]>
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> |
9984de1a | 58 | #include <linux/export.h> |
fd5eea42 | 59 | #include <linux/magic.h> |
b488893a PE |
60 | #include <linux/pid.h> |
61 | #include <linux/nsproxy.h> | |
bdbb776f | 62 | #include <linux/ptrace.h> |
8bd75c77 | 63 | #include <linux/sched/rt.h> |
13d60f4b | 64 | #include <linux/hugetlb.h> |
b488893a | 65 | |
4732efbe | 66 | #include <asm/futex.h> |
1da177e4 | 67 | |
c87e2837 IM |
68 | #include "rtmutex_common.h" |
69 | ||
a0c1e907 TG |
70 | int __read_mostly futex_cmpxchg_enabled; |
71 | ||
1da177e4 LT |
72 | #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8) |
73 | ||
b41277dc DH |
74 | /* |
75 | * Futex flags used to encode options to functions and preserve them across | |
76 | * restarts. | |
77 | */ | |
78 | #define FLAGS_SHARED 0x01 | |
79 | #define FLAGS_CLOCKRT 0x02 | |
80 | #define FLAGS_HAS_TIMEOUT 0x04 | |
81 | ||
c87e2837 IM |
82 | /* |
83 | * Priority Inheritance state: | |
84 | */ | |
85 | struct futex_pi_state { | |
86 | /* | |
87 | * list of 'owned' pi_state instances - these have to be | |
88 | * cleaned up in do_exit() if the task exits prematurely: | |
89 | */ | |
90 | struct list_head list; | |
91 | ||
92 | /* | |
93 | * The PI object: | |
94 | */ | |
95 | struct rt_mutex pi_mutex; | |
96 | ||
97 | struct task_struct *owner; | |
98 | atomic_t refcount; | |
99 | ||
100 | union futex_key key; | |
101 | }; | |
102 | ||
d8d88fbb DH |
103 | /** |
104 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 105 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
106 | * @task: the task waiting on the futex |
107 | * @lock_ptr: the hash bucket lock | |
108 | * @key: the key the futex is hashed on | |
109 | * @pi_state: optional priority inheritance state | |
110 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
111 | * @requeue_pi_key: the requeue_pi target futex key | |
112 | * @bitset: bitset for the optional bitmasked wakeup | |
113 | * | |
114 | * We use this hashed waitqueue, instead of a normal wait_queue_t, so | |
1da177e4 LT |
115 | * we can wake only the relevant ones (hashed queues may be shared). |
116 | * | |
117 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 118 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 119 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
120 | * the second. |
121 | * | |
122 | * PI futexes are typically woken before they are removed from the hash list via | |
123 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
124 | */ |
125 | struct futex_q { | |
ec92d082 | 126 | struct plist_node list; |
1da177e4 | 127 | |
d8d88fbb | 128 | struct task_struct *task; |
1da177e4 | 129 | spinlock_t *lock_ptr; |
1da177e4 | 130 | union futex_key key; |
c87e2837 | 131 | struct futex_pi_state *pi_state; |
52400ba9 | 132 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 133 | union futex_key *requeue_pi_key; |
cd689985 | 134 | u32 bitset; |
1da177e4 LT |
135 | }; |
136 | ||
5bdb05f9 DH |
137 | static const struct futex_q futex_q_init = { |
138 | /* list gets initialized in queue_me()*/ | |
139 | .key = FUTEX_KEY_INIT, | |
140 | .bitset = FUTEX_BITSET_MATCH_ANY | |
141 | }; | |
142 | ||
1da177e4 | 143 | /* |
b2d0994b DH |
144 | * Hash buckets are shared by all the futex_keys that hash to the same |
145 | * location. Each key may have multiple futex_q structures, one for each task | |
146 | * waiting on a futex. | |
1da177e4 LT |
147 | */ |
148 | struct futex_hash_bucket { | |
ec92d082 PP |
149 | spinlock_t lock; |
150 | struct plist_head chain; | |
1da177e4 LT |
151 | }; |
152 | ||
153 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; | |
154 | ||
1da177e4 LT |
155 | /* |
156 | * We hash on the keys returned from get_futex_key (see below). | |
157 | */ | |
158 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
159 | { | |
160 | u32 hash = jhash2((u32*)&key->both.word, | |
161 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
162 | key->both.offset); | |
163 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; | |
164 | } | |
165 | ||
166 | /* | |
167 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
168 | */ | |
169 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
170 | { | |
2bc87203 DH |
171 | return (key1 && key2 |
172 | && key1->both.word == key2->both.word | |
1da177e4 LT |
173 | && key1->both.ptr == key2->both.ptr |
174 | && key1->both.offset == key2->both.offset); | |
175 | } | |
176 | ||
38d47c1b PZ |
177 | /* |
178 | * Take a reference to the resource addressed by a key. | |
179 | * Can be called while holding spinlocks. | |
180 | * | |
181 | */ | |
182 | static void get_futex_key_refs(union futex_key *key) | |
183 | { | |
184 | if (!key->both.ptr) | |
185 | return; | |
186 | ||
187 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
188 | case FUT_OFF_INODE: | |
7de9c6ee | 189 | ihold(key->shared.inode); |
38d47c1b PZ |
190 | break; |
191 | case FUT_OFF_MMSHARED: | |
192 | atomic_inc(&key->private.mm->mm_count); | |
193 | break; | |
194 | } | |
195 | } | |
196 | ||
197 | /* | |
198 | * Drop a reference to the resource addressed by a key. | |
199 | * The hash bucket spinlock must not be held. | |
200 | */ | |
201 | static void drop_futex_key_refs(union futex_key *key) | |
202 | { | |
90621c40 DH |
203 | if (!key->both.ptr) { |
204 | /* If we're here then we tried to put a key we failed to get */ | |
205 | WARN_ON_ONCE(1); | |
38d47c1b | 206 | return; |
90621c40 | 207 | } |
38d47c1b PZ |
208 | |
209 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
210 | case FUT_OFF_INODE: | |
211 | iput(key->shared.inode); | |
212 | break; | |
213 | case FUT_OFF_MMSHARED: | |
214 | mmdrop(key->private.mm); | |
215 | break; | |
216 | } | |
217 | } | |
218 | ||
34f01cc1 | 219 | /** |
d96ee56c DH |
220 | * get_futex_key() - Get parameters which are the keys for a futex |
221 | * @uaddr: virtual address of the futex | |
222 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED | |
223 | * @key: address where result is stored. | |
9ea71503 SB |
224 | * @rw: mapping needs to be read/write (values: VERIFY_READ, |
225 | * VERIFY_WRITE) | |
34f01cc1 | 226 | * |
6c23cbbd RD |
227 | * Return: a negative error code or 0 |
228 | * | |
34f01cc1 | 229 | * The key words are stored in *key on success. |
1da177e4 | 230 | * |
6131ffaa | 231 | * For shared mappings, it's (page->index, file_inode(vma->vm_file), |
1da177e4 LT |
232 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
233 | * We can usually work out the index without swapping in the page. | |
234 | * | |
b2d0994b | 235 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 236 | */ |
64d1304a | 237 | static int |
9ea71503 | 238 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) |
1da177e4 | 239 | { |
e2970f2f | 240 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 241 | struct mm_struct *mm = current->mm; |
a5b338f2 | 242 | struct page *page, *page_head; |
9ea71503 | 243 | int err, ro = 0; |
1da177e4 LT |
244 | |
245 | /* | |
246 | * The futex address must be "naturally" aligned. | |
247 | */ | |
e2970f2f | 248 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 249 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 250 | return -EINVAL; |
e2970f2f | 251 | address -= key->both.offset; |
1da177e4 | 252 | |
34f01cc1 ED |
253 | /* |
254 | * PROCESS_PRIVATE futexes are fast. | |
255 | * As the mm cannot disappear under us and the 'key' only needs | |
256 | * virtual address, we dont even have to find the underlying vma. | |
257 | * Note : We do have to check 'uaddr' is a valid user address, | |
258 | * but access_ok() should be faster than find_vma() | |
259 | */ | |
260 | if (!fshared) { | |
7485d0d3 | 261 | if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32)))) |
34f01cc1 ED |
262 | return -EFAULT; |
263 | key->private.mm = mm; | |
264 | key->private.address = address; | |
42569c39 | 265 | get_futex_key_refs(key); |
34f01cc1 ED |
266 | return 0; |
267 | } | |
1da177e4 | 268 | |
38d47c1b | 269 | again: |
7485d0d3 | 270 | err = get_user_pages_fast(address, 1, 1, &page); |
9ea71503 SB |
271 | /* |
272 | * If write access is not required (eg. FUTEX_WAIT), try | |
273 | * and get read-only access. | |
274 | */ | |
275 | if (err == -EFAULT && rw == VERIFY_READ) { | |
276 | err = get_user_pages_fast(address, 1, 0, &page); | |
277 | ro = 1; | |
278 | } | |
38d47c1b PZ |
279 | if (err < 0) |
280 | return err; | |
9ea71503 SB |
281 | else |
282 | err = 0; | |
38d47c1b | 283 | |
a5b338f2 AA |
284 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
285 | page_head = page; | |
286 | if (unlikely(PageTail(page))) { | |
38d47c1b | 287 | put_page(page); |
a5b338f2 AA |
288 | /* serialize against __split_huge_page_splitting() */ |
289 | local_irq_disable(); | |
290 | if (likely(__get_user_pages_fast(address, 1, 1, &page) == 1)) { | |
291 | page_head = compound_head(page); | |
292 | /* | |
293 | * page_head is valid pointer but we must pin | |
294 | * it before taking the PG_lock and/or | |
295 | * PG_compound_lock. The moment we re-enable | |
296 | * irqs __split_huge_page_splitting() can | |
297 | * return and the head page can be freed from | |
298 | * under us. We can't take the PG_lock and/or | |
299 | * PG_compound_lock on a page that could be | |
300 | * freed from under us. | |
301 | */ | |
302 | if (page != page_head) { | |
303 | get_page(page_head); | |
304 | put_page(page); | |
305 | } | |
306 | local_irq_enable(); | |
307 | } else { | |
308 | local_irq_enable(); | |
309 | goto again; | |
310 | } | |
311 | } | |
312 | #else | |
313 | page_head = compound_head(page); | |
314 | if (page != page_head) { | |
315 | get_page(page_head); | |
316 | put_page(page); | |
317 | } | |
318 | #endif | |
319 | ||
320 | lock_page(page_head); | |
e6780f72 HD |
321 | |
322 | /* | |
323 | * If page_head->mapping is NULL, then it cannot be a PageAnon | |
324 | * page; but it might be the ZERO_PAGE or in the gate area or | |
325 | * in a special mapping (all cases which we are happy to fail); | |
326 | * or it may have been a good file page when get_user_pages_fast | |
327 | * found it, but truncated or holepunched or subjected to | |
328 | * invalidate_complete_page2 before we got the page lock (also | |
329 | * cases which we are happy to fail). And we hold a reference, | |
330 | * so refcount care in invalidate_complete_page's remove_mapping | |
331 | * prevents drop_caches from setting mapping to NULL beneath us. | |
332 | * | |
333 | * The case we do have to guard against is when memory pressure made | |
334 | * shmem_writepage move it from filecache to swapcache beneath us: | |
335 | * an unlikely race, but we do need to retry for page_head->mapping. | |
336 | */ | |
a5b338f2 | 337 | if (!page_head->mapping) { |
e6780f72 | 338 | int shmem_swizzled = PageSwapCache(page_head); |
a5b338f2 AA |
339 | unlock_page(page_head); |
340 | put_page(page_head); | |
e6780f72 HD |
341 | if (shmem_swizzled) |
342 | goto again; | |
343 | return -EFAULT; | |
38d47c1b | 344 | } |
1da177e4 LT |
345 | |
346 | /* | |
347 | * Private mappings are handled in a simple way. | |
348 | * | |
349 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
350 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 351 | * the object not the particular process. |
1da177e4 | 352 | */ |
a5b338f2 | 353 | if (PageAnon(page_head)) { |
9ea71503 SB |
354 | /* |
355 | * A RO anonymous page will never change and thus doesn't make | |
356 | * sense for futex operations. | |
357 | */ | |
358 | if (ro) { | |
359 | err = -EFAULT; | |
360 | goto out; | |
361 | } | |
362 | ||
38d47c1b | 363 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 364 | key->private.mm = mm; |
e2970f2f | 365 | key->private.address = address; |
38d47c1b PZ |
366 | } else { |
367 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ | |
a5b338f2 | 368 | key->shared.inode = page_head->mapping->host; |
13d60f4b | 369 | key->shared.pgoff = basepage_index(page); |
1da177e4 LT |
370 | } |
371 | ||
38d47c1b | 372 | get_futex_key_refs(key); |
1da177e4 | 373 | |
9ea71503 | 374 | out: |
a5b338f2 AA |
375 | unlock_page(page_head); |
376 | put_page(page_head); | |
9ea71503 | 377 | return err; |
1da177e4 LT |
378 | } |
379 | ||
ae791a2d | 380 | static inline void put_futex_key(union futex_key *key) |
1da177e4 | 381 | { |
38d47c1b | 382 | drop_futex_key_refs(key); |
1da177e4 LT |
383 | } |
384 | ||
d96ee56c DH |
385 | /** |
386 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
387 | * @uaddr: pointer to faulting user space address |
388 | * | |
389 | * Slow path to fixup the fault we just took in the atomic write | |
390 | * access to @uaddr. | |
391 | * | |
fb62db2b | 392 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
393 | * user address. We know that we faulted in the atomic pagefault |
394 | * disabled section so we can as well avoid the #PF overhead by | |
395 | * calling get_user_pages() right away. | |
396 | */ | |
397 | static int fault_in_user_writeable(u32 __user *uaddr) | |
398 | { | |
722d0172 AK |
399 | struct mm_struct *mm = current->mm; |
400 | int ret; | |
401 | ||
402 | down_read(&mm->mmap_sem); | |
2efaca92 BH |
403 | ret = fixup_user_fault(current, mm, (unsigned long)uaddr, |
404 | FAULT_FLAG_WRITE); | |
722d0172 AK |
405 | up_read(&mm->mmap_sem); |
406 | ||
d0725992 TG |
407 | return ret < 0 ? ret : 0; |
408 | } | |
409 | ||
4b1c486b DH |
410 | /** |
411 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
412 | * @hb: the hash bucket the futex_q's reside in |
413 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
414 | * |
415 | * Must be called with the hb lock held. | |
416 | */ | |
417 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
418 | union futex_key *key) | |
419 | { | |
420 | struct futex_q *this; | |
421 | ||
422 | plist_for_each_entry(this, &hb->chain, list) { | |
423 | if (match_futex(&this->key, key)) | |
424 | return this; | |
425 | } | |
426 | return NULL; | |
427 | } | |
428 | ||
37a9d912 ML |
429 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
430 | u32 uval, u32 newval) | |
36cf3b5c | 431 | { |
37a9d912 | 432 | int ret; |
36cf3b5c TG |
433 | |
434 | pagefault_disable(); | |
37a9d912 | 435 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
436 | pagefault_enable(); |
437 | ||
37a9d912 | 438 | return ret; |
36cf3b5c TG |
439 | } |
440 | ||
441 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
442 | { |
443 | int ret; | |
444 | ||
a866374a | 445 | pagefault_disable(); |
e2970f2f | 446 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 447 | pagefault_enable(); |
1da177e4 LT |
448 | |
449 | return ret ? -EFAULT : 0; | |
450 | } | |
451 | ||
c87e2837 IM |
452 | |
453 | /* | |
454 | * PI code: | |
455 | */ | |
456 | static int refill_pi_state_cache(void) | |
457 | { | |
458 | struct futex_pi_state *pi_state; | |
459 | ||
460 | if (likely(current->pi_state_cache)) | |
461 | return 0; | |
462 | ||
4668edc3 | 463 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
464 | |
465 | if (!pi_state) | |
466 | return -ENOMEM; | |
467 | ||
c87e2837 IM |
468 | INIT_LIST_HEAD(&pi_state->list); |
469 | /* pi_mutex gets initialized later */ | |
470 | pi_state->owner = NULL; | |
471 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 472 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
473 | |
474 | current->pi_state_cache = pi_state; | |
475 | ||
476 | return 0; | |
477 | } | |
478 | ||
479 | static struct futex_pi_state * alloc_pi_state(void) | |
480 | { | |
481 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
482 | ||
483 | WARN_ON(!pi_state); | |
484 | current->pi_state_cache = NULL; | |
485 | ||
486 | return pi_state; | |
487 | } | |
488 | ||
489 | static void free_pi_state(struct futex_pi_state *pi_state) | |
490 | { | |
491 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
492 | return; | |
493 | ||
494 | /* | |
495 | * If pi_state->owner is NULL, the owner is most probably dying | |
496 | * and has cleaned up the pi_state already | |
497 | */ | |
498 | if (pi_state->owner) { | |
1d615482 | 499 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
c87e2837 | 500 | list_del_init(&pi_state->list); |
1d615482 | 501 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
c87e2837 IM |
502 | |
503 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
504 | } | |
505 | ||
506 | if (current->pi_state_cache) | |
507 | kfree(pi_state); | |
508 | else { | |
509 | /* | |
510 | * pi_state->list is already empty. | |
511 | * clear pi_state->owner. | |
512 | * refcount is at 0 - put it back to 1. | |
513 | */ | |
514 | pi_state->owner = NULL; | |
515 | atomic_set(&pi_state->refcount, 1); | |
516 | current->pi_state_cache = pi_state; | |
517 | } | |
518 | } | |
519 | ||
520 | /* | |
521 | * Look up the task based on what TID userspace gave us. | |
522 | * We dont trust it. | |
523 | */ | |
524 | static struct task_struct * futex_find_get_task(pid_t pid) | |
525 | { | |
526 | struct task_struct *p; | |
527 | ||
d359b549 | 528 | rcu_read_lock(); |
228ebcbe | 529 | p = find_task_by_vpid(pid); |
7a0ea09a MH |
530 | if (p) |
531 | get_task_struct(p); | |
a06381fe | 532 | |
d359b549 | 533 | rcu_read_unlock(); |
c87e2837 IM |
534 | |
535 | return p; | |
536 | } | |
537 | ||
538 | /* | |
539 | * This task is holding PI mutexes at exit time => bad. | |
540 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
541 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
542 | */ | |
543 | void exit_pi_state_list(struct task_struct *curr) | |
544 | { | |
c87e2837 IM |
545 | struct list_head *next, *head = &curr->pi_state_list; |
546 | struct futex_pi_state *pi_state; | |
627371d7 | 547 | struct futex_hash_bucket *hb; |
38d47c1b | 548 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 549 | |
a0c1e907 TG |
550 | if (!futex_cmpxchg_enabled) |
551 | return; | |
c87e2837 IM |
552 | /* |
553 | * We are a ZOMBIE and nobody can enqueue itself on | |
554 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 555 | * versus waiters unqueueing themselves: |
c87e2837 | 556 | */ |
1d615482 | 557 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 IM |
558 | while (!list_empty(head)) { |
559 | ||
560 | next = head->next; | |
561 | pi_state = list_entry(next, struct futex_pi_state, list); | |
562 | key = pi_state->key; | |
627371d7 | 563 | hb = hash_futex(&key); |
1d615482 | 564 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 565 | |
c87e2837 IM |
566 | spin_lock(&hb->lock); |
567 | ||
1d615482 | 568 | raw_spin_lock_irq(&curr->pi_lock); |
627371d7 IM |
569 | /* |
570 | * We dropped the pi-lock, so re-check whether this | |
571 | * task still owns the PI-state: | |
572 | */ | |
c87e2837 IM |
573 | if (head->next != next) { |
574 | spin_unlock(&hb->lock); | |
575 | continue; | |
576 | } | |
577 | ||
c87e2837 | 578 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
579 | WARN_ON(list_empty(&pi_state->list)); |
580 | list_del_init(&pi_state->list); | |
c87e2837 | 581 | pi_state->owner = NULL; |
1d615482 | 582 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
583 | |
584 | rt_mutex_unlock(&pi_state->pi_mutex); | |
585 | ||
586 | spin_unlock(&hb->lock); | |
587 | ||
1d615482 | 588 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 589 | } |
1d615482 | 590 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
591 | } |
592 | ||
593 | static int | |
d0aa7a70 PP |
594 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
595 | union futex_key *key, struct futex_pi_state **ps) | |
c87e2837 IM |
596 | { |
597 | struct futex_pi_state *pi_state = NULL; | |
598 | struct futex_q *this, *next; | |
ec92d082 | 599 | struct plist_head *head; |
c87e2837 | 600 | struct task_struct *p; |
778e9a9c | 601 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 IM |
602 | |
603 | head = &hb->chain; | |
604 | ||
ec92d082 | 605 | plist_for_each_entry_safe(this, next, head, list) { |
d0aa7a70 | 606 | if (match_futex(&this->key, key)) { |
c87e2837 IM |
607 | /* |
608 | * Another waiter already exists - bump up | |
609 | * the refcount and return its pi_state: | |
610 | */ | |
611 | pi_state = this->pi_state; | |
06a9ec29 | 612 | /* |
fb62db2b | 613 | * Userspace might have messed up non-PI and PI futexes |
06a9ec29 TG |
614 | */ |
615 | if (unlikely(!pi_state)) | |
616 | return -EINVAL; | |
617 | ||
627371d7 | 618 | WARN_ON(!atomic_read(&pi_state->refcount)); |
59647b6a TG |
619 | |
620 | /* | |
621 | * When pi_state->owner is NULL then the owner died | |
622 | * and another waiter is on the fly. pi_state->owner | |
623 | * is fixed up by the task which acquires | |
624 | * pi_state->rt_mutex. | |
625 | * | |
626 | * We do not check for pid == 0 which can happen when | |
627 | * the owner died and robust_list_exit() cleared the | |
628 | * TID. | |
629 | */ | |
630 | if (pid && pi_state->owner) { | |
631 | /* | |
632 | * Bail out if user space manipulated the | |
633 | * futex value. | |
634 | */ | |
635 | if (pid != task_pid_vnr(pi_state->owner)) | |
636 | return -EINVAL; | |
637 | } | |
627371d7 | 638 | |
c87e2837 | 639 | atomic_inc(&pi_state->refcount); |
d0aa7a70 | 640 | *ps = pi_state; |
c87e2837 IM |
641 | |
642 | return 0; | |
643 | } | |
644 | } | |
645 | ||
646 | /* | |
e3f2ddea | 647 | * We are the first waiter - try to look up the real owner and attach |
778e9a9c | 648 | * the new pi_state to it, but bail out when TID = 0 |
c87e2837 | 649 | */ |
778e9a9c | 650 | if (!pid) |
e3f2ddea | 651 | return -ESRCH; |
c87e2837 | 652 | p = futex_find_get_task(pid); |
7a0ea09a MH |
653 | if (!p) |
654 | return -ESRCH; | |
778e9a9c AK |
655 | |
656 | /* | |
657 | * We need to look at the task state flags to figure out, | |
658 | * whether the task is exiting. To protect against the do_exit | |
659 | * change of the task flags, we do this protected by | |
660 | * p->pi_lock: | |
661 | */ | |
1d615482 | 662 | raw_spin_lock_irq(&p->pi_lock); |
778e9a9c AK |
663 | if (unlikely(p->flags & PF_EXITING)) { |
664 | /* | |
665 | * The task is on the way out. When PF_EXITPIDONE is | |
666 | * set, we know that the task has finished the | |
667 | * cleanup: | |
668 | */ | |
669 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
670 | ||
1d615482 | 671 | raw_spin_unlock_irq(&p->pi_lock); |
778e9a9c AK |
672 | put_task_struct(p); |
673 | return ret; | |
674 | } | |
c87e2837 IM |
675 | |
676 | pi_state = alloc_pi_state(); | |
677 | ||
678 | /* | |
679 | * Initialize the pi_mutex in locked state and make 'p' | |
680 | * the owner of it: | |
681 | */ | |
682 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
683 | ||
684 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 685 | pi_state->key = *key; |
c87e2837 | 686 | |
627371d7 | 687 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
688 | list_add(&pi_state->list, &p->pi_state_list); |
689 | pi_state->owner = p; | |
1d615482 | 690 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
691 | |
692 | put_task_struct(p); | |
693 | ||
d0aa7a70 | 694 | *ps = pi_state; |
c87e2837 IM |
695 | |
696 | return 0; | |
697 | } | |
698 | ||
1a52084d | 699 | /** |
d96ee56c | 700 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
701 | * @uaddr: the pi futex user address |
702 | * @hb: the pi futex hash bucket | |
703 | * @key: the futex key associated with uaddr and hb | |
704 | * @ps: the pi_state pointer where we store the result of the | |
705 | * lookup | |
706 | * @task: the task to perform the atomic lock work for. This will | |
707 | * be "current" except in the case of requeue pi. | |
708 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d | 709 | * |
6c23cbbd RD |
710 | * Return: |
711 | * 0 - ready to wait; | |
712 | * 1 - acquired the lock; | |
1a52084d DH |
713 | * <0 - error |
714 | * | |
715 | * The hb->lock and futex_key refs shall be held by the caller. | |
716 | */ | |
717 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
718 | union futex_key *key, | |
719 | struct futex_pi_state **ps, | |
bab5bc9e | 720 | struct task_struct *task, int set_waiters) |
1a52084d | 721 | { |
59fa6245 | 722 | int lock_taken, ret, force_take = 0; |
c0c9ed15 | 723 | u32 uval, newval, curval, vpid = task_pid_vnr(task); |
1a52084d DH |
724 | |
725 | retry: | |
726 | ret = lock_taken = 0; | |
727 | ||
728 | /* | |
729 | * To avoid races, we attempt to take the lock here again | |
730 | * (by doing a 0 -> TID atomic cmpxchg), while holding all | |
731 | * the locks. It will most likely not succeed. | |
732 | */ | |
c0c9ed15 | 733 | newval = vpid; |
bab5bc9e DH |
734 | if (set_waiters) |
735 | newval |= FUTEX_WAITERS; | |
1a52084d | 736 | |
37a9d912 | 737 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval))) |
1a52084d DH |
738 | return -EFAULT; |
739 | ||
740 | /* | |
741 | * Detect deadlocks. | |
742 | */ | |
c0c9ed15 | 743 | if ((unlikely((curval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
744 | return -EDEADLK; |
745 | ||
746 | /* | |
747 | * Surprise - we got the lock. Just return to userspace: | |
748 | */ | |
749 | if (unlikely(!curval)) | |
750 | return 1; | |
751 | ||
752 | uval = curval; | |
753 | ||
754 | /* | |
755 | * Set the FUTEX_WAITERS flag, so the owner will know it has someone | |
756 | * to wake at the next unlock. | |
757 | */ | |
758 | newval = curval | FUTEX_WAITERS; | |
759 | ||
760 | /* | |
59fa6245 | 761 | * Should we force take the futex? See below. |
1a52084d | 762 | */ |
59fa6245 TG |
763 | if (unlikely(force_take)) { |
764 | /* | |
765 | * Keep the OWNER_DIED and the WAITERS bit and set the | |
766 | * new TID value. | |
767 | */ | |
c0c9ed15 | 768 | newval = (curval & ~FUTEX_TID_MASK) | vpid; |
59fa6245 | 769 | force_take = 0; |
1a52084d DH |
770 | lock_taken = 1; |
771 | } | |
772 | ||
37a9d912 | 773 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))) |
1a52084d DH |
774 | return -EFAULT; |
775 | if (unlikely(curval != uval)) | |
776 | goto retry; | |
777 | ||
778 | /* | |
59fa6245 | 779 | * We took the lock due to forced take over. |
1a52084d DH |
780 | */ |
781 | if (unlikely(lock_taken)) | |
782 | return 1; | |
783 | ||
784 | /* | |
785 | * We dont have the lock. Look up the PI state (or create it if | |
786 | * we are the first waiter): | |
787 | */ | |
788 | ret = lookup_pi_state(uval, hb, key, ps); | |
789 | ||
790 | if (unlikely(ret)) { | |
791 | switch (ret) { | |
792 | case -ESRCH: | |
793 | /* | |
59fa6245 TG |
794 | * We failed to find an owner for this |
795 | * futex. So we have no pi_state to block | |
796 | * on. This can happen in two cases: | |
797 | * | |
798 | * 1) The owner died | |
799 | * 2) A stale FUTEX_WAITERS bit | |
800 | * | |
801 | * Re-read the futex value. | |
1a52084d DH |
802 | */ |
803 | if (get_futex_value_locked(&curval, uaddr)) | |
804 | return -EFAULT; | |
805 | ||
806 | /* | |
59fa6245 TG |
807 | * If the owner died or we have a stale |
808 | * WAITERS bit the owner TID in the user space | |
809 | * futex is 0. | |
1a52084d | 810 | */ |
59fa6245 TG |
811 | if (!(curval & FUTEX_TID_MASK)) { |
812 | force_take = 1; | |
1a52084d DH |
813 | goto retry; |
814 | } | |
815 | default: | |
816 | break; | |
817 | } | |
818 | } | |
819 | ||
820 | return ret; | |
821 | } | |
822 | ||
2e12978a LJ |
823 | /** |
824 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
825 | * @q: The futex_q to unqueue | |
826 | * | |
827 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
828 | */ | |
829 | static void __unqueue_futex(struct futex_q *q) | |
830 | { | |
831 | struct futex_hash_bucket *hb; | |
832 | ||
29096202 SR |
833 | if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr)) |
834 | || WARN_ON(plist_node_empty(&q->list))) | |
2e12978a LJ |
835 | return; |
836 | ||
837 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
838 | plist_del(&q->list, &hb->chain); | |
839 | } | |
840 | ||
1da177e4 LT |
841 | /* |
842 | * The hash bucket lock must be held when this is called. | |
843 | * Afterwards, the futex_q must not be accessed. | |
844 | */ | |
845 | static void wake_futex(struct futex_q *q) | |
846 | { | |
f1a11e05 TG |
847 | struct task_struct *p = q->task; |
848 | ||
aa10990e DH |
849 | if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n")) |
850 | return; | |
851 | ||
1da177e4 | 852 | /* |
f1a11e05 | 853 | * We set q->lock_ptr = NULL _before_ we wake up the task. If |
fb62db2b RD |
854 | * a non-futex wake up happens on another CPU then the task |
855 | * might exit and p would dereference a non-existing task | |
f1a11e05 TG |
856 | * struct. Prevent this by holding a reference on p across the |
857 | * wake up. | |
1da177e4 | 858 | */ |
f1a11e05 TG |
859 | get_task_struct(p); |
860 | ||
2e12978a | 861 | __unqueue_futex(q); |
1da177e4 | 862 | /* |
f1a11e05 TG |
863 | * The waiting task can free the futex_q as soon as |
864 | * q->lock_ptr = NULL is written, without taking any locks. A | |
865 | * memory barrier is required here to prevent the following | |
866 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 867 | */ |
ccdea2f8 | 868 | smp_wmb(); |
1da177e4 | 869 | q->lock_ptr = NULL; |
f1a11e05 TG |
870 | |
871 | wake_up_state(p, TASK_NORMAL); | |
872 | put_task_struct(p); | |
1da177e4 LT |
873 | } |
874 | ||
c87e2837 IM |
875 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
876 | { | |
877 | struct task_struct *new_owner; | |
878 | struct futex_pi_state *pi_state = this->pi_state; | |
7cfdaf38 | 879 | u32 uninitialized_var(curval), newval; |
c87e2837 IM |
880 | |
881 | if (!pi_state) | |
882 | return -EINVAL; | |
883 | ||
51246bfd TG |
884 | /* |
885 | * If current does not own the pi_state then the futex is | |
886 | * inconsistent and user space fiddled with the futex value. | |
887 | */ | |
888 | if (pi_state->owner != current) | |
889 | return -EINVAL; | |
890 | ||
d209d74d | 891 | raw_spin_lock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
892 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
893 | ||
894 | /* | |
f123c98e SR |
895 | * It is possible that the next waiter (the one that brought |
896 | * this owner to the kernel) timed out and is no longer | |
897 | * waiting on the lock. | |
c87e2837 IM |
898 | */ |
899 | if (!new_owner) | |
900 | new_owner = this->task; | |
901 | ||
902 | /* | |
903 | * We pass it to the next owner. (The WAITERS bit is always | |
904 | * kept enabled while there is PI state around. We must also | |
905 | * preserve the owner died bit.) | |
906 | */ | |
e3f2ddea | 907 | if (!(uval & FUTEX_OWNER_DIED)) { |
778e9a9c AK |
908 | int ret = 0; |
909 | ||
b488893a | 910 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 911 | |
37a9d912 | 912 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
778e9a9c | 913 | ret = -EFAULT; |
cde898fa | 914 | else if (curval != uval) |
778e9a9c AK |
915 | ret = -EINVAL; |
916 | if (ret) { | |
d209d74d | 917 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
778e9a9c AK |
918 | return ret; |
919 | } | |
e3f2ddea | 920 | } |
c87e2837 | 921 | |
1d615482 | 922 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
627371d7 IM |
923 | WARN_ON(list_empty(&pi_state->list)); |
924 | list_del_init(&pi_state->list); | |
1d615482 | 925 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
627371d7 | 926 | |
1d615482 | 927 | raw_spin_lock_irq(&new_owner->pi_lock); |
627371d7 | 928 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
929 | list_add(&pi_state->list, &new_owner->pi_state_list); |
930 | pi_state->owner = new_owner; | |
1d615482 | 931 | raw_spin_unlock_irq(&new_owner->pi_lock); |
627371d7 | 932 | |
d209d74d | 933 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
934 | rt_mutex_unlock(&pi_state->pi_mutex); |
935 | ||
936 | return 0; | |
937 | } | |
938 | ||
939 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) | |
940 | { | |
7cfdaf38 | 941 | u32 uninitialized_var(oldval); |
c87e2837 IM |
942 | |
943 | /* | |
944 | * There is no waiter, so we unlock the futex. The owner died | |
945 | * bit has not to be preserved here. We are the owner: | |
946 | */ | |
37a9d912 ML |
947 | if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0)) |
948 | return -EFAULT; | |
c87e2837 IM |
949 | if (oldval != uval) |
950 | return -EAGAIN; | |
951 | ||
952 | return 0; | |
953 | } | |
954 | ||
8b8f319f IM |
955 | /* |
956 | * Express the locking dependencies for lockdep: | |
957 | */ | |
958 | static inline void | |
959 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
960 | { | |
961 | if (hb1 <= hb2) { | |
962 | spin_lock(&hb1->lock); | |
963 | if (hb1 < hb2) | |
964 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
965 | } else { /* hb1 > hb2 */ | |
966 | spin_lock(&hb2->lock); | |
967 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
968 | } | |
969 | } | |
970 | ||
5eb3dc62 DH |
971 | static inline void |
972 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
973 | { | |
f061d351 | 974 | spin_unlock(&hb1->lock); |
88f502fe IM |
975 | if (hb1 != hb2) |
976 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
977 | } |
978 | ||
1da177e4 | 979 | /* |
b2d0994b | 980 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 981 | */ |
b41277dc DH |
982 | static int |
983 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 984 | { |
e2970f2f | 985 | struct futex_hash_bucket *hb; |
1da177e4 | 986 | struct futex_q *this, *next; |
ec92d082 | 987 | struct plist_head *head; |
38d47c1b | 988 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 LT |
989 | int ret; |
990 | ||
cd689985 TG |
991 | if (!bitset) |
992 | return -EINVAL; | |
993 | ||
9ea71503 | 994 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ); |
1da177e4 LT |
995 | if (unlikely(ret != 0)) |
996 | goto out; | |
997 | ||
e2970f2f IM |
998 | hb = hash_futex(&key); |
999 | spin_lock(&hb->lock); | |
1000 | head = &hb->chain; | |
1da177e4 | 1001 | |
ec92d082 | 1002 | plist_for_each_entry_safe(this, next, head, list) { |
1da177e4 | 1003 | if (match_futex (&this->key, &key)) { |
52400ba9 | 1004 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
1005 | ret = -EINVAL; |
1006 | break; | |
1007 | } | |
cd689985 TG |
1008 | |
1009 | /* Check if one of the bits is set in both bitsets */ | |
1010 | if (!(this->bitset & bitset)) | |
1011 | continue; | |
1012 | ||
1da177e4 LT |
1013 | wake_futex(this); |
1014 | if (++ret >= nr_wake) | |
1015 | break; | |
1016 | } | |
1017 | } | |
1018 | ||
e2970f2f | 1019 | spin_unlock(&hb->lock); |
ae791a2d | 1020 | put_futex_key(&key); |
42d35d48 | 1021 | out: |
1da177e4 LT |
1022 | return ret; |
1023 | } | |
1024 | ||
4732efbe JJ |
1025 | /* |
1026 | * Wake up all waiters hashed on the physical page that is mapped | |
1027 | * to this virtual address: | |
1028 | */ | |
e2970f2f | 1029 | static int |
b41277dc | 1030 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 1031 | int nr_wake, int nr_wake2, int op) |
4732efbe | 1032 | { |
38d47c1b | 1033 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 1034 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 1035 | struct plist_head *head; |
4732efbe | 1036 | struct futex_q *this, *next; |
e4dc5b7a | 1037 | int ret, op_ret; |
4732efbe | 1038 | |
e4dc5b7a | 1039 | retry: |
9ea71503 | 1040 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
4732efbe JJ |
1041 | if (unlikely(ret != 0)) |
1042 | goto out; | |
9ea71503 | 1043 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
4732efbe | 1044 | if (unlikely(ret != 0)) |
42d35d48 | 1045 | goto out_put_key1; |
4732efbe | 1046 | |
e2970f2f IM |
1047 | hb1 = hash_futex(&key1); |
1048 | hb2 = hash_futex(&key2); | |
4732efbe | 1049 | |
e4dc5b7a | 1050 | retry_private: |
eaaea803 | 1051 | double_lock_hb(hb1, hb2); |
e2970f2f | 1052 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1053 | if (unlikely(op_ret < 0)) { |
4732efbe | 1054 | |
5eb3dc62 | 1055 | double_unlock_hb(hb1, hb2); |
4732efbe | 1056 | |
7ee1dd3f | 1057 | #ifndef CONFIG_MMU |
e2970f2f IM |
1058 | /* |
1059 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
1060 | * but we might get them from range checking | |
1061 | */ | |
7ee1dd3f | 1062 | ret = op_ret; |
42d35d48 | 1063 | goto out_put_keys; |
7ee1dd3f DH |
1064 | #endif |
1065 | ||
796f8d9b DG |
1066 | if (unlikely(op_ret != -EFAULT)) { |
1067 | ret = op_ret; | |
42d35d48 | 1068 | goto out_put_keys; |
796f8d9b DG |
1069 | } |
1070 | ||
d0725992 | 1071 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 1072 | if (ret) |
de87fcc1 | 1073 | goto out_put_keys; |
4732efbe | 1074 | |
b41277dc | 1075 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1076 | goto retry_private; |
1077 | ||
ae791a2d TG |
1078 | put_futex_key(&key2); |
1079 | put_futex_key(&key1); | |
e4dc5b7a | 1080 | goto retry; |
4732efbe JJ |
1081 | } |
1082 | ||
e2970f2f | 1083 | head = &hb1->chain; |
4732efbe | 1084 | |
ec92d082 | 1085 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe | 1086 | if (match_futex (&this->key, &key1)) { |
aa10990e DH |
1087 | if (this->pi_state || this->rt_waiter) { |
1088 | ret = -EINVAL; | |
1089 | goto out_unlock; | |
1090 | } | |
4732efbe JJ |
1091 | wake_futex(this); |
1092 | if (++ret >= nr_wake) | |
1093 | break; | |
1094 | } | |
1095 | } | |
1096 | ||
1097 | if (op_ret > 0) { | |
e2970f2f | 1098 | head = &hb2->chain; |
4732efbe JJ |
1099 | |
1100 | op_ret = 0; | |
ec92d082 | 1101 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe | 1102 | if (match_futex (&this->key, &key2)) { |
aa10990e DH |
1103 | if (this->pi_state || this->rt_waiter) { |
1104 | ret = -EINVAL; | |
1105 | goto out_unlock; | |
1106 | } | |
4732efbe JJ |
1107 | wake_futex(this); |
1108 | if (++op_ret >= nr_wake2) | |
1109 | break; | |
1110 | } | |
1111 | } | |
1112 | ret += op_ret; | |
1113 | } | |
1114 | ||
aa10990e | 1115 | out_unlock: |
5eb3dc62 | 1116 | double_unlock_hb(hb1, hb2); |
42d35d48 | 1117 | out_put_keys: |
ae791a2d | 1118 | put_futex_key(&key2); |
42d35d48 | 1119 | out_put_key1: |
ae791a2d | 1120 | put_futex_key(&key1); |
42d35d48 | 1121 | out: |
4732efbe JJ |
1122 | return ret; |
1123 | } | |
1124 | ||
9121e478 DH |
1125 | /** |
1126 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1127 | * @q: the futex_q to requeue | |
1128 | * @hb1: the source hash_bucket | |
1129 | * @hb2: the target hash_bucket | |
1130 | * @key2: the new key for the requeued futex_q | |
1131 | */ | |
1132 | static inline | |
1133 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1134 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1135 | { | |
1136 | ||
1137 | /* | |
1138 | * If key1 and key2 hash to the same bucket, no need to | |
1139 | * requeue. | |
1140 | */ | |
1141 | if (likely(&hb1->chain != &hb2->chain)) { | |
1142 | plist_del(&q->list, &hb1->chain); | |
1143 | plist_add(&q->list, &hb2->chain); | |
1144 | q->lock_ptr = &hb2->lock; | |
9121e478 DH |
1145 | } |
1146 | get_futex_key_refs(key2); | |
1147 | q->key = *key2; | |
1148 | } | |
1149 | ||
52400ba9 DH |
1150 | /** |
1151 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1152 | * @q: the futex_q |
1153 | * @key: the key of the requeue target futex | |
1154 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1155 | * |
1156 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1157 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1158 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1159 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1160 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1161 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1162 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1163 | */ |
1164 | static inline | |
beda2c7e DH |
1165 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1166 | struct futex_hash_bucket *hb) | |
52400ba9 | 1167 | { |
52400ba9 DH |
1168 | get_futex_key_refs(key); |
1169 | q->key = *key; | |
1170 | ||
2e12978a | 1171 | __unqueue_futex(q); |
52400ba9 DH |
1172 | |
1173 | WARN_ON(!q->rt_waiter); | |
1174 | q->rt_waiter = NULL; | |
1175 | ||
beda2c7e | 1176 | q->lock_ptr = &hb->lock; |
beda2c7e | 1177 | |
f1a11e05 | 1178 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1179 | } |
1180 | ||
1181 | /** | |
1182 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1183 | * @pifutex: the user address of the to futex |
1184 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1185 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1186 | * @key1: the from futex key | |
1187 | * @key2: the to futex key | |
1188 | * @ps: address to store the pi_state pointer | |
1189 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1190 | * |
1191 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1192 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1193 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1194 | * hb1 and hb2 must be held by the caller. | |
52400ba9 | 1195 | * |
6c23cbbd RD |
1196 | * Return: |
1197 | * 0 - failed to acquire the lock atomically; | |
1198 | * 1 - acquired the lock; | |
52400ba9 DH |
1199 | * <0 - error |
1200 | */ | |
1201 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1202 | struct futex_hash_bucket *hb1, | |
1203 | struct futex_hash_bucket *hb2, | |
1204 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1205 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1206 | { |
bab5bc9e | 1207 | struct futex_q *top_waiter = NULL; |
52400ba9 DH |
1208 | u32 curval; |
1209 | int ret; | |
1210 | ||
1211 | if (get_futex_value_locked(&curval, pifutex)) | |
1212 | return -EFAULT; | |
1213 | ||
bab5bc9e DH |
1214 | /* |
1215 | * Find the top_waiter and determine if there are additional waiters. | |
1216 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1217 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1218 | * as we have means to handle the possible fault. If not, don't set | |
1219 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1220 | * the kernel. | |
1221 | */ | |
52400ba9 DH |
1222 | top_waiter = futex_top_waiter(hb1, key1); |
1223 | ||
1224 | /* There are no waiters, nothing for us to do. */ | |
1225 | if (!top_waiter) | |
1226 | return 0; | |
1227 | ||
84bc4af5 DH |
1228 | /* Ensure we requeue to the expected futex. */ |
1229 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1230 | return -EINVAL; | |
1231 | ||
52400ba9 | 1232 | /* |
bab5bc9e DH |
1233 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1234 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1235 | * in ps in contended cases. | |
52400ba9 | 1236 | */ |
bab5bc9e DH |
1237 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1238 | set_waiters); | |
52400ba9 | 1239 | if (ret == 1) |
beda2c7e | 1240 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
52400ba9 DH |
1241 | |
1242 | return ret; | |
1243 | } | |
1244 | ||
1245 | /** | |
1246 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1247 | * @uaddr1: source futex user address |
b41277dc | 1248 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1249 | * @uaddr2: target futex user address |
1250 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1251 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1252 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1253 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1254 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1255 | * |
1256 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1257 | * uaddr2 atomically on behalf of the top waiter. | |
1258 | * | |
6c23cbbd RD |
1259 | * Return: |
1260 | * >=0 - on success, the number of tasks requeued or woken; | |
52400ba9 | 1261 | * <0 - on error |
1da177e4 | 1262 | */ |
b41277dc DH |
1263 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
1264 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
1265 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 1266 | { |
38d47c1b | 1267 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1268 | int drop_count = 0, task_count = 0, ret; |
1269 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1270 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 1271 | struct plist_head *head1; |
1da177e4 | 1272 | struct futex_q *this, *next; |
52400ba9 DH |
1273 | u32 curval2; |
1274 | ||
1275 | if (requeue_pi) { | |
1276 | /* | |
1277 | * requeue_pi requires a pi_state, try to allocate it now | |
1278 | * without any locks in case it fails. | |
1279 | */ | |
1280 | if (refill_pi_state_cache()) | |
1281 | return -ENOMEM; | |
1282 | /* | |
1283 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1284 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1285 | * returning to userspace, so as to not leave the rt_mutex with | |
1286 | * waiters and no owner. However, second and third wake-ups | |
1287 | * cannot be predicted as they involve race conditions with the | |
1288 | * first wake and a fault while looking up the pi_state. Both | |
1289 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1290 | * use nr_wake=1. | |
1291 | */ | |
1292 | if (nr_wake != 1) | |
1293 | return -EINVAL; | |
1294 | } | |
1da177e4 | 1295 | |
42d35d48 | 1296 | retry: |
52400ba9 DH |
1297 | if (pi_state != NULL) { |
1298 | /* | |
1299 | * We will have to lookup the pi_state again, so free this one | |
1300 | * to keep the accounting correct. | |
1301 | */ | |
1302 | free_pi_state(pi_state); | |
1303 | pi_state = NULL; | |
1304 | } | |
1305 | ||
9ea71503 | 1306 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
1da177e4 LT |
1307 | if (unlikely(ret != 0)) |
1308 | goto out; | |
9ea71503 SB |
1309 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, |
1310 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1311 | if (unlikely(ret != 0)) |
42d35d48 | 1312 | goto out_put_key1; |
1da177e4 | 1313 | |
e2970f2f IM |
1314 | hb1 = hash_futex(&key1); |
1315 | hb2 = hash_futex(&key2); | |
1da177e4 | 1316 | |
e4dc5b7a | 1317 | retry_private: |
8b8f319f | 1318 | double_lock_hb(hb1, hb2); |
1da177e4 | 1319 | |
e2970f2f IM |
1320 | if (likely(cmpval != NULL)) { |
1321 | u32 curval; | |
1da177e4 | 1322 | |
e2970f2f | 1323 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1324 | |
1325 | if (unlikely(ret)) { | |
5eb3dc62 | 1326 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1327 | |
e2970f2f | 1328 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1329 | if (ret) |
1330 | goto out_put_keys; | |
1da177e4 | 1331 | |
b41277dc | 1332 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 1333 | goto retry_private; |
1da177e4 | 1334 | |
ae791a2d TG |
1335 | put_futex_key(&key2); |
1336 | put_futex_key(&key1); | |
e4dc5b7a | 1337 | goto retry; |
1da177e4 | 1338 | } |
e2970f2f | 1339 | if (curval != *cmpval) { |
1da177e4 LT |
1340 | ret = -EAGAIN; |
1341 | goto out_unlock; | |
1342 | } | |
1343 | } | |
1344 | ||
52400ba9 | 1345 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1346 | /* |
1347 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1348 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1349 | * bit. We force this here where we are able to easily handle | |
1350 | * faults rather in the requeue loop below. | |
1351 | */ | |
52400ba9 | 1352 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1353 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1354 | |
1355 | /* | |
1356 | * At this point the top_waiter has either taken uaddr2 or is | |
1357 | * waiting on it. If the former, then the pi_state will not | |
1358 | * exist yet, look it up one more time to ensure we have a | |
1359 | * reference to it. | |
1360 | */ | |
1361 | if (ret == 1) { | |
1362 | WARN_ON(pi_state); | |
89061d3d | 1363 | drop_count++; |
52400ba9 DH |
1364 | task_count++; |
1365 | ret = get_futex_value_locked(&curval2, uaddr2); | |
1366 | if (!ret) | |
1367 | ret = lookup_pi_state(curval2, hb2, &key2, | |
1368 | &pi_state); | |
1369 | } | |
1370 | ||
1371 | switch (ret) { | |
1372 | case 0: | |
1373 | break; | |
1374 | case -EFAULT: | |
1375 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1376 | put_futex_key(&key2); |
1377 | put_futex_key(&key1); | |
d0725992 | 1378 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
1379 | if (!ret) |
1380 | goto retry; | |
1381 | goto out; | |
1382 | case -EAGAIN: | |
1383 | /* The owner was exiting, try again. */ | |
1384 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1385 | put_futex_key(&key2); |
1386 | put_futex_key(&key1); | |
52400ba9 DH |
1387 | cond_resched(); |
1388 | goto retry; | |
1389 | default: | |
1390 | goto out_unlock; | |
1391 | } | |
1392 | } | |
1393 | ||
e2970f2f | 1394 | head1 = &hb1->chain; |
ec92d082 | 1395 | plist_for_each_entry_safe(this, next, head1, list) { |
52400ba9 DH |
1396 | if (task_count - nr_wake >= nr_requeue) |
1397 | break; | |
1398 | ||
1399 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1400 | continue; |
52400ba9 | 1401 | |
392741e0 DH |
1402 | /* |
1403 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
1404 | * be paired with each other and no other futex ops. | |
aa10990e DH |
1405 | * |
1406 | * We should never be requeueing a futex_q with a pi_state, | |
1407 | * which is awaiting a futex_unlock_pi(). | |
392741e0 DH |
1408 | */ |
1409 | if ((requeue_pi && !this->rt_waiter) || | |
aa10990e DH |
1410 | (!requeue_pi && this->rt_waiter) || |
1411 | this->pi_state) { | |
392741e0 DH |
1412 | ret = -EINVAL; |
1413 | break; | |
1414 | } | |
52400ba9 DH |
1415 | |
1416 | /* | |
1417 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1418 | * lock, we already woke the top_waiter. If not, it will be | |
1419 | * woken by futex_unlock_pi(). | |
1420 | */ | |
1421 | if (++task_count <= nr_wake && !requeue_pi) { | |
1da177e4 | 1422 | wake_futex(this); |
52400ba9 DH |
1423 | continue; |
1424 | } | |
1da177e4 | 1425 | |
84bc4af5 DH |
1426 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
1427 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
1428 | ret = -EINVAL; | |
1429 | break; | |
1430 | } | |
1431 | ||
52400ba9 DH |
1432 | /* |
1433 | * Requeue nr_requeue waiters and possibly one more in the case | |
1434 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1435 | */ | |
1436 | if (requeue_pi) { | |
1437 | /* Prepare the waiter to take the rt_mutex. */ | |
1438 | atomic_inc(&pi_state->refcount); | |
1439 | this->pi_state = pi_state; | |
1440 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1441 | this->rt_waiter, | |
1442 | this->task, 1); | |
1443 | if (ret == 1) { | |
1444 | /* We got the lock. */ | |
beda2c7e | 1445 | requeue_pi_wake_futex(this, &key2, hb2); |
89061d3d | 1446 | drop_count++; |
52400ba9 DH |
1447 | continue; |
1448 | } else if (ret) { | |
1449 | /* -EDEADLK */ | |
1450 | this->pi_state = NULL; | |
1451 | free_pi_state(pi_state); | |
1452 | goto out_unlock; | |
1453 | } | |
1da177e4 | 1454 | } |
52400ba9 DH |
1455 | requeue_futex(this, hb1, hb2, &key2); |
1456 | drop_count++; | |
1da177e4 LT |
1457 | } |
1458 | ||
1459 | out_unlock: | |
5eb3dc62 | 1460 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1461 | |
cd84a42f DH |
1462 | /* |
1463 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1464 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1465 | * one at key2 and updated their key pointer. We no longer need to | |
1466 | * hold the references to key1. | |
1467 | */ | |
1da177e4 | 1468 | while (--drop_count >= 0) |
9adef58b | 1469 | drop_futex_key_refs(&key1); |
1da177e4 | 1470 | |
42d35d48 | 1471 | out_put_keys: |
ae791a2d | 1472 | put_futex_key(&key2); |
42d35d48 | 1473 | out_put_key1: |
ae791a2d | 1474 | put_futex_key(&key1); |
42d35d48 | 1475 | out: |
52400ba9 DH |
1476 | if (pi_state != NULL) |
1477 | free_pi_state(pi_state); | |
1478 | return ret ? ret : task_count; | |
1da177e4 LT |
1479 | } |
1480 | ||
1481 | /* The key must be already stored in q->key. */ | |
82af7aca | 1482 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 1483 | __acquires(&hb->lock) |
1da177e4 | 1484 | { |
e2970f2f | 1485 | struct futex_hash_bucket *hb; |
1da177e4 | 1486 | |
e2970f2f IM |
1487 | hb = hash_futex(&q->key); |
1488 | q->lock_ptr = &hb->lock; | |
1da177e4 | 1489 | |
e2970f2f IM |
1490 | spin_lock(&hb->lock); |
1491 | return hb; | |
1da177e4 LT |
1492 | } |
1493 | ||
d40d65c8 DH |
1494 | static inline void |
1495 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) | |
15e408cd | 1496 | __releases(&hb->lock) |
d40d65c8 DH |
1497 | { |
1498 | spin_unlock(&hb->lock); | |
d40d65c8 DH |
1499 | } |
1500 | ||
1501 | /** | |
1502 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
1503 | * @q: The futex_q to enqueue | |
1504 | * @hb: The destination hash bucket | |
1505 | * | |
1506 | * The hb->lock must be held by the caller, and is released here. A call to | |
1507 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
1508 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
1509 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
1510 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
1511 | * an example). | |
1512 | */ | |
82af7aca | 1513 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
15e408cd | 1514 | __releases(&hb->lock) |
1da177e4 | 1515 | { |
ec92d082 PP |
1516 | int prio; |
1517 | ||
1518 | /* | |
1519 | * The priority used to register this element is | |
1520 | * - either the real thread-priority for the real-time threads | |
1521 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
1522 | * - or MAX_RT_PRIO for non-RT threads. | |
1523 | * Thus, all RT-threads are woken first in priority order, and | |
1524 | * the others are woken last, in FIFO order. | |
1525 | */ | |
1526 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1527 | ||
1528 | plist_node_init(&q->list, prio); | |
ec92d082 | 1529 | plist_add(&q->list, &hb->chain); |
c87e2837 | 1530 | q->task = current; |
e2970f2f | 1531 | spin_unlock(&hb->lock); |
1da177e4 LT |
1532 | } |
1533 | ||
d40d65c8 DH |
1534 | /** |
1535 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
1536 | * @q: The futex_q to unqueue | |
1537 | * | |
1538 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
1539 | * be paired with exactly one earlier call to queue_me(). | |
1540 | * | |
6c23cbbd RD |
1541 | * Return: |
1542 | * 1 - if the futex_q was still queued (and we removed unqueued it); | |
d40d65c8 | 1543 | * 0 - if the futex_q was already removed by the waking thread |
1da177e4 | 1544 | */ |
1da177e4 LT |
1545 | static int unqueue_me(struct futex_q *q) |
1546 | { | |
1da177e4 | 1547 | spinlock_t *lock_ptr; |
e2970f2f | 1548 | int ret = 0; |
1da177e4 LT |
1549 | |
1550 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 1551 | retry: |
1da177e4 | 1552 | lock_ptr = q->lock_ptr; |
e91467ec | 1553 | barrier(); |
c80544dc | 1554 | if (lock_ptr != NULL) { |
1da177e4 LT |
1555 | spin_lock(lock_ptr); |
1556 | /* | |
1557 | * q->lock_ptr can change between reading it and | |
1558 | * spin_lock(), causing us to take the wrong lock. This | |
1559 | * corrects the race condition. | |
1560 | * | |
1561 | * Reasoning goes like this: if we have the wrong lock, | |
1562 | * q->lock_ptr must have changed (maybe several times) | |
1563 | * between reading it and the spin_lock(). It can | |
1564 | * change again after the spin_lock() but only if it was | |
1565 | * already changed before the spin_lock(). It cannot, | |
1566 | * however, change back to the original value. Therefore | |
1567 | * we can detect whether we acquired the correct lock. | |
1568 | */ | |
1569 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
1570 | spin_unlock(lock_ptr); | |
1571 | goto retry; | |
1572 | } | |
2e12978a | 1573 | __unqueue_futex(q); |
c87e2837 IM |
1574 | |
1575 | BUG_ON(q->pi_state); | |
1576 | ||
1da177e4 LT |
1577 | spin_unlock(lock_ptr); |
1578 | ret = 1; | |
1579 | } | |
1580 | ||
9adef58b | 1581 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1582 | return ret; |
1583 | } | |
1584 | ||
c87e2837 IM |
1585 | /* |
1586 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
1587 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
1588 | * and dropped here. | |
c87e2837 | 1589 | */ |
d0aa7a70 | 1590 | static void unqueue_me_pi(struct futex_q *q) |
15e408cd | 1591 | __releases(q->lock_ptr) |
c87e2837 | 1592 | { |
2e12978a | 1593 | __unqueue_futex(q); |
c87e2837 IM |
1594 | |
1595 | BUG_ON(!q->pi_state); | |
1596 | free_pi_state(q->pi_state); | |
1597 | q->pi_state = NULL; | |
1598 | ||
d0aa7a70 | 1599 | spin_unlock(q->lock_ptr); |
c87e2837 IM |
1600 | } |
1601 | ||
d0aa7a70 | 1602 | /* |
cdf71a10 | 1603 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 1604 | * |
778e9a9c AK |
1605 | * Must be called with hash bucket lock held and mm->sem held for non |
1606 | * private futexes. | |
d0aa7a70 | 1607 | */ |
778e9a9c | 1608 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
ae791a2d | 1609 | struct task_struct *newowner) |
d0aa7a70 | 1610 | { |
cdf71a10 | 1611 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 1612 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 1613 | struct task_struct *oldowner = pi_state->owner; |
7cfdaf38 | 1614 | u32 uval, uninitialized_var(curval), newval; |
e4dc5b7a | 1615 | int ret; |
d0aa7a70 PP |
1616 | |
1617 | /* Owner died? */ | |
1b7558e4 TG |
1618 | if (!pi_state->owner) |
1619 | newtid |= FUTEX_OWNER_DIED; | |
1620 | ||
1621 | /* | |
1622 | * We are here either because we stole the rtmutex from the | |
8161239a LJ |
1623 | * previous highest priority waiter or we are the highest priority |
1624 | * waiter but failed to get the rtmutex the first time. | |
1625 | * We have to replace the newowner TID in the user space variable. | |
1626 | * This must be atomic as we have to preserve the owner died bit here. | |
1b7558e4 | 1627 | * |
b2d0994b DH |
1628 | * Note: We write the user space value _before_ changing the pi_state |
1629 | * because we can fault here. Imagine swapped out pages or a fork | |
1630 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
1631 | * |
1632 | * Modifying pi_state _before_ the user space value would | |
1633 | * leave the pi_state in an inconsistent state when we fault | |
1634 | * here, because we need to drop the hash bucket lock to | |
1635 | * handle the fault. This might be observed in the PID check | |
1636 | * in lookup_pi_state. | |
1637 | */ | |
1638 | retry: | |
1639 | if (get_futex_value_locked(&uval, uaddr)) | |
1640 | goto handle_fault; | |
1641 | ||
1642 | while (1) { | |
1643 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
1644 | ||
37a9d912 | 1645 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1b7558e4 TG |
1646 | goto handle_fault; |
1647 | if (curval == uval) | |
1648 | break; | |
1649 | uval = curval; | |
1650 | } | |
1651 | ||
1652 | /* | |
1653 | * We fixed up user space. Now we need to fix the pi_state | |
1654 | * itself. | |
1655 | */ | |
d0aa7a70 | 1656 | if (pi_state->owner != NULL) { |
1d615482 | 1657 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
d0aa7a70 PP |
1658 | WARN_ON(list_empty(&pi_state->list)); |
1659 | list_del_init(&pi_state->list); | |
1d615482 | 1660 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
1b7558e4 | 1661 | } |
d0aa7a70 | 1662 | |
cdf71a10 | 1663 | pi_state->owner = newowner; |
d0aa7a70 | 1664 | |
1d615482 | 1665 | raw_spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 1666 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 | 1667 | list_add(&pi_state->list, &newowner->pi_state_list); |
1d615482 | 1668 | raw_spin_unlock_irq(&newowner->pi_lock); |
1b7558e4 | 1669 | return 0; |
d0aa7a70 | 1670 | |
d0aa7a70 | 1671 | /* |
1b7558e4 | 1672 | * To handle the page fault we need to drop the hash bucket |
8161239a LJ |
1673 | * lock here. That gives the other task (either the highest priority |
1674 | * waiter itself or the task which stole the rtmutex) the | |
1b7558e4 TG |
1675 | * chance to try the fixup of the pi_state. So once we are |
1676 | * back from handling the fault we need to check the pi_state | |
1677 | * after reacquiring the hash bucket lock and before trying to | |
1678 | * do another fixup. When the fixup has been done already we | |
1679 | * simply return. | |
d0aa7a70 | 1680 | */ |
1b7558e4 TG |
1681 | handle_fault: |
1682 | spin_unlock(q->lock_ptr); | |
778e9a9c | 1683 | |
d0725992 | 1684 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 1685 | |
1b7558e4 | 1686 | spin_lock(q->lock_ptr); |
778e9a9c | 1687 | |
1b7558e4 TG |
1688 | /* |
1689 | * Check if someone else fixed it for us: | |
1690 | */ | |
1691 | if (pi_state->owner != oldowner) | |
1692 | return 0; | |
1693 | ||
1694 | if (ret) | |
1695 | return ret; | |
1696 | ||
1697 | goto retry; | |
d0aa7a70 PP |
1698 | } |
1699 | ||
72c1bbf3 | 1700 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 1701 | |
dd973998 DH |
1702 | /** |
1703 | * fixup_owner() - Post lock pi_state and corner case management | |
1704 | * @uaddr: user address of the futex | |
dd973998 DH |
1705 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
1706 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
1707 | * | |
1708 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
1709 | * the pi_state owner as well as handle race conditions that may allow us to | |
1710 | * acquire the lock. Must be called with the hb lock held. | |
1711 | * | |
6c23cbbd RD |
1712 | * Return: |
1713 | * 1 - success, lock taken; | |
1714 | * 0 - success, lock not taken; | |
dd973998 DH |
1715 | * <0 - on error (-EFAULT) |
1716 | */ | |
ae791a2d | 1717 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 DH |
1718 | { |
1719 | struct task_struct *owner; | |
1720 | int ret = 0; | |
1721 | ||
1722 | if (locked) { | |
1723 | /* | |
1724 | * Got the lock. We might not be the anticipated owner if we | |
1725 | * did a lock-steal - fix up the PI-state in that case: | |
1726 | */ | |
1727 | if (q->pi_state->owner != current) | |
ae791a2d | 1728 | ret = fixup_pi_state_owner(uaddr, q, current); |
dd973998 DH |
1729 | goto out; |
1730 | } | |
1731 | ||
1732 | /* | |
1733 | * Catch the rare case, where the lock was released when we were on the | |
1734 | * way back before we locked the hash bucket. | |
1735 | */ | |
1736 | if (q->pi_state->owner == current) { | |
1737 | /* | |
1738 | * Try to get the rt_mutex now. This might fail as some other | |
1739 | * task acquired the rt_mutex after we removed ourself from the | |
1740 | * rt_mutex waiters list. | |
1741 | */ | |
1742 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
1743 | locked = 1; | |
1744 | goto out; | |
1745 | } | |
1746 | ||
1747 | /* | |
1748 | * pi_state is incorrect, some other task did a lock steal and | |
1749 | * we returned due to timeout or signal without taking the | |
8161239a | 1750 | * rt_mutex. Too late. |
dd973998 | 1751 | */ |
8161239a | 1752 | raw_spin_lock(&q->pi_state->pi_mutex.wait_lock); |
dd973998 | 1753 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); |
8161239a LJ |
1754 | if (!owner) |
1755 | owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); | |
1756 | raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock); | |
ae791a2d | 1757 | ret = fixup_pi_state_owner(uaddr, q, owner); |
dd973998 DH |
1758 | goto out; |
1759 | } | |
1760 | ||
1761 | /* | |
1762 | * Paranoia check. If we did not take the lock, then we should not be | |
8161239a | 1763 | * the owner of the rt_mutex. |
dd973998 DH |
1764 | */ |
1765 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
1766 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
1767 | "pi-state %p\n", ret, | |
1768 | q->pi_state->pi_mutex.owner, | |
1769 | q->pi_state->owner); | |
1770 | ||
1771 | out: | |
1772 | return ret ? ret : locked; | |
1773 | } | |
1774 | ||
ca5f9524 DH |
1775 | /** |
1776 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
1777 | * @hb: the futex hash bucket, must be locked by the caller | |
1778 | * @q: the futex_q to queue up on | |
1779 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
1780 | */ |
1781 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 1782 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 1783 | { |
9beba3c5 DH |
1784 | /* |
1785 | * The task state is guaranteed to be set before another task can | |
1786 | * wake it. set_current_state() is implemented using set_mb() and | |
1787 | * queue_me() calls spin_unlock() upon completion, both serializing | |
1788 | * access to the hash list and forcing another memory barrier. | |
1789 | */ | |
f1a11e05 | 1790 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 1791 | queue_me(q, hb); |
ca5f9524 DH |
1792 | |
1793 | /* Arm the timer */ | |
1794 | if (timeout) { | |
1795 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); | |
1796 | if (!hrtimer_active(&timeout->timer)) | |
1797 | timeout->task = NULL; | |
1798 | } | |
1799 | ||
1800 | /* | |
0729e196 DH |
1801 | * If we have been removed from the hash list, then another task |
1802 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
1803 | */ |
1804 | if (likely(!plist_node_empty(&q->list))) { | |
1805 | /* | |
1806 | * If the timer has already expired, current will already be | |
1807 | * flagged for rescheduling. Only call schedule if there | |
1808 | * is no timeout, or if it has yet to expire. | |
1809 | */ | |
1810 | if (!timeout || timeout->task) | |
1811 | schedule(); | |
1812 | } | |
1813 | __set_current_state(TASK_RUNNING); | |
1814 | } | |
1815 | ||
f801073f DH |
1816 | /** |
1817 | * futex_wait_setup() - Prepare to wait on a futex | |
1818 | * @uaddr: the futex userspace address | |
1819 | * @val: the expected value | |
b41277dc | 1820 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
1821 | * @q: the associated futex_q |
1822 | * @hb: storage for hash_bucket pointer to be returned to caller | |
1823 | * | |
1824 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
1825 | * compare it with the expected value. Handle atomic faults internally. | |
1826 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
1827 | * with no q.key reference on failure. | |
1828 | * | |
6c23cbbd RD |
1829 | * Return: |
1830 | * 0 - uaddr contains val and hb has been locked; | |
ca4a04cf | 1831 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked |
f801073f | 1832 | */ |
b41277dc | 1833 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 1834 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 1835 | { |
e2970f2f IM |
1836 | u32 uval; |
1837 | int ret; | |
1da177e4 | 1838 | |
1da177e4 | 1839 | /* |
b2d0994b | 1840 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
1841 | * Order is important: |
1842 | * | |
1843 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
1844 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
1845 | * | |
1846 | * The basic logical guarantee of a futex is that it blocks ONLY | |
1847 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
1848 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
1849 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
1850 | * cond(var) false, which would violate the guarantee. |
1851 | * | |
8fe8f545 ML |
1852 | * On the other hand, we insert q and release the hash-bucket only |
1853 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
1854 | * absorb a wakeup if *uaddr does not match the desired values | |
1855 | * while the syscall executes. | |
1da177e4 | 1856 | */ |
f801073f | 1857 | retry: |
9ea71503 | 1858 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ); |
f801073f | 1859 | if (unlikely(ret != 0)) |
a5a2a0c7 | 1860 | return ret; |
f801073f DH |
1861 | |
1862 | retry_private: | |
1863 | *hb = queue_lock(q); | |
1864 | ||
e2970f2f | 1865 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 1866 | |
f801073f DH |
1867 | if (ret) { |
1868 | queue_unlock(q, *hb); | |
1da177e4 | 1869 | |
e2970f2f | 1870 | ret = get_user(uval, uaddr); |
e4dc5b7a | 1871 | if (ret) |
f801073f | 1872 | goto out; |
1da177e4 | 1873 | |
b41277dc | 1874 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1875 | goto retry_private; |
1876 | ||
ae791a2d | 1877 | put_futex_key(&q->key); |
e4dc5b7a | 1878 | goto retry; |
1da177e4 | 1879 | } |
ca5f9524 | 1880 | |
f801073f DH |
1881 | if (uval != val) { |
1882 | queue_unlock(q, *hb); | |
1883 | ret = -EWOULDBLOCK; | |
2fff78c7 | 1884 | } |
1da177e4 | 1885 | |
f801073f DH |
1886 | out: |
1887 | if (ret) | |
ae791a2d | 1888 | put_futex_key(&q->key); |
f801073f DH |
1889 | return ret; |
1890 | } | |
1891 | ||
b41277dc DH |
1892 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
1893 | ktime_t *abs_time, u32 bitset) | |
f801073f DH |
1894 | { |
1895 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
1896 | struct restart_block *restart; |
1897 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 1898 | struct futex_q q = futex_q_init; |
f801073f DH |
1899 | int ret; |
1900 | ||
1901 | if (!bitset) | |
1902 | return -EINVAL; | |
f801073f DH |
1903 | q.bitset = bitset; |
1904 | ||
1905 | if (abs_time) { | |
1906 | to = &timeout; | |
1907 | ||
b41277dc DH |
1908 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
1909 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
1910 | HRTIMER_MODE_ABS); | |
f801073f DH |
1911 | hrtimer_init_sleeper(to, current); |
1912 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
1913 | current->timer_slack_ns); | |
1914 | } | |
1915 | ||
d58e6576 | 1916 | retry: |
7ada876a DH |
1917 | /* |
1918 | * Prepare to wait on uaddr. On success, holds hb lock and increments | |
1919 | * q.key refs. | |
1920 | */ | |
b41277dc | 1921 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
1922 | if (ret) |
1923 | goto out; | |
1924 | ||
ca5f9524 | 1925 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 1926 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
1927 | |
1928 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 1929 | ret = 0; |
7ada876a | 1930 | /* unqueue_me() drops q.key ref */ |
1da177e4 | 1931 | if (!unqueue_me(&q)) |
7ada876a | 1932 | goto out; |
2fff78c7 | 1933 | ret = -ETIMEDOUT; |
ca5f9524 | 1934 | if (to && !to->task) |
7ada876a | 1935 | goto out; |
72c1bbf3 | 1936 | |
e2970f2f | 1937 | /* |
d58e6576 TG |
1938 | * We expect signal_pending(current), but we might be the |
1939 | * victim of a spurious wakeup as well. | |
e2970f2f | 1940 | */ |
7ada876a | 1941 | if (!signal_pending(current)) |
d58e6576 | 1942 | goto retry; |
d58e6576 | 1943 | |
2fff78c7 | 1944 | ret = -ERESTARTSYS; |
c19384b5 | 1945 | if (!abs_time) |
7ada876a | 1946 | goto out; |
1da177e4 | 1947 | |
2fff78c7 PZ |
1948 | restart = ¤t_thread_info()->restart_block; |
1949 | restart->fn = futex_wait_restart; | |
a3c74c52 | 1950 | restart->futex.uaddr = uaddr; |
2fff78c7 PZ |
1951 | restart->futex.val = val; |
1952 | restart->futex.time = abs_time->tv64; | |
1953 | restart->futex.bitset = bitset; | |
0cd9c649 | 1954 | restart->futex.flags = flags | FLAGS_HAS_TIMEOUT; |
42d35d48 | 1955 | |
2fff78c7 PZ |
1956 | ret = -ERESTART_RESTARTBLOCK; |
1957 | ||
42d35d48 | 1958 | out: |
ca5f9524 DH |
1959 | if (to) { |
1960 | hrtimer_cancel(&to->timer); | |
1961 | destroy_hrtimer_on_stack(&to->timer); | |
1962 | } | |
c87e2837 IM |
1963 | return ret; |
1964 | } | |
1965 | ||
72c1bbf3 NP |
1966 | |
1967 | static long futex_wait_restart(struct restart_block *restart) | |
1968 | { | |
a3c74c52 | 1969 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 1970 | ktime_t t, *tp = NULL; |
72c1bbf3 | 1971 | |
a72188d8 DH |
1972 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
1973 | t.tv64 = restart->futex.time; | |
1974 | tp = &t; | |
1975 | } | |
72c1bbf3 | 1976 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
1977 | |
1978 | return (long)futex_wait(uaddr, restart->futex.flags, | |
1979 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
1980 | } |
1981 | ||
1982 | ||
c87e2837 IM |
1983 | /* |
1984 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
1985 | * and failed. The kernel side here does the whole locking operation: | |
1986 | * if there are waiters then it will block, it does PI, etc. (Due to | |
1987 | * races the kernel might see a 0 value of the futex too.) | |
1988 | */ | |
b41277dc DH |
1989 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect, |
1990 | ktime_t *time, int trylock) | |
c87e2837 | 1991 | { |
c5780e97 | 1992 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 1993 | struct futex_hash_bucket *hb; |
5bdb05f9 | 1994 | struct futex_q q = futex_q_init; |
dd973998 | 1995 | int res, ret; |
c87e2837 IM |
1996 | |
1997 | if (refill_pi_state_cache()) | |
1998 | return -ENOMEM; | |
1999 | ||
c19384b5 | 2000 | if (time) { |
c5780e97 | 2001 | to = &timeout; |
237fc6e7 TG |
2002 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
2003 | HRTIMER_MODE_ABS); | |
c5780e97 | 2004 | hrtimer_init_sleeper(to, current); |
cc584b21 | 2005 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
2006 | } |
2007 | ||
42d35d48 | 2008 | retry: |
9ea71503 | 2009 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE); |
c87e2837 | 2010 | if (unlikely(ret != 0)) |
42d35d48 | 2011 | goto out; |
c87e2837 | 2012 | |
e4dc5b7a | 2013 | retry_private: |
82af7aca | 2014 | hb = queue_lock(&q); |
c87e2837 | 2015 | |
bab5bc9e | 2016 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 2017 | if (unlikely(ret)) { |
778e9a9c | 2018 | switch (ret) { |
1a52084d DH |
2019 | case 1: |
2020 | /* We got the lock. */ | |
2021 | ret = 0; | |
2022 | goto out_unlock_put_key; | |
2023 | case -EFAULT: | |
2024 | goto uaddr_faulted; | |
778e9a9c AK |
2025 | case -EAGAIN: |
2026 | /* | |
2027 | * Task is exiting and we just wait for the | |
2028 | * exit to complete. | |
2029 | */ | |
2030 | queue_unlock(&q, hb); | |
ae791a2d | 2031 | put_futex_key(&q.key); |
778e9a9c AK |
2032 | cond_resched(); |
2033 | goto retry; | |
778e9a9c | 2034 | default: |
42d35d48 | 2035 | goto out_unlock_put_key; |
c87e2837 | 2036 | } |
c87e2837 IM |
2037 | } |
2038 | ||
2039 | /* | |
2040 | * Only actually queue now that the atomic ops are done: | |
2041 | */ | |
82af7aca | 2042 | queue_me(&q, hb); |
c87e2837 | 2043 | |
c87e2837 IM |
2044 | WARN_ON(!q.pi_state); |
2045 | /* | |
2046 | * Block on the PI mutex: | |
2047 | */ | |
2048 | if (!trylock) | |
2049 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); | |
2050 | else { | |
2051 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); | |
2052 | /* Fixup the trylock return value: */ | |
2053 | ret = ret ? 0 : -EWOULDBLOCK; | |
2054 | } | |
2055 | ||
a99e4e41 | 2056 | spin_lock(q.lock_ptr); |
dd973998 DH |
2057 | /* |
2058 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2059 | * haven't already. | |
2060 | */ | |
ae791a2d | 2061 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 DH |
2062 | /* |
2063 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
2064 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
2065 | */ | |
2066 | if (res) | |
2067 | ret = (res < 0) ? res : 0; | |
c87e2837 | 2068 | |
e8f6386c | 2069 | /* |
dd973998 DH |
2070 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
2071 | * it and return the fault to userspace. | |
e8f6386c DH |
2072 | */ |
2073 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
2074 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
2075 | ||
778e9a9c AK |
2076 | /* Unqueue and drop the lock */ |
2077 | unqueue_me_pi(&q); | |
c87e2837 | 2078 | |
5ecb01cf | 2079 | goto out_put_key; |
c87e2837 | 2080 | |
42d35d48 | 2081 | out_unlock_put_key: |
c87e2837 IM |
2082 | queue_unlock(&q, hb); |
2083 | ||
42d35d48 | 2084 | out_put_key: |
ae791a2d | 2085 | put_futex_key(&q.key); |
42d35d48 | 2086 | out: |
237fc6e7 TG |
2087 | if (to) |
2088 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 2089 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 2090 | |
42d35d48 | 2091 | uaddr_faulted: |
778e9a9c AK |
2092 | queue_unlock(&q, hb); |
2093 | ||
d0725992 | 2094 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
2095 | if (ret) |
2096 | goto out_put_key; | |
c87e2837 | 2097 | |
b41277dc | 2098 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2099 | goto retry_private; |
2100 | ||
ae791a2d | 2101 | put_futex_key(&q.key); |
e4dc5b7a | 2102 | goto retry; |
c87e2837 IM |
2103 | } |
2104 | ||
c87e2837 IM |
2105 | /* |
2106 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
2107 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
2108 | * and do the rt-mutex unlock. | |
2109 | */ | |
b41277dc | 2110 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 IM |
2111 | { |
2112 | struct futex_hash_bucket *hb; | |
2113 | struct futex_q *this, *next; | |
ec92d082 | 2114 | struct plist_head *head; |
38d47c1b | 2115 | union futex_key key = FUTEX_KEY_INIT; |
c0c9ed15 | 2116 | u32 uval, vpid = task_pid_vnr(current); |
e4dc5b7a | 2117 | int ret; |
c87e2837 IM |
2118 | |
2119 | retry: | |
2120 | if (get_user(uval, uaddr)) | |
2121 | return -EFAULT; | |
2122 | /* | |
2123 | * We release only a lock we actually own: | |
2124 | */ | |
c0c9ed15 | 2125 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 2126 | return -EPERM; |
c87e2837 | 2127 | |
9ea71503 | 2128 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE); |
c87e2837 IM |
2129 | if (unlikely(ret != 0)) |
2130 | goto out; | |
2131 | ||
2132 | hb = hash_futex(&key); | |
2133 | spin_lock(&hb->lock); | |
2134 | ||
c87e2837 IM |
2135 | /* |
2136 | * To avoid races, try to do the TID -> 0 atomic transition | |
2137 | * again. If it succeeds then we can return without waking | |
2138 | * anyone else up: | |
2139 | */ | |
37a9d912 ML |
2140 | if (!(uval & FUTEX_OWNER_DIED) && |
2141 | cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0)) | |
c87e2837 IM |
2142 | goto pi_faulted; |
2143 | /* | |
2144 | * Rare case: we managed to release the lock atomically, | |
2145 | * no need to wake anyone else up: | |
2146 | */ | |
c0c9ed15 | 2147 | if (unlikely(uval == vpid)) |
c87e2837 IM |
2148 | goto out_unlock; |
2149 | ||
2150 | /* | |
2151 | * Ok, other tasks may need to be woken up - check waiters | |
2152 | * and do the wakeup if necessary: | |
2153 | */ | |
2154 | head = &hb->chain; | |
2155 | ||
ec92d082 | 2156 | plist_for_each_entry_safe(this, next, head, list) { |
c87e2837 IM |
2157 | if (!match_futex (&this->key, &key)) |
2158 | continue; | |
2159 | ret = wake_futex_pi(uaddr, uval, this); | |
2160 | /* | |
2161 | * The atomic access to the futex value | |
2162 | * generated a pagefault, so retry the | |
2163 | * user-access and the wakeup: | |
2164 | */ | |
2165 | if (ret == -EFAULT) | |
2166 | goto pi_faulted; | |
2167 | goto out_unlock; | |
2168 | } | |
2169 | /* | |
2170 | * No waiters - kernel unlocks the futex: | |
2171 | */ | |
e3f2ddea IM |
2172 | if (!(uval & FUTEX_OWNER_DIED)) { |
2173 | ret = unlock_futex_pi(uaddr, uval); | |
2174 | if (ret == -EFAULT) | |
2175 | goto pi_faulted; | |
2176 | } | |
c87e2837 IM |
2177 | |
2178 | out_unlock: | |
2179 | spin_unlock(&hb->lock); | |
ae791a2d | 2180 | put_futex_key(&key); |
c87e2837 | 2181 | |
42d35d48 | 2182 | out: |
c87e2837 IM |
2183 | return ret; |
2184 | ||
2185 | pi_faulted: | |
778e9a9c | 2186 | spin_unlock(&hb->lock); |
ae791a2d | 2187 | put_futex_key(&key); |
c87e2837 | 2188 | |
d0725992 | 2189 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2190 | if (!ret) |
c87e2837 IM |
2191 | goto retry; |
2192 | ||
1da177e4 LT |
2193 | return ret; |
2194 | } | |
2195 | ||
52400ba9 DH |
2196 | /** |
2197 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2198 | * @hb: the hash_bucket futex_q was original enqueued on | |
2199 | * @q: the futex_q woken while waiting to be requeued | |
2200 | * @key2: the futex_key of the requeue target futex | |
2201 | * @timeout: the timeout associated with the wait (NULL if none) | |
2202 | * | |
2203 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2204 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2205 | * the wakeup and return the appropriate error code to the caller. Must be | |
2206 | * called with the hb lock held. | |
2207 | * | |
6c23cbbd RD |
2208 | * Return: |
2209 | * 0 = no early wakeup detected; | |
2210 | * <0 = -ETIMEDOUT or -ERESTARTNOINTR | |
52400ba9 DH |
2211 | */ |
2212 | static inline | |
2213 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2214 | struct futex_q *q, union futex_key *key2, | |
2215 | struct hrtimer_sleeper *timeout) | |
2216 | { | |
2217 | int ret = 0; | |
2218 | ||
2219 | /* | |
2220 | * With the hb lock held, we avoid races while we process the wakeup. | |
2221 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2222 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2223 | * It can't be requeued from uaddr2 to something else since we don't | |
2224 | * support a PI aware source futex for requeue. | |
2225 | */ | |
2226 | if (!match_futex(&q->key, key2)) { | |
2227 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2228 | /* | |
2229 | * We were woken prior to requeue by a timeout or a signal. | |
2230 | * Unqueue the futex_q and determine which it was. | |
2231 | */ | |
2e12978a | 2232 | plist_del(&q->list, &hb->chain); |
52400ba9 | 2233 | |
d58e6576 | 2234 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 2235 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2236 | if (timeout && !timeout->task) |
2237 | ret = -ETIMEDOUT; | |
d58e6576 | 2238 | else if (signal_pending(current)) |
1c840c14 | 2239 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
2240 | } |
2241 | return ret; | |
2242 | } | |
2243 | ||
2244 | /** | |
2245 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 2246 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 2247 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
52400ba9 DH |
2248 | * the same type, no requeueing from private to shared, etc. |
2249 | * @val: the expected value of uaddr | |
2250 | * @abs_time: absolute timeout | |
56ec1607 | 2251 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
2252 | * @uaddr2: the pi futex we will take prior to returning to user-space |
2253 | * | |
2254 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
6f7b0a2a DH |
2255 | * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake |
2256 | * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to | |
2257 | * userspace. This ensures the rt_mutex maintains an owner when it has waiters; | |
2258 | * without one, the pi logic would not know which task to boost/deboost, if | |
2259 | * there was a need to. | |
52400ba9 DH |
2260 | * |
2261 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
6c23cbbd | 2262 | * via the following-- |
52400ba9 | 2263 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() |
cc6db4e6 DH |
2264 | * 2) wakeup on uaddr2 after a requeue |
2265 | * 3) signal | |
2266 | * 4) timeout | |
52400ba9 | 2267 | * |
cc6db4e6 | 2268 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
2269 | * |
2270 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2271 | * 5) successful lock | |
2272 | * 6) signal | |
2273 | * 7) timeout | |
2274 | * 8) other lock acquisition failure | |
2275 | * | |
cc6db4e6 | 2276 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
2277 | * |
2278 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2279 | * | |
6c23cbbd RD |
2280 | * Return: |
2281 | * 0 - On success; | |
52400ba9 DH |
2282 | * <0 - On error |
2283 | */ | |
b41277dc | 2284 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 2285 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 2286 | u32 __user *uaddr2) |
52400ba9 DH |
2287 | { |
2288 | struct hrtimer_sleeper timeout, *to = NULL; | |
2289 | struct rt_mutex_waiter rt_waiter; | |
2290 | struct rt_mutex *pi_mutex = NULL; | |
52400ba9 | 2291 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
2292 | union futex_key key2 = FUTEX_KEY_INIT; |
2293 | struct futex_q q = futex_q_init; | |
52400ba9 | 2294 | int res, ret; |
52400ba9 | 2295 | |
6f7b0a2a DH |
2296 | if (uaddr == uaddr2) |
2297 | return -EINVAL; | |
2298 | ||
52400ba9 DH |
2299 | if (!bitset) |
2300 | return -EINVAL; | |
2301 | ||
2302 | if (abs_time) { | |
2303 | to = &timeout; | |
b41277dc DH |
2304 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2305 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2306 | HRTIMER_MODE_ABS); | |
52400ba9 DH |
2307 | hrtimer_init_sleeper(to, current); |
2308 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2309 | current->timer_slack_ns); | |
2310 | } | |
2311 | ||
2312 | /* | |
2313 | * The waiter is allocated on our stack, manipulated by the requeue | |
2314 | * code while we sleep on uaddr. | |
2315 | */ | |
2316 | debug_rt_mutex_init_waiter(&rt_waiter); | |
2317 | rt_waiter.task = NULL; | |
2318 | ||
9ea71503 | 2319 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
52400ba9 DH |
2320 | if (unlikely(ret != 0)) |
2321 | goto out; | |
2322 | ||
84bc4af5 DH |
2323 | q.bitset = bitset; |
2324 | q.rt_waiter = &rt_waiter; | |
2325 | q.requeue_pi_key = &key2; | |
2326 | ||
7ada876a DH |
2327 | /* |
2328 | * Prepare to wait on uaddr. On success, increments q.key (key1) ref | |
2329 | * count. | |
2330 | */ | |
b41277dc | 2331 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 TG |
2332 | if (ret) |
2333 | goto out_key2; | |
52400ba9 DH |
2334 | |
2335 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ | |
f1a11e05 | 2336 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2337 | |
2338 | spin_lock(&hb->lock); | |
2339 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2340 | spin_unlock(&hb->lock); | |
2341 | if (ret) | |
2342 | goto out_put_keys; | |
2343 | ||
2344 | /* | |
2345 | * In order for us to be here, we know our q.key == key2, and since | |
2346 | * we took the hb->lock above, we also know that futex_requeue() has | |
2347 | * completed and we no longer have to concern ourselves with a wakeup | |
7ada876a DH |
2348 | * race with the atomic proxy lock acquisition by the requeue code. The |
2349 | * futex_requeue dropped our key1 reference and incremented our key2 | |
2350 | * reference count. | |
52400ba9 DH |
2351 | */ |
2352 | ||
2353 | /* Check if the requeue code acquired the second futex for us. */ | |
2354 | if (!q.rt_waiter) { | |
2355 | /* | |
2356 | * Got the lock. We might not be the anticipated owner if we | |
2357 | * did a lock-steal - fix up the PI-state in that case. | |
2358 | */ | |
2359 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2360 | spin_lock(q.lock_ptr); | |
ae791a2d | 2361 | ret = fixup_pi_state_owner(uaddr2, &q, current); |
52400ba9 DH |
2362 | spin_unlock(q.lock_ptr); |
2363 | } | |
2364 | } else { | |
2365 | /* | |
2366 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2367 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2368 | * the pi_state. | |
2369 | */ | |
f27071cb | 2370 | WARN_ON(!q.pi_state); |
52400ba9 DH |
2371 | pi_mutex = &q.pi_state->pi_mutex; |
2372 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1); | |
2373 | debug_rt_mutex_free_waiter(&rt_waiter); | |
2374 | ||
2375 | spin_lock(q.lock_ptr); | |
2376 | /* | |
2377 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2378 | * haven't already. | |
2379 | */ | |
ae791a2d | 2380 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 DH |
2381 | /* |
2382 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 2383 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
2384 | */ |
2385 | if (res) | |
2386 | ret = (res < 0) ? res : 0; | |
2387 | ||
2388 | /* Unqueue and drop the lock. */ | |
2389 | unqueue_me_pi(&q); | |
2390 | } | |
2391 | ||
2392 | /* | |
2393 | * If fixup_pi_state_owner() faulted and was unable to handle the | |
2394 | * fault, unlock the rt_mutex and return the fault to userspace. | |
2395 | */ | |
2396 | if (ret == -EFAULT) { | |
b6070a8d | 2397 | if (pi_mutex && rt_mutex_owner(pi_mutex) == current) |
52400ba9 DH |
2398 | rt_mutex_unlock(pi_mutex); |
2399 | } else if (ret == -EINTR) { | |
52400ba9 | 2400 | /* |
cc6db4e6 DH |
2401 | * We've already been requeued, but cannot restart by calling |
2402 | * futex_lock_pi() directly. We could restart this syscall, but | |
2403 | * it would detect that the user space "val" changed and return | |
2404 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
2405 | * -EWOULDBLOCK directly. | |
52400ba9 | 2406 | */ |
2070887f | 2407 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2408 | } |
2409 | ||
2410 | out_put_keys: | |
ae791a2d | 2411 | put_futex_key(&q.key); |
c8b15a70 | 2412 | out_key2: |
ae791a2d | 2413 | put_futex_key(&key2); |
52400ba9 DH |
2414 | |
2415 | out: | |
2416 | if (to) { | |
2417 | hrtimer_cancel(&to->timer); | |
2418 | destroy_hrtimer_on_stack(&to->timer); | |
2419 | } | |
2420 | return ret; | |
2421 | } | |
2422 | ||
0771dfef IM |
2423 | /* |
2424 | * Support for robust futexes: the kernel cleans up held futexes at | |
2425 | * thread exit time. | |
2426 | * | |
2427 | * Implementation: user-space maintains a per-thread list of locks it | |
2428 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2429 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2430 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2431 | * always manipulated with the lock held, so the list is private and |
2432 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2433 | * field, to allow the kernel to clean up if the thread dies after | |
2434 | * acquiring the lock, but just before it could have added itself to | |
2435 | * the list. There can only be one such pending lock. | |
2436 | */ | |
2437 | ||
2438 | /** | |
d96ee56c DH |
2439 | * sys_set_robust_list() - Set the robust-futex list head of a task |
2440 | * @head: pointer to the list-head | |
2441 | * @len: length of the list-head, as userspace expects | |
0771dfef | 2442 | */ |
836f92ad HC |
2443 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2444 | size_t, len) | |
0771dfef | 2445 | { |
a0c1e907 TG |
2446 | if (!futex_cmpxchg_enabled) |
2447 | return -ENOSYS; | |
0771dfef IM |
2448 | /* |
2449 | * The kernel knows only one size for now: | |
2450 | */ | |
2451 | if (unlikely(len != sizeof(*head))) | |
2452 | return -EINVAL; | |
2453 | ||
2454 | current->robust_list = head; | |
2455 | ||
2456 | return 0; | |
2457 | } | |
2458 | ||
2459 | /** | |
d96ee56c DH |
2460 | * sys_get_robust_list() - Get the robust-futex list head of a task |
2461 | * @pid: pid of the process [zero for current task] | |
2462 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
2463 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 2464 | */ |
836f92ad HC |
2465 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
2466 | struct robust_list_head __user * __user *, head_ptr, | |
2467 | size_t __user *, len_ptr) | |
0771dfef | 2468 | { |
ba46df98 | 2469 | struct robust_list_head __user *head; |
0771dfef | 2470 | unsigned long ret; |
bdbb776f | 2471 | struct task_struct *p; |
0771dfef | 2472 | |
a0c1e907 TG |
2473 | if (!futex_cmpxchg_enabled) |
2474 | return -ENOSYS; | |
2475 | ||
bdbb776f KC |
2476 | rcu_read_lock(); |
2477 | ||
2478 | ret = -ESRCH; | |
0771dfef | 2479 | if (!pid) |
bdbb776f | 2480 | p = current; |
0771dfef | 2481 | else { |
228ebcbe | 2482 | p = find_task_by_vpid(pid); |
0771dfef IM |
2483 | if (!p) |
2484 | goto err_unlock; | |
0771dfef IM |
2485 | } |
2486 | ||
bdbb776f KC |
2487 | ret = -EPERM; |
2488 | if (!ptrace_may_access(p, PTRACE_MODE_READ)) | |
2489 | goto err_unlock; | |
2490 | ||
2491 | head = p->robust_list; | |
2492 | rcu_read_unlock(); | |
2493 | ||
0771dfef IM |
2494 | if (put_user(sizeof(*head), len_ptr)) |
2495 | return -EFAULT; | |
2496 | return put_user(head, head_ptr); | |
2497 | ||
2498 | err_unlock: | |
aaa2a97e | 2499 | rcu_read_unlock(); |
0771dfef IM |
2500 | |
2501 | return ret; | |
2502 | } | |
2503 | ||
2504 | /* | |
2505 | * Process a futex-list entry, check whether it's owned by the | |
2506 | * dying task, and do notification if so: | |
2507 | */ | |
e3f2ddea | 2508 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 2509 | { |
7cfdaf38 | 2510 | u32 uval, uninitialized_var(nval), mval; |
0771dfef | 2511 | |
8f17d3a5 IM |
2512 | retry: |
2513 | if (get_user(uval, uaddr)) | |
0771dfef IM |
2514 | return -1; |
2515 | ||
b488893a | 2516 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
2517 | /* |
2518 | * Ok, this dying thread is truly holding a futex | |
2519 | * of interest. Set the OWNER_DIED bit atomically | |
2520 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
2521 | * set, wake up a waiter (if any). (We have to do a | |
2522 | * futex_wake() even if OWNER_DIED is already set - | |
2523 | * to handle the rare but possible case of recursive | |
2524 | * thread-death.) The rest of the cleanup is done in | |
2525 | * userspace. | |
2526 | */ | |
e3f2ddea | 2527 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
6e0aa9f8 TG |
2528 | /* |
2529 | * We are not holding a lock here, but we want to have | |
2530 | * the pagefault_disable/enable() protection because | |
2531 | * we want to handle the fault gracefully. If the | |
2532 | * access fails we try to fault in the futex with R/W | |
2533 | * verification via get_user_pages. get_user() above | |
2534 | * does not guarantee R/W access. If that fails we | |
2535 | * give up and leave the futex locked. | |
2536 | */ | |
2537 | if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) { | |
2538 | if (fault_in_user_writeable(uaddr)) | |
2539 | return -1; | |
2540 | goto retry; | |
2541 | } | |
c87e2837 | 2542 | if (nval != uval) |
8f17d3a5 | 2543 | goto retry; |
0771dfef | 2544 | |
e3f2ddea IM |
2545 | /* |
2546 | * Wake robust non-PI futexes here. The wakeup of | |
2547 | * PI futexes happens in exit_pi_state(): | |
2548 | */ | |
36cf3b5c | 2549 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 2550 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
2551 | } |
2552 | return 0; | |
2553 | } | |
2554 | ||
e3f2ddea IM |
2555 | /* |
2556 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
2557 | */ | |
2558 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 2559 | struct robust_list __user * __user *head, |
1dcc41bb | 2560 | unsigned int *pi) |
e3f2ddea IM |
2561 | { |
2562 | unsigned long uentry; | |
2563 | ||
ba46df98 | 2564 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
2565 | return -EFAULT; |
2566 | ||
ba46df98 | 2567 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
2568 | *pi = uentry & 1; |
2569 | ||
2570 | return 0; | |
2571 | } | |
2572 | ||
0771dfef IM |
2573 | /* |
2574 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
2575 | * and mark any locks found there dead, and notify any waiters. | |
2576 | * | |
2577 | * We silently return on any sign of list-walking problem. | |
2578 | */ | |
2579 | void exit_robust_list(struct task_struct *curr) | |
2580 | { | |
2581 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 2582 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 DH |
2583 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
2584 | unsigned int uninitialized_var(next_pi); | |
0771dfef | 2585 | unsigned long futex_offset; |
9f96cb1e | 2586 | int rc; |
0771dfef | 2587 | |
a0c1e907 TG |
2588 | if (!futex_cmpxchg_enabled) |
2589 | return; | |
2590 | ||
0771dfef IM |
2591 | /* |
2592 | * Fetch the list head (which was registered earlier, via | |
2593 | * sys_set_robust_list()): | |
2594 | */ | |
e3f2ddea | 2595 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
2596 | return; |
2597 | /* | |
2598 | * Fetch the relative futex offset: | |
2599 | */ | |
2600 | if (get_user(futex_offset, &head->futex_offset)) | |
2601 | return; | |
2602 | /* | |
2603 | * Fetch any possibly pending lock-add first, and handle it | |
2604 | * if it exists: | |
2605 | */ | |
e3f2ddea | 2606 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 2607 | return; |
e3f2ddea | 2608 | |
9f96cb1e | 2609 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 2610 | while (entry != &head->list) { |
9f96cb1e MS |
2611 | /* |
2612 | * Fetch the next entry in the list before calling | |
2613 | * handle_futex_death: | |
2614 | */ | |
2615 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
2616 | /* |
2617 | * A pending lock might already be on the list, so | |
c87e2837 | 2618 | * don't process it twice: |
0771dfef IM |
2619 | */ |
2620 | if (entry != pending) | |
ba46df98 | 2621 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 2622 | curr, pi)) |
0771dfef | 2623 | return; |
9f96cb1e | 2624 | if (rc) |
0771dfef | 2625 | return; |
9f96cb1e MS |
2626 | entry = next_entry; |
2627 | pi = next_pi; | |
0771dfef IM |
2628 | /* |
2629 | * Avoid excessively long or circular lists: | |
2630 | */ | |
2631 | if (!--limit) | |
2632 | break; | |
2633 | ||
2634 | cond_resched(); | |
2635 | } | |
9f96cb1e MS |
2636 | |
2637 | if (pending) | |
2638 | handle_futex_death((void __user *)pending + futex_offset, | |
2639 | curr, pip); | |
0771dfef IM |
2640 | } |
2641 | ||
c19384b5 | 2642 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 2643 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 2644 | { |
81b40539 | 2645 | int cmd = op & FUTEX_CMD_MASK; |
b41277dc | 2646 | unsigned int flags = 0; |
34f01cc1 ED |
2647 | |
2648 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 2649 | flags |= FLAGS_SHARED; |
1da177e4 | 2650 | |
b41277dc DH |
2651 | if (op & FUTEX_CLOCK_REALTIME) { |
2652 | flags |= FLAGS_CLOCKRT; | |
2653 | if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI) | |
2654 | return -ENOSYS; | |
2655 | } | |
1da177e4 | 2656 | |
59263b51 TG |
2657 | switch (cmd) { |
2658 | case FUTEX_LOCK_PI: | |
2659 | case FUTEX_UNLOCK_PI: | |
2660 | case FUTEX_TRYLOCK_PI: | |
2661 | case FUTEX_WAIT_REQUEUE_PI: | |
2662 | case FUTEX_CMP_REQUEUE_PI: | |
2663 | if (!futex_cmpxchg_enabled) | |
2664 | return -ENOSYS; | |
2665 | } | |
2666 | ||
34f01cc1 | 2667 | switch (cmd) { |
1da177e4 | 2668 | case FUTEX_WAIT: |
cd689985 TG |
2669 | val3 = FUTEX_BITSET_MATCH_ANY; |
2670 | case FUTEX_WAIT_BITSET: | |
81b40539 | 2671 | return futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 | 2672 | case FUTEX_WAKE: |
cd689985 TG |
2673 | val3 = FUTEX_BITSET_MATCH_ANY; |
2674 | case FUTEX_WAKE_BITSET: | |
81b40539 | 2675 | return futex_wake(uaddr, flags, val, val3); |
1da177e4 | 2676 | case FUTEX_REQUEUE: |
81b40539 | 2677 | return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 | 2678 | case FUTEX_CMP_REQUEUE: |
81b40539 | 2679 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
4732efbe | 2680 | case FUTEX_WAKE_OP: |
81b40539 | 2681 | return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
c87e2837 | 2682 | case FUTEX_LOCK_PI: |
81b40539 | 2683 | return futex_lock_pi(uaddr, flags, val, timeout, 0); |
c87e2837 | 2684 | case FUTEX_UNLOCK_PI: |
81b40539 | 2685 | return futex_unlock_pi(uaddr, flags); |
c87e2837 | 2686 | case FUTEX_TRYLOCK_PI: |
81b40539 | 2687 | return futex_lock_pi(uaddr, flags, 0, timeout, 1); |
52400ba9 DH |
2688 | case FUTEX_WAIT_REQUEUE_PI: |
2689 | val3 = FUTEX_BITSET_MATCH_ANY; | |
81b40539 TG |
2690 | return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
2691 | uaddr2); | |
52400ba9 | 2692 | case FUTEX_CMP_REQUEUE_PI: |
81b40539 | 2693 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
1da177e4 | 2694 | } |
81b40539 | 2695 | return -ENOSYS; |
1da177e4 LT |
2696 | } |
2697 | ||
2698 | ||
17da2bd9 HC |
2699 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
2700 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
2701 | u32, val3) | |
1da177e4 | 2702 | { |
c19384b5 PP |
2703 | struct timespec ts; |
2704 | ktime_t t, *tp = NULL; | |
e2970f2f | 2705 | u32 val2 = 0; |
34f01cc1 | 2706 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 2707 | |
cd689985 | 2708 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
2709 | cmd == FUTEX_WAIT_BITSET || |
2710 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
c19384b5 | 2711 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 2712 | return -EFAULT; |
c19384b5 | 2713 | if (!timespec_valid(&ts)) |
9741ef96 | 2714 | return -EINVAL; |
c19384b5 PP |
2715 | |
2716 | t = timespec_to_ktime(ts); | |
34f01cc1 | 2717 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 2718 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 2719 | tp = &t; |
1da177e4 LT |
2720 | } |
2721 | /* | |
52400ba9 | 2722 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 2723 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 2724 | */ |
f54f0986 | 2725 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 2726 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 2727 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 2728 | |
c19384b5 | 2729 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
2730 | } |
2731 | ||
f6d107fb | 2732 | static int __init futex_init(void) |
1da177e4 | 2733 | { |
a0c1e907 | 2734 | u32 curval; |
3e4ab747 | 2735 | int i; |
95362fa9 | 2736 | |
a0c1e907 TG |
2737 | /* |
2738 | * This will fail and we want it. Some arch implementations do | |
2739 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
2740 | * functionality. We want to know that before we call in any | |
2741 | * of the complex code paths. Also we want to prevent | |
2742 | * registration of robust lists in that case. NULL is | |
2743 | * guaranteed to fault and we get -EFAULT on functional | |
fb62db2b | 2744 | * implementation, the non-functional ones will return |
a0c1e907 TG |
2745 | * -ENOSYS. |
2746 | */ | |
37a9d912 | 2747 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) |
a0c1e907 TG |
2748 | futex_cmpxchg_enabled = 1; |
2749 | ||
3e4ab747 | 2750 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { |
732375c6 | 2751 | plist_head_init(&futex_queues[i].chain); |
3e4ab747 TG |
2752 | spin_lock_init(&futex_queues[i].lock); |
2753 | } | |
2754 | ||
1da177e4 LT |
2755 | return 0; |
2756 | } | |
f6d107fb | 2757 | __initcall(futex_init); |