]>
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 | ||
84bc4af5 DH |
118 | /* The expected requeue pi target futex key: */ |
119 | union futex_key *requeue_pi_key; | |
120 | ||
cd689985 TG |
121 | /* Bitset for the optional bitmasked wakeup */ |
122 | u32 bitset; | |
1da177e4 LT |
123 | }; |
124 | ||
125 | /* | |
b2d0994b DH |
126 | * Hash buckets are shared by all the futex_keys that hash to the same |
127 | * location. Each key may have multiple futex_q structures, one for each task | |
128 | * waiting on a futex. | |
1da177e4 LT |
129 | */ |
130 | struct futex_hash_bucket { | |
ec92d082 PP |
131 | spinlock_t lock; |
132 | struct plist_head chain; | |
1da177e4 LT |
133 | }; |
134 | ||
135 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; | |
136 | ||
1da177e4 LT |
137 | /* |
138 | * We hash on the keys returned from get_futex_key (see below). | |
139 | */ | |
140 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
141 | { | |
142 | u32 hash = jhash2((u32*)&key->both.word, | |
143 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
144 | key->both.offset); | |
145 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; | |
146 | } | |
147 | ||
148 | /* | |
149 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
150 | */ | |
151 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
152 | { | |
153 | return (key1->both.word == key2->both.word | |
154 | && key1->both.ptr == key2->both.ptr | |
155 | && key1->both.offset == key2->both.offset); | |
156 | } | |
157 | ||
38d47c1b PZ |
158 | /* |
159 | * Take a reference to the resource addressed by a key. | |
160 | * Can be called while holding spinlocks. | |
161 | * | |
162 | */ | |
163 | static void get_futex_key_refs(union futex_key *key) | |
164 | { | |
165 | if (!key->both.ptr) | |
166 | return; | |
167 | ||
168 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
169 | case FUT_OFF_INODE: | |
170 | atomic_inc(&key->shared.inode->i_count); | |
171 | break; | |
172 | case FUT_OFF_MMSHARED: | |
173 | atomic_inc(&key->private.mm->mm_count); | |
174 | break; | |
175 | } | |
176 | } | |
177 | ||
178 | /* | |
179 | * Drop a reference to the resource addressed by a key. | |
180 | * The hash bucket spinlock must not be held. | |
181 | */ | |
182 | static void drop_futex_key_refs(union futex_key *key) | |
183 | { | |
90621c40 DH |
184 | if (!key->both.ptr) { |
185 | /* If we're here then we tried to put a key we failed to get */ | |
186 | WARN_ON_ONCE(1); | |
38d47c1b | 187 | return; |
90621c40 | 188 | } |
38d47c1b PZ |
189 | |
190 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
191 | case FUT_OFF_INODE: | |
192 | iput(key->shared.inode); | |
193 | break; | |
194 | case FUT_OFF_MMSHARED: | |
195 | mmdrop(key->private.mm); | |
196 | break; | |
197 | } | |
198 | } | |
199 | ||
34f01cc1 ED |
200 | /** |
201 | * get_futex_key - Get parameters which are the keys for a futex. | |
202 | * @uaddr: virtual address of the futex | |
b2d0994b | 203 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED |
34f01cc1 | 204 | * @key: address where result is stored. |
64d1304a | 205 | * @rw: mapping needs to be read/write (values: VERIFY_READ, VERIFY_WRITE) |
34f01cc1 ED |
206 | * |
207 | * Returns a negative error code or 0 | |
208 | * The key words are stored in *key on success. | |
1da177e4 | 209 | * |
f3a43f3f | 210 | * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode, |
1da177e4 LT |
211 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
212 | * We can usually work out the index without swapping in the page. | |
213 | * | |
b2d0994b | 214 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 215 | */ |
64d1304a TG |
216 | static int |
217 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) | |
1da177e4 | 218 | { |
e2970f2f | 219 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 220 | struct mm_struct *mm = current->mm; |
1da177e4 LT |
221 | struct page *page; |
222 | int err; | |
223 | ||
224 | /* | |
225 | * The futex address must be "naturally" aligned. | |
226 | */ | |
e2970f2f | 227 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 228 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 229 | return -EINVAL; |
e2970f2f | 230 | address -= key->both.offset; |
1da177e4 | 231 | |
34f01cc1 ED |
232 | /* |
233 | * PROCESS_PRIVATE futexes are fast. | |
234 | * As the mm cannot disappear under us and the 'key' only needs | |
235 | * virtual address, we dont even have to find the underlying vma. | |
236 | * Note : We do have to check 'uaddr' is a valid user address, | |
237 | * but access_ok() should be faster than find_vma() | |
238 | */ | |
239 | if (!fshared) { | |
64d1304a | 240 | if (unlikely(!access_ok(rw, uaddr, sizeof(u32)))) |
34f01cc1 ED |
241 | return -EFAULT; |
242 | key->private.mm = mm; | |
243 | key->private.address = address; | |
42569c39 | 244 | get_futex_key_refs(key); |
34f01cc1 ED |
245 | return 0; |
246 | } | |
1da177e4 | 247 | |
38d47c1b | 248 | again: |
64d1304a | 249 | err = get_user_pages_fast(address, 1, rw == VERIFY_WRITE, &page); |
38d47c1b PZ |
250 | if (err < 0) |
251 | return err; | |
252 | ||
ce2ae53b | 253 | page = compound_head(page); |
38d47c1b PZ |
254 | lock_page(page); |
255 | if (!page->mapping) { | |
256 | unlock_page(page); | |
257 | put_page(page); | |
258 | goto again; | |
259 | } | |
1da177e4 LT |
260 | |
261 | /* | |
262 | * Private mappings are handled in a simple way. | |
263 | * | |
264 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
265 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 266 | * the object not the particular process. |
1da177e4 | 267 | */ |
38d47c1b PZ |
268 | if (PageAnon(page)) { |
269 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ | |
1da177e4 | 270 | key->private.mm = mm; |
e2970f2f | 271 | key->private.address = address; |
38d47c1b PZ |
272 | } else { |
273 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ | |
274 | key->shared.inode = page->mapping->host; | |
275 | key->shared.pgoff = page->index; | |
1da177e4 LT |
276 | } |
277 | ||
38d47c1b | 278 | get_futex_key_refs(key); |
1da177e4 | 279 | |
38d47c1b PZ |
280 | unlock_page(page); |
281 | put_page(page); | |
282 | return 0; | |
1da177e4 LT |
283 | } |
284 | ||
38d47c1b | 285 | static inline |
c2f9f201 | 286 | void put_futex_key(int fshared, union futex_key *key) |
1da177e4 | 287 | { |
38d47c1b | 288 | drop_futex_key_refs(key); |
1da177e4 LT |
289 | } |
290 | ||
d0725992 TG |
291 | /* |
292 | * fault_in_user_writeable - fault in user address and verify RW access | |
293 | * @uaddr: pointer to faulting user space address | |
294 | * | |
295 | * Slow path to fixup the fault we just took in the atomic write | |
296 | * access to @uaddr. | |
297 | * | |
298 | * We have no generic implementation of a non destructive write to the | |
299 | * user address. We know that we faulted in the atomic pagefault | |
300 | * disabled section so we can as well avoid the #PF overhead by | |
301 | * calling get_user_pages() right away. | |
302 | */ | |
303 | static int fault_in_user_writeable(u32 __user *uaddr) | |
304 | { | |
305 | int ret = get_user_pages(current, current->mm, (unsigned long)uaddr, | |
aa715284 | 306 | 1, 1, 0, NULL, NULL); |
d0725992 TG |
307 | return ret < 0 ? ret : 0; |
308 | } | |
309 | ||
4b1c486b DH |
310 | /** |
311 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
312 | * @hb: the hash bucket the futex_q's reside in | |
313 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
314 | * | |
315 | * Must be called with the hb lock held. | |
316 | */ | |
317 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
318 | union futex_key *key) | |
319 | { | |
320 | struct futex_q *this; | |
321 | ||
322 | plist_for_each_entry(this, &hb->chain, list) { | |
323 | if (match_futex(&this->key, key)) | |
324 | return this; | |
325 | } | |
326 | return NULL; | |
327 | } | |
328 | ||
36cf3b5c TG |
329 | static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval) |
330 | { | |
331 | u32 curval; | |
332 | ||
333 | pagefault_disable(); | |
334 | curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval); | |
335 | pagefault_enable(); | |
336 | ||
337 | return curval; | |
338 | } | |
339 | ||
340 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
341 | { |
342 | int ret; | |
343 | ||
a866374a | 344 | pagefault_disable(); |
e2970f2f | 345 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 346 | pagefault_enable(); |
1da177e4 LT |
347 | |
348 | return ret ? -EFAULT : 0; | |
349 | } | |
350 | ||
c87e2837 IM |
351 | |
352 | /* | |
353 | * PI code: | |
354 | */ | |
355 | static int refill_pi_state_cache(void) | |
356 | { | |
357 | struct futex_pi_state *pi_state; | |
358 | ||
359 | if (likely(current->pi_state_cache)) | |
360 | return 0; | |
361 | ||
4668edc3 | 362 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
363 | |
364 | if (!pi_state) | |
365 | return -ENOMEM; | |
366 | ||
c87e2837 IM |
367 | INIT_LIST_HEAD(&pi_state->list); |
368 | /* pi_mutex gets initialized later */ | |
369 | pi_state->owner = NULL; | |
370 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 371 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
372 | |
373 | current->pi_state_cache = pi_state; | |
374 | ||
375 | return 0; | |
376 | } | |
377 | ||
378 | static struct futex_pi_state * alloc_pi_state(void) | |
379 | { | |
380 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
381 | ||
382 | WARN_ON(!pi_state); | |
383 | current->pi_state_cache = NULL; | |
384 | ||
385 | return pi_state; | |
386 | } | |
387 | ||
388 | static void free_pi_state(struct futex_pi_state *pi_state) | |
389 | { | |
390 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
391 | return; | |
392 | ||
393 | /* | |
394 | * If pi_state->owner is NULL, the owner is most probably dying | |
395 | * and has cleaned up the pi_state already | |
396 | */ | |
397 | if (pi_state->owner) { | |
398 | spin_lock_irq(&pi_state->owner->pi_lock); | |
399 | list_del_init(&pi_state->list); | |
400 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
401 | ||
402 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
403 | } | |
404 | ||
405 | if (current->pi_state_cache) | |
406 | kfree(pi_state); | |
407 | else { | |
408 | /* | |
409 | * pi_state->list is already empty. | |
410 | * clear pi_state->owner. | |
411 | * refcount is at 0 - put it back to 1. | |
412 | */ | |
413 | pi_state->owner = NULL; | |
414 | atomic_set(&pi_state->refcount, 1); | |
415 | current->pi_state_cache = pi_state; | |
416 | } | |
417 | } | |
418 | ||
419 | /* | |
420 | * Look up the task based on what TID userspace gave us. | |
421 | * We dont trust it. | |
422 | */ | |
423 | static struct task_struct * futex_find_get_task(pid_t pid) | |
424 | { | |
425 | struct task_struct *p; | |
c69e8d9c | 426 | const struct cred *cred = current_cred(), *pcred; |
c87e2837 | 427 | |
d359b549 | 428 | rcu_read_lock(); |
228ebcbe | 429 | p = find_task_by_vpid(pid); |
c69e8d9c | 430 | if (!p) { |
a06381fe | 431 | p = ERR_PTR(-ESRCH); |
c69e8d9c DH |
432 | } else { |
433 | pcred = __task_cred(p); | |
434 | if (cred->euid != pcred->euid && | |
435 | cred->euid != pcred->uid) | |
436 | p = ERR_PTR(-ESRCH); | |
437 | else | |
438 | get_task_struct(p); | |
439 | } | |
a06381fe | 440 | |
d359b549 | 441 | rcu_read_unlock(); |
c87e2837 IM |
442 | |
443 | return p; | |
444 | } | |
445 | ||
446 | /* | |
447 | * This task is holding PI mutexes at exit time => bad. | |
448 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
449 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
450 | */ | |
451 | void exit_pi_state_list(struct task_struct *curr) | |
452 | { | |
c87e2837 IM |
453 | struct list_head *next, *head = &curr->pi_state_list; |
454 | struct futex_pi_state *pi_state; | |
627371d7 | 455 | struct futex_hash_bucket *hb; |
38d47c1b | 456 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 457 | |
a0c1e907 TG |
458 | if (!futex_cmpxchg_enabled) |
459 | return; | |
c87e2837 IM |
460 | /* |
461 | * We are a ZOMBIE and nobody can enqueue itself on | |
462 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 463 | * versus waiters unqueueing themselves: |
c87e2837 IM |
464 | */ |
465 | spin_lock_irq(&curr->pi_lock); | |
466 | while (!list_empty(head)) { | |
467 | ||
468 | next = head->next; | |
469 | pi_state = list_entry(next, struct futex_pi_state, list); | |
470 | key = pi_state->key; | |
627371d7 | 471 | hb = hash_futex(&key); |
c87e2837 IM |
472 | spin_unlock_irq(&curr->pi_lock); |
473 | ||
c87e2837 IM |
474 | spin_lock(&hb->lock); |
475 | ||
476 | spin_lock_irq(&curr->pi_lock); | |
627371d7 IM |
477 | /* |
478 | * We dropped the pi-lock, so re-check whether this | |
479 | * task still owns the PI-state: | |
480 | */ | |
c87e2837 IM |
481 | if (head->next != next) { |
482 | spin_unlock(&hb->lock); | |
483 | continue; | |
484 | } | |
485 | ||
c87e2837 | 486 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
487 | WARN_ON(list_empty(&pi_state->list)); |
488 | list_del_init(&pi_state->list); | |
c87e2837 IM |
489 | pi_state->owner = NULL; |
490 | spin_unlock_irq(&curr->pi_lock); | |
491 | ||
492 | rt_mutex_unlock(&pi_state->pi_mutex); | |
493 | ||
494 | spin_unlock(&hb->lock); | |
495 | ||
496 | spin_lock_irq(&curr->pi_lock); | |
497 | } | |
498 | spin_unlock_irq(&curr->pi_lock); | |
499 | } | |
500 | ||
501 | static int | |
d0aa7a70 PP |
502 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
503 | union futex_key *key, struct futex_pi_state **ps) | |
c87e2837 IM |
504 | { |
505 | struct futex_pi_state *pi_state = NULL; | |
506 | struct futex_q *this, *next; | |
ec92d082 | 507 | struct plist_head *head; |
c87e2837 | 508 | struct task_struct *p; |
778e9a9c | 509 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 IM |
510 | |
511 | head = &hb->chain; | |
512 | ||
ec92d082 | 513 | plist_for_each_entry_safe(this, next, head, list) { |
d0aa7a70 | 514 | if (match_futex(&this->key, key)) { |
c87e2837 IM |
515 | /* |
516 | * Another waiter already exists - bump up | |
517 | * the refcount and return its pi_state: | |
518 | */ | |
519 | pi_state = this->pi_state; | |
06a9ec29 TG |
520 | /* |
521 | * Userspace might have messed up non PI and PI futexes | |
522 | */ | |
523 | if (unlikely(!pi_state)) | |
524 | return -EINVAL; | |
525 | ||
627371d7 | 526 | WARN_ON(!atomic_read(&pi_state->refcount)); |
778e9a9c AK |
527 | WARN_ON(pid && pi_state->owner && |
528 | pi_state->owner->pid != pid); | |
627371d7 | 529 | |
c87e2837 | 530 | atomic_inc(&pi_state->refcount); |
d0aa7a70 | 531 | *ps = pi_state; |
c87e2837 IM |
532 | |
533 | return 0; | |
534 | } | |
535 | } | |
536 | ||
537 | /* | |
e3f2ddea | 538 | * We are the first waiter - try to look up the real owner and attach |
778e9a9c | 539 | * the new pi_state to it, but bail out when TID = 0 |
c87e2837 | 540 | */ |
778e9a9c | 541 | if (!pid) |
e3f2ddea | 542 | return -ESRCH; |
c87e2837 | 543 | p = futex_find_get_task(pid); |
778e9a9c AK |
544 | if (IS_ERR(p)) |
545 | return PTR_ERR(p); | |
546 | ||
547 | /* | |
548 | * We need to look at the task state flags to figure out, | |
549 | * whether the task is exiting. To protect against the do_exit | |
550 | * change of the task flags, we do this protected by | |
551 | * p->pi_lock: | |
552 | */ | |
553 | spin_lock_irq(&p->pi_lock); | |
554 | if (unlikely(p->flags & PF_EXITING)) { | |
555 | /* | |
556 | * The task is on the way out. When PF_EXITPIDONE is | |
557 | * set, we know that the task has finished the | |
558 | * cleanup: | |
559 | */ | |
560 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
561 | ||
562 | spin_unlock_irq(&p->pi_lock); | |
563 | put_task_struct(p); | |
564 | return ret; | |
565 | } | |
c87e2837 IM |
566 | |
567 | pi_state = alloc_pi_state(); | |
568 | ||
569 | /* | |
570 | * Initialize the pi_mutex in locked state and make 'p' | |
571 | * the owner of it: | |
572 | */ | |
573 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
574 | ||
575 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 576 | pi_state->key = *key; |
c87e2837 | 577 | |
627371d7 | 578 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
579 | list_add(&pi_state->list, &p->pi_state_list); |
580 | pi_state->owner = p; | |
581 | spin_unlock_irq(&p->pi_lock); | |
582 | ||
583 | put_task_struct(p); | |
584 | ||
d0aa7a70 | 585 | *ps = pi_state; |
c87e2837 IM |
586 | |
587 | return 0; | |
588 | } | |
589 | ||
1a52084d DH |
590 | /** |
591 | * futex_lock_pi_atomic() - atomic work required to acquire a pi aware futex | |
bab5bc9e DH |
592 | * @uaddr: the pi futex user address |
593 | * @hb: the pi futex hash bucket | |
594 | * @key: the futex key associated with uaddr and hb | |
595 | * @ps: the pi_state pointer where we store the result of the | |
596 | * lookup | |
597 | * @task: the task to perform the atomic lock work for. This will | |
598 | * be "current" except in the case of requeue pi. | |
599 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d DH |
600 | * |
601 | * Returns: | |
602 | * 0 - ready to wait | |
603 | * 1 - acquired the lock | |
604 | * <0 - error | |
605 | * | |
606 | * The hb->lock and futex_key refs shall be held by the caller. | |
607 | */ | |
608 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
609 | union futex_key *key, | |
610 | struct futex_pi_state **ps, | |
bab5bc9e | 611 | struct task_struct *task, int set_waiters) |
1a52084d DH |
612 | { |
613 | int lock_taken, ret, ownerdied = 0; | |
614 | u32 uval, newval, curval; | |
615 | ||
616 | retry: | |
617 | ret = lock_taken = 0; | |
618 | ||
619 | /* | |
620 | * To avoid races, we attempt to take the lock here again | |
621 | * (by doing a 0 -> TID atomic cmpxchg), while holding all | |
622 | * the locks. It will most likely not succeed. | |
623 | */ | |
624 | newval = task_pid_vnr(task); | |
bab5bc9e DH |
625 | if (set_waiters) |
626 | newval |= FUTEX_WAITERS; | |
1a52084d DH |
627 | |
628 | curval = cmpxchg_futex_value_locked(uaddr, 0, newval); | |
629 | ||
630 | if (unlikely(curval == -EFAULT)) | |
631 | return -EFAULT; | |
632 | ||
633 | /* | |
634 | * Detect deadlocks. | |
635 | */ | |
636 | if ((unlikely((curval & FUTEX_TID_MASK) == task_pid_vnr(task)))) | |
637 | return -EDEADLK; | |
638 | ||
639 | /* | |
640 | * Surprise - we got the lock. Just return to userspace: | |
641 | */ | |
642 | if (unlikely(!curval)) | |
643 | return 1; | |
644 | ||
645 | uval = curval; | |
646 | ||
647 | /* | |
648 | * Set the FUTEX_WAITERS flag, so the owner will know it has someone | |
649 | * to wake at the next unlock. | |
650 | */ | |
651 | newval = curval | FUTEX_WAITERS; | |
652 | ||
653 | /* | |
654 | * There are two cases, where a futex might have no owner (the | |
655 | * owner TID is 0): OWNER_DIED. We take over the futex in this | |
656 | * case. We also do an unconditional take over, when the owner | |
657 | * of the futex died. | |
658 | * | |
659 | * This is safe as we are protected by the hash bucket lock ! | |
660 | */ | |
661 | if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { | |
662 | /* Keep the OWNER_DIED bit */ | |
663 | newval = (curval & ~FUTEX_TID_MASK) | task_pid_vnr(task); | |
664 | ownerdied = 0; | |
665 | lock_taken = 1; | |
666 | } | |
667 | ||
668 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); | |
669 | ||
670 | if (unlikely(curval == -EFAULT)) | |
671 | return -EFAULT; | |
672 | if (unlikely(curval != uval)) | |
673 | goto retry; | |
674 | ||
675 | /* | |
676 | * We took the lock due to owner died take over. | |
677 | */ | |
678 | if (unlikely(lock_taken)) | |
679 | return 1; | |
680 | ||
681 | /* | |
682 | * We dont have the lock. Look up the PI state (or create it if | |
683 | * we are the first waiter): | |
684 | */ | |
685 | ret = lookup_pi_state(uval, hb, key, ps); | |
686 | ||
687 | if (unlikely(ret)) { | |
688 | switch (ret) { | |
689 | case -ESRCH: | |
690 | /* | |
691 | * No owner found for this futex. Check if the | |
692 | * OWNER_DIED bit is set to figure out whether | |
693 | * this is a robust futex or not. | |
694 | */ | |
695 | if (get_futex_value_locked(&curval, uaddr)) | |
696 | return -EFAULT; | |
697 | ||
698 | /* | |
699 | * We simply start over in case of a robust | |
700 | * futex. The code above will take the futex | |
701 | * and return happy. | |
702 | */ | |
703 | if (curval & FUTEX_OWNER_DIED) { | |
704 | ownerdied = 1; | |
705 | goto retry; | |
706 | } | |
707 | default: | |
708 | break; | |
709 | } | |
710 | } | |
711 | ||
712 | return ret; | |
713 | } | |
714 | ||
1da177e4 LT |
715 | /* |
716 | * The hash bucket lock must be held when this is called. | |
717 | * Afterwards, the futex_q must not be accessed. | |
718 | */ | |
719 | static void wake_futex(struct futex_q *q) | |
720 | { | |
f1a11e05 TG |
721 | struct task_struct *p = q->task; |
722 | ||
1da177e4 | 723 | /* |
f1a11e05 TG |
724 | * We set q->lock_ptr = NULL _before_ we wake up the task. If |
725 | * a non futex wake up happens on another CPU then the task | |
726 | * might exit and p would dereference a non existing task | |
727 | * struct. Prevent this by holding a reference on p across the | |
728 | * wake up. | |
1da177e4 | 729 | */ |
f1a11e05 TG |
730 | get_task_struct(p); |
731 | ||
732 | plist_del(&q->list, &q->list.plist); | |
1da177e4 | 733 | /* |
f1a11e05 TG |
734 | * The waiting task can free the futex_q as soon as |
735 | * q->lock_ptr = NULL is written, without taking any locks. A | |
736 | * memory barrier is required here to prevent the following | |
737 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 738 | */ |
ccdea2f8 | 739 | smp_wmb(); |
1da177e4 | 740 | q->lock_ptr = NULL; |
f1a11e05 TG |
741 | |
742 | wake_up_state(p, TASK_NORMAL); | |
743 | put_task_struct(p); | |
1da177e4 LT |
744 | } |
745 | ||
c87e2837 IM |
746 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
747 | { | |
748 | struct task_struct *new_owner; | |
749 | struct futex_pi_state *pi_state = this->pi_state; | |
750 | u32 curval, newval; | |
751 | ||
752 | if (!pi_state) | |
753 | return -EINVAL; | |
754 | ||
21778867 | 755 | spin_lock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
756 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
757 | ||
758 | /* | |
759 | * This happens when we have stolen the lock and the original | |
760 | * pending owner did not enqueue itself back on the rt_mutex. | |
761 | * Thats not a tragedy. We know that way, that a lock waiter | |
762 | * is on the fly. We make the futex_q waiter the pending owner. | |
763 | */ | |
764 | if (!new_owner) | |
765 | new_owner = this->task; | |
766 | ||
767 | /* | |
768 | * We pass it to the next owner. (The WAITERS bit is always | |
769 | * kept enabled while there is PI state around. We must also | |
770 | * preserve the owner died bit.) | |
771 | */ | |
e3f2ddea | 772 | if (!(uval & FUTEX_OWNER_DIED)) { |
778e9a9c AK |
773 | int ret = 0; |
774 | ||
b488893a | 775 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 776 | |
36cf3b5c | 777 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
778e9a9c | 778 | |
e3f2ddea | 779 | if (curval == -EFAULT) |
778e9a9c | 780 | ret = -EFAULT; |
cde898fa | 781 | else if (curval != uval) |
778e9a9c AK |
782 | ret = -EINVAL; |
783 | if (ret) { | |
784 | spin_unlock(&pi_state->pi_mutex.wait_lock); | |
785 | return ret; | |
786 | } | |
e3f2ddea | 787 | } |
c87e2837 | 788 | |
627371d7 IM |
789 | spin_lock_irq(&pi_state->owner->pi_lock); |
790 | WARN_ON(list_empty(&pi_state->list)); | |
791 | list_del_init(&pi_state->list); | |
792 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
793 | ||
794 | spin_lock_irq(&new_owner->pi_lock); | |
795 | WARN_ON(!list_empty(&pi_state->list)); | |
c87e2837 IM |
796 | list_add(&pi_state->list, &new_owner->pi_state_list); |
797 | pi_state->owner = new_owner; | |
627371d7 IM |
798 | spin_unlock_irq(&new_owner->pi_lock); |
799 | ||
21778867 | 800 | spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
801 | rt_mutex_unlock(&pi_state->pi_mutex); |
802 | ||
803 | return 0; | |
804 | } | |
805 | ||
806 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) | |
807 | { | |
808 | u32 oldval; | |
809 | ||
810 | /* | |
811 | * There is no waiter, so we unlock the futex. The owner died | |
812 | * bit has not to be preserved here. We are the owner: | |
813 | */ | |
36cf3b5c | 814 | oldval = cmpxchg_futex_value_locked(uaddr, uval, 0); |
c87e2837 IM |
815 | |
816 | if (oldval == -EFAULT) | |
817 | return oldval; | |
818 | if (oldval != uval) | |
819 | return -EAGAIN; | |
820 | ||
821 | return 0; | |
822 | } | |
823 | ||
8b8f319f IM |
824 | /* |
825 | * Express the locking dependencies for lockdep: | |
826 | */ | |
827 | static inline void | |
828 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
829 | { | |
830 | if (hb1 <= hb2) { | |
831 | spin_lock(&hb1->lock); | |
832 | if (hb1 < hb2) | |
833 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
834 | } else { /* hb1 > hb2 */ | |
835 | spin_lock(&hb2->lock); | |
836 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
837 | } | |
838 | } | |
839 | ||
5eb3dc62 DH |
840 | static inline void |
841 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
842 | { | |
f061d351 | 843 | spin_unlock(&hb1->lock); |
88f502fe IM |
844 | if (hb1 != hb2) |
845 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
846 | } |
847 | ||
1da177e4 | 848 | /* |
b2d0994b | 849 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 850 | */ |
c2f9f201 | 851 | static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset) |
1da177e4 | 852 | { |
e2970f2f | 853 | struct futex_hash_bucket *hb; |
1da177e4 | 854 | struct futex_q *this, *next; |
ec92d082 | 855 | struct plist_head *head; |
38d47c1b | 856 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 LT |
857 | int ret; |
858 | ||
cd689985 TG |
859 | if (!bitset) |
860 | return -EINVAL; | |
861 | ||
64d1304a | 862 | ret = get_futex_key(uaddr, fshared, &key, VERIFY_READ); |
1da177e4 LT |
863 | if (unlikely(ret != 0)) |
864 | goto out; | |
865 | ||
e2970f2f IM |
866 | hb = hash_futex(&key); |
867 | spin_lock(&hb->lock); | |
868 | head = &hb->chain; | |
1da177e4 | 869 | |
ec92d082 | 870 | plist_for_each_entry_safe(this, next, head, list) { |
1da177e4 | 871 | if (match_futex (&this->key, &key)) { |
52400ba9 | 872 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
873 | ret = -EINVAL; |
874 | break; | |
875 | } | |
cd689985 TG |
876 | |
877 | /* Check if one of the bits is set in both bitsets */ | |
878 | if (!(this->bitset & bitset)) | |
879 | continue; | |
880 | ||
1da177e4 LT |
881 | wake_futex(this); |
882 | if (++ret >= nr_wake) | |
883 | break; | |
884 | } | |
885 | } | |
886 | ||
e2970f2f | 887 | spin_unlock(&hb->lock); |
38d47c1b | 888 | put_futex_key(fshared, &key); |
42d35d48 | 889 | out: |
1da177e4 LT |
890 | return ret; |
891 | } | |
892 | ||
4732efbe JJ |
893 | /* |
894 | * Wake up all waiters hashed on the physical page that is mapped | |
895 | * to this virtual address: | |
896 | */ | |
e2970f2f | 897 | static int |
c2f9f201 | 898 | futex_wake_op(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, |
e2970f2f | 899 | int nr_wake, int nr_wake2, int op) |
4732efbe | 900 | { |
38d47c1b | 901 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 902 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 903 | struct plist_head *head; |
4732efbe | 904 | struct futex_q *this, *next; |
e4dc5b7a | 905 | int ret, op_ret; |
4732efbe | 906 | |
e4dc5b7a | 907 | retry: |
64d1304a | 908 | ret = get_futex_key(uaddr1, fshared, &key1, VERIFY_READ); |
4732efbe JJ |
909 | if (unlikely(ret != 0)) |
910 | goto out; | |
64d1304a | 911 | ret = get_futex_key(uaddr2, fshared, &key2, VERIFY_WRITE); |
4732efbe | 912 | if (unlikely(ret != 0)) |
42d35d48 | 913 | goto out_put_key1; |
4732efbe | 914 | |
e2970f2f IM |
915 | hb1 = hash_futex(&key1); |
916 | hb2 = hash_futex(&key2); | |
4732efbe | 917 | |
8b8f319f | 918 | double_lock_hb(hb1, hb2); |
e4dc5b7a | 919 | retry_private: |
e2970f2f | 920 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 921 | if (unlikely(op_ret < 0)) { |
4732efbe | 922 | |
5eb3dc62 | 923 | double_unlock_hb(hb1, hb2); |
4732efbe | 924 | |
7ee1dd3f | 925 | #ifndef CONFIG_MMU |
e2970f2f IM |
926 | /* |
927 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
928 | * but we might get them from range checking | |
929 | */ | |
7ee1dd3f | 930 | ret = op_ret; |
42d35d48 | 931 | goto out_put_keys; |
7ee1dd3f DH |
932 | #endif |
933 | ||
796f8d9b DG |
934 | if (unlikely(op_ret != -EFAULT)) { |
935 | ret = op_ret; | |
42d35d48 | 936 | goto out_put_keys; |
796f8d9b DG |
937 | } |
938 | ||
d0725992 | 939 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 940 | if (ret) |
de87fcc1 | 941 | goto out_put_keys; |
4732efbe | 942 | |
e4dc5b7a DH |
943 | if (!fshared) |
944 | goto retry_private; | |
945 | ||
de87fcc1 DH |
946 | put_futex_key(fshared, &key2); |
947 | put_futex_key(fshared, &key1); | |
e4dc5b7a | 948 | goto retry; |
4732efbe JJ |
949 | } |
950 | ||
e2970f2f | 951 | head = &hb1->chain; |
4732efbe | 952 | |
ec92d082 | 953 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
954 | if (match_futex (&this->key, &key1)) { |
955 | wake_futex(this); | |
956 | if (++ret >= nr_wake) | |
957 | break; | |
958 | } | |
959 | } | |
960 | ||
961 | if (op_ret > 0) { | |
e2970f2f | 962 | head = &hb2->chain; |
4732efbe JJ |
963 | |
964 | op_ret = 0; | |
ec92d082 | 965 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
966 | if (match_futex (&this->key, &key2)) { |
967 | wake_futex(this); | |
968 | if (++op_ret >= nr_wake2) | |
969 | break; | |
970 | } | |
971 | } | |
972 | ret += op_ret; | |
973 | } | |
974 | ||
5eb3dc62 | 975 | double_unlock_hb(hb1, hb2); |
42d35d48 | 976 | out_put_keys: |
38d47c1b | 977 | put_futex_key(fshared, &key2); |
42d35d48 | 978 | out_put_key1: |
38d47c1b | 979 | put_futex_key(fshared, &key1); |
42d35d48 | 980 | out: |
4732efbe JJ |
981 | return ret; |
982 | } | |
983 | ||
9121e478 DH |
984 | /** |
985 | * requeue_futex() - Requeue a futex_q from one hb to another | |
986 | * @q: the futex_q to requeue | |
987 | * @hb1: the source hash_bucket | |
988 | * @hb2: the target hash_bucket | |
989 | * @key2: the new key for the requeued futex_q | |
990 | */ | |
991 | static inline | |
992 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
993 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
994 | { | |
995 | ||
996 | /* | |
997 | * If key1 and key2 hash to the same bucket, no need to | |
998 | * requeue. | |
999 | */ | |
1000 | if (likely(&hb1->chain != &hb2->chain)) { | |
1001 | plist_del(&q->list, &hb1->chain); | |
1002 | plist_add(&q->list, &hb2->chain); | |
1003 | q->lock_ptr = &hb2->lock; | |
1004 | #ifdef CONFIG_DEBUG_PI_LIST | |
1005 | q->list.plist.lock = &hb2->lock; | |
1006 | #endif | |
1007 | } | |
1008 | get_futex_key_refs(key2); | |
1009 | q->key = *key2; | |
1010 | } | |
1011 | ||
52400ba9 DH |
1012 | /** |
1013 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
1014 | * q: the futex_q | |
1015 | * key: the key of the requeue target futex | |
beda2c7e | 1016 | * hb: the hash_bucket of the requeue target futex |
52400ba9 DH |
1017 | * |
1018 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1019 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1020 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1021 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1022 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1023 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1024 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1025 | */ |
1026 | static inline | |
beda2c7e DH |
1027 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1028 | struct futex_hash_bucket *hb) | |
52400ba9 DH |
1029 | { |
1030 | drop_futex_key_refs(&q->key); | |
1031 | get_futex_key_refs(key); | |
1032 | q->key = *key; | |
1033 | ||
1034 | WARN_ON(plist_node_empty(&q->list)); | |
1035 | plist_del(&q->list, &q->list.plist); | |
1036 | ||
1037 | WARN_ON(!q->rt_waiter); | |
1038 | q->rt_waiter = NULL; | |
1039 | ||
beda2c7e DH |
1040 | q->lock_ptr = &hb->lock; |
1041 | #ifdef CONFIG_DEBUG_PI_LIST | |
1042 | q->list.plist.lock = &hb->lock; | |
1043 | #endif | |
1044 | ||
f1a11e05 | 1045 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1046 | } |
1047 | ||
1048 | /** | |
1049 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1050 | * @pifutex: the user address of the to futex |
1051 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1052 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1053 | * @key1: the from futex key | |
1054 | * @key2: the to futex key | |
1055 | * @ps: address to store the pi_state pointer | |
1056 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1057 | * |
1058 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1059 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1060 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1061 | * hb1 and hb2 must be held by the caller. | |
52400ba9 DH |
1062 | * |
1063 | * Returns: | |
1064 | * 0 - failed to acquire the lock atomicly | |
1065 | * 1 - acquired the lock | |
1066 | * <0 - error | |
1067 | */ | |
1068 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1069 | struct futex_hash_bucket *hb1, | |
1070 | struct futex_hash_bucket *hb2, | |
1071 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1072 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1073 | { |
bab5bc9e | 1074 | struct futex_q *top_waiter = NULL; |
52400ba9 DH |
1075 | u32 curval; |
1076 | int ret; | |
1077 | ||
1078 | if (get_futex_value_locked(&curval, pifutex)) | |
1079 | return -EFAULT; | |
1080 | ||
bab5bc9e DH |
1081 | /* |
1082 | * Find the top_waiter and determine if there are additional waiters. | |
1083 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1084 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1085 | * as we have means to handle the possible fault. If not, don't set | |
1086 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1087 | * the kernel. | |
1088 | */ | |
52400ba9 DH |
1089 | top_waiter = futex_top_waiter(hb1, key1); |
1090 | ||
1091 | /* There are no waiters, nothing for us to do. */ | |
1092 | if (!top_waiter) | |
1093 | return 0; | |
1094 | ||
84bc4af5 DH |
1095 | /* Ensure we requeue to the expected futex. */ |
1096 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1097 | return -EINVAL; | |
1098 | ||
52400ba9 | 1099 | /* |
bab5bc9e DH |
1100 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1101 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1102 | * in ps in contended cases. | |
52400ba9 | 1103 | */ |
bab5bc9e DH |
1104 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1105 | set_waiters); | |
52400ba9 | 1106 | if (ret == 1) |
beda2c7e | 1107 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
52400ba9 DH |
1108 | |
1109 | return ret; | |
1110 | } | |
1111 | ||
1112 | /** | |
1113 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
1114 | * uaddr1: source futex user address | |
1115 | * uaddr2: target futex user address | |
1116 | * nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1117 | * nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1118 | * requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
1119 | * pi futex (pi to pi requeue is not supported) | |
1120 | * | |
1121 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1122 | * uaddr2 atomically on behalf of the top waiter. | |
1123 | * | |
1124 | * Returns: | |
1125 | * >=0 - on success, the number of tasks requeued or woken | |
1126 | * <0 - on error | |
1da177e4 | 1127 | */ |
c2f9f201 | 1128 | static int futex_requeue(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, |
52400ba9 DH |
1129 | int nr_wake, int nr_requeue, u32 *cmpval, |
1130 | int requeue_pi) | |
1da177e4 | 1131 | { |
38d47c1b | 1132 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1133 | int drop_count = 0, task_count = 0, ret; |
1134 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1135 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 1136 | struct plist_head *head1; |
1da177e4 | 1137 | struct futex_q *this, *next; |
52400ba9 DH |
1138 | u32 curval2; |
1139 | ||
1140 | if (requeue_pi) { | |
1141 | /* | |
1142 | * requeue_pi requires a pi_state, try to allocate it now | |
1143 | * without any locks in case it fails. | |
1144 | */ | |
1145 | if (refill_pi_state_cache()) | |
1146 | return -ENOMEM; | |
1147 | /* | |
1148 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1149 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1150 | * returning to userspace, so as to not leave the rt_mutex with | |
1151 | * waiters and no owner. However, second and third wake-ups | |
1152 | * cannot be predicted as they involve race conditions with the | |
1153 | * first wake and a fault while looking up the pi_state. Both | |
1154 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1155 | * use nr_wake=1. | |
1156 | */ | |
1157 | if (nr_wake != 1) | |
1158 | return -EINVAL; | |
1159 | } | |
1da177e4 | 1160 | |
42d35d48 | 1161 | retry: |
52400ba9 DH |
1162 | if (pi_state != NULL) { |
1163 | /* | |
1164 | * We will have to lookup the pi_state again, so free this one | |
1165 | * to keep the accounting correct. | |
1166 | */ | |
1167 | free_pi_state(pi_state); | |
1168 | pi_state = NULL; | |
1169 | } | |
1170 | ||
64d1304a | 1171 | ret = get_futex_key(uaddr1, fshared, &key1, VERIFY_READ); |
1da177e4 LT |
1172 | if (unlikely(ret != 0)) |
1173 | goto out; | |
521c1808 TG |
1174 | ret = get_futex_key(uaddr2, fshared, &key2, |
1175 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1176 | if (unlikely(ret != 0)) |
42d35d48 | 1177 | goto out_put_key1; |
1da177e4 | 1178 | |
e2970f2f IM |
1179 | hb1 = hash_futex(&key1); |
1180 | hb2 = hash_futex(&key2); | |
1da177e4 | 1181 | |
e4dc5b7a | 1182 | retry_private: |
8b8f319f | 1183 | double_lock_hb(hb1, hb2); |
1da177e4 | 1184 | |
e2970f2f IM |
1185 | if (likely(cmpval != NULL)) { |
1186 | u32 curval; | |
1da177e4 | 1187 | |
e2970f2f | 1188 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1189 | |
1190 | if (unlikely(ret)) { | |
5eb3dc62 | 1191 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1192 | |
e2970f2f | 1193 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1194 | if (ret) |
1195 | goto out_put_keys; | |
1da177e4 | 1196 | |
e4dc5b7a DH |
1197 | if (!fshared) |
1198 | goto retry_private; | |
1da177e4 | 1199 | |
e4dc5b7a DH |
1200 | put_futex_key(fshared, &key2); |
1201 | put_futex_key(fshared, &key1); | |
1202 | goto retry; | |
1da177e4 | 1203 | } |
e2970f2f | 1204 | if (curval != *cmpval) { |
1da177e4 LT |
1205 | ret = -EAGAIN; |
1206 | goto out_unlock; | |
1207 | } | |
1208 | } | |
1209 | ||
52400ba9 | 1210 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1211 | /* |
1212 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1213 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1214 | * bit. We force this here where we are able to easily handle | |
1215 | * faults rather in the requeue loop below. | |
1216 | */ | |
52400ba9 | 1217 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1218 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1219 | |
1220 | /* | |
1221 | * At this point the top_waiter has either taken uaddr2 or is | |
1222 | * waiting on it. If the former, then the pi_state will not | |
1223 | * exist yet, look it up one more time to ensure we have a | |
1224 | * reference to it. | |
1225 | */ | |
1226 | if (ret == 1) { | |
1227 | WARN_ON(pi_state); | |
1228 | task_count++; | |
1229 | ret = get_futex_value_locked(&curval2, uaddr2); | |
1230 | if (!ret) | |
1231 | ret = lookup_pi_state(curval2, hb2, &key2, | |
1232 | &pi_state); | |
1233 | } | |
1234 | ||
1235 | switch (ret) { | |
1236 | case 0: | |
1237 | break; | |
1238 | case -EFAULT: | |
1239 | double_unlock_hb(hb1, hb2); | |
1240 | put_futex_key(fshared, &key2); | |
1241 | put_futex_key(fshared, &key1); | |
d0725992 | 1242 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
1243 | if (!ret) |
1244 | goto retry; | |
1245 | goto out; | |
1246 | case -EAGAIN: | |
1247 | /* The owner was exiting, try again. */ | |
1248 | double_unlock_hb(hb1, hb2); | |
1249 | put_futex_key(fshared, &key2); | |
1250 | put_futex_key(fshared, &key1); | |
1251 | cond_resched(); | |
1252 | goto retry; | |
1253 | default: | |
1254 | goto out_unlock; | |
1255 | } | |
1256 | } | |
1257 | ||
e2970f2f | 1258 | head1 = &hb1->chain; |
ec92d082 | 1259 | plist_for_each_entry_safe(this, next, head1, list) { |
52400ba9 DH |
1260 | if (task_count - nr_wake >= nr_requeue) |
1261 | break; | |
1262 | ||
1263 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1264 | continue; |
52400ba9 | 1265 | |
392741e0 DH |
1266 | /* |
1267 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
1268 | * be paired with each other and no other futex ops. | |
1269 | */ | |
1270 | if ((requeue_pi && !this->rt_waiter) || | |
1271 | (!requeue_pi && this->rt_waiter)) { | |
1272 | ret = -EINVAL; | |
1273 | break; | |
1274 | } | |
52400ba9 DH |
1275 | |
1276 | /* | |
1277 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1278 | * lock, we already woke the top_waiter. If not, it will be | |
1279 | * woken by futex_unlock_pi(). | |
1280 | */ | |
1281 | if (++task_count <= nr_wake && !requeue_pi) { | |
1da177e4 | 1282 | wake_futex(this); |
52400ba9 DH |
1283 | continue; |
1284 | } | |
1da177e4 | 1285 | |
84bc4af5 DH |
1286 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
1287 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
1288 | ret = -EINVAL; | |
1289 | break; | |
1290 | } | |
1291 | ||
52400ba9 DH |
1292 | /* |
1293 | * Requeue nr_requeue waiters and possibly one more in the case | |
1294 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1295 | */ | |
1296 | if (requeue_pi) { | |
1297 | /* Prepare the waiter to take the rt_mutex. */ | |
1298 | atomic_inc(&pi_state->refcount); | |
1299 | this->pi_state = pi_state; | |
1300 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1301 | this->rt_waiter, | |
1302 | this->task, 1); | |
1303 | if (ret == 1) { | |
1304 | /* We got the lock. */ | |
beda2c7e | 1305 | requeue_pi_wake_futex(this, &key2, hb2); |
52400ba9 DH |
1306 | continue; |
1307 | } else if (ret) { | |
1308 | /* -EDEADLK */ | |
1309 | this->pi_state = NULL; | |
1310 | free_pi_state(pi_state); | |
1311 | goto out_unlock; | |
1312 | } | |
1da177e4 | 1313 | } |
52400ba9 DH |
1314 | requeue_futex(this, hb1, hb2, &key2); |
1315 | drop_count++; | |
1da177e4 LT |
1316 | } |
1317 | ||
1318 | out_unlock: | |
5eb3dc62 | 1319 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1320 | |
cd84a42f DH |
1321 | /* |
1322 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1323 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1324 | * one at key2 and updated their key pointer. We no longer need to | |
1325 | * hold the references to key1. | |
1326 | */ | |
1da177e4 | 1327 | while (--drop_count >= 0) |
9adef58b | 1328 | drop_futex_key_refs(&key1); |
1da177e4 | 1329 | |
42d35d48 | 1330 | out_put_keys: |
38d47c1b | 1331 | put_futex_key(fshared, &key2); |
42d35d48 | 1332 | out_put_key1: |
38d47c1b | 1333 | put_futex_key(fshared, &key1); |
42d35d48 | 1334 | out: |
52400ba9 DH |
1335 | if (pi_state != NULL) |
1336 | free_pi_state(pi_state); | |
1337 | return ret ? ret : task_count; | |
1da177e4 LT |
1338 | } |
1339 | ||
1340 | /* The key must be already stored in q->key. */ | |
82af7aca | 1341 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
1da177e4 | 1342 | { |
e2970f2f | 1343 | struct futex_hash_bucket *hb; |
1da177e4 | 1344 | |
9adef58b | 1345 | get_futex_key_refs(&q->key); |
e2970f2f IM |
1346 | hb = hash_futex(&q->key); |
1347 | q->lock_ptr = &hb->lock; | |
1da177e4 | 1348 | |
e2970f2f IM |
1349 | spin_lock(&hb->lock); |
1350 | return hb; | |
1da177e4 LT |
1351 | } |
1352 | ||
d40d65c8 DH |
1353 | static inline void |
1354 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) | |
1355 | { | |
1356 | spin_unlock(&hb->lock); | |
1357 | drop_futex_key_refs(&q->key); | |
1358 | } | |
1359 | ||
1360 | /** | |
1361 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
1362 | * @q: The futex_q to enqueue | |
1363 | * @hb: The destination hash bucket | |
1364 | * | |
1365 | * The hb->lock must be held by the caller, and is released here. A call to | |
1366 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
1367 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
1368 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
1369 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
1370 | * an example). | |
1371 | */ | |
82af7aca | 1372 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 1373 | { |
ec92d082 PP |
1374 | int prio; |
1375 | ||
1376 | /* | |
1377 | * The priority used to register this element is | |
1378 | * - either the real thread-priority for the real-time threads | |
1379 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
1380 | * - or MAX_RT_PRIO for non-RT threads. | |
1381 | * Thus, all RT-threads are woken first in priority order, and | |
1382 | * the others are woken last, in FIFO order. | |
1383 | */ | |
1384 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1385 | ||
1386 | plist_node_init(&q->list, prio); | |
1387 | #ifdef CONFIG_DEBUG_PI_LIST | |
1388 | q->list.plist.lock = &hb->lock; | |
1389 | #endif | |
1390 | plist_add(&q->list, &hb->chain); | |
c87e2837 | 1391 | q->task = current; |
e2970f2f | 1392 | spin_unlock(&hb->lock); |
1da177e4 LT |
1393 | } |
1394 | ||
d40d65c8 DH |
1395 | /** |
1396 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
1397 | * @q: The futex_q to unqueue | |
1398 | * | |
1399 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
1400 | * be paired with exactly one earlier call to queue_me(). | |
1401 | * | |
1402 | * Returns: | |
1403 | * 1 - if the futex_q was still queued (and we removed unqueued it) | |
1404 | * 0 - if the futex_q was already removed by the waking thread | |
1da177e4 | 1405 | */ |
1da177e4 LT |
1406 | static int unqueue_me(struct futex_q *q) |
1407 | { | |
1da177e4 | 1408 | spinlock_t *lock_ptr; |
e2970f2f | 1409 | int ret = 0; |
1da177e4 LT |
1410 | |
1411 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 1412 | retry: |
1da177e4 | 1413 | lock_ptr = q->lock_ptr; |
e91467ec | 1414 | barrier(); |
c80544dc | 1415 | if (lock_ptr != NULL) { |
1da177e4 LT |
1416 | spin_lock(lock_ptr); |
1417 | /* | |
1418 | * q->lock_ptr can change between reading it and | |
1419 | * spin_lock(), causing us to take the wrong lock. This | |
1420 | * corrects the race condition. | |
1421 | * | |
1422 | * Reasoning goes like this: if we have the wrong lock, | |
1423 | * q->lock_ptr must have changed (maybe several times) | |
1424 | * between reading it and the spin_lock(). It can | |
1425 | * change again after the spin_lock() but only if it was | |
1426 | * already changed before the spin_lock(). It cannot, | |
1427 | * however, change back to the original value. Therefore | |
1428 | * we can detect whether we acquired the correct lock. | |
1429 | */ | |
1430 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
1431 | spin_unlock(lock_ptr); | |
1432 | goto retry; | |
1433 | } | |
ec92d082 PP |
1434 | WARN_ON(plist_node_empty(&q->list)); |
1435 | plist_del(&q->list, &q->list.plist); | |
c87e2837 IM |
1436 | |
1437 | BUG_ON(q->pi_state); | |
1438 | ||
1da177e4 LT |
1439 | spin_unlock(lock_ptr); |
1440 | ret = 1; | |
1441 | } | |
1442 | ||
9adef58b | 1443 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1444 | return ret; |
1445 | } | |
1446 | ||
c87e2837 IM |
1447 | /* |
1448 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
1449 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
1450 | * and dropped here. | |
c87e2837 | 1451 | */ |
d0aa7a70 | 1452 | static void unqueue_me_pi(struct futex_q *q) |
c87e2837 | 1453 | { |
ec92d082 PP |
1454 | WARN_ON(plist_node_empty(&q->list)); |
1455 | plist_del(&q->list, &q->list.plist); | |
c87e2837 IM |
1456 | |
1457 | BUG_ON(!q->pi_state); | |
1458 | free_pi_state(q->pi_state); | |
1459 | q->pi_state = NULL; | |
1460 | ||
d0aa7a70 | 1461 | spin_unlock(q->lock_ptr); |
c87e2837 | 1462 | |
9adef58b | 1463 | drop_futex_key_refs(&q->key); |
c87e2837 IM |
1464 | } |
1465 | ||
d0aa7a70 | 1466 | /* |
cdf71a10 | 1467 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 1468 | * |
778e9a9c AK |
1469 | * Must be called with hash bucket lock held and mm->sem held for non |
1470 | * private futexes. | |
d0aa7a70 | 1471 | */ |
778e9a9c | 1472 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
c2f9f201 | 1473 | struct task_struct *newowner, int fshared) |
d0aa7a70 | 1474 | { |
cdf71a10 | 1475 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 1476 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 1477 | struct task_struct *oldowner = pi_state->owner; |
d0aa7a70 | 1478 | u32 uval, curval, newval; |
e4dc5b7a | 1479 | int ret; |
d0aa7a70 PP |
1480 | |
1481 | /* Owner died? */ | |
1b7558e4 TG |
1482 | if (!pi_state->owner) |
1483 | newtid |= FUTEX_OWNER_DIED; | |
1484 | ||
1485 | /* | |
1486 | * We are here either because we stole the rtmutex from the | |
1487 | * pending owner or we are the pending owner which failed to | |
1488 | * get the rtmutex. We have to replace the pending owner TID | |
1489 | * in the user space variable. This must be atomic as we have | |
1490 | * to preserve the owner died bit here. | |
1491 | * | |
b2d0994b DH |
1492 | * Note: We write the user space value _before_ changing the pi_state |
1493 | * because we can fault here. Imagine swapped out pages or a fork | |
1494 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
1495 | * |
1496 | * Modifying pi_state _before_ the user space value would | |
1497 | * leave the pi_state in an inconsistent state when we fault | |
1498 | * here, because we need to drop the hash bucket lock to | |
1499 | * handle the fault. This might be observed in the PID check | |
1500 | * in lookup_pi_state. | |
1501 | */ | |
1502 | retry: | |
1503 | if (get_futex_value_locked(&uval, uaddr)) | |
1504 | goto handle_fault; | |
1505 | ||
1506 | while (1) { | |
1507 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
1508 | ||
1509 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); | |
1510 | ||
1511 | if (curval == -EFAULT) | |
1512 | goto handle_fault; | |
1513 | if (curval == uval) | |
1514 | break; | |
1515 | uval = curval; | |
1516 | } | |
1517 | ||
1518 | /* | |
1519 | * We fixed up user space. Now we need to fix the pi_state | |
1520 | * itself. | |
1521 | */ | |
d0aa7a70 PP |
1522 | if (pi_state->owner != NULL) { |
1523 | spin_lock_irq(&pi_state->owner->pi_lock); | |
1524 | WARN_ON(list_empty(&pi_state->list)); | |
1525 | list_del_init(&pi_state->list); | |
1526 | spin_unlock_irq(&pi_state->owner->pi_lock); | |
1b7558e4 | 1527 | } |
d0aa7a70 | 1528 | |
cdf71a10 | 1529 | pi_state->owner = newowner; |
d0aa7a70 | 1530 | |
cdf71a10 | 1531 | spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 1532 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 TG |
1533 | list_add(&pi_state->list, &newowner->pi_state_list); |
1534 | spin_unlock_irq(&newowner->pi_lock); | |
1b7558e4 | 1535 | return 0; |
d0aa7a70 | 1536 | |
d0aa7a70 | 1537 | /* |
1b7558e4 TG |
1538 | * To handle the page fault we need to drop the hash bucket |
1539 | * lock here. That gives the other task (either the pending | |
1540 | * owner itself or the task which stole the rtmutex) the | |
1541 | * chance to try the fixup of the pi_state. So once we are | |
1542 | * back from handling the fault we need to check the pi_state | |
1543 | * after reacquiring the hash bucket lock and before trying to | |
1544 | * do another fixup. When the fixup has been done already we | |
1545 | * simply return. | |
d0aa7a70 | 1546 | */ |
1b7558e4 TG |
1547 | handle_fault: |
1548 | spin_unlock(q->lock_ptr); | |
778e9a9c | 1549 | |
d0725992 | 1550 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 1551 | |
1b7558e4 | 1552 | spin_lock(q->lock_ptr); |
778e9a9c | 1553 | |
1b7558e4 TG |
1554 | /* |
1555 | * Check if someone else fixed it for us: | |
1556 | */ | |
1557 | if (pi_state->owner != oldowner) | |
1558 | return 0; | |
1559 | ||
1560 | if (ret) | |
1561 | return ret; | |
1562 | ||
1563 | goto retry; | |
d0aa7a70 PP |
1564 | } |
1565 | ||
34f01cc1 ED |
1566 | /* |
1567 | * In case we must use restart_block to restart a futex_wait, | |
ce6bd420 | 1568 | * we encode in the 'flags' shared capability |
34f01cc1 | 1569 | */ |
1acdac10 TG |
1570 | #define FLAGS_SHARED 0x01 |
1571 | #define FLAGS_CLOCKRT 0x02 | |
a72188d8 | 1572 | #define FLAGS_HAS_TIMEOUT 0x04 |
34f01cc1 | 1573 | |
72c1bbf3 | 1574 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 1575 | |
dd973998 DH |
1576 | /** |
1577 | * fixup_owner() - Post lock pi_state and corner case management | |
1578 | * @uaddr: user address of the futex | |
1579 | * @fshared: whether the futex is shared (1) or not (0) | |
1580 | * @q: futex_q (contains pi_state and access to the rt_mutex) | |
1581 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
1582 | * | |
1583 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
1584 | * the pi_state owner as well as handle race conditions that may allow us to | |
1585 | * acquire the lock. Must be called with the hb lock held. | |
1586 | * | |
1587 | * Returns: | |
1588 | * 1 - success, lock taken | |
1589 | * 0 - success, lock not taken | |
1590 | * <0 - on error (-EFAULT) | |
1591 | */ | |
1592 | static int fixup_owner(u32 __user *uaddr, int fshared, struct futex_q *q, | |
1593 | int locked) | |
1594 | { | |
1595 | struct task_struct *owner; | |
1596 | int ret = 0; | |
1597 | ||
1598 | if (locked) { | |
1599 | /* | |
1600 | * Got the lock. We might not be the anticipated owner if we | |
1601 | * did a lock-steal - fix up the PI-state in that case: | |
1602 | */ | |
1603 | if (q->pi_state->owner != current) | |
1604 | ret = fixup_pi_state_owner(uaddr, q, current, fshared); | |
1605 | goto out; | |
1606 | } | |
1607 | ||
1608 | /* | |
1609 | * Catch the rare case, where the lock was released when we were on the | |
1610 | * way back before we locked the hash bucket. | |
1611 | */ | |
1612 | if (q->pi_state->owner == current) { | |
1613 | /* | |
1614 | * Try to get the rt_mutex now. This might fail as some other | |
1615 | * task acquired the rt_mutex after we removed ourself from the | |
1616 | * rt_mutex waiters list. | |
1617 | */ | |
1618 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
1619 | locked = 1; | |
1620 | goto out; | |
1621 | } | |
1622 | ||
1623 | /* | |
1624 | * pi_state is incorrect, some other task did a lock steal and | |
1625 | * we returned due to timeout or signal without taking the | |
1626 | * rt_mutex. Too late. We can access the rt_mutex_owner without | |
1627 | * locking, as the other task is now blocked on the hash bucket | |
1628 | * lock. Fix the state up. | |
1629 | */ | |
1630 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); | |
1631 | ret = fixup_pi_state_owner(uaddr, q, owner, fshared); | |
1632 | goto out; | |
1633 | } | |
1634 | ||
1635 | /* | |
1636 | * Paranoia check. If we did not take the lock, then we should not be | |
1637 | * the owner, nor the pending owner, of the rt_mutex. | |
1638 | */ | |
1639 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
1640 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
1641 | "pi-state %p\n", ret, | |
1642 | q->pi_state->pi_mutex.owner, | |
1643 | q->pi_state->owner); | |
1644 | ||
1645 | out: | |
1646 | return ret ? ret : locked; | |
1647 | } | |
1648 | ||
ca5f9524 DH |
1649 | /** |
1650 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
1651 | * @hb: the futex hash bucket, must be locked by the caller | |
1652 | * @q: the futex_q to queue up on | |
1653 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
1654 | */ |
1655 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 1656 | struct hrtimer_sleeper *timeout) |
ca5f9524 DH |
1657 | { |
1658 | queue_me(q, hb); | |
1659 | ||
1660 | /* | |
1661 | * There might have been scheduling since the queue_me(), as we | |
1662 | * cannot hold a spinlock across the get_user() in case it | |
1663 | * faults, and we cannot just set TASK_INTERRUPTIBLE state when | |
f1a11e05 | 1664 | * queueing ourselves into the futex hash. This code thus has to |
ca5f9524 DH |
1665 | * rely on the futex_wake() code removing us from hash when it |
1666 | * wakes us up. | |
1667 | */ | |
f1a11e05 | 1668 | set_current_state(TASK_INTERRUPTIBLE); |
ca5f9524 DH |
1669 | |
1670 | /* Arm the timer */ | |
1671 | if (timeout) { | |
1672 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); | |
1673 | if (!hrtimer_active(&timeout->timer)) | |
1674 | timeout->task = NULL; | |
1675 | } | |
1676 | ||
1677 | /* | |
1678 | * !plist_node_empty() is safe here without any lock. | |
1679 | * q.lock_ptr != 0 is not safe, because of ordering against wakeup. | |
1680 | */ | |
1681 | if (likely(!plist_node_empty(&q->list))) { | |
1682 | /* | |
1683 | * If the timer has already expired, current will already be | |
1684 | * flagged for rescheduling. Only call schedule if there | |
1685 | * is no timeout, or if it has yet to expire. | |
1686 | */ | |
1687 | if (!timeout || timeout->task) | |
1688 | schedule(); | |
1689 | } | |
1690 | __set_current_state(TASK_RUNNING); | |
1691 | } | |
1692 | ||
f801073f DH |
1693 | /** |
1694 | * futex_wait_setup() - Prepare to wait on a futex | |
1695 | * @uaddr: the futex userspace address | |
1696 | * @val: the expected value | |
1697 | * @fshared: whether the futex is shared (1) or not (0) | |
1698 | * @q: the associated futex_q | |
1699 | * @hb: storage for hash_bucket pointer to be returned to caller | |
1700 | * | |
1701 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
1702 | * compare it with the expected value. Handle atomic faults internally. | |
1703 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
1704 | * with no q.key reference on failure. | |
1705 | * | |
1706 | * Returns: | |
1707 | * 0 - uaddr contains val and hb has been locked | |
1708 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlcoked | |
1709 | */ | |
1710 | static int futex_wait_setup(u32 __user *uaddr, u32 val, int fshared, | |
1711 | struct futex_q *q, struct futex_hash_bucket **hb) | |
1da177e4 | 1712 | { |
e2970f2f IM |
1713 | u32 uval; |
1714 | int ret; | |
1da177e4 | 1715 | |
1da177e4 | 1716 | /* |
b2d0994b | 1717 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
1718 | * Order is important: |
1719 | * | |
1720 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
1721 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
1722 | * | |
1723 | * The basic logical guarantee of a futex is that it blocks ONLY | |
1724 | * if cond(var) is known to be true at the time of blocking, for | |
1725 | * any cond. If we queued after testing *uaddr, that would open | |
1726 | * a race condition where we could block indefinitely with | |
1727 | * cond(var) false, which would violate the guarantee. | |
1728 | * | |
1729 | * A consequence is that futex_wait() can return zero and absorb | |
1730 | * a wakeup when *uaddr != val on entry to the syscall. This is | |
1731 | * rare, but normal. | |
1da177e4 | 1732 | */ |
f801073f DH |
1733 | retry: |
1734 | q->key = FUTEX_KEY_INIT; | |
521c1808 | 1735 | ret = get_futex_key(uaddr, fshared, &q->key, VERIFY_READ); |
f801073f | 1736 | if (unlikely(ret != 0)) |
a5a2a0c7 | 1737 | return ret; |
f801073f DH |
1738 | |
1739 | retry_private: | |
1740 | *hb = queue_lock(q); | |
1741 | ||
e2970f2f | 1742 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 1743 | |
f801073f DH |
1744 | if (ret) { |
1745 | queue_unlock(q, *hb); | |
1da177e4 | 1746 | |
e2970f2f | 1747 | ret = get_user(uval, uaddr); |
e4dc5b7a | 1748 | if (ret) |
f801073f | 1749 | goto out; |
1da177e4 | 1750 | |
e4dc5b7a DH |
1751 | if (!fshared) |
1752 | goto retry_private; | |
1753 | ||
f801073f | 1754 | put_futex_key(fshared, &q->key); |
e4dc5b7a | 1755 | goto retry; |
1da177e4 | 1756 | } |
ca5f9524 | 1757 | |
f801073f DH |
1758 | if (uval != val) { |
1759 | queue_unlock(q, *hb); | |
1760 | ret = -EWOULDBLOCK; | |
2fff78c7 | 1761 | } |
1da177e4 | 1762 | |
f801073f DH |
1763 | out: |
1764 | if (ret) | |
1765 | put_futex_key(fshared, &q->key); | |
1766 | return ret; | |
1767 | } | |
1768 | ||
1769 | static int futex_wait(u32 __user *uaddr, int fshared, | |
1770 | u32 val, ktime_t *abs_time, u32 bitset, int clockrt) | |
1771 | { | |
1772 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
1773 | struct restart_block *restart; |
1774 | struct futex_hash_bucket *hb; | |
1775 | struct futex_q q; | |
1776 | int ret; | |
1777 | ||
1778 | if (!bitset) | |
1779 | return -EINVAL; | |
1780 | ||
1781 | q.pi_state = NULL; | |
1782 | q.bitset = bitset; | |
52400ba9 | 1783 | q.rt_waiter = NULL; |
84bc4af5 | 1784 | q.requeue_pi_key = NULL; |
f801073f DH |
1785 | |
1786 | if (abs_time) { | |
1787 | to = &timeout; | |
1788 | ||
1789 | hrtimer_init_on_stack(&to->timer, clockrt ? CLOCK_REALTIME : | |
1790 | CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1791 | hrtimer_init_sleeper(to, current); | |
1792 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
1793 | current->timer_slack_ns); | |
1794 | } | |
1795 | ||
1796 | /* Prepare to wait on uaddr. */ | |
1797 | ret = futex_wait_setup(uaddr, val, fshared, &q, &hb); | |
1798 | if (ret) | |
1799 | goto out; | |
1800 | ||
ca5f9524 | 1801 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 1802 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
1803 | |
1804 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 1805 | ret = 0; |
1da177e4 | 1806 | if (!unqueue_me(&q)) |
2fff78c7 PZ |
1807 | goto out_put_key; |
1808 | ret = -ETIMEDOUT; | |
ca5f9524 | 1809 | if (to && !to->task) |
2fff78c7 | 1810 | goto out_put_key; |
72c1bbf3 | 1811 | |
e2970f2f IM |
1812 | /* |
1813 | * We expect signal_pending(current), but another thread may | |
1814 | * have handled it for us already. | |
1815 | */ | |
2fff78c7 | 1816 | ret = -ERESTARTSYS; |
c19384b5 | 1817 | if (!abs_time) |
2fff78c7 | 1818 | goto out_put_key; |
1da177e4 | 1819 | |
2fff78c7 PZ |
1820 | restart = ¤t_thread_info()->restart_block; |
1821 | restart->fn = futex_wait_restart; | |
1822 | restart->futex.uaddr = (u32 *)uaddr; | |
1823 | restart->futex.val = val; | |
1824 | restart->futex.time = abs_time->tv64; | |
1825 | restart->futex.bitset = bitset; | |
a72188d8 | 1826 | restart->futex.flags = FLAGS_HAS_TIMEOUT; |
2fff78c7 PZ |
1827 | |
1828 | if (fshared) | |
1829 | restart->futex.flags |= FLAGS_SHARED; | |
1830 | if (clockrt) | |
1831 | restart->futex.flags |= FLAGS_CLOCKRT; | |
42d35d48 | 1832 | |
2fff78c7 PZ |
1833 | ret = -ERESTART_RESTARTBLOCK; |
1834 | ||
1835 | out_put_key: | |
1836 | put_futex_key(fshared, &q.key); | |
42d35d48 | 1837 | out: |
ca5f9524 DH |
1838 | if (to) { |
1839 | hrtimer_cancel(&to->timer); | |
1840 | destroy_hrtimer_on_stack(&to->timer); | |
1841 | } | |
c87e2837 IM |
1842 | return ret; |
1843 | } | |
1844 | ||
72c1bbf3 NP |
1845 | |
1846 | static long futex_wait_restart(struct restart_block *restart) | |
1847 | { | |
ce6bd420 | 1848 | u32 __user *uaddr = (u32 __user *)restart->futex.uaddr; |
c2f9f201 | 1849 | int fshared = 0; |
a72188d8 | 1850 | ktime_t t, *tp = NULL; |
72c1bbf3 | 1851 | |
a72188d8 DH |
1852 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
1853 | t.tv64 = restart->futex.time; | |
1854 | tp = &t; | |
1855 | } | |
72c1bbf3 | 1856 | restart->fn = do_no_restart_syscall; |
ce6bd420 | 1857 | if (restart->futex.flags & FLAGS_SHARED) |
c2f9f201 | 1858 | fshared = 1; |
a72188d8 | 1859 | return (long)futex_wait(uaddr, fshared, restart->futex.val, tp, |
1acdac10 TG |
1860 | restart->futex.bitset, |
1861 | restart->futex.flags & FLAGS_CLOCKRT); | |
72c1bbf3 NP |
1862 | } |
1863 | ||
1864 | ||
c87e2837 IM |
1865 | /* |
1866 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
1867 | * and failed. The kernel side here does the whole locking operation: | |
1868 | * if there are waiters then it will block, it does PI, etc. (Due to | |
1869 | * races the kernel might see a 0 value of the futex too.) | |
1870 | */ | |
c2f9f201 | 1871 | static int futex_lock_pi(u32 __user *uaddr, int fshared, |
34f01cc1 | 1872 | int detect, ktime_t *time, int trylock) |
c87e2837 | 1873 | { |
c5780e97 | 1874 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 1875 | struct futex_hash_bucket *hb; |
c87e2837 | 1876 | struct futex_q q; |
dd973998 | 1877 | int res, ret; |
c87e2837 IM |
1878 | |
1879 | if (refill_pi_state_cache()) | |
1880 | return -ENOMEM; | |
1881 | ||
c19384b5 | 1882 | if (time) { |
c5780e97 | 1883 | to = &timeout; |
237fc6e7 TG |
1884 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
1885 | HRTIMER_MODE_ABS); | |
c5780e97 | 1886 | hrtimer_init_sleeper(to, current); |
cc584b21 | 1887 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
1888 | } |
1889 | ||
c87e2837 | 1890 | q.pi_state = NULL; |
52400ba9 | 1891 | q.rt_waiter = NULL; |
84bc4af5 | 1892 | q.requeue_pi_key = NULL; |
42d35d48 | 1893 | retry: |
38d47c1b | 1894 | q.key = FUTEX_KEY_INIT; |
64d1304a | 1895 | ret = get_futex_key(uaddr, fshared, &q.key, VERIFY_WRITE); |
c87e2837 | 1896 | if (unlikely(ret != 0)) |
42d35d48 | 1897 | goto out; |
c87e2837 | 1898 | |
e4dc5b7a | 1899 | retry_private: |
82af7aca | 1900 | hb = queue_lock(&q); |
c87e2837 | 1901 | |
bab5bc9e | 1902 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 1903 | if (unlikely(ret)) { |
778e9a9c | 1904 | switch (ret) { |
1a52084d DH |
1905 | case 1: |
1906 | /* We got the lock. */ | |
1907 | ret = 0; | |
1908 | goto out_unlock_put_key; | |
1909 | case -EFAULT: | |
1910 | goto uaddr_faulted; | |
778e9a9c AK |
1911 | case -EAGAIN: |
1912 | /* | |
1913 | * Task is exiting and we just wait for the | |
1914 | * exit to complete. | |
1915 | */ | |
1916 | queue_unlock(&q, hb); | |
de87fcc1 | 1917 | put_futex_key(fshared, &q.key); |
778e9a9c AK |
1918 | cond_resched(); |
1919 | goto retry; | |
778e9a9c | 1920 | default: |
42d35d48 | 1921 | goto out_unlock_put_key; |
c87e2837 | 1922 | } |
c87e2837 IM |
1923 | } |
1924 | ||
1925 | /* | |
1926 | * Only actually queue now that the atomic ops are done: | |
1927 | */ | |
82af7aca | 1928 | queue_me(&q, hb); |
c87e2837 | 1929 | |
c87e2837 IM |
1930 | WARN_ON(!q.pi_state); |
1931 | /* | |
1932 | * Block on the PI mutex: | |
1933 | */ | |
1934 | if (!trylock) | |
1935 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); | |
1936 | else { | |
1937 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); | |
1938 | /* Fixup the trylock return value: */ | |
1939 | ret = ret ? 0 : -EWOULDBLOCK; | |
1940 | } | |
1941 | ||
a99e4e41 | 1942 | spin_lock(q.lock_ptr); |
dd973998 DH |
1943 | /* |
1944 | * Fixup the pi_state owner and possibly acquire the lock if we | |
1945 | * haven't already. | |
1946 | */ | |
1947 | res = fixup_owner(uaddr, fshared, &q, !ret); | |
1948 | /* | |
1949 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
1950 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
1951 | */ | |
1952 | if (res) | |
1953 | ret = (res < 0) ? res : 0; | |
c87e2837 | 1954 | |
e8f6386c | 1955 | /* |
dd973998 DH |
1956 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
1957 | * it and return the fault to userspace. | |
e8f6386c DH |
1958 | */ |
1959 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
1960 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
1961 | ||
778e9a9c AK |
1962 | /* Unqueue and drop the lock */ |
1963 | unqueue_me_pi(&q); | |
c87e2837 | 1964 | |
dd973998 | 1965 | goto out; |
c87e2837 | 1966 | |
42d35d48 | 1967 | out_unlock_put_key: |
c87e2837 IM |
1968 | queue_unlock(&q, hb); |
1969 | ||
42d35d48 | 1970 | out_put_key: |
38d47c1b | 1971 | put_futex_key(fshared, &q.key); |
42d35d48 | 1972 | out: |
237fc6e7 TG |
1973 | if (to) |
1974 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 1975 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 1976 | |
42d35d48 | 1977 | uaddr_faulted: |
778e9a9c AK |
1978 | queue_unlock(&q, hb); |
1979 | ||
d0725992 | 1980 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
1981 | if (ret) |
1982 | goto out_put_key; | |
c87e2837 | 1983 | |
e4dc5b7a DH |
1984 | if (!fshared) |
1985 | goto retry_private; | |
1986 | ||
1987 | put_futex_key(fshared, &q.key); | |
1988 | goto retry; | |
c87e2837 IM |
1989 | } |
1990 | ||
c87e2837 IM |
1991 | /* |
1992 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
1993 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
1994 | * and do the rt-mutex unlock. | |
1995 | */ | |
c2f9f201 | 1996 | static int futex_unlock_pi(u32 __user *uaddr, int fshared) |
c87e2837 IM |
1997 | { |
1998 | struct futex_hash_bucket *hb; | |
1999 | struct futex_q *this, *next; | |
2000 | u32 uval; | |
ec92d082 | 2001 | struct plist_head *head; |
38d47c1b | 2002 | union futex_key key = FUTEX_KEY_INIT; |
e4dc5b7a | 2003 | int ret; |
c87e2837 IM |
2004 | |
2005 | retry: | |
2006 | if (get_user(uval, uaddr)) | |
2007 | return -EFAULT; | |
2008 | /* | |
2009 | * We release only a lock we actually own: | |
2010 | */ | |
b488893a | 2011 | if ((uval & FUTEX_TID_MASK) != task_pid_vnr(current)) |
c87e2837 | 2012 | return -EPERM; |
c87e2837 | 2013 | |
64d1304a | 2014 | ret = get_futex_key(uaddr, fshared, &key, VERIFY_WRITE); |
c87e2837 IM |
2015 | if (unlikely(ret != 0)) |
2016 | goto out; | |
2017 | ||
2018 | hb = hash_futex(&key); | |
2019 | spin_lock(&hb->lock); | |
2020 | ||
c87e2837 IM |
2021 | /* |
2022 | * To avoid races, try to do the TID -> 0 atomic transition | |
2023 | * again. If it succeeds then we can return without waking | |
2024 | * anyone else up: | |
2025 | */ | |
36cf3b5c | 2026 | if (!(uval & FUTEX_OWNER_DIED)) |
b488893a | 2027 | uval = cmpxchg_futex_value_locked(uaddr, task_pid_vnr(current), 0); |
36cf3b5c | 2028 | |
c87e2837 IM |
2029 | |
2030 | if (unlikely(uval == -EFAULT)) | |
2031 | goto pi_faulted; | |
2032 | /* | |
2033 | * Rare case: we managed to release the lock atomically, | |
2034 | * no need to wake anyone else up: | |
2035 | */ | |
b488893a | 2036 | if (unlikely(uval == task_pid_vnr(current))) |
c87e2837 IM |
2037 | goto out_unlock; |
2038 | ||
2039 | /* | |
2040 | * Ok, other tasks may need to be woken up - check waiters | |
2041 | * and do the wakeup if necessary: | |
2042 | */ | |
2043 | head = &hb->chain; | |
2044 | ||
ec92d082 | 2045 | plist_for_each_entry_safe(this, next, head, list) { |
c87e2837 IM |
2046 | if (!match_futex (&this->key, &key)) |
2047 | continue; | |
2048 | ret = wake_futex_pi(uaddr, uval, this); | |
2049 | /* | |
2050 | * The atomic access to the futex value | |
2051 | * generated a pagefault, so retry the | |
2052 | * user-access and the wakeup: | |
2053 | */ | |
2054 | if (ret == -EFAULT) | |
2055 | goto pi_faulted; | |
2056 | goto out_unlock; | |
2057 | } | |
2058 | /* | |
2059 | * No waiters - kernel unlocks the futex: | |
2060 | */ | |
e3f2ddea IM |
2061 | if (!(uval & FUTEX_OWNER_DIED)) { |
2062 | ret = unlock_futex_pi(uaddr, uval); | |
2063 | if (ret == -EFAULT) | |
2064 | goto pi_faulted; | |
2065 | } | |
c87e2837 IM |
2066 | |
2067 | out_unlock: | |
2068 | spin_unlock(&hb->lock); | |
38d47c1b | 2069 | put_futex_key(fshared, &key); |
c87e2837 | 2070 | |
42d35d48 | 2071 | out: |
c87e2837 IM |
2072 | return ret; |
2073 | ||
2074 | pi_faulted: | |
778e9a9c | 2075 | spin_unlock(&hb->lock); |
e4dc5b7a | 2076 | put_futex_key(fshared, &key); |
c87e2837 | 2077 | |
d0725992 | 2078 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2079 | if (!ret) |
c87e2837 IM |
2080 | goto retry; |
2081 | ||
1da177e4 LT |
2082 | return ret; |
2083 | } | |
2084 | ||
52400ba9 DH |
2085 | /** |
2086 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2087 | * @hb: the hash_bucket futex_q was original enqueued on | |
2088 | * @q: the futex_q woken while waiting to be requeued | |
2089 | * @key2: the futex_key of the requeue target futex | |
2090 | * @timeout: the timeout associated with the wait (NULL if none) | |
2091 | * | |
2092 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2093 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2094 | * the wakeup and return the appropriate error code to the caller. Must be | |
2095 | * called with the hb lock held. | |
2096 | * | |
2097 | * Returns | |
2098 | * 0 - no early wakeup detected | |
1c840c14 | 2099 | * <0 - -ETIMEDOUT or -ERESTARTNOINTR |
52400ba9 DH |
2100 | */ |
2101 | static inline | |
2102 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2103 | struct futex_q *q, union futex_key *key2, | |
2104 | struct hrtimer_sleeper *timeout) | |
2105 | { | |
2106 | int ret = 0; | |
2107 | ||
2108 | /* | |
2109 | * With the hb lock held, we avoid races while we process the wakeup. | |
2110 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2111 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2112 | * It can't be requeued from uaddr2 to something else since we don't | |
2113 | * support a PI aware source futex for requeue. | |
2114 | */ | |
2115 | if (!match_futex(&q->key, key2)) { | |
2116 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2117 | /* | |
2118 | * We were woken prior to requeue by a timeout or a signal. | |
2119 | * Unqueue the futex_q and determine which it was. | |
2120 | */ | |
2121 | plist_del(&q->list, &q->list.plist); | |
2122 | drop_futex_key_refs(&q->key); | |
2123 | ||
2124 | if (timeout && !timeout->task) | |
2125 | ret = -ETIMEDOUT; | |
1c840c14 TG |
2126 | else |
2127 | ret = -ERESTARTNOINTR; | |
52400ba9 DH |
2128 | } |
2129 | return ret; | |
2130 | } | |
2131 | ||
2132 | /** | |
2133 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 2134 | * @uaddr: the futex we initially wait on (non-pi) |
52400ba9 DH |
2135 | * @fshared: whether the futexes are shared (1) or not (0). They must be |
2136 | * the same type, no requeueing from private to shared, etc. | |
2137 | * @val: the expected value of uaddr | |
2138 | * @abs_time: absolute timeout | |
56ec1607 | 2139 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
2140 | * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0) |
2141 | * @uaddr2: the pi futex we will take prior to returning to user-space | |
2142 | * | |
2143 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
2144 | * uaddr2 which must be PI aware. Normal wakeup will wake on uaddr2 and | |
2145 | * complete the acquisition of the rt_mutex prior to returning to userspace. | |
2146 | * This ensures the rt_mutex maintains an owner when it has waiters; without | |
2147 | * one, the pi logic wouldn't know which task to boost/deboost, if there was a | |
2148 | * need to. | |
2149 | * | |
2150 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
2151 | * via the following: | |
2152 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() | |
cc6db4e6 DH |
2153 | * 2) wakeup on uaddr2 after a requeue |
2154 | * 3) signal | |
2155 | * 4) timeout | |
52400ba9 | 2156 | * |
cc6db4e6 | 2157 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
2158 | * |
2159 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2160 | * 5) successful lock | |
2161 | * 6) signal | |
2162 | * 7) timeout | |
2163 | * 8) other lock acquisition failure | |
2164 | * | |
cc6db4e6 | 2165 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
2166 | * |
2167 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2168 | * | |
2169 | * Returns: | |
2170 | * 0 - On success | |
2171 | * <0 - On error | |
2172 | */ | |
2173 | static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared, | |
2174 | u32 val, ktime_t *abs_time, u32 bitset, | |
2175 | int clockrt, u32 __user *uaddr2) | |
2176 | { | |
2177 | struct hrtimer_sleeper timeout, *to = NULL; | |
2178 | struct rt_mutex_waiter rt_waiter; | |
2179 | struct rt_mutex *pi_mutex = NULL; | |
52400ba9 DH |
2180 | struct futex_hash_bucket *hb; |
2181 | union futex_key key2; | |
2182 | struct futex_q q; | |
2183 | int res, ret; | |
52400ba9 DH |
2184 | |
2185 | if (!bitset) | |
2186 | return -EINVAL; | |
2187 | ||
2188 | if (abs_time) { | |
2189 | to = &timeout; | |
2190 | hrtimer_init_on_stack(&to->timer, clockrt ? CLOCK_REALTIME : | |
2191 | CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
2192 | hrtimer_init_sleeper(to, current); | |
2193 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2194 | current->timer_slack_ns); | |
2195 | } | |
2196 | ||
2197 | /* | |
2198 | * The waiter is allocated on our stack, manipulated by the requeue | |
2199 | * code while we sleep on uaddr. | |
2200 | */ | |
2201 | debug_rt_mutex_init_waiter(&rt_waiter); | |
2202 | rt_waiter.task = NULL; | |
2203 | ||
52400ba9 | 2204 | key2 = FUTEX_KEY_INIT; |
521c1808 | 2205 | ret = get_futex_key(uaddr2, fshared, &key2, VERIFY_WRITE); |
52400ba9 DH |
2206 | if (unlikely(ret != 0)) |
2207 | goto out; | |
2208 | ||
84bc4af5 DH |
2209 | q.pi_state = NULL; |
2210 | q.bitset = bitset; | |
2211 | q.rt_waiter = &rt_waiter; | |
2212 | q.requeue_pi_key = &key2; | |
2213 | ||
52400ba9 DH |
2214 | /* Prepare to wait on uaddr. */ |
2215 | ret = futex_wait_setup(uaddr, val, fshared, &q, &hb); | |
c8b15a70 TG |
2216 | if (ret) |
2217 | goto out_key2; | |
52400ba9 DH |
2218 | |
2219 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ | |
f1a11e05 | 2220 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2221 | |
2222 | spin_lock(&hb->lock); | |
2223 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2224 | spin_unlock(&hb->lock); | |
2225 | if (ret) | |
2226 | goto out_put_keys; | |
2227 | ||
2228 | /* | |
2229 | * In order for us to be here, we know our q.key == key2, and since | |
2230 | * we took the hb->lock above, we also know that futex_requeue() has | |
2231 | * completed and we no longer have to concern ourselves with a wakeup | |
2232 | * race with the atomic proxy lock acquition by the requeue code. | |
2233 | */ | |
2234 | ||
2235 | /* Check if the requeue code acquired the second futex for us. */ | |
2236 | if (!q.rt_waiter) { | |
2237 | /* | |
2238 | * Got the lock. We might not be the anticipated owner if we | |
2239 | * did a lock-steal - fix up the PI-state in that case. | |
2240 | */ | |
2241 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2242 | spin_lock(q.lock_ptr); | |
2243 | ret = fixup_pi_state_owner(uaddr2, &q, current, | |
2244 | fshared); | |
2245 | spin_unlock(q.lock_ptr); | |
2246 | } | |
2247 | } else { | |
2248 | /* | |
2249 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2250 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2251 | * the pi_state. | |
2252 | */ | |
2253 | WARN_ON(!&q.pi_state); | |
2254 | pi_mutex = &q.pi_state->pi_mutex; | |
2255 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1); | |
2256 | debug_rt_mutex_free_waiter(&rt_waiter); | |
2257 | ||
2258 | spin_lock(q.lock_ptr); | |
2259 | /* | |
2260 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2261 | * haven't already. | |
2262 | */ | |
2263 | res = fixup_owner(uaddr2, fshared, &q, !ret); | |
2264 | /* | |
2265 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 2266 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
2267 | */ |
2268 | if (res) | |
2269 | ret = (res < 0) ? res : 0; | |
2270 | ||
2271 | /* Unqueue and drop the lock. */ | |
2272 | unqueue_me_pi(&q); | |
2273 | } | |
2274 | ||
2275 | /* | |
2276 | * If fixup_pi_state_owner() faulted and was unable to handle the | |
2277 | * fault, unlock the rt_mutex and return the fault to userspace. | |
2278 | */ | |
2279 | if (ret == -EFAULT) { | |
2280 | if (rt_mutex_owner(pi_mutex) == current) | |
2281 | rt_mutex_unlock(pi_mutex); | |
2282 | } else if (ret == -EINTR) { | |
52400ba9 | 2283 | /* |
cc6db4e6 DH |
2284 | * We've already been requeued, but cannot restart by calling |
2285 | * futex_lock_pi() directly. We could restart this syscall, but | |
2286 | * it would detect that the user space "val" changed and return | |
2287 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
2288 | * -EWOULDBLOCK directly. | |
52400ba9 | 2289 | */ |
2070887f | 2290 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2291 | } |
2292 | ||
2293 | out_put_keys: | |
2294 | put_futex_key(fshared, &q.key); | |
c8b15a70 | 2295 | out_key2: |
52400ba9 DH |
2296 | put_futex_key(fshared, &key2); |
2297 | ||
2298 | out: | |
2299 | if (to) { | |
2300 | hrtimer_cancel(&to->timer); | |
2301 | destroy_hrtimer_on_stack(&to->timer); | |
2302 | } | |
2303 | return ret; | |
2304 | } | |
2305 | ||
0771dfef IM |
2306 | /* |
2307 | * Support for robust futexes: the kernel cleans up held futexes at | |
2308 | * thread exit time. | |
2309 | * | |
2310 | * Implementation: user-space maintains a per-thread list of locks it | |
2311 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2312 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2313 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2314 | * always manipulated with the lock held, so the list is private and |
2315 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2316 | * field, to allow the kernel to clean up if the thread dies after | |
2317 | * acquiring the lock, but just before it could have added itself to | |
2318 | * the list. There can only be one such pending lock. | |
2319 | */ | |
2320 | ||
2321 | /** | |
2322 | * sys_set_robust_list - set the robust-futex list head of a task | |
2323 | * @head: pointer to the list-head | |
2324 | * @len: length of the list-head, as userspace expects | |
2325 | */ | |
836f92ad HC |
2326 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2327 | size_t, len) | |
0771dfef | 2328 | { |
a0c1e907 TG |
2329 | if (!futex_cmpxchg_enabled) |
2330 | return -ENOSYS; | |
0771dfef IM |
2331 | /* |
2332 | * The kernel knows only one size for now: | |
2333 | */ | |
2334 | if (unlikely(len != sizeof(*head))) | |
2335 | return -EINVAL; | |
2336 | ||
2337 | current->robust_list = head; | |
2338 | ||
2339 | return 0; | |
2340 | } | |
2341 | ||
2342 | /** | |
2343 | * sys_get_robust_list - get the robust-futex list head of a task | |
2344 | * @pid: pid of the process [zero for current task] | |
2345 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
2346 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
2347 | */ | |
836f92ad HC |
2348 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
2349 | struct robust_list_head __user * __user *, head_ptr, | |
2350 | size_t __user *, len_ptr) | |
0771dfef | 2351 | { |
ba46df98 | 2352 | struct robust_list_head __user *head; |
0771dfef | 2353 | unsigned long ret; |
c69e8d9c | 2354 | const struct cred *cred = current_cred(), *pcred; |
0771dfef | 2355 | |
a0c1e907 TG |
2356 | if (!futex_cmpxchg_enabled) |
2357 | return -ENOSYS; | |
2358 | ||
0771dfef IM |
2359 | if (!pid) |
2360 | head = current->robust_list; | |
2361 | else { | |
2362 | struct task_struct *p; | |
2363 | ||
2364 | ret = -ESRCH; | |
aaa2a97e | 2365 | rcu_read_lock(); |
228ebcbe | 2366 | p = find_task_by_vpid(pid); |
0771dfef IM |
2367 | if (!p) |
2368 | goto err_unlock; | |
2369 | ret = -EPERM; | |
c69e8d9c DH |
2370 | pcred = __task_cred(p); |
2371 | if (cred->euid != pcred->euid && | |
2372 | cred->euid != pcred->uid && | |
76aac0e9 | 2373 | !capable(CAP_SYS_PTRACE)) |
0771dfef IM |
2374 | goto err_unlock; |
2375 | head = p->robust_list; | |
aaa2a97e | 2376 | rcu_read_unlock(); |
0771dfef IM |
2377 | } |
2378 | ||
2379 | if (put_user(sizeof(*head), len_ptr)) | |
2380 | return -EFAULT; | |
2381 | return put_user(head, head_ptr); | |
2382 | ||
2383 | err_unlock: | |
aaa2a97e | 2384 | rcu_read_unlock(); |
0771dfef IM |
2385 | |
2386 | return ret; | |
2387 | } | |
2388 | ||
2389 | /* | |
2390 | * Process a futex-list entry, check whether it's owned by the | |
2391 | * dying task, and do notification if so: | |
2392 | */ | |
e3f2ddea | 2393 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 2394 | { |
e3f2ddea | 2395 | u32 uval, nval, mval; |
0771dfef | 2396 | |
8f17d3a5 IM |
2397 | retry: |
2398 | if (get_user(uval, uaddr)) | |
0771dfef IM |
2399 | return -1; |
2400 | ||
b488893a | 2401 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
2402 | /* |
2403 | * Ok, this dying thread is truly holding a futex | |
2404 | * of interest. Set the OWNER_DIED bit atomically | |
2405 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
2406 | * set, wake up a waiter (if any). (We have to do a | |
2407 | * futex_wake() even if OWNER_DIED is already set - | |
2408 | * to handle the rare but possible case of recursive | |
2409 | * thread-death.) The rest of the cleanup is done in | |
2410 | * userspace. | |
2411 | */ | |
e3f2ddea IM |
2412 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
2413 | nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval); | |
2414 | ||
c87e2837 IM |
2415 | if (nval == -EFAULT) |
2416 | return -1; | |
2417 | ||
2418 | if (nval != uval) | |
8f17d3a5 | 2419 | goto retry; |
0771dfef | 2420 | |
e3f2ddea IM |
2421 | /* |
2422 | * Wake robust non-PI futexes here. The wakeup of | |
2423 | * PI futexes happens in exit_pi_state(): | |
2424 | */ | |
36cf3b5c | 2425 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 2426 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
2427 | } |
2428 | return 0; | |
2429 | } | |
2430 | ||
e3f2ddea IM |
2431 | /* |
2432 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
2433 | */ | |
2434 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 AV |
2435 | struct robust_list __user * __user *head, |
2436 | int *pi) | |
e3f2ddea IM |
2437 | { |
2438 | unsigned long uentry; | |
2439 | ||
ba46df98 | 2440 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
2441 | return -EFAULT; |
2442 | ||
ba46df98 | 2443 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
2444 | *pi = uentry & 1; |
2445 | ||
2446 | return 0; | |
2447 | } | |
2448 | ||
0771dfef IM |
2449 | /* |
2450 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
2451 | * and mark any locks found there dead, and notify any waiters. | |
2452 | * | |
2453 | * We silently return on any sign of list-walking problem. | |
2454 | */ | |
2455 | void exit_robust_list(struct task_struct *curr) | |
2456 | { | |
2457 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e MS |
2458 | struct robust_list __user *entry, *next_entry, *pending; |
2459 | unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip; | |
0771dfef | 2460 | unsigned long futex_offset; |
9f96cb1e | 2461 | int rc; |
0771dfef | 2462 | |
a0c1e907 TG |
2463 | if (!futex_cmpxchg_enabled) |
2464 | return; | |
2465 | ||
0771dfef IM |
2466 | /* |
2467 | * Fetch the list head (which was registered earlier, via | |
2468 | * sys_set_robust_list()): | |
2469 | */ | |
e3f2ddea | 2470 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
2471 | return; |
2472 | /* | |
2473 | * Fetch the relative futex offset: | |
2474 | */ | |
2475 | if (get_user(futex_offset, &head->futex_offset)) | |
2476 | return; | |
2477 | /* | |
2478 | * Fetch any possibly pending lock-add first, and handle it | |
2479 | * if it exists: | |
2480 | */ | |
e3f2ddea | 2481 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 2482 | return; |
e3f2ddea | 2483 | |
9f96cb1e | 2484 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 2485 | while (entry != &head->list) { |
9f96cb1e MS |
2486 | /* |
2487 | * Fetch the next entry in the list before calling | |
2488 | * handle_futex_death: | |
2489 | */ | |
2490 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
2491 | /* |
2492 | * A pending lock might already be on the list, so | |
c87e2837 | 2493 | * don't process it twice: |
0771dfef IM |
2494 | */ |
2495 | if (entry != pending) | |
ba46df98 | 2496 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 2497 | curr, pi)) |
0771dfef | 2498 | return; |
9f96cb1e | 2499 | if (rc) |
0771dfef | 2500 | return; |
9f96cb1e MS |
2501 | entry = next_entry; |
2502 | pi = next_pi; | |
0771dfef IM |
2503 | /* |
2504 | * Avoid excessively long or circular lists: | |
2505 | */ | |
2506 | if (!--limit) | |
2507 | break; | |
2508 | ||
2509 | cond_resched(); | |
2510 | } | |
9f96cb1e MS |
2511 | |
2512 | if (pending) | |
2513 | handle_futex_death((void __user *)pending + futex_offset, | |
2514 | curr, pip); | |
0771dfef IM |
2515 | } |
2516 | ||
c19384b5 | 2517 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 2518 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 2519 | { |
1acdac10 | 2520 | int clockrt, ret = -ENOSYS; |
34f01cc1 | 2521 | int cmd = op & FUTEX_CMD_MASK; |
c2f9f201 | 2522 | int fshared = 0; |
34f01cc1 ED |
2523 | |
2524 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
c2f9f201 | 2525 | fshared = 1; |
1da177e4 | 2526 | |
1acdac10 | 2527 | clockrt = op & FUTEX_CLOCK_REALTIME; |
52400ba9 | 2528 | if (clockrt && cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI) |
1acdac10 | 2529 | return -ENOSYS; |
1da177e4 | 2530 | |
34f01cc1 | 2531 | switch (cmd) { |
1da177e4 | 2532 | case FUTEX_WAIT: |
cd689985 TG |
2533 | val3 = FUTEX_BITSET_MATCH_ANY; |
2534 | case FUTEX_WAIT_BITSET: | |
1acdac10 | 2535 | ret = futex_wait(uaddr, fshared, val, timeout, val3, clockrt); |
1da177e4 LT |
2536 | break; |
2537 | case FUTEX_WAKE: | |
cd689985 TG |
2538 | val3 = FUTEX_BITSET_MATCH_ANY; |
2539 | case FUTEX_WAKE_BITSET: | |
2540 | ret = futex_wake(uaddr, fshared, val, val3); | |
1da177e4 | 2541 | break; |
1da177e4 | 2542 | case FUTEX_REQUEUE: |
52400ba9 | 2543 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL, 0); |
1da177e4 LT |
2544 | break; |
2545 | case FUTEX_CMP_REQUEUE: | |
52400ba9 DH |
2546 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3, |
2547 | 0); | |
1da177e4 | 2548 | break; |
4732efbe | 2549 | case FUTEX_WAKE_OP: |
34f01cc1 | 2550 | ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3); |
4732efbe | 2551 | break; |
c87e2837 | 2552 | case FUTEX_LOCK_PI: |
a0c1e907 TG |
2553 | if (futex_cmpxchg_enabled) |
2554 | ret = futex_lock_pi(uaddr, fshared, val, timeout, 0); | |
c87e2837 IM |
2555 | break; |
2556 | case FUTEX_UNLOCK_PI: | |
a0c1e907 TG |
2557 | if (futex_cmpxchg_enabled) |
2558 | ret = futex_unlock_pi(uaddr, fshared); | |
c87e2837 IM |
2559 | break; |
2560 | case FUTEX_TRYLOCK_PI: | |
a0c1e907 TG |
2561 | if (futex_cmpxchg_enabled) |
2562 | ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1); | |
c87e2837 | 2563 | break; |
52400ba9 DH |
2564 | case FUTEX_WAIT_REQUEUE_PI: |
2565 | val3 = FUTEX_BITSET_MATCH_ANY; | |
2566 | ret = futex_wait_requeue_pi(uaddr, fshared, val, timeout, val3, | |
2567 | clockrt, uaddr2); | |
2568 | break; | |
52400ba9 DH |
2569 | case FUTEX_CMP_REQUEUE_PI: |
2570 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3, | |
2571 | 1); | |
2572 | break; | |
1da177e4 LT |
2573 | default: |
2574 | ret = -ENOSYS; | |
2575 | } | |
2576 | return ret; | |
2577 | } | |
2578 | ||
2579 | ||
17da2bd9 HC |
2580 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
2581 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
2582 | u32, val3) | |
1da177e4 | 2583 | { |
c19384b5 PP |
2584 | struct timespec ts; |
2585 | ktime_t t, *tp = NULL; | |
e2970f2f | 2586 | u32 val2 = 0; |
34f01cc1 | 2587 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 2588 | |
cd689985 | 2589 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
2590 | cmd == FUTEX_WAIT_BITSET || |
2591 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
c19384b5 | 2592 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 2593 | return -EFAULT; |
c19384b5 | 2594 | if (!timespec_valid(&ts)) |
9741ef96 | 2595 | return -EINVAL; |
c19384b5 PP |
2596 | |
2597 | t = timespec_to_ktime(ts); | |
34f01cc1 | 2598 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 2599 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 2600 | tp = &t; |
1da177e4 LT |
2601 | } |
2602 | /* | |
52400ba9 | 2603 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 2604 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 2605 | */ |
f54f0986 | 2606 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 2607 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 2608 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 2609 | |
c19384b5 | 2610 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
2611 | } |
2612 | ||
f6d107fb | 2613 | static int __init futex_init(void) |
1da177e4 | 2614 | { |
a0c1e907 | 2615 | u32 curval; |
3e4ab747 | 2616 | int i; |
95362fa9 | 2617 | |
a0c1e907 TG |
2618 | /* |
2619 | * This will fail and we want it. Some arch implementations do | |
2620 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
2621 | * functionality. We want to know that before we call in any | |
2622 | * of the complex code paths. Also we want to prevent | |
2623 | * registration of robust lists in that case. NULL is | |
2624 | * guaranteed to fault and we get -EFAULT on functional | |
2625 | * implementation, the non functional ones will return | |
2626 | * -ENOSYS. | |
2627 | */ | |
2628 | curval = cmpxchg_futex_value_locked(NULL, 0, 0); | |
2629 | if (curval == -EFAULT) | |
2630 | futex_cmpxchg_enabled = 1; | |
2631 | ||
3e4ab747 TG |
2632 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { |
2633 | plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock); | |
2634 | spin_lock_init(&futex_queues[i].lock); | |
2635 | } | |
2636 | ||
1da177e4 LT |
2637 | return 0; |
2638 | } | |
f6d107fb | 2639 | __initcall(futex_init); |