]>
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 | ||
b41277dc DH |
71 | /* |
72 | * Futex flags used to encode options to functions and preserve them across | |
73 | * restarts. | |
74 | */ | |
75 | #define FLAGS_SHARED 0x01 | |
76 | #define FLAGS_CLOCKRT 0x02 | |
77 | #define FLAGS_HAS_TIMEOUT 0x04 | |
78 | ||
c87e2837 IM |
79 | /* |
80 | * Priority Inheritance state: | |
81 | */ | |
82 | struct futex_pi_state { | |
83 | /* | |
84 | * list of 'owned' pi_state instances - these have to be | |
85 | * cleaned up in do_exit() if the task exits prematurely: | |
86 | */ | |
87 | struct list_head list; | |
88 | ||
89 | /* | |
90 | * The PI object: | |
91 | */ | |
92 | struct rt_mutex pi_mutex; | |
93 | ||
94 | struct task_struct *owner; | |
95 | atomic_t refcount; | |
96 | ||
97 | union futex_key key; | |
98 | }; | |
99 | ||
d8d88fbb DH |
100 | /** |
101 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 102 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
103 | * @task: the task waiting on the futex |
104 | * @lock_ptr: the hash bucket lock | |
105 | * @key: the key the futex is hashed on | |
106 | * @pi_state: optional priority inheritance state | |
107 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
108 | * @requeue_pi_key: the requeue_pi target futex key | |
109 | * @bitset: bitset for the optional bitmasked wakeup | |
110 | * | |
111 | * We use this hashed waitqueue, instead of a normal wait_queue_t, so | |
1da177e4 LT |
112 | * we can wake only the relevant ones (hashed queues may be shared). |
113 | * | |
114 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 115 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 116 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
117 | * the second. |
118 | * | |
119 | * PI futexes are typically woken before they are removed from the hash list via | |
120 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
121 | */ |
122 | struct futex_q { | |
ec92d082 | 123 | struct plist_node list; |
1da177e4 | 124 | |
d8d88fbb | 125 | struct task_struct *task; |
1da177e4 | 126 | spinlock_t *lock_ptr; |
1da177e4 | 127 | union futex_key key; |
c87e2837 | 128 | struct futex_pi_state *pi_state; |
52400ba9 | 129 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 130 | union futex_key *requeue_pi_key; |
cd689985 | 131 | u32 bitset; |
1da177e4 LT |
132 | }; |
133 | ||
5bdb05f9 DH |
134 | static const struct futex_q futex_q_init = { |
135 | /* list gets initialized in queue_me()*/ | |
136 | .key = FUTEX_KEY_INIT, | |
137 | .bitset = FUTEX_BITSET_MATCH_ANY | |
138 | }; | |
139 | ||
1da177e4 | 140 | /* |
b2d0994b DH |
141 | * Hash buckets are shared by all the futex_keys that hash to the same |
142 | * location. Each key may have multiple futex_q structures, one for each task | |
143 | * waiting on a futex. | |
1da177e4 LT |
144 | */ |
145 | struct futex_hash_bucket { | |
ec92d082 PP |
146 | spinlock_t lock; |
147 | struct plist_head chain; | |
1da177e4 LT |
148 | }; |
149 | ||
150 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; | |
151 | ||
1da177e4 LT |
152 | /* |
153 | * We hash on the keys returned from get_futex_key (see below). | |
154 | */ | |
155 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
156 | { | |
157 | u32 hash = jhash2((u32*)&key->both.word, | |
158 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
159 | key->both.offset); | |
160 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; | |
161 | } | |
162 | ||
163 | /* | |
164 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
165 | */ | |
166 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
167 | { | |
2bc87203 DH |
168 | return (key1 && key2 |
169 | && key1->both.word == key2->both.word | |
1da177e4 LT |
170 | && key1->both.ptr == key2->both.ptr |
171 | && key1->both.offset == key2->both.offset); | |
172 | } | |
173 | ||
38d47c1b PZ |
174 | /* |
175 | * Take a reference to the resource addressed by a key. | |
176 | * Can be called while holding spinlocks. | |
177 | * | |
178 | */ | |
179 | static void get_futex_key_refs(union futex_key *key) | |
180 | { | |
181 | if (!key->both.ptr) | |
182 | return; | |
183 | ||
184 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
185 | case FUT_OFF_INODE: | |
7de9c6ee | 186 | ihold(key->shared.inode); |
38d47c1b PZ |
187 | break; |
188 | case FUT_OFF_MMSHARED: | |
189 | atomic_inc(&key->private.mm->mm_count); | |
190 | break; | |
191 | } | |
192 | } | |
193 | ||
194 | /* | |
195 | * Drop a reference to the resource addressed by a key. | |
196 | * The hash bucket spinlock must not be held. | |
197 | */ | |
198 | static void drop_futex_key_refs(union futex_key *key) | |
199 | { | |
90621c40 DH |
200 | if (!key->both.ptr) { |
201 | /* If we're here then we tried to put a key we failed to get */ | |
202 | WARN_ON_ONCE(1); | |
38d47c1b | 203 | return; |
90621c40 | 204 | } |
38d47c1b PZ |
205 | |
206 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
207 | case FUT_OFF_INODE: | |
208 | iput(key->shared.inode); | |
209 | break; | |
210 | case FUT_OFF_MMSHARED: | |
211 | mmdrop(key->private.mm); | |
212 | break; | |
213 | } | |
214 | } | |
215 | ||
34f01cc1 | 216 | /** |
d96ee56c DH |
217 | * get_futex_key() - Get parameters which are the keys for a futex |
218 | * @uaddr: virtual address of the futex | |
219 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED | |
220 | * @key: address where result is stored. | |
34f01cc1 ED |
221 | * |
222 | * Returns a negative error code or 0 | |
223 | * The key words are stored in *key on success. | |
1da177e4 | 224 | * |
f3a43f3f | 225 | * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode, |
1da177e4 LT |
226 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
227 | * We can usually work out the index without swapping in the page. | |
228 | * | |
b2d0994b | 229 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 230 | */ |
64d1304a | 231 | static int |
7485d0d3 | 232 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key) |
1da177e4 | 233 | { |
e2970f2f | 234 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 235 | struct mm_struct *mm = current->mm; |
a5b338f2 | 236 | struct page *page, *page_head; |
1da177e4 LT |
237 | int err; |
238 | ||
239 | /* | |
240 | * The futex address must be "naturally" aligned. | |
241 | */ | |
e2970f2f | 242 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 243 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 244 | return -EINVAL; |
e2970f2f | 245 | address -= key->both.offset; |
1da177e4 | 246 | |
34f01cc1 ED |
247 | /* |
248 | * PROCESS_PRIVATE futexes are fast. | |
249 | * As the mm cannot disappear under us and the 'key' only needs | |
250 | * virtual address, we dont even have to find the underlying vma. | |
251 | * Note : We do have to check 'uaddr' is a valid user address, | |
252 | * but access_ok() should be faster than find_vma() | |
253 | */ | |
254 | if (!fshared) { | |
7485d0d3 | 255 | if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32)))) |
34f01cc1 ED |
256 | return -EFAULT; |
257 | key->private.mm = mm; | |
258 | key->private.address = address; | |
42569c39 | 259 | get_futex_key_refs(key); |
34f01cc1 ED |
260 | return 0; |
261 | } | |
1da177e4 | 262 | |
38d47c1b | 263 | again: |
7485d0d3 | 264 | err = get_user_pages_fast(address, 1, 1, &page); |
38d47c1b PZ |
265 | if (err < 0) |
266 | return err; | |
267 | ||
a5b338f2 AA |
268 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
269 | page_head = page; | |
270 | if (unlikely(PageTail(page))) { | |
38d47c1b | 271 | put_page(page); |
a5b338f2 AA |
272 | /* serialize against __split_huge_page_splitting() */ |
273 | local_irq_disable(); | |
274 | if (likely(__get_user_pages_fast(address, 1, 1, &page) == 1)) { | |
275 | page_head = compound_head(page); | |
276 | /* | |
277 | * page_head is valid pointer but we must pin | |
278 | * it before taking the PG_lock and/or | |
279 | * PG_compound_lock. The moment we re-enable | |
280 | * irqs __split_huge_page_splitting() can | |
281 | * return and the head page can be freed from | |
282 | * under us. We can't take the PG_lock and/or | |
283 | * PG_compound_lock on a page that could be | |
284 | * freed from under us. | |
285 | */ | |
286 | if (page != page_head) { | |
287 | get_page(page_head); | |
288 | put_page(page); | |
289 | } | |
290 | local_irq_enable(); | |
291 | } else { | |
292 | local_irq_enable(); | |
293 | goto again; | |
294 | } | |
295 | } | |
296 | #else | |
297 | page_head = compound_head(page); | |
298 | if (page != page_head) { | |
299 | get_page(page_head); | |
300 | put_page(page); | |
301 | } | |
302 | #endif | |
303 | ||
304 | lock_page(page_head); | |
305 | if (!page_head->mapping) { | |
306 | unlock_page(page_head); | |
307 | put_page(page_head); | |
38d47c1b PZ |
308 | goto again; |
309 | } | |
1da177e4 LT |
310 | |
311 | /* | |
312 | * Private mappings are handled in a simple way. | |
313 | * | |
314 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
315 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 316 | * the object not the particular process. |
1da177e4 | 317 | */ |
a5b338f2 | 318 | if (PageAnon(page_head)) { |
38d47c1b | 319 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 320 | key->private.mm = mm; |
e2970f2f | 321 | key->private.address = address; |
38d47c1b PZ |
322 | } else { |
323 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ | |
a5b338f2 AA |
324 | key->shared.inode = page_head->mapping->host; |
325 | key->shared.pgoff = page_head->index; | |
1da177e4 LT |
326 | } |
327 | ||
38d47c1b | 328 | get_futex_key_refs(key); |
1da177e4 | 329 | |
a5b338f2 AA |
330 | unlock_page(page_head); |
331 | put_page(page_head); | |
38d47c1b | 332 | return 0; |
1da177e4 LT |
333 | } |
334 | ||
ae791a2d | 335 | static inline void put_futex_key(union futex_key *key) |
1da177e4 | 336 | { |
38d47c1b | 337 | drop_futex_key_refs(key); |
1da177e4 LT |
338 | } |
339 | ||
d96ee56c DH |
340 | /** |
341 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
342 | * @uaddr: pointer to faulting user space address |
343 | * | |
344 | * Slow path to fixup the fault we just took in the atomic write | |
345 | * access to @uaddr. | |
346 | * | |
fb62db2b | 347 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
348 | * user address. We know that we faulted in the atomic pagefault |
349 | * disabled section so we can as well avoid the #PF overhead by | |
350 | * calling get_user_pages() right away. | |
351 | */ | |
352 | static int fault_in_user_writeable(u32 __user *uaddr) | |
353 | { | |
722d0172 AK |
354 | struct mm_struct *mm = current->mm; |
355 | int ret; | |
356 | ||
357 | down_read(&mm->mmap_sem); | |
358 | ret = get_user_pages(current, mm, (unsigned long)uaddr, | |
359 | 1, 1, 0, NULL, NULL); | |
360 | up_read(&mm->mmap_sem); | |
361 | ||
d0725992 TG |
362 | return ret < 0 ? ret : 0; |
363 | } | |
364 | ||
4b1c486b DH |
365 | /** |
366 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
367 | * @hb: the hash bucket the futex_q's reside in |
368 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
369 | * |
370 | * Must be called with the hb lock held. | |
371 | */ | |
372 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
373 | union futex_key *key) | |
374 | { | |
375 | struct futex_q *this; | |
376 | ||
377 | plist_for_each_entry(this, &hb->chain, list) { | |
378 | if (match_futex(&this->key, key)) | |
379 | return this; | |
380 | } | |
381 | return NULL; | |
382 | } | |
383 | ||
37a9d912 ML |
384 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
385 | u32 uval, u32 newval) | |
36cf3b5c | 386 | { |
37a9d912 | 387 | int ret; |
36cf3b5c TG |
388 | |
389 | pagefault_disable(); | |
37a9d912 | 390 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
391 | pagefault_enable(); |
392 | ||
37a9d912 | 393 | return ret; |
36cf3b5c TG |
394 | } |
395 | ||
396 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
397 | { |
398 | int ret; | |
399 | ||
a866374a | 400 | pagefault_disable(); |
e2970f2f | 401 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 402 | pagefault_enable(); |
1da177e4 LT |
403 | |
404 | return ret ? -EFAULT : 0; | |
405 | } | |
406 | ||
c87e2837 IM |
407 | |
408 | /* | |
409 | * PI code: | |
410 | */ | |
411 | static int refill_pi_state_cache(void) | |
412 | { | |
413 | struct futex_pi_state *pi_state; | |
414 | ||
415 | if (likely(current->pi_state_cache)) | |
416 | return 0; | |
417 | ||
4668edc3 | 418 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
419 | |
420 | if (!pi_state) | |
421 | return -ENOMEM; | |
422 | ||
c87e2837 IM |
423 | INIT_LIST_HEAD(&pi_state->list); |
424 | /* pi_mutex gets initialized later */ | |
425 | pi_state->owner = NULL; | |
426 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 427 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
428 | |
429 | current->pi_state_cache = pi_state; | |
430 | ||
431 | return 0; | |
432 | } | |
433 | ||
434 | static struct futex_pi_state * alloc_pi_state(void) | |
435 | { | |
436 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
437 | ||
438 | WARN_ON(!pi_state); | |
439 | current->pi_state_cache = NULL; | |
440 | ||
441 | return pi_state; | |
442 | } | |
443 | ||
444 | static void free_pi_state(struct futex_pi_state *pi_state) | |
445 | { | |
446 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
447 | return; | |
448 | ||
449 | /* | |
450 | * If pi_state->owner is NULL, the owner is most probably dying | |
451 | * and has cleaned up the pi_state already | |
452 | */ | |
453 | if (pi_state->owner) { | |
1d615482 | 454 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
c87e2837 | 455 | list_del_init(&pi_state->list); |
1d615482 | 456 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
c87e2837 IM |
457 | |
458 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
459 | } | |
460 | ||
461 | if (current->pi_state_cache) | |
462 | kfree(pi_state); | |
463 | else { | |
464 | /* | |
465 | * pi_state->list is already empty. | |
466 | * clear pi_state->owner. | |
467 | * refcount is at 0 - put it back to 1. | |
468 | */ | |
469 | pi_state->owner = NULL; | |
470 | atomic_set(&pi_state->refcount, 1); | |
471 | current->pi_state_cache = pi_state; | |
472 | } | |
473 | } | |
474 | ||
475 | /* | |
476 | * Look up the task based on what TID userspace gave us. | |
477 | * We dont trust it. | |
478 | */ | |
479 | static struct task_struct * futex_find_get_task(pid_t pid) | |
480 | { | |
481 | struct task_struct *p; | |
482 | ||
d359b549 | 483 | rcu_read_lock(); |
228ebcbe | 484 | p = find_task_by_vpid(pid); |
7a0ea09a MH |
485 | if (p) |
486 | get_task_struct(p); | |
a06381fe | 487 | |
d359b549 | 488 | rcu_read_unlock(); |
c87e2837 IM |
489 | |
490 | return p; | |
491 | } | |
492 | ||
493 | /* | |
494 | * This task is holding PI mutexes at exit time => bad. | |
495 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
496 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
497 | */ | |
498 | void exit_pi_state_list(struct task_struct *curr) | |
499 | { | |
c87e2837 IM |
500 | struct list_head *next, *head = &curr->pi_state_list; |
501 | struct futex_pi_state *pi_state; | |
627371d7 | 502 | struct futex_hash_bucket *hb; |
38d47c1b | 503 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 504 | |
a0c1e907 TG |
505 | if (!futex_cmpxchg_enabled) |
506 | return; | |
c87e2837 IM |
507 | /* |
508 | * We are a ZOMBIE and nobody can enqueue itself on | |
509 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 510 | * versus waiters unqueueing themselves: |
c87e2837 | 511 | */ |
1d615482 | 512 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 IM |
513 | while (!list_empty(head)) { |
514 | ||
515 | next = head->next; | |
516 | pi_state = list_entry(next, struct futex_pi_state, list); | |
517 | key = pi_state->key; | |
627371d7 | 518 | hb = hash_futex(&key); |
1d615482 | 519 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 520 | |
c87e2837 IM |
521 | spin_lock(&hb->lock); |
522 | ||
1d615482 | 523 | raw_spin_lock_irq(&curr->pi_lock); |
627371d7 IM |
524 | /* |
525 | * We dropped the pi-lock, so re-check whether this | |
526 | * task still owns the PI-state: | |
527 | */ | |
c87e2837 IM |
528 | if (head->next != next) { |
529 | spin_unlock(&hb->lock); | |
530 | continue; | |
531 | } | |
532 | ||
c87e2837 | 533 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
534 | WARN_ON(list_empty(&pi_state->list)); |
535 | list_del_init(&pi_state->list); | |
c87e2837 | 536 | pi_state->owner = NULL; |
1d615482 | 537 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
538 | |
539 | rt_mutex_unlock(&pi_state->pi_mutex); | |
540 | ||
541 | spin_unlock(&hb->lock); | |
542 | ||
1d615482 | 543 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 544 | } |
1d615482 | 545 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
546 | } |
547 | ||
548 | static int | |
d0aa7a70 PP |
549 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
550 | union futex_key *key, struct futex_pi_state **ps) | |
c87e2837 IM |
551 | { |
552 | struct futex_pi_state *pi_state = NULL; | |
553 | struct futex_q *this, *next; | |
ec92d082 | 554 | struct plist_head *head; |
c87e2837 | 555 | struct task_struct *p; |
778e9a9c | 556 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 IM |
557 | |
558 | head = &hb->chain; | |
559 | ||
ec92d082 | 560 | plist_for_each_entry_safe(this, next, head, list) { |
d0aa7a70 | 561 | if (match_futex(&this->key, key)) { |
c87e2837 IM |
562 | /* |
563 | * Another waiter already exists - bump up | |
564 | * the refcount and return its pi_state: | |
565 | */ | |
566 | pi_state = this->pi_state; | |
06a9ec29 | 567 | /* |
fb62db2b | 568 | * Userspace might have messed up non-PI and PI futexes |
06a9ec29 TG |
569 | */ |
570 | if (unlikely(!pi_state)) | |
571 | return -EINVAL; | |
572 | ||
627371d7 | 573 | WARN_ON(!atomic_read(&pi_state->refcount)); |
59647b6a TG |
574 | |
575 | /* | |
576 | * When pi_state->owner is NULL then the owner died | |
577 | * and another waiter is on the fly. pi_state->owner | |
578 | * is fixed up by the task which acquires | |
579 | * pi_state->rt_mutex. | |
580 | * | |
581 | * We do not check for pid == 0 which can happen when | |
582 | * the owner died and robust_list_exit() cleared the | |
583 | * TID. | |
584 | */ | |
585 | if (pid && pi_state->owner) { | |
586 | /* | |
587 | * Bail out if user space manipulated the | |
588 | * futex value. | |
589 | */ | |
590 | if (pid != task_pid_vnr(pi_state->owner)) | |
591 | return -EINVAL; | |
592 | } | |
627371d7 | 593 | |
c87e2837 | 594 | atomic_inc(&pi_state->refcount); |
d0aa7a70 | 595 | *ps = pi_state; |
c87e2837 IM |
596 | |
597 | return 0; | |
598 | } | |
599 | } | |
600 | ||
601 | /* | |
e3f2ddea | 602 | * We are the first waiter - try to look up the real owner and attach |
778e9a9c | 603 | * the new pi_state to it, but bail out when TID = 0 |
c87e2837 | 604 | */ |
778e9a9c | 605 | if (!pid) |
e3f2ddea | 606 | return -ESRCH; |
c87e2837 | 607 | p = futex_find_get_task(pid); |
7a0ea09a MH |
608 | if (!p) |
609 | return -ESRCH; | |
778e9a9c AK |
610 | |
611 | /* | |
612 | * We need to look at the task state flags to figure out, | |
613 | * whether the task is exiting. To protect against the do_exit | |
614 | * change of the task flags, we do this protected by | |
615 | * p->pi_lock: | |
616 | */ | |
1d615482 | 617 | raw_spin_lock_irq(&p->pi_lock); |
778e9a9c AK |
618 | if (unlikely(p->flags & PF_EXITING)) { |
619 | /* | |
620 | * The task is on the way out. When PF_EXITPIDONE is | |
621 | * set, we know that the task has finished the | |
622 | * cleanup: | |
623 | */ | |
624 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
625 | ||
1d615482 | 626 | raw_spin_unlock_irq(&p->pi_lock); |
778e9a9c AK |
627 | put_task_struct(p); |
628 | return ret; | |
629 | } | |
c87e2837 IM |
630 | |
631 | pi_state = alloc_pi_state(); | |
632 | ||
633 | /* | |
634 | * Initialize the pi_mutex in locked state and make 'p' | |
635 | * the owner of it: | |
636 | */ | |
637 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
638 | ||
639 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 640 | pi_state->key = *key; |
c87e2837 | 641 | |
627371d7 | 642 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
643 | list_add(&pi_state->list, &p->pi_state_list); |
644 | pi_state->owner = p; | |
1d615482 | 645 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
646 | |
647 | put_task_struct(p); | |
648 | ||
d0aa7a70 | 649 | *ps = pi_state; |
c87e2837 IM |
650 | |
651 | return 0; | |
652 | } | |
653 | ||
1a52084d | 654 | /** |
d96ee56c | 655 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
656 | * @uaddr: the pi futex user address |
657 | * @hb: the pi futex hash bucket | |
658 | * @key: the futex key associated with uaddr and hb | |
659 | * @ps: the pi_state pointer where we store the result of the | |
660 | * lookup | |
661 | * @task: the task to perform the atomic lock work for. This will | |
662 | * be "current" except in the case of requeue pi. | |
663 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d DH |
664 | * |
665 | * Returns: | |
666 | * 0 - ready to wait | |
667 | * 1 - acquired the lock | |
668 | * <0 - error | |
669 | * | |
670 | * The hb->lock and futex_key refs shall be held by the caller. | |
671 | */ | |
672 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
673 | union futex_key *key, | |
674 | struct futex_pi_state **ps, | |
bab5bc9e | 675 | struct task_struct *task, int set_waiters) |
1a52084d DH |
676 | { |
677 | int lock_taken, ret, ownerdied = 0; | |
c0c9ed15 | 678 | u32 uval, newval, curval, vpid = task_pid_vnr(task); |
1a52084d DH |
679 | |
680 | retry: | |
681 | ret = lock_taken = 0; | |
682 | ||
683 | /* | |
684 | * To avoid races, we attempt to take the lock here again | |
685 | * (by doing a 0 -> TID atomic cmpxchg), while holding all | |
686 | * the locks. It will most likely not succeed. | |
687 | */ | |
c0c9ed15 | 688 | newval = vpid; |
bab5bc9e DH |
689 | if (set_waiters) |
690 | newval |= FUTEX_WAITERS; | |
1a52084d | 691 | |
37a9d912 | 692 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval))) |
1a52084d DH |
693 | return -EFAULT; |
694 | ||
695 | /* | |
696 | * Detect deadlocks. | |
697 | */ | |
c0c9ed15 | 698 | if ((unlikely((curval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
699 | return -EDEADLK; |
700 | ||
701 | /* | |
702 | * Surprise - we got the lock. Just return to userspace: | |
703 | */ | |
704 | if (unlikely(!curval)) | |
705 | return 1; | |
706 | ||
707 | uval = curval; | |
708 | ||
709 | /* | |
710 | * Set the FUTEX_WAITERS flag, so the owner will know it has someone | |
711 | * to wake at the next unlock. | |
712 | */ | |
713 | newval = curval | FUTEX_WAITERS; | |
714 | ||
715 | /* | |
716 | * There are two cases, where a futex might have no owner (the | |
717 | * owner TID is 0): OWNER_DIED. We take over the futex in this | |
718 | * case. We also do an unconditional take over, when the owner | |
719 | * of the futex died. | |
720 | * | |
721 | * This is safe as we are protected by the hash bucket lock ! | |
722 | */ | |
723 | if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { | |
724 | /* Keep the OWNER_DIED bit */ | |
c0c9ed15 | 725 | newval = (curval & ~FUTEX_TID_MASK) | vpid; |
1a52084d DH |
726 | ownerdied = 0; |
727 | lock_taken = 1; | |
728 | } | |
729 | ||
37a9d912 | 730 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))) |
1a52084d DH |
731 | return -EFAULT; |
732 | if (unlikely(curval != uval)) | |
733 | goto retry; | |
734 | ||
735 | /* | |
736 | * We took the lock due to owner died take over. | |
737 | */ | |
738 | if (unlikely(lock_taken)) | |
739 | return 1; | |
740 | ||
741 | /* | |
742 | * We dont have the lock. Look up the PI state (or create it if | |
743 | * we are the first waiter): | |
744 | */ | |
745 | ret = lookup_pi_state(uval, hb, key, ps); | |
746 | ||
747 | if (unlikely(ret)) { | |
748 | switch (ret) { | |
749 | case -ESRCH: | |
750 | /* | |
751 | * No owner found for this futex. Check if the | |
752 | * OWNER_DIED bit is set to figure out whether | |
753 | * this is a robust futex or not. | |
754 | */ | |
755 | if (get_futex_value_locked(&curval, uaddr)) | |
756 | return -EFAULT; | |
757 | ||
758 | /* | |
759 | * We simply start over in case of a robust | |
760 | * futex. The code above will take the futex | |
761 | * and return happy. | |
762 | */ | |
763 | if (curval & FUTEX_OWNER_DIED) { | |
764 | ownerdied = 1; | |
765 | goto retry; | |
766 | } | |
767 | default: | |
768 | break; | |
769 | } | |
770 | } | |
771 | ||
772 | return ret; | |
773 | } | |
774 | ||
2e12978a LJ |
775 | /** |
776 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
777 | * @q: The futex_q to unqueue | |
778 | * | |
779 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
780 | */ | |
781 | static void __unqueue_futex(struct futex_q *q) | |
782 | { | |
783 | struct futex_hash_bucket *hb; | |
784 | ||
785 | if (WARN_ON(!q->lock_ptr || !spin_is_locked(q->lock_ptr) | |
786 | || plist_node_empty(&q->list))) | |
787 | return; | |
788 | ||
789 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
790 | plist_del(&q->list, &hb->chain); | |
791 | } | |
792 | ||
1da177e4 LT |
793 | /* |
794 | * The hash bucket lock must be held when this is called. | |
795 | * Afterwards, the futex_q must not be accessed. | |
796 | */ | |
797 | static void wake_futex(struct futex_q *q) | |
798 | { | |
f1a11e05 TG |
799 | struct task_struct *p = q->task; |
800 | ||
1da177e4 | 801 | /* |
f1a11e05 | 802 | * We set q->lock_ptr = NULL _before_ we wake up the task. If |
fb62db2b RD |
803 | * a non-futex wake up happens on another CPU then the task |
804 | * might exit and p would dereference a non-existing task | |
f1a11e05 TG |
805 | * struct. Prevent this by holding a reference on p across the |
806 | * wake up. | |
1da177e4 | 807 | */ |
f1a11e05 TG |
808 | get_task_struct(p); |
809 | ||
2e12978a | 810 | __unqueue_futex(q); |
1da177e4 | 811 | /* |
f1a11e05 TG |
812 | * The waiting task can free the futex_q as soon as |
813 | * q->lock_ptr = NULL is written, without taking any locks. A | |
814 | * memory barrier is required here to prevent the following | |
815 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 816 | */ |
ccdea2f8 | 817 | smp_wmb(); |
1da177e4 | 818 | q->lock_ptr = NULL; |
f1a11e05 TG |
819 | |
820 | wake_up_state(p, TASK_NORMAL); | |
821 | put_task_struct(p); | |
1da177e4 LT |
822 | } |
823 | ||
c87e2837 IM |
824 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
825 | { | |
826 | struct task_struct *new_owner; | |
827 | struct futex_pi_state *pi_state = this->pi_state; | |
828 | u32 curval, newval; | |
829 | ||
830 | if (!pi_state) | |
831 | return -EINVAL; | |
832 | ||
51246bfd TG |
833 | /* |
834 | * If current does not own the pi_state then the futex is | |
835 | * inconsistent and user space fiddled with the futex value. | |
836 | */ | |
837 | if (pi_state->owner != current) | |
838 | return -EINVAL; | |
839 | ||
d209d74d | 840 | raw_spin_lock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
841 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
842 | ||
843 | /* | |
f123c98e SR |
844 | * It is possible that the next waiter (the one that brought |
845 | * this owner to the kernel) timed out and is no longer | |
846 | * waiting on the lock. | |
c87e2837 IM |
847 | */ |
848 | if (!new_owner) | |
849 | new_owner = this->task; | |
850 | ||
851 | /* | |
852 | * We pass it to the next owner. (The WAITERS bit is always | |
853 | * kept enabled while there is PI state around. We must also | |
854 | * preserve the owner died bit.) | |
855 | */ | |
e3f2ddea | 856 | if (!(uval & FUTEX_OWNER_DIED)) { |
778e9a9c AK |
857 | int ret = 0; |
858 | ||
b488893a | 859 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 860 | |
37a9d912 | 861 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
778e9a9c | 862 | ret = -EFAULT; |
cde898fa | 863 | else if (curval != uval) |
778e9a9c AK |
864 | ret = -EINVAL; |
865 | if (ret) { | |
d209d74d | 866 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
778e9a9c AK |
867 | return ret; |
868 | } | |
e3f2ddea | 869 | } |
c87e2837 | 870 | |
1d615482 | 871 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
627371d7 IM |
872 | WARN_ON(list_empty(&pi_state->list)); |
873 | list_del_init(&pi_state->list); | |
1d615482 | 874 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
627371d7 | 875 | |
1d615482 | 876 | raw_spin_lock_irq(&new_owner->pi_lock); |
627371d7 | 877 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
878 | list_add(&pi_state->list, &new_owner->pi_state_list); |
879 | pi_state->owner = new_owner; | |
1d615482 | 880 | raw_spin_unlock_irq(&new_owner->pi_lock); |
627371d7 | 881 | |
d209d74d | 882 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
883 | rt_mutex_unlock(&pi_state->pi_mutex); |
884 | ||
885 | return 0; | |
886 | } | |
887 | ||
888 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) | |
889 | { | |
890 | u32 oldval; | |
891 | ||
892 | /* | |
893 | * There is no waiter, so we unlock the futex. The owner died | |
894 | * bit has not to be preserved here. We are the owner: | |
895 | */ | |
37a9d912 ML |
896 | if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0)) |
897 | return -EFAULT; | |
c87e2837 IM |
898 | if (oldval != uval) |
899 | return -EAGAIN; | |
900 | ||
901 | return 0; | |
902 | } | |
903 | ||
8b8f319f IM |
904 | /* |
905 | * Express the locking dependencies for lockdep: | |
906 | */ | |
907 | static inline void | |
908 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
909 | { | |
910 | if (hb1 <= hb2) { | |
911 | spin_lock(&hb1->lock); | |
912 | if (hb1 < hb2) | |
913 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
914 | } else { /* hb1 > hb2 */ | |
915 | spin_lock(&hb2->lock); | |
916 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
917 | } | |
918 | } | |
919 | ||
5eb3dc62 DH |
920 | static inline void |
921 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
922 | { | |
f061d351 | 923 | spin_unlock(&hb1->lock); |
88f502fe IM |
924 | if (hb1 != hb2) |
925 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
926 | } |
927 | ||
1da177e4 | 928 | /* |
b2d0994b | 929 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 930 | */ |
b41277dc DH |
931 | static int |
932 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 933 | { |
e2970f2f | 934 | struct futex_hash_bucket *hb; |
1da177e4 | 935 | struct futex_q *this, *next; |
ec92d082 | 936 | struct plist_head *head; |
38d47c1b | 937 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 LT |
938 | int ret; |
939 | ||
cd689985 TG |
940 | if (!bitset) |
941 | return -EINVAL; | |
942 | ||
b41277dc | 943 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key); |
1da177e4 LT |
944 | if (unlikely(ret != 0)) |
945 | goto out; | |
946 | ||
e2970f2f IM |
947 | hb = hash_futex(&key); |
948 | spin_lock(&hb->lock); | |
949 | head = &hb->chain; | |
1da177e4 | 950 | |
ec92d082 | 951 | plist_for_each_entry_safe(this, next, head, list) { |
1da177e4 | 952 | if (match_futex (&this->key, &key)) { |
52400ba9 | 953 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
954 | ret = -EINVAL; |
955 | break; | |
956 | } | |
cd689985 TG |
957 | |
958 | /* Check if one of the bits is set in both bitsets */ | |
959 | if (!(this->bitset & bitset)) | |
960 | continue; | |
961 | ||
1da177e4 LT |
962 | wake_futex(this); |
963 | if (++ret >= nr_wake) | |
964 | break; | |
965 | } | |
966 | } | |
967 | ||
e2970f2f | 968 | spin_unlock(&hb->lock); |
ae791a2d | 969 | put_futex_key(&key); |
42d35d48 | 970 | out: |
1da177e4 LT |
971 | return ret; |
972 | } | |
973 | ||
4732efbe JJ |
974 | /* |
975 | * Wake up all waiters hashed on the physical page that is mapped | |
976 | * to this virtual address: | |
977 | */ | |
e2970f2f | 978 | static int |
b41277dc | 979 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 980 | int nr_wake, int nr_wake2, int op) |
4732efbe | 981 | { |
38d47c1b | 982 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 983 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 984 | struct plist_head *head; |
4732efbe | 985 | struct futex_q *this, *next; |
e4dc5b7a | 986 | int ret, op_ret; |
4732efbe | 987 | |
e4dc5b7a | 988 | retry: |
b41277dc | 989 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1); |
4732efbe JJ |
990 | if (unlikely(ret != 0)) |
991 | goto out; | |
b41277dc | 992 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2); |
4732efbe | 993 | if (unlikely(ret != 0)) |
42d35d48 | 994 | goto out_put_key1; |
4732efbe | 995 | |
e2970f2f IM |
996 | hb1 = hash_futex(&key1); |
997 | hb2 = hash_futex(&key2); | |
4732efbe | 998 | |
e4dc5b7a | 999 | retry_private: |
eaaea803 | 1000 | double_lock_hb(hb1, hb2); |
e2970f2f | 1001 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1002 | if (unlikely(op_ret < 0)) { |
4732efbe | 1003 | |
5eb3dc62 | 1004 | double_unlock_hb(hb1, hb2); |
4732efbe | 1005 | |
7ee1dd3f | 1006 | #ifndef CONFIG_MMU |
e2970f2f IM |
1007 | /* |
1008 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
1009 | * but we might get them from range checking | |
1010 | */ | |
7ee1dd3f | 1011 | ret = op_ret; |
42d35d48 | 1012 | goto out_put_keys; |
7ee1dd3f DH |
1013 | #endif |
1014 | ||
796f8d9b DG |
1015 | if (unlikely(op_ret != -EFAULT)) { |
1016 | ret = op_ret; | |
42d35d48 | 1017 | goto out_put_keys; |
796f8d9b DG |
1018 | } |
1019 | ||
d0725992 | 1020 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 1021 | if (ret) |
de87fcc1 | 1022 | goto out_put_keys; |
4732efbe | 1023 | |
b41277dc | 1024 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1025 | goto retry_private; |
1026 | ||
ae791a2d TG |
1027 | put_futex_key(&key2); |
1028 | put_futex_key(&key1); | |
e4dc5b7a | 1029 | goto retry; |
4732efbe JJ |
1030 | } |
1031 | ||
e2970f2f | 1032 | head = &hb1->chain; |
4732efbe | 1033 | |
ec92d082 | 1034 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
1035 | if (match_futex (&this->key, &key1)) { |
1036 | wake_futex(this); | |
1037 | if (++ret >= nr_wake) | |
1038 | break; | |
1039 | } | |
1040 | } | |
1041 | ||
1042 | if (op_ret > 0) { | |
e2970f2f | 1043 | head = &hb2->chain; |
4732efbe JJ |
1044 | |
1045 | op_ret = 0; | |
ec92d082 | 1046 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
1047 | if (match_futex (&this->key, &key2)) { |
1048 | wake_futex(this); | |
1049 | if (++op_ret >= nr_wake2) | |
1050 | break; | |
1051 | } | |
1052 | } | |
1053 | ret += op_ret; | |
1054 | } | |
1055 | ||
5eb3dc62 | 1056 | double_unlock_hb(hb1, hb2); |
42d35d48 | 1057 | out_put_keys: |
ae791a2d | 1058 | put_futex_key(&key2); |
42d35d48 | 1059 | out_put_key1: |
ae791a2d | 1060 | put_futex_key(&key1); |
42d35d48 | 1061 | out: |
4732efbe JJ |
1062 | return ret; |
1063 | } | |
1064 | ||
9121e478 DH |
1065 | /** |
1066 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1067 | * @q: the futex_q to requeue | |
1068 | * @hb1: the source hash_bucket | |
1069 | * @hb2: the target hash_bucket | |
1070 | * @key2: the new key for the requeued futex_q | |
1071 | */ | |
1072 | static inline | |
1073 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1074 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1075 | { | |
1076 | ||
1077 | /* | |
1078 | * If key1 and key2 hash to the same bucket, no need to | |
1079 | * requeue. | |
1080 | */ | |
1081 | if (likely(&hb1->chain != &hb2->chain)) { | |
1082 | plist_del(&q->list, &hb1->chain); | |
1083 | plist_add(&q->list, &hb2->chain); | |
1084 | q->lock_ptr = &hb2->lock; | |
9121e478 DH |
1085 | } |
1086 | get_futex_key_refs(key2); | |
1087 | q->key = *key2; | |
1088 | } | |
1089 | ||
52400ba9 DH |
1090 | /** |
1091 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1092 | * @q: the futex_q |
1093 | * @key: the key of the requeue target futex | |
1094 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1095 | * |
1096 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1097 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1098 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1099 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1100 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1101 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1102 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1103 | */ |
1104 | static inline | |
beda2c7e DH |
1105 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1106 | struct futex_hash_bucket *hb) | |
52400ba9 | 1107 | { |
52400ba9 DH |
1108 | get_futex_key_refs(key); |
1109 | q->key = *key; | |
1110 | ||
2e12978a | 1111 | __unqueue_futex(q); |
52400ba9 DH |
1112 | |
1113 | WARN_ON(!q->rt_waiter); | |
1114 | q->rt_waiter = NULL; | |
1115 | ||
beda2c7e | 1116 | q->lock_ptr = &hb->lock; |
beda2c7e | 1117 | |
f1a11e05 | 1118 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1119 | } |
1120 | ||
1121 | /** | |
1122 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1123 | * @pifutex: the user address of the to futex |
1124 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1125 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1126 | * @key1: the from futex key | |
1127 | * @key2: the to futex key | |
1128 | * @ps: address to store the pi_state pointer | |
1129 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1130 | * |
1131 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1132 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1133 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1134 | * hb1 and hb2 must be held by the caller. | |
52400ba9 DH |
1135 | * |
1136 | * Returns: | |
1137 | * 0 - failed to acquire the lock atomicly | |
1138 | * 1 - acquired the lock | |
1139 | * <0 - error | |
1140 | */ | |
1141 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1142 | struct futex_hash_bucket *hb1, | |
1143 | struct futex_hash_bucket *hb2, | |
1144 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1145 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1146 | { |
bab5bc9e | 1147 | struct futex_q *top_waiter = NULL; |
52400ba9 DH |
1148 | u32 curval; |
1149 | int ret; | |
1150 | ||
1151 | if (get_futex_value_locked(&curval, pifutex)) | |
1152 | return -EFAULT; | |
1153 | ||
bab5bc9e DH |
1154 | /* |
1155 | * Find the top_waiter and determine if there are additional waiters. | |
1156 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1157 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1158 | * as we have means to handle the possible fault. If not, don't set | |
1159 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1160 | * the kernel. | |
1161 | */ | |
52400ba9 DH |
1162 | top_waiter = futex_top_waiter(hb1, key1); |
1163 | ||
1164 | /* There are no waiters, nothing for us to do. */ | |
1165 | if (!top_waiter) | |
1166 | return 0; | |
1167 | ||
84bc4af5 DH |
1168 | /* Ensure we requeue to the expected futex. */ |
1169 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1170 | return -EINVAL; | |
1171 | ||
52400ba9 | 1172 | /* |
bab5bc9e DH |
1173 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1174 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1175 | * in ps in contended cases. | |
52400ba9 | 1176 | */ |
bab5bc9e DH |
1177 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1178 | set_waiters); | |
52400ba9 | 1179 | if (ret == 1) |
beda2c7e | 1180 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
52400ba9 DH |
1181 | |
1182 | return ret; | |
1183 | } | |
1184 | ||
1185 | /** | |
1186 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1187 | * @uaddr1: source futex user address |
b41277dc | 1188 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1189 | * @uaddr2: target futex user address |
1190 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1191 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1192 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1193 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1194 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1195 | * |
1196 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1197 | * uaddr2 atomically on behalf of the top waiter. | |
1198 | * | |
1199 | * Returns: | |
1200 | * >=0 - on success, the number of tasks requeued or woken | |
1201 | * <0 - on error | |
1da177e4 | 1202 | */ |
b41277dc DH |
1203 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
1204 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
1205 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 1206 | { |
38d47c1b | 1207 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1208 | int drop_count = 0, task_count = 0, ret; |
1209 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1210 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 1211 | struct plist_head *head1; |
1da177e4 | 1212 | struct futex_q *this, *next; |
52400ba9 DH |
1213 | u32 curval2; |
1214 | ||
1215 | if (requeue_pi) { | |
1216 | /* | |
1217 | * requeue_pi requires a pi_state, try to allocate it now | |
1218 | * without any locks in case it fails. | |
1219 | */ | |
1220 | if (refill_pi_state_cache()) | |
1221 | return -ENOMEM; | |
1222 | /* | |
1223 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1224 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1225 | * returning to userspace, so as to not leave the rt_mutex with | |
1226 | * waiters and no owner. However, second and third wake-ups | |
1227 | * cannot be predicted as they involve race conditions with the | |
1228 | * first wake and a fault while looking up the pi_state. Both | |
1229 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1230 | * use nr_wake=1. | |
1231 | */ | |
1232 | if (nr_wake != 1) | |
1233 | return -EINVAL; | |
1234 | } | |
1da177e4 | 1235 | |
42d35d48 | 1236 | retry: |
52400ba9 DH |
1237 | if (pi_state != NULL) { |
1238 | /* | |
1239 | * We will have to lookup the pi_state again, so free this one | |
1240 | * to keep the accounting correct. | |
1241 | */ | |
1242 | free_pi_state(pi_state); | |
1243 | pi_state = NULL; | |
1244 | } | |
1245 | ||
b41277dc | 1246 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1); |
1da177e4 LT |
1247 | if (unlikely(ret != 0)) |
1248 | goto out; | |
b41277dc | 1249 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2); |
1da177e4 | 1250 | if (unlikely(ret != 0)) |
42d35d48 | 1251 | goto out_put_key1; |
1da177e4 | 1252 | |
e2970f2f IM |
1253 | hb1 = hash_futex(&key1); |
1254 | hb2 = hash_futex(&key2); | |
1da177e4 | 1255 | |
e4dc5b7a | 1256 | retry_private: |
8b8f319f | 1257 | double_lock_hb(hb1, hb2); |
1da177e4 | 1258 | |
e2970f2f IM |
1259 | if (likely(cmpval != NULL)) { |
1260 | u32 curval; | |
1da177e4 | 1261 | |
e2970f2f | 1262 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1263 | |
1264 | if (unlikely(ret)) { | |
5eb3dc62 | 1265 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1266 | |
e2970f2f | 1267 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1268 | if (ret) |
1269 | goto out_put_keys; | |
1da177e4 | 1270 | |
b41277dc | 1271 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 1272 | goto retry_private; |
1da177e4 | 1273 | |
ae791a2d TG |
1274 | put_futex_key(&key2); |
1275 | put_futex_key(&key1); | |
e4dc5b7a | 1276 | goto retry; |
1da177e4 | 1277 | } |
e2970f2f | 1278 | if (curval != *cmpval) { |
1da177e4 LT |
1279 | ret = -EAGAIN; |
1280 | goto out_unlock; | |
1281 | } | |
1282 | } | |
1283 | ||
52400ba9 | 1284 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1285 | /* |
1286 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1287 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1288 | * bit. We force this here where we are able to easily handle | |
1289 | * faults rather in the requeue loop below. | |
1290 | */ | |
52400ba9 | 1291 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1292 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1293 | |
1294 | /* | |
1295 | * At this point the top_waiter has either taken uaddr2 or is | |
1296 | * waiting on it. If the former, then the pi_state will not | |
1297 | * exist yet, look it up one more time to ensure we have a | |
1298 | * reference to it. | |
1299 | */ | |
1300 | if (ret == 1) { | |
1301 | WARN_ON(pi_state); | |
89061d3d | 1302 | drop_count++; |
52400ba9 DH |
1303 | task_count++; |
1304 | ret = get_futex_value_locked(&curval2, uaddr2); | |
1305 | if (!ret) | |
1306 | ret = lookup_pi_state(curval2, hb2, &key2, | |
1307 | &pi_state); | |
1308 | } | |
1309 | ||
1310 | switch (ret) { | |
1311 | case 0: | |
1312 | break; | |
1313 | case -EFAULT: | |
1314 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1315 | put_futex_key(&key2); |
1316 | put_futex_key(&key1); | |
d0725992 | 1317 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
1318 | if (!ret) |
1319 | goto retry; | |
1320 | goto out; | |
1321 | case -EAGAIN: | |
1322 | /* The owner was exiting, try again. */ | |
1323 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1324 | put_futex_key(&key2); |
1325 | put_futex_key(&key1); | |
52400ba9 DH |
1326 | cond_resched(); |
1327 | goto retry; | |
1328 | default: | |
1329 | goto out_unlock; | |
1330 | } | |
1331 | } | |
1332 | ||
e2970f2f | 1333 | head1 = &hb1->chain; |
ec92d082 | 1334 | plist_for_each_entry_safe(this, next, head1, list) { |
52400ba9 DH |
1335 | if (task_count - nr_wake >= nr_requeue) |
1336 | break; | |
1337 | ||
1338 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1339 | continue; |
52400ba9 | 1340 | |
392741e0 DH |
1341 | /* |
1342 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
1343 | * be paired with each other and no other futex ops. | |
1344 | */ | |
1345 | if ((requeue_pi && !this->rt_waiter) || | |
1346 | (!requeue_pi && this->rt_waiter)) { | |
1347 | ret = -EINVAL; | |
1348 | break; | |
1349 | } | |
52400ba9 DH |
1350 | |
1351 | /* | |
1352 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1353 | * lock, we already woke the top_waiter. If not, it will be | |
1354 | * woken by futex_unlock_pi(). | |
1355 | */ | |
1356 | if (++task_count <= nr_wake && !requeue_pi) { | |
1da177e4 | 1357 | wake_futex(this); |
52400ba9 DH |
1358 | continue; |
1359 | } | |
1da177e4 | 1360 | |
84bc4af5 DH |
1361 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
1362 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
1363 | ret = -EINVAL; | |
1364 | break; | |
1365 | } | |
1366 | ||
52400ba9 DH |
1367 | /* |
1368 | * Requeue nr_requeue waiters and possibly one more in the case | |
1369 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1370 | */ | |
1371 | if (requeue_pi) { | |
1372 | /* Prepare the waiter to take the rt_mutex. */ | |
1373 | atomic_inc(&pi_state->refcount); | |
1374 | this->pi_state = pi_state; | |
1375 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1376 | this->rt_waiter, | |
1377 | this->task, 1); | |
1378 | if (ret == 1) { | |
1379 | /* We got the lock. */ | |
beda2c7e | 1380 | requeue_pi_wake_futex(this, &key2, hb2); |
89061d3d | 1381 | drop_count++; |
52400ba9 DH |
1382 | continue; |
1383 | } else if (ret) { | |
1384 | /* -EDEADLK */ | |
1385 | this->pi_state = NULL; | |
1386 | free_pi_state(pi_state); | |
1387 | goto out_unlock; | |
1388 | } | |
1da177e4 | 1389 | } |
52400ba9 DH |
1390 | requeue_futex(this, hb1, hb2, &key2); |
1391 | drop_count++; | |
1da177e4 LT |
1392 | } |
1393 | ||
1394 | out_unlock: | |
5eb3dc62 | 1395 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1396 | |
cd84a42f DH |
1397 | /* |
1398 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1399 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1400 | * one at key2 and updated their key pointer. We no longer need to | |
1401 | * hold the references to key1. | |
1402 | */ | |
1da177e4 | 1403 | while (--drop_count >= 0) |
9adef58b | 1404 | drop_futex_key_refs(&key1); |
1da177e4 | 1405 | |
42d35d48 | 1406 | out_put_keys: |
ae791a2d | 1407 | put_futex_key(&key2); |
42d35d48 | 1408 | out_put_key1: |
ae791a2d | 1409 | put_futex_key(&key1); |
42d35d48 | 1410 | out: |
52400ba9 DH |
1411 | if (pi_state != NULL) |
1412 | free_pi_state(pi_state); | |
1413 | return ret ? ret : task_count; | |
1da177e4 LT |
1414 | } |
1415 | ||
1416 | /* The key must be already stored in q->key. */ | |
82af7aca | 1417 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 1418 | __acquires(&hb->lock) |
1da177e4 | 1419 | { |
e2970f2f | 1420 | struct futex_hash_bucket *hb; |
1da177e4 | 1421 | |
e2970f2f IM |
1422 | hb = hash_futex(&q->key); |
1423 | q->lock_ptr = &hb->lock; | |
1da177e4 | 1424 | |
e2970f2f IM |
1425 | spin_lock(&hb->lock); |
1426 | return hb; | |
1da177e4 LT |
1427 | } |
1428 | ||
d40d65c8 DH |
1429 | static inline void |
1430 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) | |
15e408cd | 1431 | __releases(&hb->lock) |
d40d65c8 DH |
1432 | { |
1433 | spin_unlock(&hb->lock); | |
d40d65c8 DH |
1434 | } |
1435 | ||
1436 | /** | |
1437 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
1438 | * @q: The futex_q to enqueue | |
1439 | * @hb: The destination hash bucket | |
1440 | * | |
1441 | * The hb->lock must be held by the caller, and is released here. A call to | |
1442 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
1443 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
1444 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
1445 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
1446 | * an example). | |
1447 | */ | |
82af7aca | 1448 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
15e408cd | 1449 | __releases(&hb->lock) |
1da177e4 | 1450 | { |
ec92d082 PP |
1451 | int prio; |
1452 | ||
1453 | /* | |
1454 | * The priority used to register this element is | |
1455 | * - either the real thread-priority for the real-time threads | |
1456 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
1457 | * - or MAX_RT_PRIO for non-RT threads. | |
1458 | * Thus, all RT-threads are woken first in priority order, and | |
1459 | * the others are woken last, in FIFO order. | |
1460 | */ | |
1461 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1462 | ||
1463 | plist_node_init(&q->list, prio); | |
ec92d082 | 1464 | plist_add(&q->list, &hb->chain); |
c87e2837 | 1465 | q->task = current; |
e2970f2f | 1466 | spin_unlock(&hb->lock); |
1da177e4 LT |
1467 | } |
1468 | ||
d40d65c8 DH |
1469 | /** |
1470 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
1471 | * @q: The futex_q to unqueue | |
1472 | * | |
1473 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
1474 | * be paired with exactly one earlier call to queue_me(). | |
1475 | * | |
1476 | * Returns: | |
1477 | * 1 - if the futex_q was still queued (and we removed unqueued it) | |
1478 | * 0 - if the futex_q was already removed by the waking thread | |
1da177e4 | 1479 | */ |
1da177e4 LT |
1480 | static int unqueue_me(struct futex_q *q) |
1481 | { | |
1da177e4 | 1482 | spinlock_t *lock_ptr; |
e2970f2f | 1483 | int ret = 0; |
1da177e4 LT |
1484 | |
1485 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 1486 | retry: |
1da177e4 | 1487 | lock_ptr = q->lock_ptr; |
e91467ec | 1488 | barrier(); |
c80544dc | 1489 | if (lock_ptr != NULL) { |
1da177e4 LT |
1490 | spin_lock(lock_ptr); |
1491 | /* | |
1492 | * q->lock_ptr can change between reading it and | |
1493 | * spin_lock(), causing us to take the wrong lock. This | |
1494 | * corrects the race condition. | |
1495 | * | |
1496 | * Reasoning goes like this: if we have the wrong lock, | |
1497 | * q->lock_ptr must have changed (maybe several times) | |
1498 | * between reading it and the spin_lock(). It can | |
1499 | * change again after the spin_lock() but only if it was | |
1500 | * already changed before the spin_lock(). It cannot, | |
1501 | * however, change back to the original value. Therefore | |
1502 | * we can detect whether we acquired the correct lock. | |
1503 | */ | |
1504 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
1505 | spin_unlock(lock_ptr); | |
1506 | goto retry; | |
1507 | } | |
2e12978a | 1508 | __unqueue_futex(q); |
c87e2837 IM |
1509 | |
1510 | BUG_ON(q->pi_state); | |
1511 | ||
1da177e4 LT |
1512 | spin_unlock(lock_ptr); |
1513 | ret = 1; | |
1514 | } | |
1515 | ||
9adef58b | 1516 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1517 | return ret; |
1518 | } | |
1519 | ||
c87e2837 IM |
1520 | /* |
1521 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
1522 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
1523 | * and dropped here. | |
c87e2837 | 1524 | */ |
d0aa7a70 | 1525 | static void unqueue_me_pi(struct futex_q *q) |
15e408cd | 1526 | __releases(q->lock_ptr) |
c87e2837 | 1527 | { |
2e12978a | 1528 | __unqueue_futex(q); |
c87e2837 IM |
1529 | |
1530 | BUG_ON(!q->pi_state); | |
1531 | free_pi_state(q->pi_state); | |
1532 | q->pi_state = NULL; | |
1533 | ||
d0aa7a70 | 1534 | spin_unlock(q->lock_ptr); |
c87e2837 IM |
1535 | } |
1536 | ||
d0aa7a70 | 1537 | /* |
cdf71a10 | 1538 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 1539 | * |
778e9a9c AK |
1540 | * Must be called with hash bucket lock held and mm->sem held for non |
1541 | * private futexes. | |
d0aa7a70 | 1542 | */ |
778e9a9c | 1543 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
ae791a2d | 1544 | struct task_struct *newowner) |
d0aa7a70 | 1545 | { |
cdf71a10 | 1546 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 1547 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 1548 | struct task_struct *oldowner = pi_state->owner; |
d0aa7a70 | 1549 | u32 uval, curval, newval; |
e4dc5b7a | 1550 | int ret; |
d0aa7a70 PP |
1551 | |
1552 | /* Owner died? */ | |
1b7558e4 TG |
1553 | if (!pi_state->owner) |
1554 | newtid |= FUTEX_OWNER_DIED; | |
1555 | ||
1556 | /* | |
1557 | * We are here either because we stole the rtmutex from the | |
8161239a LJ |
1558 | * previous highest priority waiter or we are the highest priority |
1559 | * waiter but failed to get the rtmutex the first time. | |
1560 | * We have to replace the newowner TID in the user space variable. | |
1561 | * This must be atomic as we have to preserve the owner died bit here. | |
1b7558e4 | 1562 | * |
b2d0994b DH |
1563 | * Note: We write the user space value _before_ changing the pi_state |
1564 | * because we can fault here. Imagine swapped out pages or a fork | |
1565 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
1566 | * |
1567 | * Modifying pi_state _before_ the user space value would | |
1568 | * leave the pi_state in an inconsistent state when we fault | |
1569 | * here, because we need to drop the hash bucket lock to | |
1570 | * handle the fault. This might be observed in the PID check | |
1571 | * in lookup_pi_state. | |
1572 | */ | |
1573 | retry: | |
1574 | if (get_futex_value_locked(&uval, uaddr)) | |
1575 | goto handle_fault; | |
1576 | ||
1577 | while (1) { | |
1578 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
1579 | ||
37a9d912 | 1580 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1b7558e4 TG |
1581 | goto handle_fault; |
1582 | if (curval == uval) | |
1583 | break; | |
1584 | uval = curval; | |
1585 | } | |
1586 | ||
1587 | /* | |
1588 | * We fixed up user space. Now we need to fix the pi_state | |
1589 | * itself. | |
1590 | */ | |
d0aa7a70 | 1591 | if (pi_state->owner != NULL) { |
1d615482 | 1592 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
d0aa7a70 PP |
1593 | WARN_ON(list_empty(&pi_state->list)); |
1594 | list_del_init(&pi_state->list); | |
1d615482 | 1595 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
1b7558e4 | 1596 | } |
d0aa7a70 | 1597 | |
cdf71a10 | 1598 | pi_state->owner = newowner; |
d0aa7a70 | 1599 | |
1d615482 | 1600 | raw_spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 1601 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 | 1602 | list_add(&pi_state->list, &newowner->pi_state_list); |
1d615482 | 1603 | raw_spin_unlock_irq(&newowner->pi_lock); |
1b7558e4 | 1604 | return 0; |
d0aa7a70 | 1605 | |
d0aa7a70 | 1606 | /* |
1b7558e4 | 1607 | * To handle the page fault we need to drop the hash bucket |
8161239a LJ |
1608 | * lock here. That gives the other task (either the highest priority |
1609 | * waiter itself or the task which stole the rtmutex) the | |
1b7558e4 TG |
1610 | * chance to try the fixup of the pi_state. So once we are |
1611 | * back from handling the fault we need to check the pi_state | |
1612 | * after reacquiring the hash bucket lock and before trying to | |
1613 | * do another fixup. When the fixup has been done already we | |
1614 | * simply return. | |
d0aa7a70 | 1615 | */ |
1b7558e4 TG |
1616 | handle_fault: |
1617 | spin_unlock(q->lock_ptr); | |
778e9a9c | 1618 | |
d0725992 | 1619 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 1620 | |
1b7558e4 | 1621 | spin_lock(q->lock_ptr); |
778e9a9c | 1622 | |
1b7558e4 TG |
1623 | /* |
1624 | * Check if someone else fixed it for us: | |
1625 | */ | |
1626 | if (pi_state->owner != oldowner) | |
1627 | return 0; | |
1628 | ||
1629 | if (ret) | |
1630 | return ret; | |
1631 | ||
1632 | goto retry; | |
d0aa7a70 PP |
1633 | } |
1634 | ||
72c1bbf3 | 1635 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 1636 | |
dd973998 DH |
1637 | /** |
1638 | * fixup_owner() - Post lock pi_state and corner case management | |
1639 | * @uaddr: user address of the futex | |
dd973998 DH |
1640 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
1641 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
1642 | * | |
1643 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
1644 | * the pi_state owner as well as handle race conditions that may allow us to | |
1645 | * acquire the lock. Must be called with the hb lock held. | |
1646 | * | |
1647 | * Returns: | |
1648 | * 1 - success, lock taken | |
1649 | * 0 - success, lock not taken | |
1650 | * <0 - on error (-EFAULT) | |
1651 | */ | |
ae791a2d | 1652 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 DH |
1653 | { |
1654 | struct task_struct *owner; | |
1655 | int ret = 0; | |
1656 | ||
1657 | if (locked) { | |
1658 | /* | |
1659 | * Got the lock. We might not be the anticipated owner if we | |
1660 | * did a lock-steal - fix up the PI-state in that case: | |
1661 | */ | |
1662 | if (q->pi_state->owner != current) | |
ae791a2d | 1663 | ret = fixup_pi_state_owner(uaddr, q, current); |
dd973998 DH |
1664 | goto out; |
1665 | } | |
1666 | ||
1667 | /* | |
1668 | * Catch the rare case, where the lock was released when we were on the | |
1669 | * way back before we locked the hash bucket. | |
1670 | */ | |
1671 | if (q->pi_state->owner == current) { | |
1672 | /* | |
1673 | * Try to get the rt_mutex now. This might fail as some other | |
1674 | * task acquired the rt_mutex after we removed ourself from the | |
1675 | * rt_mutex waiters list. | |
1676 | */ | |
1677 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
1678 | locked = 1; | |
1679 | goto out; | |
1680 | } | |
1681 | ||
1682 | /* | |
1683 | * pi_state is incorrect, some other task did a lock steal and | |
1684 | * we returned due to timeout or signal without taking the | |
8161239a | 1685 | * rt_mutex. Too late. |
dd973998 | 1686 | */ |
8161239a | 1687 | raw_spin_lock(&q->pi_state->pi_mutex.wait_lock); |
dd973998 | 1688 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); |
8161239a LJ |
1689 | if (!owner) |
1690 | owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); | |
1691 | raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock); | |
ae791a2d | 1692 | ret = fixup_pi_state_owner(uaddr, q, owner); |
dd973998 DH |
1693 | goto out; |
1694 | } | |
1695 | ||
1696 | /* | |
1697 | * Paranoia check. If we did not take the lock, then we should not be | |
8161239a | 1698 | * the owner of the rt_mutex. |
dd973998 DH |
1699 | */ |
1700 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
1701 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
1702 | "pi-state %p\n", ret, | |
1703 | q->pi_state->pi_mutex.owner, | |
1704 | q->pi_state->owner); | |
1705 | ||
1706 | out: | |
1707 | return ret ? ret : locked; | |
1708 | } | |
1709 | ||
ca5f9524 DH |
1710 | /** |
1711 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
1712 | * @hb: the futex hash bucket, must be locked by the caller | |
1713 | * @q: the futex_q to queue up on | |
1714 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
1715 | */ |
1716 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 1717 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 1718 | { |
9beba3c5 DH |
1719 | /* |
1720 | * The task state is guaranteed to be set before another task can | |
1721 | * wake it. set_current_state() is implemented using set_mb() and | |
1722 | * queue_me() calls spin_unlock() upon completion, both serializing | |
1723 | * access to the hash list and forcing another memory barrier. | |
1724 | */ | |
f1a11e05 | 1725 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 1726 | queue_me(q, hb); |
ca5f9524 DH |
1727 | |
1728 | /* Arm the timer */ | |
1729 | if (timeout) { | |
1730 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); | |
1731 | if (!hrtimer_active(&timeout->timer)) | |
1732 | timeout->task = NULL; | |
1733 | } | |
1734 | ||
1735 | /* | |
0729e196 DH |
1736 | * If we have been removed from the hash list, then another task |
1737 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
1738 | */ |
1739 | if (likely(!plist_node_empty(&q->list))) { | |
1740 | /* | |
1741 | * If the timer has already expired, current will already be | |
1742 | * flagged for rescheduling. Only call schedule if there | |
1743 | * is no timeout, or if it has yet to expire. | |
1744 | */ | |
1745 | if (!timeout || timeout->task) | |
1746 | schedule(); | |
1747 | } | |
1748 | __set_current_state(TASK_RUNNING); | |
1749 | } | |
1750 | ||
f801073f DH |
1751 | /** |
1752 | * futex_wait_setup() - Prepare to wait on a futex | |
1753 | * @uaddr: the futex userspace address | |
1754 | * @val: the expected value | |
b41277dc | 1755 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
1756 | * @q: the associated futex_q |
1757 | * @hb: storage for hash_bucket pointer to be returned to caller | |
1758 | * | |
1759 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
1760 | * compare it with the expected value. Handle atomic faults internally. | |
1761 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
1762 | * with no q.key reference on failure. | |
1763 | * | |
1764 | * Returns: | |
1765 | * 0 - uaddr contains val and hb has been locked | |
1766 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlcoked | |
1767 | */ | |
b41277dc | 1768 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 1769 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 1770 | { |
e2970f2f IM |
1771 | u32 uval; |
1772 | int ret; | |
1da177e4 | 1773 | |
1da177e4 | 1774 | /* |
b2d0994b | 1775 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
1776 | * Order is important: |
1777 | * | |
1778 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
1779 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
1780 | * | |
1781 | * The basic logical guarantee of a futex is that it blocks ONLY | |
1782 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
1783 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
1784 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
1785 | * cond(var) false, which would violate the guarantee. |
1786 | * | |
8fe8f545 ML |
1787 | * On the other hand, we insert q and release the hash-bucket only |
1788 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
1789 | * absorb a wakeup if *uaddr does not match the desired values | |
1790 | * while the syscall executes. | |
1da177e4 | 1791 | */ |
f801073f | 1792 | retry: |
b41277dc | 1793 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key); |
f801073f | 1794 | if (unlikely(ret != 0)) |
a5a2a0c7 | 1795 | return ret; |
f801073f DH |
1796 | |
1797 | retry_private: | |
1798 | *hb = queue_lock(q); | |
1799 | ||
e2970f2f | 1800 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 1801 | |
f801073f DH |
1802 | if (ret) { |
1803 | queue_unlock(q, *hb); | |
1da177e4 | 1804 | |
e2970f2f | 1805 | ret = get_user(uval, uaddr); |
e4dc5b7a | 1806 | if (ret) |
f801073f | 1807 | goto out; |
1da177e4 | 1808 | |
b41277dc | 1809 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1810 | goto retry_private; |
1811 | ||
ae791a2d | 1812 | put_futex_key(&q->key); |
e4dc5b7a | 1813 | goto retry; |
1da177e4 | 1814 | } |
ca5f9524 | 1815 | |
f801073f DH |
1816 | if (uval != val) { |
1817 | queue_unlock(q, *hb); | |
1818 | ret = -EWOULDBLOCK; | |
2fff78c7 | 1819 | } |
1da177e4 | 1820 | |
f801073f DH |
1821 | out: |
1822 | if (ret) | |
ae791a2d | 1823 | put_futex_key(&q->key); |
f801073f DH |
1824 | return ret; |
1825 | } | |
1826 | ||
b41277dc DH |
1827 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
1828 | ktime_t *abs_time, u32 bitset) | |
f801073f DH |
1829 | { |
1830 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
1831 | struct restart_block *restart; |
1832 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 1833 | struct futex_q q = futex_q_init; |
f801073f DH |
1834 | int ret; |
1835 | ||
1836 | if (!bitset) | |
1837 | return -EINVAL; | |
f801073f DH |
1838 | q.bitset = bitset; |
1839 | ||
1840 | if (abs_time) { | |
1841 | to = &timeout; | |
1842 | ||
b41277dc DH |
1843 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
1844 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
1845 | HRTIMER_MODE_ABS); | |
f801073f DH |
1846 | hrtimer_init_sleeper(to, current); |
1847 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
1848 | current->timer_slack_ns); | |
1849 | } | |
1850 | ||
d58e6576 | 1851 | retry: |
7ada876a DH |
1852 | /* |
1853 | * Prepare to wait on uaddr. On success, holds hb lock and increments | |
1854 | * q.key refs. | |
1855 | */ | |
b41277dc | 1856 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
1857 | if (ret) |
1858 | goto out; | |
1859 | ||
ca5f9524 | 1860 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 1861 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
1862 | |
1863 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 1864 | ret = 0; |
7ada876a | 1865 | /* unqueue_me() drops q.key ref */ |
1da177e4 | 1866 | if (!unqueue_me(&q)) |
7ada876a | 1867 | goto out; |
2fff78c7 | 1868 | ret = -ETIMEDOUT; |
ca5f9524 | 1869 | if (to && !to->task) |
7ada876a | 1870 | goto out; |
72c1bbf3 | 1871 | |
e2970f2f | 1872 | /* |
d58e6576 TG |
1873 | * We expect signal_pending(current), but we might be the |
1874 | * victim of a spurious wakeup as well. | |
e2970f2f | 1875 | */ |
7ada876a | 1876 | if (!signal_pending(current)) |
d58e6576 | 1877 | goto retry; |
d58e6576 | 1878 | |
2fff78c7 | 1879 | ret = -ERESTARTSYS; |
c19384b5 | 1880 | if (!abs_time) |
7ada876a | 1881 | goto out; |
1da177e4 | 1882 | |
2fff78c7 PZ |
1883 | restart = ¤t_thread_info()->restart_block; |
1884 | restart->fn = futex_wait_restart; | |
a3c74c52 | 1885 | restart->futex.uaddr = uaddr; |
2fff78c7 PZ |
1886 | restart->futex.val = val; |
1887 | restart->futex.time = abs_time->tv64; | |
1888 | restart->futex.bitset = bitset; | |
b41277dc | 1889 | restart->futex.flags = flags; |
42d35d48 | 1890 | |
2fff78c7 PZ |
1891 | ret = -ERESTART_RESTARTBLOCK; |
1892 | ||
42d35d48 | 1893 | out: |
ca5f9524 DH |
1894 | if (to) { |
1895 | hrtimer_cancel(&to->timer); | |
1896 | destroy_hrtimer_on_stack(&to->timer); | |
1897 | } | |
c87e2837 IM |
1898 | return ret; |
1899 | } | |
1900 | ||
72c1bbf3 NP |
1901 | |
1902 | static long futex_wait_restart(struct restart_block *restart) | |
1903 | { | |
a3c74c52 | 1904 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 1905 | ktime_t t, *tp = NULL; |
72c1bbf3 | 1906 | |
a72188d8 DH |
1907 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
1908 | t.tv64 = restart->futex.time; | |
1909 | tp = &t; | |
1910 | } | |
72c1bbf3 | 1911 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
1912 | |
1913 | return (long)futex_wait(uaddr, restart->futex.flags, | |
1914 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
1915 | } |
1916 | ||
1917 | ||
c87e2837 IM |
1918 | /* |
1919 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
1920 | * and failed. The kernel side here does the whole locking operation: | |
1921 | * if there are waiters then it will block, it does PI, etc. (Due to | |
1922 | * races the kernel might see a 0 value of the futex too.) | |
1923 | */ | |
b41277dc DH |
1924 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect, |
1925 | ktime_t *time, int trylock) | |
c87e2837 | 1926 | { |
c5780e97 | 1927 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 1928 | struct futex_hash_bucket *hb; |
5bdb05f9 | 1929 | struct futex_q q = futex_q_init; |
dd973998 | 1930 | int res, ret; |
c87e2837 IM |
1931 | |
1932 | if (refill_pi_state_cache()) | |
1933 | return -ENOMEM; | |
1934 | ||
c19384b5 | 1935 | if (time) { |
c5780e97 | 1936 | to = &timeout; |
237fc6e7 TG |
1937 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
1938 | HRTIMER_MODE_ABS); | |
c5780e97 | 1939 | hrtimer_init_sleeper(to, current); |
cc584b21 | 1940 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
1941 | } |
1942 | ||
42d35d48 | 1943 | retry: |
b41277dc | 1944 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key); |
c87e2837 | 1945 | if (unlikely(ret != 0)) |
42d35d48 | 1946 | goto out; |
c87e2837 | 1947 | |
e4dc5b7a | 1948 | retry_private: |
82af7aca | 1949 | hb = queue_lock(&q); |
c87e2837 | 1950 | |
bab5bc9e | 1951 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 1952 | if (unlikely(ret)) { |
778e9a9c | 1953 | switch (ret) { |
1a52084d DH |
1954 | case 1: |
1955 | /* We got the lock. */ | |
1956 | ret = 0; | |
1957 | goto out_unlock_put_key; | |
1958 | case -EFAULT: | |
1959 | goto uaddr_faulted; | |
778e9a9c AK |
1960 | case -EAGAIN: |
1961 | /* | |
1962 | * Task is exiting and we just wait for the | |
1963 | * exit to complete. | |
1964 | */ | |
1965 | queue_unlock(&q, hb); | |
ae791a2d | 1966 | put_futex_key(&q.key); |
778e9a9c AK |
1967 | cond_resched(); |
1968 | goto retry; | |
778e9a9c | 1969 | default: |
42d35d48 | 1970 | goto out_unlock_put_key; |
c87e2837 | 1971 | } |
c87e2837 IM |
1972 | } |
1973 | ||
1974 | /* | |
1975 | * Only actually queue now that the atomic ops are done: | |
1976 | */ | |
82af7aca | 1977 | queue_me(&q, hb); |
c87e2837 | 1978 | |
c87e2837 IM |
1979 | WARN_ON(!q.pi_state); |
1980 | /* | |
1981 | * Block on the PI mutex: | |
1982 | */ | |
1983 | if (!trylock) | |
1984 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); | |
1985 | else { | |
1986 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); | |
1987 | /* Fixup the trylock return value: */ | |
1988 | ret = ret ? 0 : -EWOULDBLOCK; | |
1989 | } | |
1990 | ||
a99e4e41 | 1991 | spin_lock(q.lock_ptr); |
dd973998 DH |
1992 | /* |
1993 | * Fixup the pi_state owner and possibly acquire the lock if we | |
1994 | * haven't already. | |
1995 | */ | |
ae791a2d | 1996 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 DH |
1997 | /* |
1998 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
1999 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
2000 | */ | |
2001 | if (res) | |
2002 | ret = (res < 0) ? res : 0; | |
c87e2837 | 2003 | |
e8f6386c | 2004 | /* |
dd973998 DH |
2005 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
2006 | * it and return the fault to userspace. | |
e8f6386c DH |
2007 | */ |
2008 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
2009 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
2010 | ||
778e9a9c AK |
2011 | /* Unqueue and drop the lock */ |
2012 | unqueue_me_pi(&q); | |
c87e2837 | 2013 | |
5ecb01cf | 2014 | goto out_put_key; |
c87e2837 | 2015 | |
42d35d48 | 2016 | out_unlock_put_key: |
c87e2837 IM |
2017 | queue_unlock(&q, hb); |
2018 | ||
42d35d48 | 2019 | out_put_key: |
ae791a2d | 2020 | put_futex_key(&q.key); |
42d35d48 | 2021 | out: |
237fc6e7 TG |
2022 | if (to) |
2023 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 2024 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 2025 | |
42d35d48 | 2026 | uaddr_faulted: |
778e9a9c AK |
2027 | queue_unlock(&q, hb); |
2028 | ||
d0725992 | 2029 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
2030 | if (ret) |
2031 | goto out_put_key; | |
c87e2837 | 2032 | |
b41277dc | 2033 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2034 | goto retry_private; |
2035 | ||
ae791a2d | 2036 | put_futex_key(&q.key); |
e4dc5b7a | 2037 | goto retry; |
c87e2837 IM |
2038 | } |
2039 | ||
c87e2837 IM |
2040 | /* |
2041 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
2042 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
2043 | * and do the rt-mutex unlock. | |
2044 | */ | |
b41277dc | 2045 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 IM |
2046 | { |
2047 | struct futex_hash_bucket *hb; | |
2048 | struct futex_q *this, *next; | |
ec92d082 | 2049 | struct plist_head *head; |
38d47c1b | 2050 | union futex_key key = FUTEX_KEY_INIT; |
c0c9ed15 | 2051 | u32 uval, vpid = task_pid_vnr(current); |
e4dc5b7a | 2052 | int ret; |
c87e2837 IM |
2053 | |
2054 | retry: | |
2055 | if (get_user(uval, uaddr)) | |
2056 | return -EFAULT; | |
2057 | /* | |
2058 | * We release only a lock we actually own: | |
2059 | */ | |
c0c9ed15 | 2060 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 2061 | return -EPERM; |
c87e2837 | 2062 | |
b41277dc | 2063 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key); |
c87e2837 IM |
2064 | if (unlikely(ret != 0)) |
2065 | goto out; | |
2066 | ||
2067 | hb = hash_futex(&key); | |
2068 | spin_lock(&hb->lock); | |
2069 | ||
c87e2837 IM |
2070 | /* |
2071 | * To avoid races, try to do the TID -> 0 atomic transition | |
2072 | * again. If it succeeds then we can return without waking | |
2073 | * anyone else up: | |
2074 | */ | |
37a9d912 ML |
2075 | if (!(uval & FUTEX_OWNER_DIED) && |
2076 | cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0)) | |
c87e2837 IM |
2077 | goto pi_faulted; |
2078 | /* | |
2079 | * Rare case: we managed to release the lock atomically, | |
2080 | * no need to wake anyone else up: | |
2081 | */ | |
c0c9ed15 | 2082 | if (unlikely(uval == vpid)) |
c87e2837 IM |
2083 | goto out_unlock; |
2084 | ||
2085 | /* | |
2086 | * Ok, other tasks may need to be woken up - check waiters | |
2087 | * and do the wakeup if necessary: | |
2088 | */ | |
2089 | head = &hb->chain; | |
2090 | ||
ec92d082 | 2091 | plist_for_each_entry_safe(this, next, head, list) { |
c87e2837 IM |
2092 | if (!match_futex (&this->key, &key)) |
2093 | continue; | |
2094 | ret = wake_futex_pi(uaddr, uval, this); | |
2095 | /* | |
2096 | * The atomic access to the futex value | |
2097 | * generated a pagefault, so retry the | |
2098 | * user-access and the wakeup: | |
2099 | */ | |
2100 | if (ret == -EFAULT) | |
2101 | goto pi_faulted; | |
2102 | goto out_unlock; | |
2103 | } | |
2104 | /* | |
2105 | * No waiters - kernel unlocks the futex: | |
2106 | */ | |
e3f2ddea IM |
2107 | if (!(uval & FUTEX_OWNER_DIED)) { |
2108 | ret = unlock_futex_pi(uaddr, uval); | |
2109 | if (ret == -EFAULT) | |
2110 | goto pi_faulted; | |
2111 | } | |
c87e2837 IM |
2112 | |
2113 | out_unlock: | |
2114 | spin_unlock(&hb->lock); | |
ae791a2d | 2115 | put_futex_key(&key); |
c87e2837 | 2116 | |
42d35d48 | 2117 | out: |
c87e2837 IM |
2118 | return ret; |
2119 | ||
2120 | pi_faulted: | |
778e9a9c | 2121 | spin_unlock(&hb->lock); |
ae791a2d | 2122 | put_futex_key(&key); |
c87e2837 | 2123 | |
d0725992 | 2124 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2125 | if (!ret) |
c87e2837 IM |
2126 | goto retry; |
2127 | ||
1da177e4 LT |
2128 | return ret; |
2129 | } | |
2130 | ||
52400ba9 DH |
2131 | /** |
2132 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2133 | * @hb: the hash_bucket futex_q was original enqueued on | |
2134 | * @q: the futex_q woken while waiting to be requeued | |
2135 | * @key2: the futex_key of the requeue target futex | |
2136 | * @timeout: the timeout associated with the wait (NULL if none) | |
2137 | * | |
2138 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2139 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2140 | * the wakeup and return the appropriate error code to the caller. Must be | |
2141 | * called with the hb lock held. | |
2142 | * | |
2143 | * Returns | |
2144 | * 0 - no early wakeup detected | |
1c840c14 | 2145 | * <0 - -ETIMEDOUT or -ERESTARTNOINTR |
52400ba9 DH |
2146 | */ |
2147 | static inline | |
2148 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2149 | struct futex_q *q, union futex_key *key2, | |
2150 | struct hrtimer_sleeper *timeout) | |
2151 | { | |
2152 | int ret = 0; | |
2153 | ||
2154 | /* | |
2155 | * With the hb lock held, we avoid races while we process the wakeup. | |
2156 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2157 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2158 | * It can't be requeued from uaddr2 to something else since we don't | |
2159 | * support a PI aware source futex for requeue. | |
2160 | */ | |
2161 | if (!match_futex(&q->key, key2)) { | |
2162 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2163 | /* | |
2164 | * We were woken prior to requeue by a timeout or a signal. | |
2165 | * Unqueue the futex_q and determine which it was. | |
2166 | */ | |
2e12978a | 2167 | plist_del(&q->list, &hb->chain); |
52400ba9 | 2168 | |
d58e6576 | 2169 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 2170 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2171 | if (timeout && !timeout->task) |
2172 | ret = -ETIMEDOUT; | |
d58e6576 | 2173 | else if (signal_pending(current)) |
1c840c14 | 2174 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
2175 | } |
2176 | return ret; | |
2177 | } | |
2178 | ||
2179 | /** | |
2180 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 2181 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 2182 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
52400ba9 DH |
2183 | * the same type, no requeueing from private to shared, etc. |
2184 | * @val: the expected value of uaddr | |
2185 | * @abs_time: absolute timeout | |
56ec1607 | 2186 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
2187 | * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0) |
2188 | * @uaddr2: the pi futex we will take prior to returning to user-space | |
2189 | * | |
2190 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
2191 | * uaddr2 which must be PI aware. Normal wakeup will wake on uaddr2 and | |
2192 | * complete the acquisition of the rt_mutex prior to returning to userspace. | |
2193 | * This ensures the rt_mutex maintains an owner when it has waiters; without | |
2194 | * one, the pi logic wouldn't know which task to boost/deboost, if there was a | |
2195 | * need to. | |
2196 | * | |
2197 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
2198 | * via the following: | |
2199 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() | |
cc6db4e6 DH |
2200 | * 2) wakeup on uaddr2 after a requeue |
2201 | * 3) signal | |
2202 | * 4) timeout | |
52400ba9 | 2203 | * |
cc6db4e6 | 2204 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
2205 | * |
2206 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2207 | * 5) successful lock | |
2208 | * 6) signal | |
2209 | * 7) timeout | |
2210 | * 8) other lock acquisition failure | |
2211 | * | |
cc6db4e6 | 2212 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
2213 | * |
2214 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2215 | * | |
2216 | * Returns: | |
2217 | * 0 - On success | |
2218 | * <0 - On error | |
2219 | */ | |
b41277dc | 2220 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 2221 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 2222 | u32 __user *uaddr2) |
52400ba9 DH |
2223 | { |
2224 | struct hrtimer_sleeper timeout, *to = NULL; | |
2225 | struct rt_mutex_waiter rt_waiter; | |
2226 | struct rt_mutex *pi_mutex = NULL; | |
52400ba9 | 2227 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
2228 | union futex_key key2 = FUTEX_KEY_INIT; |
2229 | struct futex_q q = futex_q_init; | |
52400ba9 | 2230 | int res, ret; |
52400ba9 DH |
2231 | |
2232 | if (!bitset) | |
2233 | return -EINVAL; | |
2234 | ||
2235 | if (abs_time) { | |
2236 | to = &timeout; | |
b41277dc DH |
2237 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2238 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2239 | HRTIMER_MODE_ABS); | |
52400ba9 DH |
2240 | hrtimer_init_sleeper(to, current); |
2241 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2242 | current->timer_slack_ns); | |
2243 | } | |
2244 | ||
2245 | /* | |
2246 | * The waiter is allocated on our stack, manipulated by the requeue | |
2247 | * code while we sleep on uaddr. | |
2248 | */ | |
2249 | debug_rt_mutex_init_waiter(&rt_waiter); | |
2250 | rt_waiter.task = NULL; | |
2251 | ||
b41277dc | 2252 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2); |
52400ba9 DH |
2253 | if (unlikely(ret != 0)) |
2254 | goto out; | |
2255 | ||
84bc4af5 DH |
2256 | q.bitset = bitset; |
2257 | q.rt_waiter = &rt_waiter; | |
2258 | q.requeue_pi_key = &key2; | |
2259 | ||
7ada876a DH |
2260 | /* |
2261 | * Prepare to wait on uaddr. On success, increments q.key (key1) ref | |
2262 | * count. | |
2263 | */ | |
b41277dc | 2264 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 TG |
2265 | if (ret) |
2266 | goto out_key2; | |
52400ba9 DH |
2267 | |
2268 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ | |
f1a11e05 | 2269 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2270 | |
2271 | spin_lock(&hb->lock); | |
2272 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2273 | spin_unlock(&hb->lock); | |
2274 | if (ret) | |
2275 | goto out_put_keys; | |
2276 | ||
2277 | /* | |
2278 | * In order for us to be here, we know our q.key == key2, and since | |
2279 | * we took the hb->lock above, we also know that futex_requeue() has | |
2280 | * completed and we no longer have to concern ourselves with a wakeup | |
7ada876a DH |
2281 | * race with the atomic proxy lock acquisition by the requeue code. The |
2282 | * futex_requeue dropped our key1 reference and incremented our key2 | |
2283 | * reference count. | |
52400ba9 DH |
2284 | */ |
2285 | ||
2286 | /* Check if the requeue code acquired the second futex for us. */ | |
2287 | if (!q.rt_waiter) { | |
2288 | /* | |
2289 | * Got the lock. We might not be the anticipated owner if we | |
2290 | * did a lock-steal - fix up the PI-state in that case. | |
2291 | */ | |
2292 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2293 | spin_lock(q.lock_ptr); | |
ae791a2d | 2294 | ret = fixup_pi_state_owner(uaddr2, &q, current); |
52400ba9 DH |
2295 | spin_unlock(q.lock_ptr); |
2296 | } | |
2297 | } else { | |
2298 | /* | |
2299 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2300 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2301 | * the pi_state. | |
2302 | */ | |
2303 | WARN_ON(!&q.pi_state); | |
2304 | pi_mutex = &q.pi_state->pi_mutex; | |
2305 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1); | |
2306 | debug_rt_mutex_free_waiter(&rt_waiter); | |
2307 | ||
2308 | spin_lock(q.lock_ptr); | |
2309 | /* | |
2310 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2311 | * haven't already. | |
2312 | */ | |
ae791a2d | 2313 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 DH |
2314 | /* |
2315 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 2316 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
2317 | */ |
2318 | if (res) | |
2319 | ret = (res < 0) ? res : 0; | |
2320 | ||
2321 | /* Unqueue and drop the lock. */ | |
2322 | unqueue_me_pi(&q); | |
2323 | } | |
2324 | ||
2325 | /* | |
2326 | * If fixup_pi_state_owner() faulted and was unable to handle the | |
2327 | * fault, unlock the rt_mutex and return the fault to userspace. | |
2328 | */ | |
2329 | if (ret == -EFAULT) { | |
2330 | if (rt_mutex_owner(pi_mutex) == current) | |
2331 | rt_mutex_unlock(pi_mutex); | |
2332 | } else if (ret == -EINTR) { | |
52400ba9 | 2333 | /* |
cc6db4e6 DH |
2334 | * We've already been requeued, but cannot restart by calling |
2335 | * futex_lock_pi() directly. We could restart this syscall, but | |
2336 | * it would detect that the user space "val" changed and return | |
2337 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
2338 | * -EWOULDBLOCK directly. | |
52400ba9 | 2339 | */ |
2070887f | 2340 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2341 | } |
2342 | ||
2343 | out_put_keys: | |
ae791a2d | 2344 | put_futex_key(&q.key); |
c8b15a70 | 2345 | out_key2: |
ae791a2d | 2346 | put_futex_key(&key2); |
52400ba9 DH |
2347 | |
2348 | out: | |
2349 | if (to) { | |
2350 | hrtimer_cancel(&to->timer); | |
2351 | destroy_hrtimer_on_stack(&to->timer); | |
2352 | } | |
2353 | return ret; | |
2354 | } | |
2355 | ||
0771dfef IM |
2356 | /* |
2357 | * Support for robust futexes: the kernel cleans up held futexes at | |
2358 | * thread exit time. | |
2359 | * | |
2360 | * Implementation: user-space maintains a per-thread list of locks it | |
2361 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2362 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2363 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2364 | * always manipulated with the lock held, so the list is private and |
2365 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2366 | * field, to allow the kernel to clean up if the thread dies after | |
2367 | * acquiring the lock, but just before it could have added itself to | |
2368 | * the list. There can only be one such pending lock. | |
2369 | */ | |
2370 | ||
2371 | /** | |
d96ee56c DH |
2372 | * sys_set_robust_list() - Set the robust-futex list head of a task |
2373 | * @head: pointer to the list-head | |
2374 | * @len: length of the list-head, as userspace expects | |
0771dfef | 2375 | */ |
836f92ad HC |
2376 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2377 | size_t, len) | |
0771dfef | 2378 | { |
a0c1e907 TG |
2379 | if (!futex_cmpxchg_enabled) |
2380 | return -ENOSYS; | |
0771dfef IM |
2381 | /* |
2382 | * The kernel knows only one size for now: | |
2383 | */ | |
2384 | if (unlikely(len != sizeof(*head))) | |
2385 | return -EINVAL; | |
2386 | ||
2387 | current->robust_list = head; | |
2388 | ||
2389 | return 0; | |
2390 | } | |
2391 | ||
2392 | /** | |
d96ee56c DH |
2393 | * sys_get_robust_list() - Get the robust-futex list head of a task |
2394 | * @pid: pid of the process [zero for current task] | |
2395 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
2396 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 2397 | */ |
836f92ad HC |
2398 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
2399 | struct robust_list_head __user * __user *, head_ptr, | |
2400 | size_t __user *, len_ptr) | |
0771dfef | 2401 | { |
ba46df98 | 2402 | struct robust_list_head __user *head; |
0771dfef | 2403 | unsigned long ret; |
c69e8d9c | 2404 | const struct cred *cred = current_cred(), *pcred; |
0771dfef | 2405 | |
a0c1e907 TG |
2406 | if (!futex_cmpxchg_enabled) |
2407 | return -ENOSYS; | |
2408 | ||
0771dfef IM |
2409 | if (!pid) |
2410 | head = current->robust_list; | |
2411 | else { | |
2412 | struct task_struct *p; | |
2413 | ||
2414 | ret = -ESRCH; | |
aaa2a97e | 2415 | rcu_read_lock(); |
228ebcbe | 2416 | p = find_task_by_vpid(pid); |
0771dfef IM |
2417 | if (!p) |
2418 | goto err_unlock; | |
2419 | ret = -EPERM; | |
c69e8d9c DH |
2420 | pcred = __task_cred(p); |
2421 | if (cred->euid != pcred->euid && | |
2422 | cred->euid != pcred->uid && | |
76aac0e9 | 2423 | !capable(CAP_SYS_PTRACE)) |
0771dfef IM |
2424 | goto err_unlock; |
2425 | head = p->robust_list; | |
aaa2a97e | 2426 | rcu_read_unlock(); |
0771dfef IM |
2427 | } |
2428 | ||
2429 | if (put_user(sizeof(*head), len_ptr)) | |
2430 | return -EFAULT; | |
2431 | return put_user(head, head_ptr); | |
2432 | ||
2433 | err_unlock: | |
aaa2a97e | 2434 | rcu_read_unlock(); |
0771dfef IM |
2435 | |
2436 | return ret; | |
2437 | } | |
2438 | ||
2439 | /* | |
2440 | * Process a futex-list entry, check whether it's owned by the | |
2441 | * dying task, and do notification if so: | |
2442 | */ | |
e3f2ddea | 2443 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 2444 | { |
e3f2ddea | 2445 | u32 uval, nval, mval; |
0771dfef | 2446 | |
8f17d3a5 IM |
2447 | retry: |
2448 | if (get_user(uval, uaddr)) | |
0771dfef IM |
2449 | return -1; |
2450 | ||
b488893a | 2451 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
2452 | /* |
2453 | * Ok, this dying thread is truly holding a futex | |
2454 | * of interest. Set the OWNER_DIED bit atomically | |
2455 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
2456 | * set, wake up a waiter (if any). (We have to do a | |
2457 | * futex_wake() even if OWNER_DIED is already set - | |
2458 | * to handle the rare but possible case of recursive | |
2459 | * thread-death.) The rest of the cleanup is done in | |
2460 | * userspace. | |
2461 | */ | |
e3f2ddea | 2462 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
6e0aa9f8 TG |
2463 | /* |
2464 | * We are not holding a lock here, but we want to have | |
2465 | * the pagefault_disable/enable() protection because | |
2466 | * we want to handle the fault gracefully. If the | |
2467 | * access fails we try to fault in the futex with R/W | |
2468 | * verification via get_user_pages. get_user() above | |
2469 | * does not guarantee R/W access. If that fails we | |
2470 | * give up and leave the futex locked. | |
2471 | */ | |
2472 | if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) { | |
2473 | if (fault_in_user_writeable(uaddr)) | |
2474 | return -1; | |
2475 | goto retry; | |
2476 | } | |
c87e2837 | 2477 | if (nval != uval) |
8f17d3a5 | 2478 | goto retry; |
0771dfef | 2479 | |
e3f2ddea IM |
2480 | /* |
2481 | * Wake robust non-PI futexes here. The wakeup of | |
2482 | * PI futexes happens in exit_pi_state(): | |
2483 | */ | |
36cf3b5c | 2484 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 2485 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
2486 | } |
2487 | return 0; | |
2488 | } | |
2489 | ||
e3f2ddea IM |
2490 | /* |
2491 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
2492 | */ | |
2493 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 2494 | struct robust_list __user * __user *head, |
1dcc41bb | 2495 | unsigned int *pi) |
e3f2ddea IM |
2496 | { |
2497 | unsigned long uentry; | |
2498 | ||
ba46df98 | 2499 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
2500 | return -EFAULT; |
2501 | ||
ba46df98 | 2502 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
2503 | *pi = uentry & 1; |
2504 | ||
2505 | return 0; | |
2506 | } | |
2507 | ||
0771dfef IM |
2508 | /* |
2509 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
2510 | * and mark any locks found there dead, and notify any waiters. | |
2511 | * | |
2512 | * We silently return on any sign of list-walking problem. | |
2513 | */ | |
2514 | void exit_robust_list(struct task_struct *curr) | |
2515 | { | |
2516 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 2517 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 DH |
2518 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
2519 | unsigned int uninitialized_var(next_pi); | |
0771dfef | 2520 | unsigned long futex_offset; |
9f96cb1e | 2521 | int rc; |
0771dfef | 2522 | |
a0c1e907 TG |
2523 | if (!futex_cmpxchg_enabled) |
2524 | return; | |
2525 | ||
0771dfef IM |
2526 | /* |
2527 | * Fetch the list head (which was registered earlier, via | |
2528 | * sys_set_robust_list()): | |
2529 | */ | |
e3f2ddea | 2530 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
2531 | return; |
2532 | /* | |
2533 | * Fetch the relative futex offset: | |
2534 | */ | |
2535 | if (get_user(futex_offset, &head->futex_offset)) | |
2536 | return; | |
2537 | /* | |
2538 | * Fetch any possibly pending lock-add first, and handle it | |
2539 | * if it exists: | |
2540 | */ | |
e3f2ddea | 2541 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 2542 | return; |
e3f2ddea | 2543 | |
9f96cb1e | 2544 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 2545 | while (entry != &head->list) { |
9f96cb1e MS |
2546 | /* |
2547 | * Fetch the next entry in the list before calling | |
2548 | * handle_futex_death: | |
2549 | */ | |
2550 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
2551 | /* |
2552 | * A pending lock might already be on the list, so | |
c87e2837 | 2553 | * don't process it twice: |
0771dfef IM |
2554 | */ |
2555 | if (entry != pending) | |
ba46df98 | 2556 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 2557 | curr, pi)) |
0771dfef | 2558 | return; |
9f96cb1e | 2559 | if (rc) |
0771dfef | 2560 | return; |
9f96cb1e MS |
2561 | entry = next_entry; |
2562 | pi = next_pi; | |
0771dfef IM |
2563 | /* |
2564 | * Avoid excessively long or circular lists: | |
2565 | */ | |
2566 | if (!--limit) | |
2567 | break; | |
2568 | ||
2569 | cond_resched(); | |
2570 | } | |
9f96cb1e MS |
2571 | |
2572 | if (pending) | |
2573 | handle_futex_death((void __user *)pending + futex_offset, | |
2574 | curr, pip); | |
0771dfef IM |
2575 | } |
2576 | ||
c19384b5 | 2577 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 2578 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 2579 | { |
b41277dc DH |
2580 | int ret = -ENOSYS, cmd = op & FUTEX_CMD_MASK; |
2581 | unsigned int flags = 0; | |
34f01cc1 ED |
2582 | |
2583 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 2584 | flags |= FLAGS_SHARED; |
1da177e4 | 2585 | |
b41277dc DH |
2586 | if (op & FUTEX_CLOCK_REALTIME) { |
2587 | flags |= FLAGS_CLOCKRT; | |
2588 | if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI) | |
2589 | return -ENOSYS; | |
2590 | } | |
1da177e4 | 2591 | |
34f01cc1 | 2592 | switch (cmd) { |
1da177e4 | 2593 | case FUTEX_WAIT: |
cd689985 TG |
2594 | val3 = FUTEX_BITSET_MATCH_ANY; |
2595 | case FUTEX_WAIT_BITSET: | |
b41277dc | 2596 | ret = futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 LT |
2597 | break; |
2598 | case FUTEX_WAKE: | |
cd689985 TG |
2599 | val3 = FUTEX_BITSET_MATCH_ANY; |
2600 | case FUTEX_WAKE_BITSET: | |
b41277dc | 2601 | ret = futex_wake(uaddr, flags, val, val3); |
1da177e4 | 2602 | break; |
1da177e4 | 2603 | case FUTEX_REQUEUE: |
b41277dc | 2604 | ret = futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 LT |
2605 | break; |
2606 | case FUTEX_CMP_REQUEUE: | |
b41277dc | 2607 | ret = futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
1da177e4 | 2608 | break; |
4732efbe | 2609 | case FUTEX_WAKE_OP: |
b41277dc | 2610 | ret = futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
4732efbe | 2611 | break; |
c87e2837 | 2612 | case FUTEX_LOCK_PI: |
a0c1e907 | 2613 | if (futex_cmpxchg_enabled) |
b41277dc | 2614 | ret = futex_lock_pi(uaddr, flags, val, timeout, 0); |
c87e2837 IM |
2615 | break; |
2616 | case FUTEX_UNLOCK_PI: | |
a0c1e907 | 2617 | if (futex_cmpxchg_enabled) |
b41277dc | 2618 | ret = futex_unlock_pi(uaddr, flags); |
c87e2837 IM |
2619 | break; |
2620 | case FUTEX_TRYLOCK_PI: | |
a0c1e907 | 2621 | if (futex_cmpxchg_enabled) |
b41277dc | 2622 | ret = futex_lock_pi(uaddr, flags, 0, timeout, 1); |
c87e2837 | 2623 | break; |
52400ba9 DH |
2624 | case FUTEX_WAIT_REQUEUE_PI: |
2625 | val3 = FUTEX_BITSET_MATCH_ANY; | |
b41277dc DH |
2626 | ret = futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
2627 | uaddr2); | |
52400ba9 | 2628 | break; |
52400ba9 | 2629 | case FUTEX_CMP_REQUEUE_PI: |
b41277dc | 2630 | ret = futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
52400ba9 | 2631 | break; |
1da177e4 LT |
2632 | default: |
2633 | ret = -ENOSYS; | |
2634 | } | |
2635 | return ret; | |
2636 | } | |
2637 | ||
2638 | ||
17da2bd9 HC |
2639 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
2640 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
2641 | u32, val3) | |
1da177e4 | 2642 | { |
c19384b5 PP |
2643 | struct timespec ts; |
2644 | ktime_t t, *tp = NULL; | |
e2970f2f | 2645 | u32 val2 = 0; |
34f01cc1 | 2646 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 2647 | |
cd689985 | 2648 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
2649 | cmd == FUTEX_WAIT_BITSET || |
2650 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
c19384b5 | 2651 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 2652 | return -EFAULT; |
c19384b5 | 2653 | if (!timespec_valid(&ts)) |
9741ef96 | 2654 | return -EINVAL; |
c19384b5 PP |
2655 | |
2656 | t = timespec_to_ktime(ts); | |
34f01cc1 | 2657 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 2658 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 2659 | tp = &t; |
1da177e4 LT |
2660 | } |
2661 | /* | |
52400ba9 | 2662 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 2663 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 2664 | */ |
f54f0986 | 2665 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 2666 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 2667 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 2668 | |
c19384b5 | 2669 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
2670 | } |
2671 | ||
f6d107fb | 2672 | static int __init futex_init(void) |
1da177e4 | 2673 | { |
a0c1e907 | 2674 | u32 curval; |
3e4ab747 | 2675 | int i; |
95362fa9 | 2676 | |
a0c1e907 TG |
2677 | /* |
2678 | * This will fail and we want it. Some arch implementations do | |
2679 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
2680 | * functionality. We want to know that before we call in any | |
2681 | * of the complex code paths. Also we want to prevent | |
2682 | * registration of robust lists in that case. NULL is | |
2683 | * guaranteed to fault and we get -EFAULT on functional | |
fb62db2b | 2684 | * implementation, the non-functional ones will return |
a0c1e907 TG |
2685 | * -ENOSYS. |
2686 | */ | |
37a9d912 | 2687 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) |
a0c1e907 TG |
2688 | futex_cmpxchg_enabled = 1; |
2689 | ||
3e4ab747 TG |
2690 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { |
2691 | plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock); | |
2692 | spin_lock_init(&futex_queues[i].lock); | |
2693 | } | |
2694 | ||
1da177e4 LT |
2695 | return 0; |
2696 | } | |
f6d107fb | 2697 | __initcall(futex_init); |