]>
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 | ||
ce2ae53b | 250 | page = compound_head(page); |
38d47c1b PZ |
251 | lock_page(page); |
252 | if (!page->mapping) { | |
253 | unlock_page(page); | |
254 | put_page(page); | |
255 | goto again; | |
256 | } | |
1da177e4 LT |
257 | |
258 | /* | |
259 | * Private mappings are handled in a simple way. | |
260 | * | |
261 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
262 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 263 | * the object not the particular process. |
1da177e4 | 264 | */ |
38d47c1b PZ |
265 | if (PageAnon(page)) { |
266 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ | |
1da177e4 | 267 | key->private.mm = mm; |
e2970f2f | 268 | key->private.address = address; |
38d47c1b PZ |
269 | } else { |
270 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ | |
271 | key->shared.inode = page->mapping->host; | |
272 | key->shared.pgoff = page->index; | |
1da177e4 LT |
273 | } |
274 | ||
38d47c1b | 275 | get_futex_key_refs(key); |
1da177e4 | 276 | |
38d47c1b PZ |
277 | unlock_page(page); |
278 | put_page(page); | |
279 | return 0; | |
1da177e4 LT |
280 | } |
281 | ||
38d47c1b | 282 | static inline |
c2f9f201 | 283 | void put_futex_key(int fshared, union futex_key *key) |
1da177e4 | 284 | { |
38d47c1b | 285 | drop_futex_key_refs(key); |
1da177e4 LT |
286 | } |
287 | ||
d0725992 TG |
288 | /* |
289 | * fault_in_user_writeable - fault in user address and verify RW access | |
290 | * @uaddr: pointer to faulting user space address | |
291 | * | |
292 | * Slow path to fixup the fault we just took in the atomic write | |
293 | * access to @uaddr. | |
294 | * | |
295 | * We have no generic implementation of a non destructive write to the | |
296 | * user address. We know that we faulted in the atomic pagefault | |
297 | * disabled section so we can as well avoid the #PF overhead by | |
298 | * calling get_user_pages() right away. | |
299 | */ | |
300 | static int fault_in_user_writeable(u32 __user *uaddr) | |
301 | { | |
302 | int ret = get_user_pages(current, current->mm, (unsigned long)uaddr, | |
aa715284 | 303 | 1, 1, 0, NULL, NULL); |
d0725992 TG |
304 | return ret < 0 ? ret : 0; |
305 | } | |
306 | ||
4b1c486b DH |
307 | /** |
308 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
309 | * @hb: the hash bucket the futex_q's reside in | |
310 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
311 | * | |
312 | * Must be called with the hb lock held. | |
313 | */ | |
314 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
315 | union futex_key *key) | |
316 | { | |
317 | struct futex_q *this; | |
318 | ||
319 | plist_for_each_entry(this, &hb->chain, list) { | |
320 | if (match_futex(&this->key, key)) | |
321 | return this; | |
322 | } | |
323 | return NULL; | |
324 | } | |
325 | ||
36cf3b5c TG |
326 | static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval) |
327 | { | |
328 | u32 curval; | |
329 | ||
330 | pagefault_disable(); | |
331 | curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval); | |
332 | pagefault_enable(); | |
333 | ||
334 | return curval; | |
335 | } | |
336 | ||
337 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
338 | { |
339 | int ret; | |
340 | ||
a866374a | 341 | pagefault_disable(); |
e2970f2f | 342 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 343 | pagefault_enable(); |
1da177e4 LT |
344 | |
345 | return ret ? -EFAULT : 0; | |
346 | } | |
347 | ||
c87e2837 IM |
348 | |
349 | /* | |
350 | * PI code: | |
351 | */ | |
352 | static int refill_pi_state_cache(void) | |
353 | { | |
354 | struct futex_pi_state *pi_state; | |
355 | ||
356 | if (likely(current->pi_state_cache)) | |
357 | return 0; | |
358 | ||
4668edc3 | 359 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
360 | |
361 | if (!pi_state) | |
362 | return -ENOMEM; | |
363 | ||
c87e2837 IM |
364 | INIT_LIST_HEAD(&pi_state->list); |
365 | /* pi_mutex gets initialized later */ | |
366 | pi_state->owner = NULL; | |
367 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 368 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
369 | |
370 | current->pi_state_cache = pi_state; | |
371 | ||
372 | return 0; | |
373 | } | |
374 | ||
375 | static struct futex_pi_state * alloc_pi_state(void) | |
376 | { | |
377 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
378 | ||
379 | WARN_ON(!pi_state); | |
380 | current->pi_state_cache = NULL; | |
381 | ||
382 | return pi_state; | |
383 | } | |
384 | ||
385 | static void free_pi_state(struct futex_pi_state *pi_state) | |
386 | { | |
387 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
388 | return; | |
389 | ||
390 | /* | |
391 | * If pi_state->owner is NULL, the owner is most probably dying | |
392 | * and has cleaned up the pi_state already | |
393 | */ | |
394 | if (pi_state->owner) { | |
395 | spin_lock_irq(&pi_state->owner->pi_lock); | |
396 | list_del_init(&pi_state->list); | |
397 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
398 | ||
399 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
400 | } | |
401 | ||
402 | if (current->pi_state_cache) | |
403 | kfree(pi_state); | |
404 | else { | |
405 | /* | |
406 | * pi_state->list is already empty. | |
407 | * clear pi_state->owner. | |
408 | * refcount is at 0 - put it back to 1. | |
409 | */ | |
410 | pi_state->owner = NULL; | |
411 | atomic_set(&pi_state->refcount, 1); | |
412 | current->pi_state_cache = pi_state; | |
413 | } | |
414 | } | |
415 | ||
416 | /* | |
417 | * Look up the task based on what TID userspace gave us. | |
418 | * We dont trust it. | |
419 | */ | |
420 | static struct task_struct * futex_find_get_task(pid_t pid) | |
421 | { | |
422 | struct task_struct *p; | |
c69e8d9c | 423 | const struct cred *cred = current_cred(), *pcred; |
c87e2837 | 424 | |
d359b549 | 425 | rcu_read_lock(); |
228ebcbe | 426 | p = find_task_by_vpid(pid); |
c69e8d9c | 427 | if (!p) { |
a06381fe | 428 | p = ERR_PTR(-ESRCH); |
c69e8d9c DH |
429 | } else { |
430 | pcred = __task_cred(p); | |
431 | if (cred->euid != pcred->euid && | |
432 | cred->euid != pcred->uid) | |
433 | p = ERR_PTR(-ESRCH); | |
434 | else | |
435 | get_task_struct(p); | |
436 | } | |
a06381fe | 437 | |
d359b549 | 438 | rcu_read_unlock(); |
c87e2837 IM |
439 | |
440 | return p; | |
441 | } | |
442 | ||
443 | /* | |
444 | * This task is holding PI mutexes at exit time => bad. | |
445 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
446 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
447 | */ | |
448 | void exit_pi_state_list(struct task_struct *curr) | |
449 | { | |
c87e2837 IM |
450 | struct list_head *next, *head = &curr->pi_state_list; |
451 | struct futex_pi_state *pi_state; | |
627371d7 | 452 | struct futex_hash_bucket *hb; |
38d47c1b | 453 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 454 | |
a0c1e907 TG |
455 | if (!futex_cmpxchg_enabled) |
456 | return; | |
c87e2837 IM |
457 | /* |
458 | * We are a ZOMBIE and nobody can enqueue itself on | |
459 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 460 | * versus waiters unqueueing themselves: |
c87e2837 IM |
461 | */ |
462 | spin_lock_irq(&curr->pi_lock); | |
463 | while (!list_empty(head)) { | |
464 | ||
465 | next = head->next; | |
466 | pi_state = list_entry(next, struct futex_pi_state, list); | |
467 | key = pi_state->key; | |
627371d7 | 468 | hb = hash_futex(&key); |
c87e2837 IM |
469 | spin_unlock_irq(&curr->pi_lock); |
470 | ||
c87e2837 IM |
471 | spin_lock(&hb->lock); |
472 | ||
473 | spin_lock_irq(&curr->pi_lock); | |
627371d7 IM |
474 | /* |
475 | * We dropped the pi-lock, so re-check whether this | |
476 | * task still owns the PI-state: | |
477 | */ | |
c87e2837 IM |
478 | if (head->next != next) { |
479 | spin_unlock(&hb->lock); | |
480 | continue; | |
481 | } | |
482 | ||
c87e2837 | 483 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
484 | WARN_ON(list_empty(&pi_state->list)); |
485 | list_del_init(&pi_state->list); | |
c87e2837 IM |
486 | pi_state->owner = NULL; |
487 | spin_unlock_irq(&curr->pi_lock); | |
488 | ||
489 | rt_mutex_unlock(&pi_state->pi_mutex); | |
490 | ||
491 | spin_unlock(&hb->lock); | |
492 | ||
493 | spin_lock_irq(&curr->pi_lock); | |
494 | } | |
495 | spin_unlock_irq(&curr->pi_lock); | |
496 | } | |
497 | ||
498 | static int | |
d0aa7a70 PP |
499 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
500 | union futex_key *key, struct futex_pi_state **ps) | |
c87e2837 IM |
501 | { |
502 | struct futex_pi_state *pi_state = NULL; | |
503 | struct futex_q *this, *next; | |
ec92d082 | 504 | struct plist_head *head; |
c87e2837 | 505 | struct task_struct *p; |
778e9a9c | 506 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 IM |
507 | |
508 | head = &hb->chain; | |
509 | ||
ec92d082 | 510 | plist_for_each_entry_safe(this, next, head, list) { |
d0aa7a70 | 511 | if (match_futex(&this->key, key)) { |
c87e2837 IM |
512 | /* |
513 | * Another waiter already exists - bump up | |
514 | * the refcount and return its pi_state: | |
515 | */ | |
516 | pi_state = this->pi_state; | |
06a9ec29 TG |
517 | /* |
518 | * Userspace might have messed up non PI and PI futexes | |
519 | */ | |
520 | if (unlikely(!pi_state)) | |
521 | return -EINVAL; | |
522 | ||
627371d7 | 523 | WARN_ON(!atomic_read(&pi_state->refcount)); |
778e9a9c AK |
524 | WARN_ON(pid && pi_state->owner && |
525 | pi_state->owner->pid != pid); | |
627371d7 | 526 | |
c87e2837 | 527 | atomic_inc(&pi_state->refcount); |
d0aa7a70 | 528 | *ps = pi_state; |
c87e2837 IM |
529 | |
530 | return 0; | |
531 | } | |
532 | } | |
533 | ||
534 | /* | |
e3f2ddea | 535 | * We are the first waiter - try to look up the real owner and attach |
778e9a9c | 536 | * the new pi_state to it, but bail out when TID = 0 |
c87e2837 | 537 | */ |
778e9a9c | 538 | if (!pid) |
e3f2ddea | 539 | return -ESRCH; |
c87e2837 | 540 | p = futex_find_get_task(pid); |
778e9a9c AK |
541 | if (IS_ERR(p)) |
542 | return PTR_ERR(p); | |
543 | ||
544 | /* | |
545 | * We need to look at the task state flags to figure out, | |
546 | * whether the task is exiting. To protect against the do_exit | |
547 | * change of the task flags, we do this protected by | |
548 | * p->pi_lock: | |
549 | */ | |
550 | spin_lock_irq(&p->pi_lock); | |
551 | if (unlikely(p->flags & PF_EXITING)) { | |
552 | /* | |
553 | * The task is on the way out. When PF_EXITPIDONE is | |
554 | * set, we know that the task has finished the | |
555 | * cleanup: | |
556 | */ | |
557 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
558 | ||
559 | spin_unlock_irq(&p->pi_lock); | |
560 | put_task_struct(p); | |
561 | return ret; | |
562 | } | |
c87e2837 IM |
563 | |
564 | pi_state = alloc_pi_state(); | |
565 | ||
566 | /* | |
567 | * Initialize the pi_mutex in locked state and make 'p' | |
568 | * the owner of it: | |
569 | */ | |
570 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
571 | ||
572 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 573 | pi_state->key = *key; |
c87e2837 | 574 | |
627371d7 | 575 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
576 | list_add(&pi_state->list, &p->pi_state_list); |
577 | pi_state->owner = p; | |
578 | spin_unlock_irq(&p->pi_lock); | |
579 | ||
580 | put_task_struct(p); | |
581 | ||
d0aa7a70 | 582 | *ps = pi_state; |
c87e2837 IM |
583 | |
584 | return 0; | |
585 | } | |
586 | ||
1a52084d DH |
587 | /** |
588 | * futex_lock_pi_atomic() - atomic work required to acquire a pi aware futex | |
bab5bc9e DH |
589 | * @uaddr: the pi futex user address |
590 | * @hb: the pi futex hash bucket | |
591 | * @key: the futex key associated with uaddr and hb | |
592 | * @ps: the pi_state pointer where we store the result of the | |
593 | * lookup | |
594 | * @task: the task to perform the atomic lock work for. This will | |
595 | * be "current" except in the case of requeue pi. | |
596 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d DH |
597 | * |
598 | * Returns: | |
599 | * 0 - ready to wait | |
600 | * 1 - acquired the lock | |
601 | * <0 - error | |
602 | * | |
603 | * The hb->lock and futex_key refs shall be held by the caller. | |
604 | */ | |
605 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
606 | union futex_key *key, | |
607 | struct futex_pi_state **ps, | |
bab5bc9e | 608 | struct task_struct *task, int set_waiters) |
1a52084d DH |
609 | { |
610 | int lock_taken, ret, ownerdied = 0; | |
611 | u32 uval, newval, curval; | |
612 | ||
613 | retry: | |
614 | ret = lock_taken = 0; | |
615 | ||
616 | /* | |
617 | * To avoid races, we attempt to take the lock here again | |
618 | * (by doing a 0 -> TID atomic cmpxchg), while holding all | |
619 | * the locks. It will most likely not succeed. | |
620 | */ | |
621 | newval = task_pid_vnr(task); | |
bab5bc9e DH |
622 | if (set_waiters) |
623 | newval |= FUTEX_WAITERS; | |
1a52084d DH |
624 | |
625 | curval = cmpxchg_futex_value_locked(uaddr, 0, newval); | |
626 | ||
627 | if (unlikely(curval == -EFAULT)) | |
628 | return -EFAULT; | |
629 | ||
630 | /* | |
631 | * Detect deadlocks. | |
632 | */ | |
633 | if ((unlikely((curval & FUTEX_TID_MASK) == task_pid_vnr(task)))) | |
634 | return -EDEADLK; | |
635 | ||
636 | /* | |
637 | * Surprise - we got the lock. Just return to userspace: | |
638 | */ | |
639 | if (unlikely(!curval)) | |
640 | return 1; | |
641 | ||
642 | uval = curval; | |
643 | ||
644 | /* | |
645 | * Set the FUTEX_WAITERS flag, so the owner will know it has someone | |
646 | * to wake at the next unlock. | |
647 | */ | |
648 | newval = curval | FUTEX_WAITERS; | |
649 | ||
650 | /* | |
651 | * There are two cases, where a futex might have no owner (the | |
652 | * owner TID is 0): OWNER_DIED. We take over the futex in this | |
653 | * case. We also do an unconditional take over, when the owner | |
654 | * of the futex died. | |
655 | * | |
656 | * This is safe as we are protected by the hash bucket lock ! | |
657 | */ | |
658 | if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { | |
659 | /* Keep the OWNER_DIED bit */ | |
660 | newval = (curval & ~FUTEX_TID_MASK) | task_pid_vnr(task); | |
661 | ownerdied = 0; | |
662 | lock_taken = 1; | |
663 | } | |
664 | ||
665 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); | |
666 | ||
667 | if (unlikely(curval == -EFAULT)) | |
668 | return -EFAULT; | |
669 | if (unlikely(curval != uval)) | |
670 | goto retry; | |
671 | ||
672 | /* | |
673 | * We took the lock due to owner died take over. | |
674 | */ | |
675 | if (unlikely(lock_taken)) | |
676 | return 1; | |
677 | ||
678 | /* | |
679 | * We dont have the lock. Look up the PI state (or create it if | |
680 | * we are the first waiter): | |
681 | */ | |
682 | ret = lookup_pi_state(uval, hb, key, ps); | |
683 | ||
684 | if (unlikely(ret)) { | |
685 | switch (ret) { | |
686 | case -ESRCH: | |
687 | /* | |
688 | * No owner found for this futex. Check if the | |
689 | * OWNER_DIED bit is set to figure out whether | |
690 | * this is a robust futex or not. | |
691 | */ | |
692 | if (get_futex_value_locked(&curval, uaddr)) | |
693 | return -EFAULT; | |
694 | ||
695 | /* | |
696 | * We simply start over in case of a robust | |
697 | * futex. The code above will take the futex | |
698 | * and return happy. | |
699 | */ | |
700 | if (curval & FUTEX_OWNER_DIED) { | |
701 | ownerdied = 1; | |
702 | goto retry; | |
703 | } | |
704 | default: | |
705 | break; | |
706 | } | |
707 | } | |
708 | ||
709 | return ret; | |
710 | } | |
711 | ||
1da177e4 LT |
712 | /* |
713 | * The hash bucket lock must be held when this is called. | |
714 | * Afterwards, the futex_q must not be accessed. | |
715 | */ | |
716 | static void wake_futex(struct futex_q *q) | |
717 | { | |
f1a11e05 TG |
718 | struct task_struct *p = q->task; |
719 | ||
1da177e4 | 720 | /* |
f1a11e05 TG |
721 | * We set q->lock_ptr = NULL _before_ we wake up the task. If |
722 | * a non futex wake up happens on another CPU then the task | |
723 | * might exit and p would dereference a non existing task | |
724 | * struct. Prevent this by holding a reference on p across the | |
725 | * wake up. | |
1da177e4 | 726 | */ |
f1a11e05 TG |
727 | get_task_struct(p); |
728 | ||
729 | plist_del(&q->list, &q->list.plist); | |
1da177e4 | 730 | /* |
f1a11e05 TG |
731 | * The waiting task can free the futex_q as soon as |
732 | * q->lock_ptr = NULL is written, without taking any locks. A | |
733 | * memory barrier is required here to prevent the following | |
734 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 735 | */ |
ccdea2f8 | 736 | smp_wmb(); |
1da177e4 | 737 | q->lock_ptr = NULL; |
f1a11e05 TG |
738 | |
739 | wake_up_state(p, TASK_NORMAL); | |
740 | put_task_struct(p); | |
1da177e4 LT |
741 | } |
742 | ||
c87e2837 IM |
743 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
744 | { | |
745 | struct task_struct *new_owner; | |
746 | struct futex_pi_state *pi_state = this->pi_state; | |
747 | u32 curval, newval; | |
748 | ||
749 | if (!pi_state) | |
750 | return -EINVAL; | |
751 | ||
21778867 | 752 | spin_lock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
753 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
754 | ||
755 | /* | |
756 | * This happens when we have stolen the lock and the original | |
757 | * pending owner did not enqueue itself back on the rt_mutex. | |
758 | * Thats not a tragedy. We know that way, that a lock waiter | |
759 | * is on the fly. We make the futex_q waiter the pending owner. | |
760 | */ | |
761 | if (!new_owner) | |
762 | new_owner = this->task; | |
763 | ||
764 | /* | |
765 | * We pass it to the next owner. (The WAITERS bit is always | |
766 | * kept enabled while there is PI state around. We must also | |
767 | * preserve the owner died bit.) | |
768 | */ | |
e3f2ddea | 769 | if (!(uval & FUTEX_OWNER_DIED)) { |
778e9a9c AK |
770 | int ret = 0; |
771 | ||
b488893a | 772 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 773 | |
36cf3b5c | 774 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
778e9a9c | 775 | |
e3f2ddea | 776 | if (curval == -EFAULT) |
778e9a9c | 777 | ret = -EFAULT; |
cde898fa | 778 | else if (curval != uval) |
778e9a9c AK |
779 | ret = -EINVAL; |
780 | if (ret) { | |
781 | spin_unlock(&pi_state->pi_mutex.wait_lock); | |
782 | return ret; | |
783 | } | |
e3f2ddea | 784 | } |
c87e2837 | 785 | |
627371d7 IM |
786 | spin_lock_irq(&pi_state->owner->pi_lock); |
787 | WARN_ON(list_empty(&pi_state->list)); | |
788 | list_del_init(&pi_state->list); | |
789 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
790 | ||
791 | spin_lock_irq(&new_owner->pi_lock); | |
792 | WARN_ON(!list_empty(&pi_state->list)); | |
c87e2837 IM |
793 | list_add(&pi_state->list, &new_owner->pi_state_list); |
794 | pi_state->owner = new_owner; | |
627371d7 IM |
795 | spin_unlock_irq(&new_owner->pi_lock); |
796 | ||
21778867 | 797 | spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
798 | rt_mutex_unlock(&pi_state->pi_mutex); |
799 | ||
800 | return 0; | |
801 | } | |
802 | ||
803 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) | |
804 | { | |
805 | u32 oldval; | |
806 | ||
807 | /* | |
808 | * There is no waiter, so we unlock the futex. The owner died | |
809 | * bit has not to be preserved here. We are the owner: | |
810 | */ | |
36cf3b5c | 811 | oldval = cmpxchg_futex_value_locked(uaddr, uval, 0); |
c87e2837 IM |
812 | |
813 | if (oldval == -EFAULT) | |
814 | return oldval; | |
815 | if (oldval != uval) | |
816 | return -EAGAIN; | |
817 | ||
818 | return 0; | |
819 | } | |
820 | ||
8b8f319f IM |
821 | /* |
822 | * Express the locking dependencies for lockdep: | |
823 | */ | |
824 | static inline void | |
825 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
826 | { | |
827 | if (hb1 <= hb2) { | |
828 | spin_lock(&hb1->lock); | |
829 | if (hb1 < hb2) | |
830 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
831 | } else { /* hb1 > hb2 */ | |
832 | spin_lock(&hb2->lock); | |
833 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
834 | } | |
835 | } | |
836 | ||
5eb3dc62 DH |
837 | static inline void |
838 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
839 | { | |
f061d351 | 840 | spin_unlock(&hb1->lock); |
88f502fe IM |
841 | if (hb1 != hb2) |
842 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
843 | } |
844 | ||
1da177e4 | 845 | /* |
b2d0994b | 846 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 847 | */ |
c2f9f201 | 848 | static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset) |
1da177e4 | 849 | { |
e2970f2f | 850 | struct futex_hash_bucket *hb; |
1da177e4 | 851 | struct futex_q *this, *next; |
ec92d082 | 852 | struct plist_head *head; |
38d47c1b | 853 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 LT |
854 | int ret; |
855 | ||
cd689985 TG |
856 | if (!bitset) |
857 | return -EINVAL; | |
858 | ||
64d1304a | 859 | ret = get_futex_key(uaddr, fshared, &key, VERIFY_READ); |
1da177e4 LT |
860 | if (unlikely(ret != 0)) |
861 | goto out; | |
862 | ||
e2970f2f IM |
863 | hb = hash_futex(&key); |
864 | spin_lock(&hb->lock); | |
865 | head = &hb->chain; | |
1da177e4 | 866 | |
ec92d082 | 867 | plist_for_each_entry_safe(this, next, head, list) { |
1da177e4 | 868 | if (match_futex (&this->key, &key)) { |
52400ba9 | 869 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
870 | ret = -EINVAL; |
871 | break; | |
872 | } | |
cd689985 TG |
873 | |
874 | /* Check if one of the bits is set in both bitsets */ | |
875 | if (!(this->bitset & bitset)) | |
876 | continue; | |
877 | ||
1da177e4 LT |
878 | wake_futex(this); |
879 | if (++ret >= nr_wake) | |
880 | break; | |
881 | } | |
882 | } | |
883 | ||
e2970f2f | 884 | spin_unlock(&hb->lock); |
38d47c1b | 885 | put_futex_key(fshared, &key); |
42d35d48 | 886 | out: |
1da177e4 LT |
887 | return ret; |
888 | } | |
889 | ||
4732efbe JJ |
890 | /* |
891 | * Wake up all waiters hashed on the physical page that is mapped | |
892 | * to this virtual address: | |
893 | */ | |
e2970f2f | 894 | static int |
c2f9f201 | 895 | futex_wake_op(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, |
e2970f2f | 896 | int nr_wake, int nr_wake2, int op) |
4732efbe | 897 | { |
38d47c1b | 898 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 899 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 900 | struct plist_head *head; |
4732efbe | 901 | struct futex_q *this, *next; |
e4dc5b7a | 902 | int ret, op_ret; |
4732efbe | 903 | |
e4dc5b7a | 904 | retry: |
64d1304a | 905 | ret = get_futex_key(uaddr1, fshared, &key1, VERIFY_READ); |
4732efbe JJ |
906 | if (unlikely(ret != 0)) |
907 | goto out; | |
64d1304a | 908 | ret = get_futex_key(uaddr2, fshared, &key2, VERIFY_WRITE); |
4732efbe | 909 | if (unlikely(ret != 0)) |
42d35d48 | 910 | goto out_put_key1; |
4732efbe | 911 | |
e2970f2f IM |
912 | hb1 = hash_futex(&key1); |
913 | hb2 = hash_futex(&key2); | |
4732efbe | 914 | |
8b8f319f | 915 | double_lock_hb(hb1, hb2); |
e4dc5b7a | 916 | retry_private: |
e2970f2f | 917 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 918 | if (unlikely(op_ret < 0)) { |
4732efbe | 919 | |
5eb3dc62 | 920 | double_unlock_hb(hb1, hb2); |
4732efbe | 921 | |
7ee1dd3f | 922 | #ifndef CONFIG_MMU |
e2970f2f IM |
923 | /* |
924 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
925 | * but we might get them from range checking | |
926 | */ | |
7ee1dd3f | 927 | ret = op_ret; |
42d35d48 | 928 | goto out_put_keys; |
7ee1dd3f DH |
929 | #endif |
930 | ||
796f8d9b DG |
931 | if (unlikely(op_ret != -EFAULT)) { |
932 | ret = op_ret; | |
42d35d48 | 933 | goto out_put_keys; |
796f8d9b DG |
934 | } |
935 | ||
d0725992 | 936 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 937 | if (ret) |
de87fcc1 | 938 | goto out_put_keys; |
4732efbe | 939 | |
e4dc5b7a DH |
940 | if (!fshared) |
941 | goto retry_private; | |
942 | ||
de87fcc1 DH |
943 | put_futex_key(fshared, &key2); |
944 | put_futex_key(fshared, &key1); | |
e4dc5b7a | 945 | goto retry; |
4732efbe JJ |
946 | } |
947 | ||
e2970f2f | 948 | head = &hb1->chain; |
4732efbe | 949 | |
ec92d082 | 950 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
951 | if (match_futex (&this->key, &key1)) { |
952 | wake_futex(this); | |
953 | if (++ret >= nr_wake) | |
954 | break; | |
955 | } | |
956 | } | |
957 | ||
958 | if (op_ret > 0) { | |
e2970f2f | 959 | head = &hb2->chain; |
4732efbe JJ |
960 | |
961 | op_ret = 0; | |
ec92d082 | 962 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
963 | if (match_futex (&this->key, &key2)) { |
964 | wake_futex(this); | |
965 | if (++op_ret >= nr_wake2) | |
966 | break; | |
967 | } | |
968 | } | |
969 | ret += op_ret; | |
970 | } | |
971 | ||
5eb3dc62 | 972 | double_unlock_hb(hb1, hb2); |
42d35d48 | 973 | out_put_keys: |
38d47c1b | 974 | put_futex_key(fshared, &key2); |
42d35d48 | 975 | out_put_key1: |
38d47c1b | 976 | put_futex_key(fshared, &key1); |
42d35d48 | 977 | out: |
4732efbe JJ |
978 | return ret; |
979 | } | |
980 | ||
9121e478 DH |
981 | /** |
982 | * requeue_futex() - Requeue a futex_q from one hb to another | |
983 | * @q: the futex_q to requeue | |
984 | * @hb1: the source hash_bucket | |
985 | * @hb2: the target hash_bucket | |
986 | * @key2: the new key for the requeued futex_q | |
987 | */ | |
988 | static inline | |
989 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
990 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
991 | { | |
992 | ||
993 | /* | |
994 | * If key1 and key2 hash to the same bucket, no need to | |
995 | * requeue. | |
996 | */ | |
997 | if (likely(&hb1->chain != &hb2->chain)) { | |
998 | plist_del(&q->list, &hb1->chain); | |
999 | plist_add(&q->list, &hb2->chain); | |
1000 | q->lock_ptr = &hb2->lock; | |
1001 | #ifdef CONFIG_DEBUG_PI_LIST | |
1002 | q->list.plist.lock = &hb2->lock; | |
1003 | #endif | |
1004 | } | |
1005 | get_futex_key_refs(key2); | |
1006 | q->key = *key2; | |
1007 | } | |
1008 | ||
52400ba9 DH |
1009 | /** |
1010 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
1011 | * q: the futex_q | |
1012 | * key: the key of the requeue target futex | |
beda2c7e | 1013 | * hb: the hash_bucket of the requeue target futex |
52400ba9 DH |
1014 | * |
1015 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1016 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1017 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1018 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1019 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1020 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1021 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1022 | */ |
1023 | static inline | |
beda2c7e DH |
1024 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1025 | struct futex_hash_bucket *hb) | |
52400ba9 DH |
1026 | { |
1027 | drop_futex_key_refs(&q->key); | |
1028 | get_futex_key_refs(key); | |
1029 | q->key = *key; | |
1030 | ||
1031 | WARN_ON(plist_node_empty(&q->list)); | |
1032 | plist_del(&q->list, &q->list.plist); | |
1033 | ||
1034 | WARN_ON(!q->rt_waiter); | |
1035 | q->rt_waiter = NULL; | |
1036 | ||
beda2c7e DH |
1037 | q->lock_ptr = &hb->lock; |
1038 | #ifdef CONFIG_DEBUG_PI_LIST | |
1039 | q->list.plist.lock = &hb->lock; | |
1040 | #endif | |
1041 | ||
f1a11e05 | 1042 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1043 | } |
1044 | ||
1045 | /** | |
1046 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1047 | * @pifutex: the user address of the to futex |
1048 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1049 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1050 | * @key1: the from futex key | |
1051 | * @key2: the to futex key | |
1052 | * @ps: address to store the pi_state pointer | |
1053 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1054 | * |
1055 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1056 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1057 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1058 | * hb1 and hb2 must be held by the caller. | |
52400ba9 DH |
1059 | * |
1060 | * Returns: | |
1061 | * 0 - failed to acquire the lock atomicly | |
1062 | * 1 - acquired the lock | |
1063 | * <0 - error | |
1064 | */ | |
1065 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1066 | struct futex_hash_bucket *hb1, | |
1067 | struct futex_hash_bucket *hb2, | |
1068 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1069 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1070 | { |
bab5bc9e | 1071 | struct futex_q *top_waiter = NULL; |
52400ba9 DH |
1072 | u32 curval; |
1073 | int ret; | |
1074 | ||
1075 | if (get_futex_value_locked(&curval, pifutex)) | |
1076 | return -EFAULT; | |
1077 | ||
bab5bc9e DH |
1078 | /* |
1079 | * Find the top_waiter and determine if there are additional waiters. | |
1080 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1081 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1082 | * as we have means to handle the possible fault. If not, don't set | |
1083 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1084 | * the kernel. | |
1085 | */ | |
52400ba9 DH |
1086 | top_waiter = futex_top_waiter(hb1, key1); |
1087 | ||
1088 | /* There are no waiters, nothing for us to do. */ | |
1089 | if (!top_waiter) | |
1090 | return 0; | |
1091 | ||
1092 | /* | |
bab5bc9e DH |
1093 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1094 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1095 | * in ps in contended cases. | |
52400ba9 | 1096 | */ |
bab5bc9e DH |
1097 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1098 | set_waiters); | |
52400ba9 | 1099 | if (ret == 1) |
beda2c7e | 1100 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
52400ba9 DH |
1101 | |
1102 | return ret; | |
1103 | } | |
1104 | ||
1105 | /** | |
1106 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
1107 | * uaddr1: source futex user address | |
1108 | * uaddr2: target futex user address | |
1109 | * nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1110 | * nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1111 | * requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
1112 | * pi futex (pi to pi requeue is not supported) | |
1113 | * | |
1114 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1115 | * uaddr2 atomically on behalf of the top waiter. | |
1116 | * | |
1117 | * Returns: | |
1118 | * >=0 - on success, the number of tasks requeued or woken | |
1119 | * <0 - on error | |
1da177e4 | 1120 | */ |
c2f9f201 | 1121 | static int futex_requeue(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, |
52400ba9 DH |
1122 | int nr_wake, int nr_requeue, u32 *cmpval, |
1123 | int requeue_pi) | |
1da177e4 | 1124 | { |
38d47c1b | 1125 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1126 | int drop_count = 0, task_count = 0, ret; |
1127 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1128 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 1129 | struct plist_head *head1; |
1da177e4 | 1130 | struct futex_q *this, *next; |
52400ba9 DH |
1131 | u32 curval2; |
1132 | ||
1133 | if (requeue_pi) { | |
1134 | /* | |
1135 | * requeue_pi requires a pi_state, try to allocate it now | |
1136 | * without any locks in case it fails. | |
1137 | */ | |
1138 | if (refill_pi_state_cache()) | |
1139 | return -ENOMEM; | |
1140 | /* | |
1141 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1142 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1143 | * returning to userspace, so as to not leave the rt_mutex with | |
1144 | * waiters and no owner. However, second and third wake-ups | |
1145 | * cannot be predicted as they involve race conditions with the | |
1146 | * first wake and a fault while looking up the pi_state. Both | |
1147 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1148 | * use nr_wake=1. | |
1149 | */ | |
1150 | if (nr_wake != 1) | |
1151 | return -EINVAL; | |
1152 | } | |
1da177e4 | 1153 | |
42d35d48 | 1154 | retry: |
52400ba9 DH |
1155 | if (pi_state != NULL) { |
1156 | /* | |
1157 | * We will have to lookup the pi_state again, so free this one | |
1158 | * to keep the accounting correct. | |
1159 | */ | |
1160 | free_pi_state(pi_state); | |
1161 | pi_state = NULL; | |
1162 | } | |
1163 | ||
64d1304a | 1164 | ret = get_futex_key(uaddr1, fshared, &key1, VERIFY_READ); |
1da177e4 LT |
1165 | if (unlikely(ret != 0)) |
1166 | goto out; | |
521c1808 TG |
1167 | ret = get_futex_key(uaddr2, fshared, &key2, |
1168 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1169 | if (unlikely(ret != 0)) |
42d35d48 | 1170 | goto out_put_key1; |
1da177e4 | 1171 | |
e2970f2f IM |
1172 | hb1 = hash_futex(&key1); |
1173 | hb2 = hash_futex(&key2); | |
1da177e4 | 1174 | |
e4dc5b7a | 1175 | retry_private: |
8b8f319f | 1176 | double_lock_hb(hb1, hb2); |
1da177e4 | 1177 | |
e2970f2f IM |
1178 | if (likely(cmpval != NULL)) { |
1179 | u32 curval; | |
1da177e4 | 1180 | |
e2970f2f | 1181 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1182 | |
1183 | if (unlikely(ret)) { | |
5eb3dc62 | 1184 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1185 | |
e2970f2f | 1186 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1187 | if (ret) |
1188 | goto out_put_keys; | |
1da177e4 | 1189 | |
e4dc5b7a DH |
1190 | if (!fshared) |
1191 | goto retry_private; | |
1da177e4 | 1192 | |
e4dc5b7a DH |
1193 | put_futex_key(fshared, &key2); |
1194 | put_futex_key(fshared, &key1); | |
1195 | goto retry; | |
1da177e4 | 1196 | } |
e2970f2f | 1197 | if (curval != *cmpval) { |
1da177e4 LT |
1198 | ret = -EAGAIN; |
1199 | goto out_unlock; | |
1200 | } | |
1201 | } | |
1202 | ||
52400ba9 | 1203 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1204 | /* |
1205 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1206 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1207 | * bit. We force this here where we are able to easily handle | |
1208 | * faults rather in the requeue loop below. | |
1209 | */ | |
52400ba9 | 1210 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1211 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1212 | |
1213 | /* | |
1214 | * At this point the top_waiter has either taken uaddr2 or is | |
1215 | * waiting on it. If the former, then the pi_state will not | |
1216 | * exist yet, look it up one more time to ensure we have a | |
1217 | * reference to it. | |
1218 | */ | |
1219 | if (ret == 1) { | |
1220 | WARN_ON(pi_state); | |
1221 | task_count++; | |
1222 | ret = get_futex_value_locked(&curval2, uaddr2); | |
1223 | if (!ret) | |
1224 | ret = lookup_pi_state(curval2, hb2, &key2, | |
1225 | &pi_state); | |
1226 | } | |
1227 | ||
1228 | switch (ret) { | |
1229 | case 0: | |
1230 | break; | |
1231 | case -EFAULT: | |
1232 | double_unlock_hb(hb1, hb2); | |
1233 | put_futex_key(fshared, &key2); | |
1234 | put_futex_key(fshared, &key1); | |
d0725992 | 1235 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
1236 | if (!ret) |
1237 | goto retry; | |
1238 | goto out; | |
1239 | case -EAGAIN: | |
1240 | /* The owner was exiting, try again. */ | |
1241 | double_unlock_hb(hb1, hb2); | |
1242 | put_futex_key(fshared, &key2); | |
1243 | put_futex_key(fshared, &key1); | |
1244 | cond_resched(); | |
1245 | goto retry; | |
1246 | default: | |
1247 | goto out_unlock; | |
1248 | } | |
1249 | } | |
1250 | ||
e2970f2f | 1251 | head1 = &hb1->chain; |
ec92d082 | 1252 | plist_for_each_entry_safe(this, next, head1, list) { |
52400ba9 DH |
1253 | if (task_count - nr_wake >= nr_requeue) |
1254 | break; | |
1255 | ||
1256 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1257 | continue; |
52400ba9 | 1258 | |
392741e0 DH |
1259 | /* |
1260 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
1261 | * be paired with each other and no other futex ops. | |
1262 | */ | |
1263 | if ((requeue_pi && !this->rt_waiter) || | |
1264 | (!requeue_pi && this->rt_waiter)) { | |
1265 | ret = -EINVAL; | |
1266 | break; | |
1267 | } | |
52400ba9 DH |
1268 | |
1269 | /* | |
1270 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1271 | * lock, we already woke the top_waiter. If not, it will be | |
1272 | * woken by futex_unlock_pi(). | |
1273 | */ | |
1274 | if (++task_count <= nr_wake && !requeue_pi) { | |
1da177e4 | 1275 | wake_futex(this); |
52400ba9 DH |
1276 | continue; |
1277 | } | |
1da177e4 | 1278 | |
52400ba9 DH |
1279 | /* |
1280 | * Requeue nr_requeue waiters and possibly one more in the case | |
1281 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1282 | */ | |
1283 | if (requeue_pi) { | |
1284 | /* Prepare the waiter to take the rt_mutex. */ | |
1285 | atomic_inc(&pi_state->refcount); | |
1286 | this->pi_state = pi_state; | |
1287 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1288 | this->rt_waiter, | |
1289 | this->task, 1); | |
1290 | if (ret == 1) { | |
1291 | /* We got the lock. */ | |
beda2c7e | 1292 | requeue_pi_wake_futex(this, &key2, hb2); |
52400ba9 DH |
1293 | continue; |
1294 | } else if (ret) { | |
1295 | /* -EDEADLK */ | |
1296 | this->pi_state = NULL; | |
1297 | free_pi_state(pi_state); | |
1298 | goto out_unlock; | |
1299 | } | |
1da177e4 | 1300 | } |
52400ba9 DH |
1301 | requeue_futex(this, hb1, hb2, &key2); |
1302 | drop_count++; | |
1da177e4 LT |
1303 | } |
1304 | ||
1305 | out_unlock: | |
5eb3dc62 | 1306 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1307 | |
cd84a42f DH |
1308 | /* |
1309 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1310 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1311 | * one at key2 and updated their key pointer. We no longer need to | |
1312 | * hold the references to key1. | |
1313 | */ | |
1da177e4 | 1314 | while (--drop_count >= 0) |
9adef58b | 1315 | drop_futex_key_refs(&key1); |
1da177e4 | 1316 | |
42d35d48 | 1317 | out_put_keys: |
38d47c1b | 1318 | put_futex_key(fshared, &key2); |
42d35d48 | 1319 | out_put_key1: |
38d47c1b | 1320 | put_futex_key(fshared, &key1); |
42d35d48 | 1321 | out: |
52400ba9 DH |
1322 | if (pi_state != NULL) |
1323 | free_pi_state(pi_state); | |
1324 | return ret ? ret : task_count; | |
1da177e4 LT |
1325 | } |
1326 | ||
1327 | /* The key must be already stored in q->key. */ | |
82af7aca | 1328 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
1da177e4 | 1329 | { |
e2970f2f | 1330 | struct futex_hash_bucket *hb; |
1da177e4 | 1331 | |
9adef58b | 1332 | get_futex_key_refs(&q->key); |
e2970f2f IM |
1333 | hb = hash_futex(&q->key); |
1334 | q->lock_ptr = &hb->lock; | |
1da177e4 | 1335 | |
e2970f2f IM |
1336 | spin_lock(&hb->lock); |
1337 | return hb; | |
1da177e4 LT |
1338 | } |
1339 | ||
82af7aca | 1340 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 1341 | { |
ec92d082 PP |
1342 | int prio; |
1343 | ||
1344 | /* | |
1345 | * The priority used to register this element is | |
1346 | * - either the real thread-priority for the real-time threads | |
1347 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
1348 | * - or MAX_RT_PRIO for non-RT threads. | |
1349 | * Thus, all RT-threads are woken first in priority order, and | |
1350 | * the others are woken last, in FIFO order. | |
1351 | */ | |
1352 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1353 | ||
1354 | plist_node_init(&q->list, prio); | |
1355 | #ifdef CONFIG_DEBUG_PI_LIST | |
1356 | q->list.plist.lock = &hb->lock; | |
1357 | #endif | |
1358 | plist_add(&q->list, &hb->chain); | |
c87e2837 | 1359 | q->task = current; |
e2970f2f | 1360 | spin_unlock(&hb->lock); |
1da177e4 LT |
1361 | } |
1362 | ||
1363 | static inline void | |
e2970f2f | 1364 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 1365 | { |
e2970f2f | 1366 | spin_unlock(&hb->lock); |
9adef58b | 1367 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1368 | } |
1369 | ||
1370 | /* | |
1371 | * queue_me and unqueue_me must be called as a pair, each | |
1372 | * exactly once. They are called with the hashed spinlock held. | |
1373 | */ | |
1374 | ||
1da177e4 LT |
1375 | /* Return 1 if we were still queued (ie. 0 means we were woken) */ |
1376 | static int unqueue_me(struct futex_q *q) | |
1377 | { | |
1da177e4 | 1378 | spinlock_t *lock_ptr; |
e2970f2f | 1379 | int ret = 0; |
1da177e4 LT |
1380 | |
1381 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 1382 | retry: |
1da177e4 | 1383 | lock_ptr = q->lock_ptr; |
e91467ec | 1384 | barrier(); |
c80544dc | 1385 | if (lock_ptr != NULL) { |
1da177e4 LT |
1386 | spin_lock(lock_ptr); |
1387 | /* | |
1388 | * q->lock_ptr can change between reading it and | |
1389 | * spin_lock(), causing us to take the wrong lock. This | |
1390 | * corrects the race condition. | |
1391 | * | |
1392 | * Reasoning goes like this: if we have the wrong lock, | |
1393 | * q->lock_ptr must have changed (maybe several times) | |
1394 | * between reading it and the spin_lock(). It can | |
1395 | * change again after the spin_lock() but only if it was | |
1396 | * already changed before the spin_lock(). It cannot, | |
1397 | * however, change back to the original value. Therefore | |
1398 | * we can detect whether we acquired the correct lock. | |
1399 | */ | |
1400 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
1401 | spin_unlock(lock_ptr); | |
1402 | goto retry; | |
1403 | } | |
ec92d082 PP |
1404 | WARN_ON(plist_node_empty(&q->list)); |
1405 | plist_del(&q->list, &q->list.plist); | |
c87e2837 IM |
1406 | |
1407 | BUG_ON(q->pi_state); | |
1408 | ||
1da177e4 LT |
1409 | spin_unlock(lock_ptr); |
1410 | ret = 1; | |
1411 | } | |
1412 | ||
9adef58b | 1413 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1414 | return ret; |
1415 | } | |
1416 | ||
c87e2837 IM |
1417 | /* |
1418 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
1419 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
1420 | * and dropped here. | |
c87e2837 | 1421 | */ |
d0aa7a70 | 1422 | static void unqueue_me_pi(struct futex_q *q) |
c87e2837 | 1423 | { |
ec92d082 PP |
1424 | WARN_ON(plist_node_empty(&q->list)); |
1425 | plist_del(&q->list, &q->list.plist); | |
c87e2837 IM |
1426 | |
1427 | BUG_ON(!q->pi_state); | |
1428 | free_pi_state(q->pi_state); | |
1429 | q->pi_state = NULL; | |
1430 | ||
d0aa7a70 | 1431 | spin_unlock(q->lock_ptr); |
c87e2837 | 1432 | |
9adef58b | 1433 | drop_futex_key_refs(&q->key); |
c87e2837 IM |
1434 | } |
1435 | ||
d0aa7a70 | 1436 | /* |
cdf71a10 | 1437 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 1438 | * |
778e9a9c AK |
1439 | * Must be called with hash bucket lock held and mm->sem held for non |
1440 | * private futexes. | |
d0aa7a70 | 1441 | */ |
778e9a9c | 1442 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
c2f9f201 | 1443 | struct task_struct *newowner, int fshared) |
d0aa7a70 | 1444 | { |
cdf71a10 | 1445 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 1446 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 1447 | struct task_struct *oldowner = pi_state->owner; |
d0aa7a70 | 1448 | u32 uval, curval, newval; |
e4dc5b7a | 1449 | int ret; |
d0aa7a70 PP |
1450 | |
1451 | /* Owner died? */ | |
1b7558e4 TG |
1452 | if (!pi_state->owner) |
1453 | newtid |= FUTEX_OWNER_DIED; | |
1454 | ||
1455 | /* | |
1456 | * We are here either because we stole the rtmutex from the | |
1457 | * pending owner or we are the pending owner which failed to | |
1458 | * get the rtmutex. We have to replace the pending owner TID | |
1459 | * in the user space variable. This must be atomic as we have | |
1460 | * to preserve the owner died bit here. | |
1461 | * | |
b2d0994b DH |
1462 | * Note: We write the user space value _before_ changing the pi_state |
1463 | * because we can fault here. Imagine swapped out pages or a fork | |
1464 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
1465 | * |
1466 | * Modifying pi_state _before_ the user space value would | |
1467 | * leave the pi_state in an inconsistent state when we fault | |
1468 | * here, because we need to drop the hash bucket lock to | |
1469 | * handle the fault. This might be observed in the PID check | |
1470 | * in lookup_pi_state. | |
1471 | */ | |
1472 | retry: | |
1473 | if (get_futex_value_locked(&uval, uaddr)) | |
1474 | goto handle_fault; | |
1475 | ||
1476 | while (1) { | |
1477 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
1478 | ||
1479 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); | |
1480 | ||
1481 | if (curval == -EFAULT) | |
1482 | goto handle_fault; | |
1483 | if (curval == uval) | |
1484 | break; | |
1485 | uval = curval; | |
1486 | } | |
1487 | ||
1488 | /* | |
1489 | * We fixed up user space. Now we need to fix the pi_state | |
1490 | * itself. | |
1491 | */ | |
d0aa7a70 PP |
1492 | if (pi_state->owner != NULL) { |
1493 | spin_lock_irq(&pi_state->owner->pi_lock); | |
1494 | WARN_ON(list_empty(&pi_state->list)); | |
1495 | list_del_init(&pi_state->list); | |
1496 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
1b7558e4 | 1497 | } |
d0aa7a70 | 1498 | |
cdf71a10 | 1499 | pi_state->owner = newowner; |
d0aa7a70 | 1500 | |
cdf71a10 | 1501 | spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 1502 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 TG |
1503 | list_add(&pi_state->list, &newowner->pi_state_list); |
1504 | spin_unlock_irq(&newowner->pi_lock); | |
1b7558e4 | 1505 | return 0; |
d0aa7a70 | 1506 | |
d0aa7a70 | 1507 | /* |
1b7558e4 TG |
1508 | * To handle the page fault we need to drop the hash bucket |
1509 | * lock here. That gives the other task (either the pending | |
1510 | * owner itself or the task which stole the rtmutex) the | |
1511 | * chance to try the fixup of the pi_state. So once we are | |
1512 | * back from handling the fault we need to check the pi_state | |
1513 | * after reacquiring the hash bucket lock and before trying to | |
1514 | * do another fixup. When the fixup has been done already we | |
1515 | * simply return. | |
d0aa7a70 | 1516 | */ |
1b7558e4 TG |
1517 | handle_fault: |
1518 | spin_unlock(q->lock_ptr); | |
778e9a9c | 1519 | |
d0725992 | 1520 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 1521 | |
1b7558e4 | 1522 | spin_lock(q->lock_ptr); |
778e9a9c | 1523 | |
1b7558e4 TG |
1524 | /* |
1525 | * Check if someone else fixed it for us: | |
1526 | */ | |
1527 | if (pi_state->owner != oldowner) | |
1528 | return 0; | |
1529 | ||
1530 | if (ret) | |
1531 | return ret; | |
1532 | ||
1533 | goto retry; | |
d0aa7a70 PP |
1534 | } |
1535 | ||
34f01cc1 ED |
1536 | /* |
1537 | * In case we must use restart_block to restart a futex_wait, | |
ce6bd420 | 1538 | * we encode in the 'flags' shared capability |
34f01cc1 | 1539 | */ |
1acdac10 TG |
1540 | #define FLAGS_SHARED 0x01 |
1541 | #define FLAGS_CLOCKRT 0x02 | |
a72188d8 | 1542 | #define FLAGS_HAS_TIMEOUT 0x04 |
34f01cc1 | 1543 | |
72c1bbf3 | 1544 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 1545 | |
dd973998 DH |
1546 | /** |
1547 | * fixup_owner() - Post lock pi_state and corner case management | |
1548 | * @uaddr: user address of the futex | |
1549 | * @fshared: whether the futex is shared (1) or not (0) | |
1550 | * @q: futex_q (contains pi_state and access to the rt_mutex) | |
1551 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
1552 | * | |
1553 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
1554 | * the pi_state owner as well as handle race conditions that may allow us to | |
1555 | * acquire the lock. Must be called with the hb lock held. | |
1556 | * | |
1557 | * Returns: | |
1558 | * 1 - success, lock taken | |
1559 | * 0 - success, lock not taken | |
1560 | * <0 - on error (-EFAULT) | |
1561 | */ | |
1562 | static int fixup_owner(u32 __user *uaddr, int fshared, struct futex_q *q, | |
1563 | int locked) | |
1564 | { | |
1565 | struct task_struct *owner; | |
1566 | int ret = 0; | |
1567 | ||
1568 | if (locked) { | |
1569 | /* | |
1570 | * Got the lock. We might not be the anticipated owner if we | |
1571 | * did a lock-steal - fix up the PI-state in that case: | |
1572 | */ | |
1573 | if (q->pi_state->owner != current) | |
1574 | ret = fixup_pi_state_owner(uaddr, q, current, fshared); | |
1575 | goto out; | |
1576 | } | |
1577 | ||
1578 | /* | |
1579 | * Catch the rare case, where the lock was released when we were on the | |
1580 | * way back before we locked the hash bucket. | |
1581 | */ | |
1582 | if (q->pi_state->owner == current) { | |
1583 | /* | |
1584 | * Try to get the rt_mutex now. This might fail as some other | |
1585 | * task acquired the rt_mutex after we removed ourself from the | |
1586 | * rt_mutex waiters list. | |
1587 | */ | |
1588 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
1589 | locked = 1; | |
1590 | goto out; | |
1591 | } | |
1592 | ||
1593 | /* | |
1594 | * pi_state is incorrect, some other task did a lock steal and | |
1595 | * we returned due to timeout or signal without taking the | |
1596 | * rt_mutex. Too late. We can access the rt_mutex_owner without | |
1597 | * locking, as the other task is now blocked on the hash bucket | |
1598 | * lock. Fix the state up. | |
1599 | */ | |
1600 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); | |
1601 | ret = fixup_pi_state_owner(uaddr, q, owner, fshared); | |
1602 | goto out; | |
1603 | } | |
1604 | ||
1605 | /* | |
1606 | * Paranoia check. If we did not take the lock, then we should not be | |
1607 | * the owner, nor the pending owner, of the rt_mutex. | |
1608 | */ | |
1609 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
1610 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
1611 | "pi-state %p\n", ret, | |
1612 | q->pi_state->pi_mutex.owner, | |
1613 | q->pi_state->owner); | |
1614 | ||
1615 | out: | |
1616 | return ret ? ret : locked; | |
1617 | } | |
1618 | ||
ca5f9524 DH |
1619 | /** |
1620 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
1621 | * @hb: the futex hash bucket, must be locked by the caller | |
1622 | * @q: the futex_q to queue up on | |
1623 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
1624 | */ |
1625 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 1626 | struct hrtimer_sleeper *timeout) |
ca5f9524 DH |
1627 | { |
1628 | queue_me(q, hb); | |
1629 | ||
1630 | /* | |
1631 | * There might have been scheduling since the queue_me(), as we | |
1632 | * cannot hold a spinlock across the get_user() in case it | |
1633 | * faults, and we cannot just set TASK_INTERRUPTIBLE state when | |
f1a11e05 | 1634 | * queueing ourselves into the futex hash. This code thus has to |
ca5f9524 DH |
1635 | * rely on the futex_wake() code removing us from hash when it |
1636 | * wakes us up. | |
1637 | */ | |
f1a11e05 | 1638 | set_current_state(TASK_INTERRUPTIBLE); |
ca5f9524 DH |
1639 | |
1640 | /* Arm the timer */ | |
1641 | if (timeout) { | |
1642 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); | |
1643 | if (!hrtimer_active(&timeout->timer)) | |
1644 | timeout->task = NULL; | |
1645 | } | |
1646 | ||
1647 | /* | |
1648 | * !plist_node_empty() is safe here without any lock. | |
1649 | * q.lock_ptr != 0 is not safe, because of ordering against wakeup. | |
1650 | */ | |
1651 | if (likely(!plist_node_empty(&q->list))) { | |
1652 | /* | |
1653 | * If the timer has already expired, current will already be | |
1654 | * flagged for rescheduling. Only call schedule if there | |
1655 | * is no timeout, or if it has yet to expire. | |
1656 | */ | |
1657 | if (!timeout || timeout->task) | |
1658 | schedule(); | |
1659 | } | |
1660 | __set_current_state(TASK_RUNNING); | |
1661 | } | |
1662 | ||
f801073f DH |
1663 | /** |
1664 | * futex_wait_setup() - Prepare to wait on a futex | |
1665 | * @uaddr: the futex userspace address | |
1666 | * @val: the expected value | |
1667 | * @fshared: whether the futex is shared (1) or not (0) | |
1668 | * @q: the associated futex_q | |
1669 | * @hb: storage for hash_bucket pointer to be returned to caller | |
1670 | * | |
1671 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
1672 | * compare it with the expected value. Handle atomic faults internally. | |
1673 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
1674 | * with no q.key reference on failure. | |
1675 | * | |
1676 | * Returns: | |
1677 | * 0 - uaddr contains val and hb has been locked | |
1678 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlcoked | |
1679 | */ | |
1680 | static int futex_wait_setup(u32 __user *uaddr, u32 val, int fshared, | |
1681 | struct futex_q *q, struct futex_hash_bucket **hb) | |
1da177e4 | 1682 | { |
e2970f2f IM |
1683 | u32 uval; |
1684 | int ret; | |
1da177e4 | 1685 | |
1da177e4 | 1686 | /* |
b2d0994b | 1687 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
1688 | * Order is important: |
1689 | * | |
1690 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
1691 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
1692 | * | |
1693 | * The basic logical guarantee of a futex is that it blocks ONLY | |
1694 | * if cond(var) is known to be true at the time of blocking, for | |
1695 | * any cond. If we queued after testing *uaddr, that would open | |
1696 | * a race condition where we could block indefinitely with | |
1697 | * cond(var) false, which would violate the guarantee. | |
1698 | * | |
1699 | * A consequence is that futex_wait() can return zero and absorb | |
1700 | * a wakeup when *uaddr != val on entry to the syscall. This is | |
1701 | * rare, but normal. | |
1da177e4 | 1702 | */ |
f801073f DH |
1703 | retry: |
1704 | q->key = FUTEX_KEY_INIT; | |
521c1808 | 1705 | ret = get_futex_key(uaddr, fshared, &q->key, VERIFY_READ); |
f801073f | 1706 | if (unlikely(ret != 0)) |
a5a2a0c7 | 1707 | return ret; |
f801073f DH |
1708 | |
1709 | retry_private: | |
1710 | *hb = queue_lock(q); | |
1711 | ||
e2970f2f | 1712 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 1713 | |
f801073f DH |
1714 | if (ret) { |
1715 | queue_unlock(q, *hb); | |
1da177e4 | 1716 | |
e2970f2f | 1717 | ret = get_user(uval, uaddr); |
e4dc5b7a | 1718 | if (ret) |
f801073f | 1719 | goto out; |
1da177e4 | 1720 | |
e4dc5b7a DH |
1721 | if (!fshared) |
1722 | goto retry_private; | |
1723 | ||
f801073f | 1724 | put_futex_key(fshared, &q->key); |
e4dc5b7a | 1725 | goto retry; |
1da177e4 | 1726 | } |
ca5f9524 | 1727 | |
f801073f DH |
1728 | if (uval != val) { |
1729 | queue_unlock(q, *hb); | |
1730 | ret = -EWOULDBLOCK; | |
2fff78c7 | 1731 | } |
1da177e4 | 1732 | |
f801073f DH |
1733 | out: |
1734 | if (ret) | |
1735 | put_futex_key(fshared, &q->key); | |
1736 | return ret; | |
1737 | } | |
1738 | ||
1739 | static int futex_wait(u32 __user *uaddr, int fshared, | |
1740 | u32 val, ktime_t *abs_time, u32 bitset, int clockrt) | |
1741 | { | |
1742 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
1743 | struct restart_block *restart; |
1744 | struct futex_hash_bucket *hb; | |
1745 | struct futex_q q; | |
1746 | int ret; | |
1747 | ||
1748 | if (!bitset) | |
1749 | return -EINVAL; | |
1750 | ||
1751 | q.pi_state = NULL; | |
1752 | q.bitset = bitset; | |
52400ba9 | 1753 | q.rt_waiter = NULL; |
f801073f DH |
1754 | |
1755 | if (abs_time) { | |
1756 | to = &timeout; | |
1757 | ||
1758 | hrtimer_init_on_stack(&to->timer, clockrt ? CLOCK_REALTIME : | |
1759 | CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1760 | hrtimer_init_sleeper(to, current); | |
1761 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
1762 | current->timer_slack_ns); | |
1763 | } | |
1764 | ||
1765 | /* Prepare to wait on uaddr. */ | |
1766 | ret = futex_wait_setup(uaddr, val, fshared, &q, &hb); | |
1767 | if (ret) | |
1768 | goto out; | |
1769 | ||
ca5f9524 | 1770 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 1771 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
1772 | |
1773 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 1774 | ret = 0; |
1da177e4 | 1775 | if (!unqueue_me(&q)) |
2fff78c7 PZ |
1776 | goto out_put_key; |
1777 | ret = -ETIMEDOUT; | |
ca5f9524 | 1778 | if (to && !to->task) |
2fff78c7 | 1779 | goto out_put_key; |
72c1bbf3 | 1780 | |
e2970f2f IM |
1781 | /* |
1782 | * We expect signal_pending(current), but another thread may | |
1783 | * have handled it for us already. | |
1784 | */ | |
2fff78c7 | 1785 | ret = -ERESTARTSYS; |
c19384b5 | 1786 | if (!abs_time) |
2fff78c7 | 1787 | goto out_put_key; |
1da177e4 | 1788 | |
2fff78c7 PZ |
1789 | restart = ¤t_thread_info()->restart_block; |
1790 | restart->fn = futex_wait_restart; | |
1791 | restart->futex.uaddr = (u32 *)uaddr; | |
1792 | restart->futex.val = val; | |
1793 | restart->futex.time = abs_time->tv64; | |
1794 | restart->futex.bitset = bitset; | |
a72188d8 | 1795 | restart->futex.flags = FLAGS_HAS_TIMEOUT; |
2fff78c7 PZ |
1796 | |
1797 | if (fshared) | |
1798 | restart->futex.flags |= FLAGS_SHARED; | |
1799 | if (clockrt) | |
1800 | restart->futex.flags |= FLAGS_CLOCKRT; | |
42d35d48 | 1801 | |
2fff78c7 PZ |
1802 | ret = -ERESTART_RESTARTBLOCK; |
1803 | ||
1804 | out_put_key: | |
1805 | put_futex_key(fshared, &q.key); | |
42d35d48 | 1806 | out: |
ca5f9524 DH |
1807 | if (to) { |
1808 | hrtimer_cancel(&to->timer); | |
1809 | destroy_hrtimer_on_stack(&to->timer); | |
1810 | } | |
c87e2837 IM |
1811 | return ret; |
1812 | } | |
1813 | ||
72c1bbf3 NP |
1814 | |
1815 | static long futex_wait_restart(struct restart_block *restart) | |
1816 | { | |
ce6bd420 | 1817 | u32 __user *uaddr = (u32 __user *)restart->futex.uaddr; |
c2f9f201 | 1818 | int fshared = 0; |
a72188d8 | 1819 | ktime_t t, *tp = NULL; |
72c1bbf3 | 1820 | |
a72188d8 DH |
1821 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
1822 | t.tv64 = restart->futex.time; | |
1823 | tp = &t; | |
1824 | } | |
72c1bbf3 | 1825 | restart->fn = do_no_restart_syscall; |
ce6bd420 | 1826 | if (restart->futex.flags & FLAGS_SHARED) |
c2f9f201 | 1827 | fshared = 1; |
a72188d8 | 1828 | return (long)futex_wait(uaddr, fshared, restart->futex.val, tp, |
1acdac10 TG |
1829 | restart->futex.bitset, |
1830 | restart->futex.flags & FLAGS_CLOCKRT); | |
72c1bbf3 NP |
1831 | } |
1832 | ||
1833 | ||
c87e2837 IM |
1834 | /* |
1835 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
1836 | * and failed. The kernel side here does the whole locking operation: | |
1837 | * if there are waiters then it will block, it does PI, etc. (Due to | |
1838 | * races the kernel might see a 0 value of the futex too.) | |
1839 | */ | |
c2f9f201 | 1840 | static int futex_lock_pi(u32 __user *uaddr, int fshared, |
34f01cc1 | 1841 | int detect, ktime_t *time, int trylock) |
c87e2837 | 1842 | { |
c5780e97 | 1843 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 1844 | struct futex_hash_bucket *hb; |
c87e2837 | 1845 | struct futex_q q; |
dd973998 | 1846 | int res, ret; |
c87e2837 IM |
1847 | |
1848 | if (refill_pi_state_cache()) | |
1849 | return -ENOMEM; | |
1850 | ||
c19384b5 | 1851 | if (time) { |
c5780e97 | 1852 | to = &timeout; |
237fc6e7 TG |
1853 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
1854 | HRTIMER_MODE_ABS); | |
c5780e97 | 1855 | hrtimer_init_sleeper(to, current); |
cc584b21 | 1856 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
1857 | } |
1858 | ||
c87e2837 | 1859 | q.pi_state = NULL; |
52400ba9 | 1860 | q.rt_waiter = NULL; |
42d35d48 | 1861 | retry: |
38d47c1b | 1862 | q.key = FUTEX_KEY_INIT; |
64d1304a | 1863 | ret = get_futex_key(uaddr, fshared, &q.key, VERIFY_WRITE); |
c87e2837 | 1864 | if (unlikely(ret != 0)) |
42d35d48 | 1865 | goto out; |
c87e2837 | 1866 | |
e4dc5b7a | 1867 | retry_private: |
82af7aca | 1868 | hb = queue_lock(&q); |
c87e2837 | 1869 | |
bab5bc9e | 1870 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 1871 | if (unlikely(ret)) { |
778e9a9c | 1872 | switch (ret) { |
1a52084d DH |
1873 | case 1: |
1874 | /* We got the lock. */ | |
1875 | ret = 0; | |
1876 | goto out_unlock_put_key; | |
1877 | case -EFAULT: | |
1878 | goto uaddr_faulted; | |
778e9a9c AK |
1879 | case -EAGAIN: |
1880 | /* | |
1881 | * Task is exiting and we just wait for the | |
1882 | * exit to complete. | |
1883 | */ | |
1884 | queue_unlock(&q, hb); | |
de87fcc1 | 1885 | put_futex_key(fshared, &q.key); |
778e9a9c AK |
1886 | cond_resched(); |
1887 | goto retry; | |
778e9a9c | 1888 | default: |
42d35d48 | 1889 | goto out_unlock_put_key; |
c87e2837 | 1890 | } |
c87e2837 IM |
1891 | } |
1892 | ||
1893 | /* | |
1894 | * Only actually queue now that the atomic ops are done: | |
1895 | */ | |
82af7aca | 1896 | queue_me(&q, hb); |
c87e2837 | 1897 | |
c87e2837 IM |
1898 | WARN_ON(!q.pi_state); |
1899 | /* | |
1900 | * Block on the PI mutex: | |
1901 | */ | |
1902 | if (!trylock) | |
1903 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); | |
1904 | else { | |
1905 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); | |
1906 | /* Fixup the trylock return value: */ | |
1907 | ret = ret ? 0 : -EWOULDBLOCK; | |
1908 | } | |
1909 | ||
a99e4e41 | 1910 | spin_lock(q.lock_ptr); |
dd973998 DH |
1911 | /* |
1912 | * Fixup the pi_state owner and possibly acquire the lock if we | |
1913 | * haven't already. | |
1914 | */ | |
1915 | res = fixup_owner(uaddr, fshared, &q, !ret); | |
1916 | /* | |
1917 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
1918 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
1919 | */ | |
1920 | if (res) | |
1921 | ret = (res < 0) ? res : 0; | |
c87e2837 | 1922 | |
e8f6386c | 1923 | /* |
dd973998 DH |
1924 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
1925 | * it and return the fault to userspace. | |
e8f6386c DH |
1926 | */ |
1927 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
1928 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
1929 | ||
778e9a9c AK |
1930 | /* Unqueue and drop the lock */ |
1931 | unqueue_me_pi(&q); | |
c87e2837 | 1932 | |
dd973998 | 1933 | goto out; |
c87e2837 | 1934 | |
42d35d48 | 1935 | out_unlock_put_key: |
c87e2837 IM |
1936 | queue_unlock(&q, hb); |
1937 | ||
42d35d48 | 1938 | out_put_key: |
38d47c1b | 1939 | put_futex_key(fshared, &q.key); |
42d35d48 | 1940 | out: |
237fc6e7 TG |
1941 | if (to) |
1942 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 1943 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 1944 | |
42d35d48 | 1945 | uaddr_faulted: |
778e9a9c AK |
1946 | queue_unlock(&q, hb); |
1947 | ||
d0725992 | 1948 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
1949 | if (ret) |
1950 | goto out_put_key; | |
c87e2837 | 1951 | |
e4dc5b7a DH |
1952 | if (!fshared) |
1953 | goto retry_private; | |
1954 | ||
1955 | put_futex_key(fshared, &q.key); | |
1956 | goto retry; | |
c87e2837 IM |
1957 | } |
1958 | ||
c87e2837 IM |
1959 | /* |
1960 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
1961 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
1962 | * and do the rt-mutex unlock. | |
1963 | */ | |
c2f9f201 | 1964 | static int futex_unlock_pi(u32 __user *uaddr, int fshared) |
c87e2837 IM |
1965 | { |
1966 | struct futex_hash_bucket *hb; | |
1967 | struct futex_q *this, *next; | |
1968 | u32 uval; | |
ec92d082 | 1969 | struct plist_head *head; |
38d47c1b | 1970 | union futex_key key = FUTEX_KEY_INIT; |
e4dc5b7a | 1971 | int ret; |
c87e2837 IM |
1972 | |
1973 | retry: | |
1974 | if (get_user(uval, uaddr)) | |
1975 | return -EFAULT; | |
1976 | /* | |
1977 | * We release only a lock we actually own: | |
1978 | */ | |
b488893a | 1979 | if ((uval & FUTEX_TID_MASK) != task_pid_vnr(current)) |
c87e2837 | 1980 | return -EPERM; |
c87e2837 | 1981 | |
64d1304a | 1982 | ret = get_futex_key(uaddr, fshared, &key, VERIFY_WRITE); |
c87e2837 IM |
1983 | if (unlikely(ret != 0)) |
1984 | goto out; | |
1985 | ||
1986 | hb = hash_futex(&key); | |
1987 | spin_lock(&hb->lock); | |
1988 | ||
c87e2837 IM |
1989 | /* |
1990 | * To avoid races, try to do the TID -> 0 atomic transition | |
1991 | * again. If it succeeds then we can return without waking | |
1992 | * anyone else up: | |
1993 | */ | |
36cf3b5c | 1994 | if (!(uval & FUTEX_OWNER_DIED)) |
b488893a | 1995 | uval = cmpxchg_futex_value_locked(uaddr, task_pid_vnr(current), 0); |
36cf3b5c | 1996 | |
c87e2837 IM |
1997 | |
1998 | if (unlikely(uval == -EFAULT)) | |
1999 | goto pi_faulted; | |
2000 | /* | |
2001 | * Rare case: we managed to release the lock atomically, | |
2002 | * no need to wake anyone else up: | |
2003 | */ | |
b488893a | 2004 | if (unlikely(uval == task_pid_vnr(current))) |
c87e2837 IM |
2005 | goto out_unlock; |
2006 | ||
2007 | /* | |
2008 | * Ok, other tasks may need to be woken up - check waiters | |
2009 | * and do the wakeup if necessary: | |
2010 | */ | |
2011 | head = &hb->chain; | |
2012 | ||
ec92d082 | 2013 | plist_for_each_entry_safe(this, next, head, list) { |
c87e2837 IM |
2014 | if (!match_futex (&this->key, &key)) |
2015 | continue; | |
2016 | ret = wake_futex_pi(uaddr, uval, this); | |
2017 | /* | |
2018 | * The atomic access to the futex value | |
2019 | * generated a pagefault, so retry the | |
2020 | * user-access and the wakeup: | |
2021 | */ | |
2022 | if (ret == -EFAULT) | |
2023 | goto pi_faulted; | |
2024 | goto out_unlock; | |
2025 | } | |
2026 | /* | |
2027 | * No waiters - kernel unlocks the futex: | |
2028 | */ | |
e3f2ddea IM |
2029 | if (!(uval & FUTEX_OWNER_DIED)) { |
2030 | ret = unlock_futex_pi(uaddr, uval); | |
2031 | if (ret == -EFAULT) | |
2032 | goto pi_faulted; | |
2033 | } | |
c87e2837 IM |
2034 | |
2035 | out_unlock: | |
2036 | spin_unlock(&hb->lock); | |
38d47c1b | 2037 | put_futex_key(fshared, &key); |
c87e2837 | 2038 | |
42d35d48 | 2039 | out: |
c87e2837 IM |
2040 | return ret; |
2041 | ||
2042 | pi_faulted: | |
778e9a9c | 2043 | spin_unlock(&hb->lock); |
e4dc5b7a | 2044 | put_futex_key(fshared, &key); |
c87e2837 | 2045 | |
d0725992 | 2046 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2047 | if (!ret) |
c87e2837 IM |
2048 | goto retry; |
2049 | ||
1da177e4 LT |
2050 | return ret; |
2051 | } | |
2052 | ||
52400ba9 DH |
2053 | /** |
2054 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2055 | * @hb: the hash_bucket futex_q was original enqueued on | |
2056 | * @q: the futex_q woken while waiting to be requeued | |
2057 | * @key2: the futex_key of the requeue target futex | |
2058 | * @timeout: the timeout associated with the wait (NULL if none) | |
2059 | * | |
2060 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2061 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2062 | * the wakeup and return the appropriate error code to the caller. Must be | |
2063 | * called with the hb lock held. | |
2064 | * | |
2065 | * Returns | |
2066 | * 0 - no early wakeup detected | |
1c840c14 | 2067 | * <0 - -ETIMEDOUT or -ERESTARTNOINTR |
52400ba9 DH |
2068 | */ |
2069 | static inline | |
2070 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2071 | struct futex_q *q, union futex_key *key2, | |
2072 | struct hrtimer_sleeper *timeout) | |
2073 | { | |
2074 | int ret = 0; | |
2075 | ||
2076 | /* | |
2077 | * With the hb lock held, we avoid races while we process the wakeup. | |
2078 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2079 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2080 | * It can't be requeued from uaddr2 to something else since we don't | |
2081 | * support a PI aware source futex for requeue. | |
2082 | */ | |
2083 | if (!match_futex(&q->key, key2)) { | |
2084 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2085 | /* | |
2086 | * We were woken prior to requeue by a timeout or a signal. | |
2087 | * Unqueue the futex_q and determine which it was. | |
2088 | */ | |
2089 | plist_del(&q->list, &q->list.plist); | |
2090 | drop_futex_key_refs(&q->key); | |
2091 | ||
2092 | if (timeout && !timeout->task) | |
2093 | ret = -ETIMEDOUT; | |
1c840c14 TG |
2094 | else |
2095 | ret = -ERESTARTNOINTR; | |
52400ba9 DH |
2096 | } |
2097 | return ret; | |
2098 | } | |
2099 | ||
2100 | /** | |
2101 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
2102 | * @uaddr: the futex we initialyl wait on (non-pi) | |
2103 | * @fshared: whether the futexes are shared (1) or not (0). They must be | |
2104 | * the same type, no requeueing from private to shared, etc. | |
2105 | * @val: the expected value of uaddr | |
2106 | * @abs_time: absolute timeout | |
2107 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all. | |
2108 | * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0) | |
2109 | * @uaddr2: the pi futex we will take prior to returning to user-space | |
2110 | * | |
2111 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
2112 | * uaddr2 which must be PI aware. Normal wakeup will wake on uaddr2 and | |
2113 | * complete the acquisition of the rt_mutex prior to returning to userspace. | |
2114 | * This ensures the rt_mutex maintains an owner when it has waiters; without | |
2115 | * one, the pi logic wouldn't know which task to boost/deboost, if there was a | |
2116 | * need to. | |
2117 | * | |
2118 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
2119 | * via the following: | |
2120 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() | |
2121 | * 2) wakeup on uaddr2 after a requeue and subsequent unlock | |
2122 | * 3) signal (before or after requeue) | |
2123 | * 4) timeout (before or after requeue) | |
2124 | * | |
2125 | * If 3, we setup a restart_block with futex_wait_requeue_pi() as the function. | |
2126 | * | |
2127 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2128 | * 5) successful lock | |
2129 | * 6) signal | |
2130 | * 7) timeout | |
2131 | * 8) other lock acquisition failure | |
2132 | * | |
2133 | * If 6, we setup a restart_block with futex_lock_pi() as the function. | |
2134 | * | |
2135 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2136 | * | |
2137 | * Returns: | |
2138 | * 0 - On success | |
2139 | * <0 - On error | |
2140 | */ | |
2141 | static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared, | |
2142 | u32 val, ktime_t *abs_time, u32 bitset, | |
2143 | int clockrt, u32 __user *uaddr2) | |
2144 | { | |
2145 | struct hrtimer_sleeper timeout, *to = NULL; | |
2146 | struct rt_mutex_waiter rt_waiter; | |
2147 | struct rt_mutex *pi_mutex = NULL; | |
52400ba9 DH |
2148 | struct futex_hash_bucket *hb; |
2149 | union futex_key key2; | |
2150 | struct futex_q q; | |
2151 | int res, ret; | |
52400ba9 DH |
2152 | |
2153 | if (!bitset) | |
2154 | return -EINVAL; | |
2155 | ||
2156 | if (abs_time) { | |
2157 | to = &timeout; | |
2158 | hrtimer_init_on_stack(&to->timer, clockrt ? CLOCK_REALTIME : | |
2159 | CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
2160 | hrtimer_init_sleeper(to, current); | |
2161 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2162 | current->timer_slack_ns); | |
2163 | } | |
2164 | ||
2165 | /* | |
2166 | * The waiter is allocated on our stack, manipulated by the requeue | |
2167 | * code while we sleep on uaddr. | |
2168 | */ | |
2169 | debug_rt_mutex_init_waiter(&rt_waiter); | |
2170 | rt_waiter.task = NULL; | |
2171 | ||
2172 | q.pi_state = NULL; | |
2173 | q.bitset = bitset; | |
2174 | q.rt_waiter = &rt_waiter; | |
2175 | ||
2176 | key2 = FUTEX_KEY_INIT; | |
521c1808 | 2177 | ret = get_futex_key(uaddr2, fshared, &key2, VERIFY_WRITE); |
52400ba9 DH |
2178 | if (unlikely(ret != 0)) |
2179 | goto out; | |
2180 | ||
2181 | /* Prepare to wait on uaddr. */ | |
2182 | ret = futex_wait_setup(uaddr, val, fshared, &q, &hb); | |
c8b15a70 TG |
2183 | if (ret) |
2184 | goto out_key2; | |
52400ba9 DH |
2185 | |
2186 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ | |
f1a11e05 | 2187 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2188 | |
2189 | spin_lock(&hb->lock); | |
2190 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2191 | spin_unlock(&hb->lock); | |
2192 | if (ret) | |
2193 | goto out_put_keys; | |
2194 | ||
2195 | /* | |
2196 | * In order for us to be here, we know our q.key == key2, and since | |
2197 | * we took the hb->lock above, we also know that futex_requeue() has | |
2198 | * completed and we no longer have to concern ourselves with a wakeup | |
2199 | * race with the atomic proxy lock acquition by the requeue code. | |
2200 | */ | |
2201 | ||
2202 | /* Check if the requeue code acquired the second futex for us. */ | |
2203 | if (!q.rt_waiter) { | |
2204 | /* | |
2205 | * Got the lock. We might not be the anticipated owner if we | |
2206 | * did a lock-steal - fix up the PI-state in that case. | |
2207 | */ | |
2208 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2209 | spin_lock(q.lock_ptr); | |
2210 | ret = fixup_pi_state_owner(uaddr2, &q, current, | |
2211 | fshared); | |
2212 | spin_unlock(q.lock_ptr); | |
2213 | } | |
2214 | } else { | |
2215 | /* | |
2216 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2217 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2218 | * the pi_state. | |
2219 | */ | |
2220 | WARN_ON(!&q.pi_state); | |
2221 | pi_mutex = &q.pi_state->pi_mutex; | |
2222 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1); | |
2223 | debug_rt_mutex_free_waiter(&rt_waiter); | |
2224 | ||
2225 | spin_lock(q.lock_ptr); | |
2226 | /* | |
2227 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2228 | * haven't already. | |
2229 | */ | |
2230 | res = fixup_owner(uaddr2, fshared, &q, !ret); | |
2231 | /* | |
2232 | * If fixup_owner() returned an error, proprogate that. If it | |
2233 | * acquired the lock, clear our -ETIMEDOUT or -EINTR. | |
2234 | */ | |
2235 | if (res) | |
2236 | ret = (res < 0) ? res : 0; | |
2237 | ||
2238 | /* Unqueue and drop the lock. */ | |
2239 | unqueue_me_pi(&q); | |
2240 | } | |
2241 | ||
2242 | /* | |
2243 | * If fixup_pi_state_owner() faulted and was unable to handle the | |
2244 | * fault, unlock the rt_mutex and return the fault to userspace. | |
2245 | */ | |
2246 | if (ret == -EFAULT) { | |
2247 | if (rt_mutex_owner(pi_mutex) == current) | |
2248 | rt_mutex_unlock(pi_mutex); | |
2249 | } else if (ret == -EINTR) { | |
52400ba9 | 2250 | /* |
2070887f TG |
2251 | * We've already been requeued, but we have no way to |
2252 | * restart by calling futex_lock_pi() directly. We | |
2253 | * could restart the syscall, but that will look at | |
2254 | * the user space value and return right away. So we | |
2255 | * drop back with EWOULDBLOCK to tell user space that | |
2256 | * "val" has been changed. That's the same what the | |
2257 | * restart of the syscall would do in | |
2258 | * futex_wait_setup(). | |
52400ba9 | 2259 | */ |
2070887f | 2260 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2261 | } |
2262 | ||
2263 | out_put_keys: | |
2264 | put_futex_key(fshared, &q.key); | |
c8b15a70 | 2265 | out_key2: |
52400ba9 DH |
2266 | put_futex_key(fshared, &key2); |
2267 | ||
2268 | out: | |
2269 | if (to) { | |
2270 | hrtimer_cancel(&to->timer); | |
2271 | destroy_hrtimer_on_stack(&to->timer); | |
2272 | } | |
2273 | return ret; | |
2274 | } | |
2275 | ||
0771dfef IM |
2276 | /* |
2277 | * Support for robust futexes: the kernel cleans up held futexes at | |
2278 | * thread exit time. | |
2279 | * | |
2280 | * Implementation: user-space maintains a per-thread list of locks it | |
2281 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2282 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2283 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2284 | * always manipulated with the lock held, so the list is private and |
2285 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2286 | * field, to allow the kernel to clean up if the thread dies after | |
2287 | * acquiring the lock, but just before it could have added itself to | |
2288 | * the list. There can only be one such pending lock. | |
2289 | */ | |
2290 | ||
2291 | /** | |
2292 | * sys_set_robust_list - set the robust-futex list head of a task | |
2293 | * @head: pointer to the list-head | |
2294 | * @len: length of the list-head, as userspace expects | |
2295 | */ | |
836f92ad HC |
2296 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2297 | size_t, len) | |
0771dfef | 2298 | { |
a0c1e907 TG |
2299 | if (!futex_cmpxchg_enabled) |
2300 | return -ENOSYS; | |
0771dfef IM |
2301 | /* |
2302 | * The kernel knows only one size for now: | |
2303 | */ | |
2304 | if (unlikely(len != sizeof(*head))) | |
2305 | return -EINVAL; | |
2306 | ||
2307 | current->robust_list = head; | |
2308 | ||
2309 | return 0; | |
2310 | } | |
2311 | ||
2312 | /** | |
2313 | * sys_get_robust_list - get the robust-futex list head of a task | |
2314 | * @pid: pid of the process [zero for current task] | |
2315 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
2316 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
2317 | */ | |
836f92ad HC |
2318 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
2319 | struct robust_list_head __user * __user *, head_ptr, | |
2320 | size_t __user *, len_ptr) | |
0771dfef | 2321 | { |
ba46df98 | 2322 | struct robust_list_head __user *head; |
0771dfef | 2323 | unsigned long ret; |
c69e8d9c | 2324 | const struct cred *cred = current_cred(), *pcred; |
0771dfef | 2325 | |
a0c1e907 TG |
2326 | if (!futex_cmpxchg_enabled) |
2327 | return -ENOSYS; | |
2328 | ||
0771dfef IM |
2329 | if (!pid) |
2330 | head = current->robust_list; | |
2331 | else { | |
2332 | struct task_struct *p; | |
2333 | ||
2334 | ret = -ESRCH; | |
aaa2a97e | 2335 | rcu_read_lock(); |
228ebcbe | 2336 | p = find_task_by_vpid(pid); |
0771dfef IM |
2337 | if (!p) |
2338 | goto err_unlock; | |
2339 | ret = -EPERM; | |
c69e8d9c DH |
2340 | pcred = __task_cred(p); |
2341 | if (cred->euid != pcred->euid && | |
2342 | cred->euid != pcred->uid && | |
76aac0e9 | 2343 | !capable(CAP_SYS_PTRACE)) |
0771dfef IM |
2344 | goto err_unlock; |
2345 | head = p->robust_list; | |
aaa2a97e | 2346 | rcu_read_unlock(); |
0771dfef IM |
2347 | } |
2348 | ||
2349 | if (put_user(sizeof(*head), len_ptr)) | |
2350 | return -EFAULT; | |
2351 | return put_user(head, head_ptr); | |
2352 | ||
2353 | err_unlock: | |
aaa2a97e | 2354 | rcu_read_unlock(); |
0771dfef IM |
2355 | |
2356 | return ret; | |
2357 | } | |
2358 | ||
2359 | /* | |
2360 | * Process a futex-list entry, check whether it's owned by the | |
2361 | * dying task, and do notification if so: | |
2362 | */ | |
e3f2ddea | 2363 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 2364 | { |
e3f2ddea | 2365 | u32 uval, nval, mval; |
0771dfef | 2366 | |
8f17d3a5 IM |
2367 | retry: |
2368 | if (get_user(uval, uaddr)) | |
0771dfef IM |
2369 | return -1; |
2370 | ||
b488893a | 2371 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
2372 | /* |
2373 | * Ok, this dying thread is truly holding a futex | |
2374 | * of interest. Set the OWNER_DIED bit atomically | |
2375 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
2376 | * set, wake up a waiter (if any). (We have to do a | |
2377 | * futex_wake() even if OWNER_DIED is already set - | |
2378 | * to handle the rare but possible case of recursive | |
2379 | * thread-death.) The rest of the cleanup is done in | |
2380 | * userspace. | |
2381 | */ | |
e3f2ddea IM |
2382 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
2383 | nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval); | |
2384 | ||
c87e2837 IM |
2385 | if (nval == -EFAULT) |
2386 | return -1; | |
2387 | ||
2388 | if (nval != uval) | |
8f17d3a5 | 2389 | goto retry; |
0771dfef | 2390 | |
e3f2ddea IM |
2391 | /* |
2392 | * Wake robust non-PI futexes here. The wakeup of | |
2393 | * PI futexes happens in exit_pi_state(): | |
2394 | */ | |
36cf3b5c | 2395 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 2396 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
2397 | } |
2398 | return 0; | |
2399 | } | |
2400 | ||
e3f2ddea IM |
2401 | /* |
2402 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
2403 | */ | |
2404 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 AV |
2405 | struct robust_list __user * __user *head, |
2406 | int *pi) | |
e3f2ddea IM |
2407 | { |
2408 | unsigned long uentry; | |
2409 | ||
ba46df98 | 2410 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
2411 | return -EFAULT; |
2412 | ||
ba46df98 | 2413 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
2414 | *pi = uentry & 1; |
2415 | ||
2416 | return 0; | |
2417 | } | |
2418 | ||
0771dfef IM |
2419 | /* |
2420 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
2421 | * and mark any locks found there dead, and notify any waiters. | |
2422 | * | |
2423 | * We silently return on any sign of list-walking problem. | |
2424 | */ | |
2425 | void exit_robust_list(struct task_struct *curr) | |
2426 | { | |
2427 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e MS |
2428 | struct robust_list __user *entry, *next_entry, *pending; |
2429 | unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip; | |
0771dfef | 2430 | unsigned long futex_offset; |
9f96cb1e | 2431 | int rc; |
0771dfef | 2432 | |
a0c1e907 TG |
2433 | if (!futex_cmpxchg_enabled) |
2434 | return; | |
2435 | ||
0771dfef IM |
2436 | /* |
2437 | * Fetch the list head (which was registered earlier, via | |
2438 | * sys_set_robust_list()): | |
2439 | */ | |
e3f2ddea | 2440 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
2441 | return; |
2442 | /* | |
2443 | * Fetch the relative futex offset: | |
2444 | */ | |
2445 | if (get_user(futex_offset, &head->futex_offset)) | |
2446 | return; | |
2447 | /* | |
2448 | * Fetch any possibly pending lock-add first, and handle it | |
2449 | * if it exists: | |
2450 | */ | |
e3f2ddea | 2451 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 2452 | return; |
e3f2ddea | 2453 | |
9f96cb1e | 2454 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 2455 | while (entry != &head->list) { |
9f96cb1e MS |
2456 | /* |
2457 | * Fetch the next entry in the list before calling | |
2458 | * handle_futex_death: | |
2459 | */ | |
2460 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
2461 | /* |
2462 | * A pending lock might already be on the list, so | |
c87e2837 | 2463 | * don't process it twice: |
0771dfef IM |
2464 | */ |
2465 | if (entry != pending) | |
ba46df98 | 2466 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 2467 | curr, pi)) |
0771dfef | 2468 | return; |
9f96cb1e | 2469 | if (rc) |
0771dfef | 2470 | return; |
9f96cb1e MS |
2471 | entry = next_entry; |
2472 | pi = next_pi; | |
0771dfef IM |
2473 | /* |
2474 | * Avoid excessively long or circular lists: | |
2475 | */ | |
2476 | if (!--limit) | |
2477 | break; | |
2478 | ||
2479 | cond_resched(); | |
2480 | } | |
9f96cb1e MS |
2481 | |
2482 | if (pending) | |
2483 | handle_futex_death((void __user *)pending + futex_offset, | |
2484 | curr, pip); | |
0771dfef IM |
2485 | } |
2486 | ||
c19384b5 | 2487 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 2488 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 2489 | { |
1acdac10 | 2490 | int clockrt, ret = -ENOSYS; |
34f01cc1 | 2491 | int cmd = op & FUTEX_CMD_MASK; |
c2f9f201 | 2492 | int fshared = 0; |
34f01cc1 ED |
2493 | |
2494 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
c2f9f201 | 2495 | fshared = 1; |
1da177e4 | 2496 | |
1acdac10 | 2497 | clockrt = op & FUTEX_CLOCK_REALTIME; |
52400ba9 | 2498 | if (clockrt && cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI) |
1acdac10 | 2499 | return -ENOSYS; |
1da177e4 | 2500 | |
34f01cc1 | 2501 | switch (cmd) { |
1da177e4 | 2502 | case FUTEX_WAIT: |
cd689985 TG |
2503 | val3 = FUTEX_BITSET_MATCH_ANY; |
2504 | case FUTEX_WAIT_BITSET: | |
1acdac10 | 2505 | ret = futex_wait(uaddr, fshared, val, timeout, val3, clockrt); |
1da177e4 LT |
2506 | break; |
2507 | case FUTEX_WAKE: | |
cd689985 TG |
2508 | val3 = FUTEX_BITSET_MATCH_ANY; |
2509 | case FUTEX_WAKE_BITSET: | |
2510 | ret = futex_wake(uaddr, fshared, val, val3); | |
1da177e4 | 2511 | break; |
1da177e4 | 2512 | case FUTEX_REQUEUE: |
52400ba9 | 2513 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL, 0); |
1da177e4 LT |
2514 | break; |
2515 | case FUTEX_CMP_REQUEUE: | |
52400ba9 DH |
2516 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3, |
2517 | 0); | |
1da177e4 | 2518 | break; |
4732efbe | 2519 | case FUTEX_WAKE_OP: |
34f01cc1 | 2520 | ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3); |
4732efbe | 2521 | break; |
c87e2837 | 2522 | case FUTEX_LOCK_PI: |
a0c1e907 TG |
2523 | if (futex_cmpxchg_enabled) |
2524 | ret = futex_lock_pi(uaddr, fshared, val, timeout, 0); | |
c87e2837 IM |
2525 | break; |
2526 | case FUTEX_UNLOCK_PI: | |
a0c1e907 TG |
2527 | if (futex_cmpxchg_enabled) |
2528 | ret = futex_unlock_pi(uaddr, fshared); | |
c87e2837 IM |
2529 | break; |
2530 | case FUTEX_TRYLOCK_PI: | |
a0c1e907 TG |
2531 | if (futex_cmpxchg_enabled) |
2532 | ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1); | |
c87e2837 | 2533 | break; |
52400ba9 DH |
2534 | case FUTEX_WAIT_REQUEUE_PI: |
2535 | val3 = FUTEX_BITSET_MATCH_ANY; | |
2536 | ret = futex_wait_requeue_pi(uaddr, fshared, val, timeout, val3, | |
2537 | clockrt, uaddr2); | |
2538 | break; | |
52400ba9 DH |
2539 | case FUTEX_CMP_REQUEUE_PI: |
2540 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3, | |
2541 | 1); | |
2542 | break; | |
1da177e4 LT |
2543 | default: |
2544 | ret = -ENOSYS; | |
2545 | } | |
2546 | return ret; | |
2547 | } | |
2548 | ||
2549 | ||
17da2bd9 HC |
2550 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
2551 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
2552 | u32, val3) | |
1da177e4 | 2553 | { |
c19384b5 PP |
2554 | struct timespec ts; |
2555 | ktime_t t, *tp = NULL; | |
e2970f2f | 2556 | u32 val2 = 0; |
34f01cc1 | 2557 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 2558 | |
cd689985 | 2559 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
2560 | cmd == FUTEX_WAIT_BITSET || |
2561 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
c19384b5 | 2562 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 2563 | return -EFAULT; |
c19384b5 | 2564 | if (!timespec_valid(&ts)) |
9741ef96 | 2565 | return -EINVAL; |
c19384b5 PP |
2566 | |
2567 | t = timespec_to_ktime(ts); | |
34f01cc1 | 2568 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 2569 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 2570 | tp = &t; |
1da177e4 LT |
2571 | } |
2572 | /* | |
52400ba9 | 2573 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 2574 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 2575 | */ |
f54f0986 | 2576 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 2577 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 2578 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 2579 | |
c19384b5 | 2580 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
2581 | } |
2582 | ||
f6d107fb | 2583 | static int __init futex_init(void) |
1da177e4 | 2584 | { |
a0c1e907 | 2585 | u32 curval; |
3e4ab747 | 2586 | int i; |
95362fa9 | 2587 | |
a0c1e907 TG |
2588 | /* |
2589 | * This will fail and we want it. Some arch implementations do | |
2590 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
2591 | * functionality. We want to know that before we call in any | |
2592 | * of the complex code paths. Also we want to prevent | |
2593 | * registration of robust lists in that case. NULL is | |
2594 | * guaranteed to fault and we get -EFAULT on functional | |
2595 | * implementation, the non functional ones will return | |
2596 | * -ENOSYS. | |
2597 | */ | |
2598 | curval = cmpxchg_futex_value_locked(NULL, 0, 0); | |
2599 | if (curval == -EFAULT) | |
2600 | futex_cmpxchg_enabled = 1; | |
2601 | ||
3e4ab747 TG |
2602 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { |
2603 | plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock); | |
2604 | spin_lock_init(&futex_queues[i].lock); | |
2605 | } | |
2606 | ||
1da177e4 LT |
2607 | return 0; |
2608 | } | |
f6d107fb | 2609 | __initcall(futex_init); |