
2015 | pi_faulted: | |
2016 | /* | |
b5686363 DH |
2017 | * We have to r/w *(int __user *)uaddr, and we have to modify it |
2018 | * atomically. Therefore, if we continue to fault after get_user() | |
2019 | * below, we need to handle the fault ourselves, while still holding | |
2020 | * the mmap_sem. This can occur if the uaddr is under contention as | |
2021 | * we have to drop the mmap_sem in order to call get_user(). | |
c87e2837 | 2022 | */ |
778e9a9c | 2023 | spin_unlock(&hb->lock); |
e4dc5b7a | 2024 | put_futex_key(fshared, &key); |
c87e2837 | 2025 | |
c87e2837 | 2026 | ret = get_user(uval, uaddr); |
b5686363 | 2027 | if (!ret) |
c87e2837 IM |
2028 | goto retry; |
2029 | ||
1da177e4 LT |
2030 | return ret; |
2031 | } | |
2032 | ||
52400ba9 DH |
2033 | /** |
2034 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2035 | * @hb: the hash_bucket futex_q was original enqueued on | |
2036 | * @q: the futex_q woken while waiting to be requeued | |
2037 | * @key2: the futex_key of the requeue target futex | |
2038 | * @timeout: the timeout associated with the wait (NULL if none) | |
2039 | * | |
2040 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2041 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2042 | * the wakeup and return the appropriate error code to the caller. Must be | |
2043 | * called with the hb lock held. | |
2044 | * | |
2045 | * Returns | |
2046 | * 0 - no early wakeup detected | |
1c840c14 | 2047 | * <0 - -ETIMEDOUT or -ERESTARTNOINTR |
52400ba9 DH |
2048 | */ |
2049 | static inline | |
2050 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2051 | struct futex_q *q, union futex_key *key2, | |
2052 | struct hrtimer_sleeper *timeout) | |
2053 | { | |
2054 | int ret = 0; | |
2055 | ||
2056 | /* | |
2057 | * With the hb lock held, we avoid races while we process the wakeup. | |
2058 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2059 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2060 | * It can't be requeued from uaddr2 to something else since we don't | |
2061 | * support a PI aware source futex for requeue. | |
2062 | */ | |
2063 | if (!match_futex(&q->key, key2)) { | |
2064 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2065 | /* | |
2066 | * We were woken prior to requeue by a timeout or a signal. | |
2067 | * Unqueue the futex_q and determine which it was. | |
2068 | */ | |
2069 | plist_del(&q->list, &q->list.plist); | |
2070 | drop_futex_key_refs(&q->key); | |
2071 | ||
2072 | if (timeout && !timeout->task) | |
2073 | ret = -ETIMEDOUT; | |
1c840c14 TG |
2074 | else |
2075 | ret = -ERESTARTNOINTR; | |
52400ba9 DH |
2076 | } |
2077 | return ret; | |
2078 | } | |
2079 | ||
2080 | /** | |
2081 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
2082 | * @uaddr: the futex we initialyl wait on (non-pi) | |
2083 | * @fshared: whether the futexes are shared (1) or not (0). They must be | |
2084 | * the same type, no requeueing from private to shared, etc. | |
2085 | * @val: the expected value of uaddr | |
2086 | * @abs_time: absolute timeout | |
2087 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all. | |
2088 | * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0) | |
2089 | * @uaddr2: the pi futex we will take prior to returning to user-space | |
2090 | * | |
2091 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
2092 | * uaddr2 which must be PI aware. Normal wakeup will wake on uaddr2 and | |
2093 | * complete the acquisition of the rt_mutex prior to returning to userspace. | |
2094 | * This ensures the rt_mutex maintains an owner when it has waiters; without | |
2095 | * one, the pi logic wouldn't know which task to boost/deboost, if there was a | |
2096 | * need to. | |
2097 | * | |
2098 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
2099 | * via the following: | |
2100 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() | |
2101 | * 2) wakeup on uaddr2 after a requeue and subsequent unlock | |
2102 | * 3) signal (before or after requeue) | |
2103 | * 4) timeout (before or after requeue) | |
2104 | * | |
2105 | * If 3, we setup a restart_block with futex_wait_requeue_pi() as the function. | |
2106 | * | |
2107 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2108 | * 5) successful lock | |
2109 | * 6) signal | |
2110 | * 7) timeout | |
2111 | * 8) other lock acquisition failure | |
2112 | * | |
2113 | * If 6, we setup a restart_block with futex_lock_pi() as the function. | |
2114 | * | |
2115 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2116 | * | |
2117 | * Returns: | |
2118 | * 0 - On success | |
2119 | * <0 - On error | |
2120 | */ | |
2121 | static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared, | |
2122 | u32 val, ktime_t *abs_time, u32 bitset, | |
2123 | int clockrt, u32 __user *uaddr2) | |
2124 | { | |
2125 | struct hrtimer_sleeper timeout, *to = NULL; | |
2126 | struct rt_mutex_waiter rt_waiter; | |
2127 | struct rt_mutex *pi_mutex = NULL; | |
52400ba9 DH |
2128 | struct futex_hash_bucket *hb; |
2129 | union futex_key key2; | |
2130 | struct futex_q q; | |
2131 | int res, ret; | |
52400ba9 DH |
2132 | |
2133 | if (!bitset) | |
2134 | return -EINVAL; | |
2135 | ||
2136 | if (abs_time) { | |
2137 | to = &timeout; | |
2138 | hrtimer_init_on_stack(&to->timer, clockrt ? CLOCK_REALTIME : | |
2139 | CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
2140 | hrtimer_init_sleeper(to, current); | |
2141 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2142 | current->timer_slack_ns); | |
2143 | } | |
2144 | ||
2145 | /* | |
2146 | * The waiter is allocated on our stack, manipulated by the requeue | |
2147 | * code while we sleep on uaddr. | |
2148 | */ | |
2149 | debug_rt_mutex_init_waiter(&rt_waiter); | |
2150 | rt_waiter.task = NULL; | |
2151 | ||
2152 | q.pi_state = NULL; | |
2153 | q.bitset = bitset; | |
2154 | q.rt_waiter = &rt_waiter; | |
2155 | ||
2156 | key2 = FUTEX_KEY_INIT; | |
521c1808 | 2157 | ret = get_futex_key(uaddr2, fshared, &key2, VERIFY_WRITE); |
52400ba9 DH |
2158 | if (unlikely(ret != 0)) |
2159 | goto out; | |
2160 | ||
2161 | /* Prepare to wait on uaddr. */ | |
2162 | ret = futex_wait_setup(uaddr, val, fshared, &q, &hb); | |
c8b15a70 TG |
2163 | if (ret) |
2164 | goto out_key2; | |
52400ba9 DH |
2165 | |
2166 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ | |
f1a11e05 | 2167 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2168 | |
2169 | spin_lock(&hb->lock); | |
2170 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2171 | spin_unlock(&hb->lock); | |
2172 | if (ret) | |
2173 | goto out_put_keys; | |
2174 | ||
2175 | /* | |
2176 | * In order for us to be here, we know our q.key == key2, and since | |
2177 | * we took the hb->lock above, we also know that futex_requeue() has | |
2178 | * completed and we no longer have to concern ourselves with a wakeup | |
2179 | * race with the atomic proxy lock acquition by the requeue code. | |
2180 | */ | |
2181 | ||
2182 | /* Check if the requeue code acquired the second futex for us. */ | |
2183 | if (!q.rt_waiter) { | |
2184 | /* | |
2185 | * Got the lock. We might not be the anticipated owner if we | |
2186 | * did a lock-steal - fix up the PI-state in that case. | |
2187 | */ | |
2188 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2189 | spin_lock(q.lock_ptr); | |
2190 | ret = fixup_pi_state_owner(uaddr2, &q, current, | |
2191 | fshared); | |
2192 | spin_unlock(q.lock_ptr); | |
2193 | } | |
2194 | } else { | |
2195 | /* | |
2196 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2197 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2198 | * the pi_state. | |
2199 | */ | |
2200 | WARN_ON(!&q.pi_state); | |
2201 | pi_mutex = &q.pi_state->pi_mutex; | |
2202 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1); | |
2203 | debug_rt_mutex_free_waiter(&rt_waiter); | |
2204 | ||
2205 | spin_lock(q.lock_ptr); | |
2206 | /* | |
2207 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2208 | * haven't already. | |
2209 | */ | |
2210 | res = fixup_owner(uaddr2, fshared, &q, !ret); | |
2211 | /* | |
2212 | * If fixup_owner() returned an error, proprogate that. If it | |
2213 | * acquired the lock, clear our -ETIMEDOUT or -EINTR. | |
2214 | */ | |
2215 | if (res) | |
2216 | ret = (res < 0) ? res : 0; | |
2217 | ||
2218 | /* Unqueue and drop the lock. */ | |
2219 | unqueue_me_pi(&q); | |
2220 | } | |
2221 | ||
2222 | /* | |
2223 | * If fixup_pi_state_owner() faulted and was unable to handle the | |
2224 | * fault, unlock the rt_mutex and return the fault to userspace. | |
2225 | */ | |
2226 | if (ret == -EFAULT) { | |
2227 | if (rt_mutex_owner(pi_mutex) == current) | |
2228 | rt_mutex_unlock(pi_mutex); | |
2229 | } else if (ret == -EINTR) { | |
52400ba9 | 2230 | /* |
2070887f TG |
2231 | * We've already been requeued, but we have no way to |
2232 | * restart by calling futex_lock_pi() directly. We | |
2233 | * could restart the syscall, but that will look at | |
2234 | * the user space value and return right away. So we | |
2235 | * drop back with EWOULDBLOCK to tell user space that | |
2236 | * "val" has been changed. That's the same what the | |
2237 | * restart of the syscall would do in | |
2238 | * futex_wait_setup(). | |
52400ba9 | 2239 | */ |
2070887f | 2240 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2241 | } |
2242 | ||
2243 | out_put_keys: | |
2244 | put_futex_key(fshared, &q.key); | |
c8b15a70 | 2245 | out_key2: |
52400ba9 DH |
2246 | put_futex_key(fshared, &key2); |
2247 | ||
2248 | out: | |
2249 | if (to) { | |
2250 | hrtimer_cancel(&to->timer); | |
2251 | destroy_hrtimer_on_stack(&to->timer); | |
2252 | } | |
2253 | return ret; | |
2254 | } | |
2255 | ||
0771dfef IM |
2256 | /* |
2257 | * Support for robust futexes: the kernel cleans up held futexes at | |
2258 | * thread exit time. | |
2259 | * | |
2260 | * Implementation: user-space maintains a per-thread list of locks it | |
2261 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2262 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2263 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2264 | * always manipulated with the lock held, so the list is private and |
2265 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2266 | * field, to allow the kernel to clean up if the thread dies after | |
2267 | * acquiring the lock, but just before it could have added itself to | |
2268 | * the list. There can only be one such pending lock. | |
2269 | */ | |
2270 | ||
2271 | /** | |
2272 | * sys_set_robust_list - set the robust-futex list head of a task | |
2273 | * @head: pointer to the list-head | |
2274 | * @len: length of the list-head, as userspace expects | |
2275 | */ | |
836f92ad HC |
2276 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2277 | size_t, len) | |
0771dfef | 2278 | { |
a0c1e907 TG |
2279 | if (!futex_cmpxchg_enabled) |
2280 | return -ENOSYS; | |
0771dfef IM |
2281 | /* |
2282 | * The kernel knows only one size for now: | |
2283 | */ | |
2284 | if (unlikely(len != sizeof(*head))) | |
2285 | return -EINVAL; | |
2286 | ||
2287 | current->robust_list = head; | |
2288 | ||
2289 | return 0; | |
2290 | } | |
2291 | ||
2292 | /** | |
2293 | * sys_get_robust_list - get the robust-futex list head of a task | |
2294 | * @pid: pid of the process [zero for current task] | |
2295 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
2296 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
2297 | */ | |
836f92ad HC |
2298 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
2299 | struct robust_list_head __user * __user *, head_ptr, | |
2300 | size_t __user *, len_ptr) | |
0771dfef | 2301 | { |
ba46df98 | 2302 | struct robust_list_head __user *head; |
0771dfef | 2303 | unsigned long ret; |
c69e8d9c | 2304 | const struct cred *cred = current_cred(), *pcred; |
0771dfef | 2305 | |
a0c1e907 TG |
2306 | if (!futex_cmpxchg_enabled) |
2307 | return -ENOSYS; | |
2308 | ||
0771dfef IM |
2309 | if (!pid) |
2310 | head = current->robust_list; | |
2311 | else { | |
2312 | struct task_struct *p; | |
2313 | ||
2314 | ret = -ESRCH; | |
aaa2a97e | 2315 | rcu_read_lock(); |
228ebcbe | 2316 | p = find_task_by_vpid(pid); |
0771dfef IM |
2317 | if (!p) |
2318 | goto err_unlock; | |
2319 | ret = -EPERM; | |
c69e8d9c DH |
2320 | pcred = __task_cred(p); |
2321 | if (cred->euid != pcred->euid && | |
2322 | cred->euid != pcred->uid && | |
76aac0e9 | 2323 | !capable(CAP_SYS_PTRACE)) |
0771dfef IM |
2324 | goto err_unlock; |
2325 | head = p->robust_list; | |
aaa2a97e | 2326 | rcu_read_unlock(); |
0771dfef IM |
2327 | } |
2328 | ||
2329 | if (put_user(sizeof(*head), len_ptr)) | |
2330 | return -EFAULT; | |
2331 | return put_user(head, head_ptr); | |
2332 | ||
2333 | err_unlock: | |
aaa2a97e | 2334 | rcu_read_unlock(); |
0771dfef IM |
2335 | |
2336 | return ret; | |
2337 | } | |
2338 | ||
2339 | /* | |
2340 | * Process a futex-list entry, check whether it's owned by the | |
2341 | * dying task, and do notification if so: | |
2342 | */ | |
e3f2ddea | 2343 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 2344 | { |
e3f2ddea | 2345 | u32 uval, nval, mval; |
0771dfef | 2346 | |
8f17d3a5 IM |
2347 | retry: |
2348 | if (get_user(uval, uaddr)) | |
0771dfef IM |
2349 | return -1; |
2350 | ||
b488893a | 2351 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
2352 | /* |
2353 | * Ok, this dying thread is truly holding a futex | |
2354 | * of interest. Set the OWNER_DIED bit atomically | |
2355 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
2356 | * set, wake up a waiter (if any). (We have to do a | |
2357 | * futex_wake() even if OWNER_DIED is already set - | |
2358 | * to handle the rare but possible case of recursive | |
2359 | * thread-death.) The rest of the cleanup is done in | |
2360 | * userspace. | |
2361 | */ | |
e3f2ddea IM |
2362 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
2363 | nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval); | |
2364 | ||
c87e2837 IM |
2365 | if (nval == -EFAULT) |
2366 | return -1; | |
2367 | ||
2368 | if (nval != uval) | |
8f17d3a5 | 2369 | goto retry; |
0771dfef | 2370 | |
e3f2ddea IM |
2371 | /* |
2372 | * Wake robust non-PI futexes here. The wakeup of | |
2373 | * PI futexes happens in exit_pi_state(): | |
2374 | */ | |
36cf3b5c | 2375 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 2376 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
2377 | } |
2378 | return 0; | |
2379 | } | |
2380 | ||
e3f2ddea IM |
2381 | /* |
2382 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
2383 | */ | |
2384 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 AV |
2385 | struct robust_list __user * __user *head, |
2386 | int *pi) | |
e3f2ddea IM |
2387 | { |
2388 | unsigned long uentry; | |
2389 | ||
ba46df98 | 2390 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
2391 | return -EFAULT; |
2392 | ||
ba46df98 | 2393 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
2394 | *pi = uentry & 1; |
2395 | ||
2396 | return 0; | |
2397 | } | |
2398 | ||
0771dfef IM |
2399 | /* |
2400 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
2401 | * and mark any locks found there dead, and notify any waiters. | |
2402 | * | |
2403 | * We silently return on any sign of list-walking problem. | |
2404 | */ | |
2405 | void exit_robust_list(struct task_struct *curr) | |
2406 | { | |
2407 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e MS |
2408 | struct robust_list __user *entry, *next_entry, *pending; |
2409 | unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip; | |
0771dfef | 2410 | unsigned long futex_offset; |
9f96cb1e | 2411 | int rc; |
0771dfef | 2412 | |
a0c1e907 TG |
2413 | if (!futex_cmpxchg_enabled) |
2414 | return; | |
2415 | ||
0771dfef IM |
2416 | /* |
2417 | * Fetch the list head (which was registered earlier, via | |
2418 | * sys_set_robust_list()): | |
2419 | */ | |
e3f2ddea | 2420 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
2421 | return; |
2422 | /* | |
2423 | * Fetch the relative futex offset: | |
2424 | */ | |
2425 | if (get_user(futex_offset, &head->futex_offset)) | |
2426 | return; | |
2427 | /* | |
2428 | * Fetch any possibly pending lock-add first, and handle it | |
2429 | * if it exists: | |
2430 | */ | |
e3f2ddea | 2431 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 2432 | return; |
e3f2ddea | 2433 | |
9f96cb1e | 2434 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 2435 | while (entry != &head->list) { |
9f96cb1e MS |
2436 | /* |
2437 | * Fetch the next entry in the list before calling | |
2438 | * handle_futex_death: | |
2439 | */ | |
2440 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
2441 | /* |
2442 | * A pending lock might already be on the list, so | |
c87e2837 | 2443 | * don't process it twice: |
0771dfef IM |
2444 | */ |
2445 | if (entry != pending) | |
ba46df98 | 2446 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 2447 | curr, pi)) |
0771dfef | 2448 | return; |
9f96cb1e | 2449 | if (rc) |
0771dfef | 2450 | return; |
9f96cb1e MS |
2451 | entry = next_entry; |
2452 | pi = next_pi; | |
0771dfef IM |
2453 | /* |
2454 | * Avoid excessively long or circular lists: | |
2455 | */ | |
2456 | if (!--limit) | |
2457 | break; | |
2458 | ||
2459 | cond_resched(); | |
2460 | } | |
9f96cb1e MS |
2461 | |
2462 | if (pending) | |
2463 | handle_futex_death((void __user *)pending + futex_offset, | |
2464 | curr, pip); | |
0771dfef IM |
2465 | } |
2466 | ||
c19384b5 | 2467 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 2468 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 2469 | { |
1acdac10 | 2470 | int clockrt, ret = -ENOSYS; |
34f01cc1 | 2471 | int cmd = op & FUTEX_CMD_MASK; |
c2f9f201 | 2472 | int fshared = 0; |
34f01cc1 ED |
2473 | |
2474 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
c2f9f201 | 2475 | fshared = 1; |
1da177e4 | 2476 | |
1acdac10 | 2477 | clockrt = op & FUTEX_CLOCK_REALTIME; |
52400ba9 | 2478 | if (clockrt && cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI) |
1acdac10 | 2479 | return -ENOSYS; |
1da177e4 | 2480 | |
34f01cc1 | 2481 | switch (cmd) { |
1da177e4 | 2482 | case FUTEX_WAIT: |
cd689985 TG |
2483 | val3 = FUTEX_BITSET_MATCH_ANY; |
2484 | case FUTEX_WAIT_BITSET: | |
1acdac10 | 2485 | ret = futex_wait(uaddr, fshared, val, timeout, val3, clockrt); |
1da177e4 LT |
2486 | break; |
2487 | case FUTEX_WAKE: | |
cd689985 TG |
2488 | val3 = FUTEX_BITSET_MATCH_ANY; |
2489 | case FUTEX_WAKE_BITSET: | |
2490 | ret = futex_wake(uaddr, fshared, val, val3); | |
1da177e4 | 2491 | break; |
1da177e4 | 2492 | case FUTEX_REQUEUE: |
52400ba9 | 2493 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL, 0); |
1da177e4 LT |
2494 | break; |
2495 | case FUTEX_CMP_REQUEUE: | |
52400ba9 DH |
2496 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3, |
2497 | 0); | |
1da177e4 | 2498 | break; |
4732efbe | 2499 | case FUTEX_WAKE_OP: |
34f01cc1 | 2500 | ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3); |
4732efbe | 2501 | break; |
c87e2837 | 2502 | case FUTEX_LOCK_PI: |
a0c1e907 TG |
2503 | if (futex_cmpxchg_enabled) |
2504 | ret = futex_lock_pi(uaddr, fshared, val, timeout, 0); | |
c87e2837 IM |
2505 | break; |
2506 | case FUTEX_UNLOCK_PI: | |
a0c1e907 TG |
2507 | if (futex_cmpxchg_enabled) |
2508 | ret = futex_unlock_pi(uaddr, fshared); | |
c87e2837 IM |
2509 | break; |
2510 | case FUTEX_TRYLOCK_PI: | |
a0c1e907 TG |
2511 | if (futex_cmpxchg_enabled) |
2512 | ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1); | |
c87e2837 | 2513 | break; |
52400ba9 DH |
2514 | case FUTEX_WAIT_REQUEUE_PI: |
2515 | val3 = FUTEX_BITSET_MATCH_ANY; | |
2516 | ret = futex_wait_requeue_pi(uaddr, fshared, val, timeout, val3, | |
2517 | clockrt, uaddr2); | |
2518 | break; | |
52400ba9 DH |
2519 | case FUTEX_CMP_REQUEUE_PI: |
2520 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3, | |
2521 | 1); | |
2522 | break; | |
1da177e4 LT |
2523 | default: |
2524 | ret = -ENOSYS; | |
2525 | } | |
2526 | return ret; | |
2527 | } | |
2528 | ||
2529 | ||
17da2bd9 HC |
2530 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
2531 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
2532 | u32, val3) | |
1da177e4 | 2533 | { |
c19384b5 PP |
2534 | struct timespec ts; |
2535 | ktime_t t, *tp = NULL; | |
e2970f2f | 2536 | u32 val2 = 0; |
34f01cc1 | 2537 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 2538 | |
cd689985 | 2539 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
2540 | cmd == FUTEX_WAIT_BITSET || |
2541 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
c19384b5 | 2542 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 2543 | return -EFAULT; |
c19384b5 | 2544 | if (!timespec_valid(&ts)) |
9741ef96 | 2545 | return -EINVAL; |
c19384b5 PP |
2546 | |
2547 | t = timespec_to_ktime(ts); | |
34f01cc1 | 2548 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 2549 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 2550 | tp = &t; |
1da177e4 LT |
2551 | } |
2552 | /* | |
52400ba9 | 2553 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 2554 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 2555 | */ |
f54f0986 | 2556 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 2557 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 2558 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 2559 | |
c19384b5 | 2560 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
2561 | } |
2562 | ||
f6d107fb | 2563 | static int __init futex_init(void) |
1da177e4 | 2564 | { |
a0c1e907 | 2565 | u32 curval; |
3e4ab747 | 2566 | int i; |
95362fa9 | 2567 | |
a0c1e907 TG |
2568 | /* |
2569 | * This will fail and we want it. Some arch implementations do | |
2570 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
2571 | * functionality. We want to know that before we call in any | |
2572 | * of the complex code paths. Also we want to prevent | |
2573 | * registration of robust lists in that case. NULL is | |
2574 | * guaranteed to fault and we get -EFAULT on functional | |
2575 | * implementation, the non functional ones will return | |
2576 | * -ENOSYS. | |
2577 | */ | |
2578 | curval = cmpxchg_futex_value_locked(NULL, 0, 0); | |
2579 | if (curval == -EFAULT) | |
2580 | futex_cmpxchg_enabled = 1; | |
2581 | ||
3e4ab747 TG |
2582 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { |
2583 | plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock); | |
2584 | spin_lock_init(&futex_queues[i].lock); | |
2585 | } | |
2586 | ||
1da177e4 LT |
2587 | return 0; |
2588 | } | |
f6d107fb | 2589 | __initcall(futex_init); |