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> |
88c8004f | 65 | #include <linux/freezer.h> |
a52b89eb | 66 | #include <linux/bootmem.h> |
b488893a | 67 | |
4732efbe | 68 | #include <asm/futex.h> |
1da177e4 | 69 | |
1696a8be | 70 | #include "locking/rtmutex_common.h" |
c87e2837 | 71 | |
99b60ce6 TG |
72 | /* |
73 | * Basic futex operation and ordering guarantees: | |
74 | * | |
75 | * The waiter reads the futex value in user space and calls | |
76 | * futex_wait(). This function computes the hash bucket and acquires | |
77 | * the hash bucket lock. After that it reads the futex user space value | |
b0c29f79 DB |
78 | * again and verifies that the data has not changed. If it has not changed |
79 | * it enqueues itself into the hash bucket, releases the hash bucket lock | |
80 | * and schedules. | |
99b60ce6 TG |
81 | * |
82 | * The waker side modifies the user space value of the futex and calls | |
b0c29f79 DB |
83 | * futex_wake(). This function computes the hash bucket and acquires the |
84 | * hash bucket lock. Then it looks for waiters on that futex in the hash | |
85 | * bucket and wakes them. | |
99b60ce6 | 86 | * |
b0c29f79 DB |
87 | * In futex wake up scenarios where no tasks are blocked on a futex, taking |
88 | * the hb spinlock can be avoided and simply return. In order for this | |
89 | * optimization to work, ordering guarantees must exist so that the waiter | |
90 | * being added to the list is acknowledged when the list is concurrently being | |
91 | * checked by the waker, avoiding scenarios like the following: | |
99b60ce6 TG |
92 | * |
93 | * CPU 0 CPU 1 | |
94 | * val = *futex; | |
95 | * sys_futex(WAIT, futex, val); | |
96 | * futex_wait(futex, val); | |
97 | * uval = *futex; | |
98 | * *futex = newval; | |
99 | * sys_futex(WAKE, futex); | |
100 | * futex_wake(futex); | |
101 | * if (queue_empty()) | |
102 | * return; | |
103 | * if (uval == val) | |
104 | * lock(hash_bucket(futex)); | |
105 | * queue(); | |
106 | * unlock(hash_bucket(futex)); | |
107 | * schedule(); | |
108 | * | |
109 | * This would cause the waiter on CPU 0 to wait forever because it | |
110 | * missed the transition of the user space value from val to newval | |
111 | * and the waker did not find the waiter in the hash bucket queue. | |
99b60ce6 | 112 | * |
b0c29f79 DB |
113 | * The correct serialization ensures that a waiter either observes |
114 | * the changed user space value before blocking or is woken by a | |
115 | * concurrent waker: | |
116 | * | |
117 | * CPU 0 CPU 1 | |
99b60ce6 TG |
118 | * val = *futex; |
119 | * sys_futex(WAIT, futex, val); | |
120 | * futex_wait(futex, val); | |
b0c29f79 DB |
121 | * |
122 | * waiters++; | |
123 | * mb(); (A) <-- paired with -. | |
124 | * | | |
125 | * lock(hash_bucket(futex)); | | |
126 | * | | |
127 | * uval = *futex; | | |
128 | * | *futex = newval; | |
129 | * | sys_futex(WAKE, futex); | |
130 | * | futex_wake(futex); | |
131 | * | | |
132 | * `-------> mb(); (B) | |
99b60ce6 | 133 | * if (uval == val) |
b0c29f79 | 134 | * queue(); |
99b60ce6 | 135 | * unlock(hash_bucket(futex)); |
b0c29f79 DB |
136 | * schedule(); if (waiters) |
137 | * lock(hash_bucket(futex)); | |
138 | * wake_waiters(futex); | |
139 | * unlock(hash_bucket(futex)); | |
140 | * | |
141 | * Where (A) orders the waiters increment and the futex value read -- this | |
142 | * is guaranteed by the head counter in the hb spinlock; and where (B) | |
143 | * orders the write to futex and the waiters read -- this is done by the | |
144 | * barriers in get_futex_key_refs(), through either ihold or atomic_inc, | |
145 | * depending on the futex type. | |
146 | * | |
147 | * This yields the following case (where X:=waiters, Y:=futex): | |
148 | * | |
149 | * X = Y = 0 | |
150 | * | |
151 | * w[X]=1 w[Y]=1 | |
152 | * MB MB | |
153 | * r[Y]=y r[X]=x | |
154 | * | |
155 | * Which guarantees that x==0 && y==0 is impossible; which translates back into | |
156 | * the guarantee that we cannot both miss the futex variable change and the | |
157 | * enqueue. | |
99b60ce6 TG |
158 | */ |
159 | ||
03b8c7b6 | 160 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 161 | int __read_mostly futex_cmpxchg_enabled; |
03b8c7b6 | 162 | #endif |
a0c1e907 | 163 | |
b41277dc DH |
164 | /* |
165 | * Futex flags used to encode options to functions and preserve them across | |
166 | * restarts. | |
167 | */ | |
168 | #define FLAGS_SHARED 0x01 | |
169 | #define FLAGS_CLOCKRT 0x02 | |
170 | #define FLAGS_HAS_TIMEOUT 0x04 | |
171 | ||
c87e2837 IM |
172 | /* |
173 | * Priority Inheritance state: | |
174 | */ | |
175 | struct futex_pi_state { | |
176 | /* | |
177 | * list of 'owned' pi_state instances - these have to be | |
178 | * cleaned up in do_exit() if the task exits prematurely: | |
179 | */ | |
180 | struct list_head list; | |
181 | ||
182 | /* | |
183 | * The PI object: | |
184 | */ | |
185 | struct rt_mutex pi_mutex; | |
186 | ||
187 | struct task_struct *owner; | |
188 | atomic_t refcount; | |
189 | ||
190 | union futex_key key; | |
191 | }; | |
192 | ||
d8d88fbb DH |
193 | /** |
194 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 195 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
196 | * @task: the task waiting on the futex |
197 | * @lock_ptr: the hash bucket lock | |
198 | * @key: the key the futex is hashed on | |
199 | * @pi_state: optional priority inheritance state | |
200 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
201 | * @requeue_pi_key: the requeue_pi target futex key | |
202 | * @bitset: bitset for the optional bitmasked wakeup | |
203 | * | |
204 | * We use this hashed waitqueue, instead of a normal wait_queue_t, so | |
1da177e4 LT |
205 | * we can wake only the relevant ones (hashed queues may be shared). |
206 | * | |
207 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 208 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 209 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
210 | * the second. |
211 | * | |
212 | * PI futexes are typically woken before they are removed from the hash list via | |
213 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
214 | */ |
215 | struct futex_q { | |
ec92d082 | 216 | struct plist_node list; |
1da177e4 | 217 | |
d8d88fbb | 218 | struct task_struct *task; |
1da177e4 | 219 | spinlock_t *lock_ptr; |
1da177e4 | 220 | union futex_key key; |
c87e2837 | 221 | struct futex_pi_state *pi_state; |
52400ba9 | 222 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 223 | union futex_key *requeue_pi_key; |
cd689985 | 224 | u32 bitset; |
1da177e4 LT |
225 | }; |
226 | ||
5bdb05f9 DH |
227 | static const struct futex_q futex_q_init = { |
228 | /* list gets initialized in queue_me()*/ | |
229 | .key = FUTEX_KEY_INIT, | |
230 | .bitset = FUTEX_BITSET_MATCH_ANY | |
231 | }; | |
232 | ||
1da177e4 | 233 | /* |
b2d0994b DH |
234 | * Hash buckets are shared by all the futex_keys that hash to the same |
235 | * location. Each key may have multiple futex_q structures, one for each task | |
236 | * waiting on a futex. | |
1da177e4 LT |
237 | */ |
238 | struct futex_hash_bucket { | |
11d4616b | 239 | atomic_t waiters; |
ec92d082 PP |
240 | spinlock_t lock; |
241 | struct plist_head chain; | |
a52b89eb | 242 | } ____cacheline_aligned_in_smp; |
1da177e4 | 243 | |
a52b89eb DB |
244 | static unsigned long __read_mostly futex_hashsize; |
245 | ||
246 | static struct futex_hash_bucket *futex_queues; | |
1da177e4 | 247 | |
b0c29f79 DB |
248 | static inline void futex_get_mm(union futex_key *key) |
249 | { | |
250 | atomic_inc(&key->private.mm->mm_count); | |
251 | /* | |
252 | * Ensure futex_get_mm() implies a full barrier such that | |
253 | * get_futex_key() implies a full barrier. This is relied upon | |
254 | * as full barrier (B), see the ordering comment above. | |
255 | */ | |
256 | smp_mb__after_atomic_inc(); | |
257 | } | |
258 | ||
11d4616b LT |
259 | /* |
260 | * Reflects a new waiter being added to the waitqueue. | |
261 | */ | |
262 | static inline void hb_waiters_inc(struct futex_hash_bucket *hb) | |
b0c29f79 DB |
263 | { |
264 | #ifdef CONFIG_SMP | |
11d4616b | 265 | atomic_inc(&hb->waiters); |
b0c29f79 | 266 | /* |
11d4616b | 267 | * Full barrier (A), see the ordering comment above. |
b0c29f79 | 268 | */ |
11d4616b LT |
269 | smp_mb__after_atomic_inc(); |
270 | #endif | |
271 | } | |
272 | ||
273 | /* | |
274 | * Reflects a waiter being removed from the waitqueue by wakeup | |
275 | * paths. | |
276 | */ | |
277 | static inline void hb_waiters_dec(struct futex_hash_bucket *hb) | |
278 | { | |
279 | #ifdef CONFIG_SMP | |
280 | atomic_dec(&hb->waiters); | |
281 | #endif | |
282 | } | |
b0c29f79 | 283 | |
11d4616b LT |
284 | static inline int hb_waiters_pending(struct futex_hash_bucket *hb) |
285 | { | |
286 | #ifdef CONFIG_SMP | |
287 | return atomic_read(&hb->waiters); | |
b0c29f79 | 288 | #else |
11d4616b | 289 | return 1; |
b0c29f79 DB |
290 | #endif |
291 | } | |
292 | ||
1da177e4 LT |
293 | /* |
294 | * We hash on the keys returned from get_futex_key (see below). | |
295 | */ | |
296 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
297 | { | |
298 | u32 hash = jhash2((u32*)&key->both.word, | |
299 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
300 | key->both.offset); | |
a52b89eb | 301 | return &futex_queues[hash & (futex_hashsize - 1)]; |
1da177e4 LT |
302 | } |
303 | ||
304 | /* | |
305 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
306 | */ | |
307 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
308 | { | |
2bc87203 DH |
309 | return (key1 && key2 |
310 | && key1->both.word == key2->both.word | |
1da177e4 LT |
311 | && key1->both.ptr == key2->both.ptr |
312 | && key1->both.offset == key2->both.offset); | |
313 | } | |
314 | ||
38d47c1b PZ |
315 | /* |
316 | * Take a reference to the resource addressed by a key. | |
317 | * Can be called while holding spinlocks. | |
318 | * | |
319 | */ | |
320 | static void get_futex_key_refs(union futex_key *key) | |
321 | { | |
322 | if (!key->both.ptr) | |
323 | return; | |
324 | ||
325 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
326 | case FUT_OFF_INODE: | |
b0c29f79 | 327 | ihold(key->shared.inode); /* implies MB (B) */ |
38d47c1b PZ |
328 | break; |
329 | case FUT_OFF_MMSHARED: | |
b0c29f79 | 330 | futex_get_mm(key); /* implies MB (B) */ |
38d47c1b PZ |
331 | break; |
332 | } | |
333 | } | |
334 | ||
335 | /* | |
336 | * Drop a reference to the resource addressed by a key. | |
337 | * The hash bucket spinlock must not be held. | |
338 | */ | |
339 | static void drop_futex_key_refs(union futex_key *key) | |
340 | { | |
90621c40 DH |
341 | if (!key->both.ptr) { |
342 | /* If we're here then we tried to put a key we failed to get */ | |
343 | WARN_ON_ONCE(1); | |
38d47c1b | 344 | return; |
90621c40 | 345 | } |
38d47c1b PZ |
346 | |
347 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
348 | case FUT_OFF_INODE: | |
349 | iput(key->shared.inode); | |
350 | break; | |
351 | case FUT_OFF_MMSHARED: | |
352 | mmdrop(key->private.mm); | |
353 | break; | |
354 | } | |
355 | } | |
356 | ||
34f01cc1 | 357 | /** |
d96ee56c DH |
358 | * get_futex_key() - Get parameters which are the keys for a futex |
359 | * @uaddr: virtual address of the futex | |
360 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED | |
361 | * @key: address where result is stored. | |
9ea71503 SB |
362 | * @rw: mapping needs to be read/write (values: VERIFY_READ, |
363 | * VERIFY_WRITE) | |
34f01cc1 | 364 | * |
6c23cbbd RD |
365 | * Return: a negative error code or 0 |
366 | * | |
34f01cc1 | 367 | * The key words are stored in *key on success. |
1da177e4 | 368 | * |
6131ffaa | 369 | * For shared mappings, it's (page->index, file_inode(vma->vm_file), |
1da177e4 LT |
370 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
371 | * We can usually work out the index without swapping in the page. | |
372 | * | |
b2d0994b | 373 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 374 | */ |
64d1304a | 375 | static int |
9ea71503 | 376 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) |
1da177e4 | 377 | { |
e2970f2f | 378 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 379 | struct mm_struct *mm = current->mm; |
a5b338f2 | 380 | struct page *page, *page_head; |
9ea71503 | 381 | int err, ro = 0; |
1da177e4 LT |
382 | |
383 | /* | |
384 | * The futex address must be "naturally" aligned. | |
385 | */ | |
e2970f2f | 386 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 387 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 388 | return -EINVAL; |
e2970f2f | 389 | address -= key->both.offset; |
1da177e4 | 390 | |
5cdec2d8 LT |
391 | if (unlikely(!access_ok(rw, uaddr, sizeof(u32)))) |
392 | return -EFAULT; | |
393 | ||
34f01cc1 ED |
394 | /* |
395 | * PROCESS_PRIVATE futexes are fast. | |
396 | * As the mm cannot disappear under us and the 'key' only needs | |
397 | * virtual address, we dont even have to find the underlying vma. | |
398 | * Note : We do have to check 'uaddr' is a valid user address, | |
399 | * but access_ok() should be faster than find_vma() | |
400 | */ | |
401 | if (!fshared) { | |
34f01cc1 ED |
402 | key->private.mm = mm; |
403 | key->private.address = address; | |
b0c29f79 | 404 | get_futex_key_refs(key); /* implies MB (B) */ |
34f01cc1 ED |
405 | return 0; |
406 | } | |
1da177e4 | 407 | |
38d47c1b | 408 | again: |
7485d0d3 | 409 | err = get_user_pages_fast(address, 1, 1, &page); |
9ea71503 SB |
410 | /* |
411 | * If write access is not required (eg. FUTEX_WAIT), try | |
412 | * and get read-only access. | |
413 | */ | |
414 | if (err == -EFAULT && rw == VERIFY_READ) { | |
415 | err = get_user_pages_fast(address, 1, 0, &page); | |
416 | ro = 1; | |
417 | } | |
38d47c1b PZ |
418 | if (err < 0) |
419 | return err; | |
9ea71503 SB |
420 | else |
421 | err = 0; | |
38d47c1b | 422 | |
a5b338f2 AA |
423 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
424 | page_head = page; | |
425 | if (unlikely(PageTail(page))) { | |
38d47c1b | 426 | put_page(page); |
a5b338f2 AA |
427 | /* serialize against __split_huge_page_splitting() */ |
428 | local_irq_disable(); | |
f12d5bfc | 429 | if (likely(__get_user_pages_fast(address, 1, !ro, &page) == 1)) { |
a5b338f2 AA |
430 | page_head = compound_head(page); |
431 | /* | |
432 | * page_head is valid pointer but we must pin | |
433 | * it before taking the PG_lock and/or | |
434 | * PG_compound_lock. The moment we re-enable | |
435 | * irqs __split_huge_page_splitting() can | |
436 | * return and the head page can be freed from | |
437 | * under us. We can't take the PG_lock and/or | |
438 | * PG_compound_lock on a page that could be | |
439 | * freed from under us. | |
440 | */ | |
441 | if (page != page_head) { | |
442 | get_page(page_head); | |
443 | put_page(page); | |
444 | } | |
445 | local_irq_enable(); | |
446 | } else { | |
447 | local_irq_enable(); | |
448 | goto again; | |
449 | } | |
450 | } | |
451 | #else | |
452 | page_head = compound_head(page); | |
453 | if (page != page_head) { | |
454 | get_page(page_head); | |
455 | put_page(page); | |
456 | } | |
457 | #endif | |
458 | ||
459 | lock_page(page_head); | |
e6780f72 HD |
460 | |
461 | /* | |
462 | * If page_head->mapping is NULL, then it cannot be a PageAnon | |
463 | * page; but it might be the ZERO_PAGE or in the gate area or | |
464 | * in a special mapping (all cases which we are happy to fail); | |
465 | * or it may have been a good file page when get_user_pages_fast | |
466 | * found it, but truncated or holepunched or subjected to | |
467 | * invalidate_complete_page2 before we got the page lock (also | |
468 | * cases which we are happy to fail). And we hold a reference, | |
469 | * so refcount care in invalidate_complete_page's remove_mapping | |
470 | * prevents drop_caches from setting mapping to NULL beneath us. | |
471 | * | |
472 | * The case we do have to guard against is when memory pressure made | |
473 | * shmem_writepage move it from filecache to swapcache beneath us: | |
474 | * an unlikely race, but we do need to retry for page_head->mapping. | |
475 | */ | |
a5b338f2 | 476 | if (!page_head->mapping) { |
e6780f72 | 477 | int shmem_swizzled = PageSwapCache(page_head); |
a5b338f2 AA |
478 | unlock_page(page_head); |
479 | put_page(page_head); | |
e6780f72 HD |
480 | if (shmem_swizzled) |
481 | goto again; | |
482 | return -EFAULT; | |
38d47c1b | 483 | } |
1da177e4 LT |
484 | |
485 | /* | |
486 | * Private mappings are handled in a simple way. | |
487 | * | |
488 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
489 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 490 | * the object not the particular process. |
1da177e4 | 491 | */ |
a5b338f2 | 492 | if (PageAnon(page_head)) { |
9ea71503 SB |
493 | /* |
494 | * A RO anonymous page will never change and thus doesn't make | |
495 | * sense for futex operations. | |
496 | */ | |
497 | if (ro) { | |
498 | err = -EFAULT; | |
499 | goto out; | |
500 | } | |
501 | ||
38d47c1b | 502 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 503 | key->private.mm = mm; |
e2970f2f | 504 | key->private.address = address; |
38d47c1b PZ |
505 | } else { |
506 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ | |
a5b338f2 | 507 | key->shared.inode = page_head->mapping->host; |
13d60f4b | 508 | key->shared.pgoff = basepage_index(page); |
1da177e4 LT |
509 | } |
510 | ||
b0c29f79 | 511 | get_futex_key_refs(key); /* implies MB (B) */ |
1da177e4 | 512 | |
9ea71503 | 513 | out: |
a5b338f2 AA |
514 | unlock_page(page_head); |
515 | put_page(page_head); | |
9ea71503 | 516 | return err; |
1da177e4 LT |
517 | } |
518 | ||
ae791a2d | 519 | static inline void put_futex_key(union futex_key *key) |
1da177e4 | 520 | { |
38d47c1b | 521 | drop_futex_key_refs(key); |
1da177e4 LT |
522 | } |
523 | ||
d96ee56c DH |
524 | /** |
525 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
526 | * @uaddr: pointer to faulting user space address |
527 | * | |
528 | * Slow path to fixup the fault we just took in the atomic write | |
529 | * access to @uaddr. | |
530 | * | |
fb62db2b | 531 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
532 | * user address. We know that we faulted in the atomic pagefault |
533 | * disabled section so we can as well avoid the #PF overhead by | |
534 | * calling get_user_pages() right away. | |
535 | */ | |
536 | static int fault_in_user_writeable(u32 __user *uaddr) | |
537 | { | |
722d0172 AK |
538 | struct mm_struct *mm = current->mm; |
539 | int ret; | |
540 | ||
541 | down_read(&mm->mmap_sem); | |
2efaca92 BH |
542 | ret = fixup_user_fault(current, mm, (unsigned long)uaddr, |
543 | FAULT_FLAG_WRITE); | |
722d0172 AK |
544 | up_read(&mm->mmap_sem); |
545 | ||
d0725992 TG |
546 | return ret < 0 ? ret : 0; |
547 | } | |
548 | ||
4b1c486b DH |
549 | /** |
550 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
551 | * @hb: the hash bucket the futex_q's reside in |
552 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
553 | * |
554 | * Must be called with the hb lock held. | |
555 | */ | |
556 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
557 | union futex_key *key) | |
558 | { | |
559 | struct futex_q *this; | |
560 | ||
561 | plist_for_each_entry(this, &hb->chain, list) { | |
562 | if (match_futex(&this->key, key)) | |
563 | return this; | |
564 | } | |
565 | return NULL; | |
566 | } | |
567 | ||
37a9d912 ML |
568 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
569 | u32 uval, u32 newval) | |
36cf3b5c | 570 | { |
37a9d912 | 571 | int ret; |
36cf3b5c TG |
572 | |
573 | pagefault_disable(); | |
37a9d912 | 574 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
575 | pagefault_enable(); |
576 | ||
37a9d912 | 577 | return ret; |
36cf3b5c TG |
578 | } |
579 | ||
580 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
581 | { |
582 | int ret; | |
583 | ||
a866374a | 584 | pagefault_disable(); |
e2970f2f | 585 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 586 | pagefault_enable(); |
1da177e4 LT |
587 | |
588 | return ret ? -EFAULT : 0; | |
589 | } | |
590 | ||
c87e2837 IM |
591 | |
592 | /* | |
593 | * PI code: | |
594 | */ | |
595 | static int refill_pi_state_cache(void) | |
596 | { | |
597 | struct futex_pi_state *pi_state; | |
598 | ||
599 | if (likely(current->pi_state_cache)) | |
600 | return 0; | |
601 | ||
4668edc3 | 602 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
603 | |
604 | if (!pi_state) | |
605 | return -ENOMEM; | |
606 | ||
c87e2837 IM |
607 | INIT_LIST_HEAD(&pi_state->list); |
608 | /* pi_mutex gets initialized later */ | |
609 | pi_state->owner = NULL; | |
610 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 611 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
612 | |
613 | current->pi_state_cache = pi_state; | |
614 | ||
615 | return 0; | |
616 | } | |
617 | ||
618 | static struct futex_pi_state * alloc_pi_state(void) | |
619 | { | |
620 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
621 | ||
622 | WARN_ON(!pi_state); | |
623 | current->pi_state_cache = NULL; | |
624 | ||
625 | return pi_state; | |
626 | } | |
627 | ||
628 | static void free_pi_state(struct futex_pi_state *pi_state) | |
629 | { | |
630 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
631 | return; | |
632 | ||
633 | /* | |
634 | * If pi_state->owner is NULL, the owner is most probably dying | |
635 | * and has cleaned up the pi_state already | |
636 | */ | |
637 | if (pi_state->owner) { | |
1d615482 | 638 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
c87e2837 | 639 | list_del_init(&pi_state->list); |
1d615482 | 640 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
c87e2837 IM |
641 | |
642 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
643 | } | |
644 | ||
645 | if (current->pi_state_cache) | |
646 | kfree(pi_state); | |
647 | else { | |
648 | /* | |
649 | * pi_state->list is already empty. | |
650 | * clear pi_state->owner. | |
651 | * refcount is at 0 - put it back to 1. | |
652 | */ | |
653 | pi_state->owner = NULL; | |
654 | atomic_set(&pi_state->refcount, 1); | |
655 | current->pi_state_cache = pi_state; | |
656 | } | |
657 | } | |
658 | ||
659 | /* | |
660 | * Look up the task based on what TID userspace gave us. | |
661 | * We dont trust it. | |
662 | */ | |
663 | static struct task_struct * futex_find_get_task(pid_t pid) | |
664 | { | |
665 | struct task_struct *p; | |
666 | ||
d359b549 | 667 | rcu_read_lock(); |
228ebcbe | 668 | p = find_task_by_vpid(pid); |
7a0ea09a MH |
669 | if (p) |
670 | get_task_struct(p); | |
a06381fe | 671 | |
d359b549 | 672 | rcu_read_unlock(); |
c87e2837 IM |
673 | |
674 | return p; | |
675 | } | |
676 | ||
677 | /* | |
678 | * This task is holding PI mutexes at exit time => bad. | |
679 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
680 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
681 | */ | |
682 | void exit_pi_state_list(struct task_struct *curr) | |
683 | { | |
c87e2837 IM |
684 | struct list_head *next, *head = &curr->pi_state_list; |
685 | struct futex_pi_state *pi_state; | |
627371d7 | 686 | struct futex_hash_bucket *hb; |
38d47c1b | 687 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 688 | |
a0c1e907 TG |
689 | if (!futex_cmpxchg_enabled) |
690 | return; | |
c87e2837 IM |
691 | /* |
692 | * We are a ZOMBIE and nobody can enqueue itself on | |
693 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 694 | * versus waiters unqueueing themselves: |
c87e2837 | 695 | */ |
1d615482 | 696 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 IM |
697 | while (!list_empty(head)) { |
698 | ||
699 | next = head->next; | |
700 | pi_state = list_entry(next, struct futex_pi_state, list); | |
701 | key = pi_state->key; | |
627371d7 | 702 | hb = hash_futex(&key); |
1d615482 | 703 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 704 | |
c87e2837 IM |
705 | spin_lock(&hb->lock); |
706 | ||
1d615482 | 707 | raw_spin_lock_irq(&curr->pi_lock); |
627371d7 IM |
708 | /* |
709 | * We dropped the pi-lock, so re-check whether this | |
710 | * task still owns the PI-state: | |
711 | */ | |
c87e2837 IM |
712 | if (head->next != next) { |
713 | spin_unlock(&hb->lock); | |
714 | continue; | |
715 | } | |
716 | ||
c87e2837 | 717 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
718 | WARN_ON(list_empty(&pi_state->list)); |
719 | list_del_init(&pi_state->list); | |
c87e2837 | 720 | pi_state->owner = NULL; |
1d615482 | 721 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
722 | |
723 | rt_mutex_unlock(&pi_state->pi_mutex); | |
724 | ||
725 | spin_unlock(&hb->lock); | |
726 | ||
1d615482 | 727 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 728 | } |
1d615482 | 729 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
730 | } |
731 | ||
732 | static int | |
d0aa7a70 PP |
733 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
734 | union futex_key *key, struct futex_pi_state **ps) | |
c87e2837 IM |
735 | { |
736 | struct futex_pi_state *pi_state = NULL; | |
737 | struct futex_q *this, *next; | |
c87e2837 | 738 | struct task_struct *p; |
778e9a9c | 739 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 | 740 | |
0d00c7b2 | 741 | plist_for_each_entry_safe(this, next, &hb->chain, list) { |
d0aa7a70 | 742 | if (match_futex(&this->key, key)) { |
c87e2837 IM |
743 | /* |
744 | * Another waiter already exists - bump up | |
745 | * the refcount and return its pi_state: | |
746 | */ | |
747 | pi_state = this->pi_state; | |
06a9ec29 | 748 | /* |
fb62db2b | 749 | * Userspace might have messed up non-PI and PI futexes |
06a9ec29 TG |
750 | */ |
751 | if (unlikely(!pi_state)) | |
752 | return -EINVAL; | |
753 | ||
627371d7 | 754 | WARN_ON(!atomic_read(&pi_state->refcount)); |
59647b6a TG |
755 | |
756 | /* | |
757 | * When pi_state->owner is NULL then the owner died | |
758 | * and another waiter is on the fly. pi_state->owner | |
759 | * is fixed up by the task which acquires | |
760 | * pi_state->rt_mutex. | |
761 | * | |
762 | * We do not check for pid == 0 which can happen when | |
763 | * the owner died and robust_list_exit() cleared the | |
764 | * TID. | |
765 | */ | |
766 | if (pid && pi_state->owner) { | |
767 | /* | |
768 | * Bail out if user space manipulated the | |
769 | * futex value. | |
770 | */ | |
771 | if (pid != task_pid_vnr(pi_state->owner)) | |
772 | return -EINVAL; | |
773 | } | |
627371d7 | 774 | |
c87e2837 | 775 | atomic_inc(&pi_state->refcount); |
d0aa7a70 | 776 | *ps = pi_state; |
c87e2837 IM |
777 | |
778 | return 0; | |
779 | } | |
780 | } | |
781 | ||
782 | /* | |
e3f2ddea | 783 | * We are the first waiter - try to look up the real owner and attach |
778e9a9c | 784 | * the new pi_state to it, but bail out when TID = 0 |
c87e2837 | 785 | */ |
778e9a9c | 786 | if (!pid) |
e3f2ddea | 787 | return -ESRCH; |
c87e2837 | 788 | p = futex_find_get_task(pid); |
7a0ea09a MH |
789 | if (!p) |
790 | return -ESRCH; | |
778e9a9c AK |
791 | |
792 | /* | |
793 | * We need to look at the task state flags to figure out, | |
794 | * whether the task is exiting. To protect against the do_exit | |
795 | * change of the task flags, we do this protected by | |
796 | * p->pi_lock: | |
797 | */ | |
1d615482 | 798 | raw_spin_lock_irq(&p->pi_lock); |
778e9a9c AK |
799 | if (unlikely(p->flags & PF_EXITING)) { |
800 | /* | |
801 | * The task is on the way out. When PF_EXITPIDONE is | |
802 | * set, we know that the task has finished the | |
803 | * cleanup: | |
804 | */ | |
805 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
806 | ||
1d615482 | 807 | raw_spin_unlock_irq(&p->pi_lock); |
778e9a9c AK |
808 | put_task_struct(p); |
809 | return ret; | |
810 | } | |
c87e2837 IM |
811 | |
812 | pi_state = alloc_pi_state(); | |
813 | ||
814 | /* | |
815 | * Initialize the pi_mutex in locked state and make 'p' | |
816 | * the owner of it: | |
817 | */ | |
818 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
819 | ||
820 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 821 | pi_state->key = *key; |
c87e2837 | 822 | |
627371d7 | 823 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
824 | list_add(&pi_state->list, &p->pi_state_list); |
825 | pi_state->owner = p; | |
1d615482 | 826 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
827 | |
828 | put_task_struct(p); | |
829 | ||
d0aa7a70 | 830 | *ps = pi_state; |
c87e2837 IM |
831 | |
832 | return 0; | |
833 | } | |
834 | ||
1a52084d | 835 | /** |
d96ee56c | 836 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
837 | * @uaddr: the pi futex user address |
838 | * @hb: the pi futex hash bucket | |
839 | * @key: the futex key associated with uaddr and hb | |
840 | * @ps: the pi_state pointer where we store the result of the | |
841 | * lookup | |
842 | * @task: the task to perform the atomic lock work for. This will | |
843 | * be "current" except in the case of requeue pi. | |
844 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d | 845 | * |
6c23cbbd RD |
846 | * Return: |
847 | * 0 - ready to wait; | |
848 | * 1 - acquired the lock; | |
1a52084d DH |
849 | * <0 - error |
850 | * | |
851 | * The hb->lock and futex_key refs shall be held by the caller. | |
852 | */ | |
853 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
854 | union futex_key *key, | |
855 | struct futex_pi_state **ps, | |
bab5bc9e | 856 | struct task_struct *task, int set_waiters) |
1a52084d | 857 | { |
59fa6245 | 858 | int lock_taken, ret, force_take = 0; |
c0c9ed15 | 859 | u32 uval, newval, curval, vpid = task_pid_vnr(task); |
1a52084d DH |
860 | |
861 | retry: | |
862 | ret = lock_taken = 0; | |
863 | ||
864 | /* | |
865 | * To avoid races, we attempt to take the lock here again | |
866 | * (by doing a 0 -> TID atomic cmpxchg), while holding all | |
867 | * the locks. It will most likely not succeed. | |
868 | */ | |
c0c9ed15 | 869 | newval = vpid; |
bab5bc9e DH |
870 | if (set_waiters) |
871 | newval |= FUTEX_WAITERS; | |
1a52084d | 872 | |
37a9d912 | 873 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval))) |
1a52084d DH |
874 | return -EFAULT; |
875 | ||
876 | /* | |
877 | * Detect deadlocks. | |
878 | */ | |
c0c9ed15 | 879 | if ((unlikely((curval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
880 | return -EDEADLK; |
881 | ||
882 | /* | |
883 | * Surprise - we got the lock. Just return to userspace: | |
884 | */ | |
885 | if (unlikely(!curval)) | |
886 | return 1; | |
887 | ||
888 | uval = curval; | |
889 | ||
890 | /* | |
891 | * Set the FUTEX_WAITERS flag, so the owner will know it has someone | |
892 | * to wake at the next unlock. | |
893 | */ | |
894 | newval = curval | FUTEX_WAITERS; | |
895 | ||
896 | /* | |
59fa6245 | 897 | * Should we force take the futex? See below. |
1a52084d | 898 | */ |
59fa6245 TG |
899 | if (unlikely(force_take)) { |
900 | /* | |
901 | * Keep the OWNER_DIED and the WAITERS bit and set the | |
902 | * new TID value. | |
903 | */ | |
c0c9ed15 | 904 | newval = (curval & ~FUTEX_TID_MASK) | vpid; |
59fa6245 | 905 | force_take = 0; |
1a52084d DH |
906 | lock_taken = 1; |
907 | } | |
908 | ||
37a9d912 | 909 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))) |
1a52084d DH |
910 | return -EFAULT; |
911 | if (unlikely(curval != uval)) | |
912 | goto retry; | |
913 | ||
914 | /* | |
59fa6245 | 915 | * We took the lock due to forced take over. |
1a52084d DH |
916 | */ |
917 | if (unlikely(lock_taken)) | |
918 | return 1; | |
919 | ||
920 | /* | |
921 | * We dont have the lock. Look up the PI state (or create it if | |
922 | * we are the first waiter): | |
923 | */ | |
924 | ret = lookup_pi_state(uval, hb, key, ps); | |
925 | ||
926 | if (unlikely(ret)) { | |
927 | switch (ret) { | |
928 | case -ESRCH: | |
929 | /* | |
59fa6245 TG |
930 | * We failed to find an owner for this |
931 | * futex. So we have no pi_state to block | |
932 | * on. This can happen in two cases: | |
933 | * | |
934 | * 1) The owner died | |
935 | * 2) A stale FUTEX_WAITERS bit | |
936 | * | |
937 | * Re-read the futex value. | |
1a52084d DH |
938 | */ |
939 | if (get_futex_value_locked(&curval, uaddr)) | |
940 | return -EFAULT; | |
941 | ||
942 | /* | |
59fa6245 TG |
943 | * If the owner died or we have a stale |
944 | * WAITERS bit the owner TID in the user space | |
945 | * futex is 0. | |
1a52084d | 946 | */ |
59fa6245 TG |
947 | if (!(curval & FUTEX_TID_MASK)) { |
948 | force_take = 1; | |
1a52084d DH |
949 | goto retry; |
950 | } | |
951 | default: | |
952 | break; | |
953 | } | |
954 | } | |
955 | ||
956 | return ret; | |
957 | } | |
958 | ||
2e12978a LJ |
959 | /** |
960 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
961 | * @q: The futex_q to unqueue | |
962 | * | |
963 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
964 | */ | |
965 | static void __unqueue_futex(struct futex_q *q) | |
966 | { | |
967 | struct futex_hash_bucket *hb; | |
968 | ||
29096202 SR |
969 | if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr)) |
970 | || WARN_ON(plist_node_empty(&q->list))) | |
2e12978a LJ |
971 | return; |
972 | ||
973 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
974 | plist_del(&q->list, &hb->chain); | |
11d4616b | 975 | hb_waiters_dec(hb); |
2e12978a LJ |
976 | } |
977 | ||
1da177e4 LT |
978 | /* |
979 | * The hash bucket lock must be held when this is called. | |
980 | * Afterwards, the futex_q must not be accessed. | |
981 | */ | |
982 | static void wake_futex(struct futex_q *q) | |
983 | { | |
f1a11e05 TG |
984 | struct task_struct *p = q->task; |
985 | ||
aa10990e DH |
986 | if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n")) |
987 | return; | |
988 | ||
1da177e4 | 989 | /* |
f1a11e05 | 990 | * We set q->lock_ptr = NULL _before_ we wake up the task. If |
fb62db2b RD |
991 | * a non-futex wake up happens on another CPU then the task |
992 | * might exit and p would dereference a non-existing task | |
f1a11e05 TG |
993 | * struct. Prevent this by holding a reference on p across the |
994 | * wake up. | |
1da177e4 | 995 | */ |
f1a11e05 TG |
996 | get_task_struct(p); |
997 | ||
2e12978a | 998 | __unqueue_futex(q); |
1da177e4 | 999 | /* |
f1a11e05 TG |
1000 | * The waiting task can free the futex_q as soon as |
1001 | * q->lock_ptr = NULL is written, without taking any locks. A | |
1002 | * memory barrier is required here to prevent the following | |
1003 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 1004 | */ |
ccdea2f8 | 1005 | smp_wmb(); |
1da177e4 | 1006 | q->lock_ptr = NULL; |
f1a11e05 TG |
1007 | |
1008 | wake_up_state(p, TASK_NORMAL); | |
1009 | put_task_struct(p); | |
1da177e4 LT |
1010 | } |
1011 | ||
c87e2837 IM |
1012 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
1013 | { | |
1014 | struct task_struct *new_owner; | |
1015 | struct futex_pi_state *pi_state = this->pi_state; | |
7cfdaf38 | 1016 | u32 uninitialized_var(curval), newval; |
c87e2837 IM |
1017 | |
1018 | if (!pi_state) | |
1019 | return -EINVAL; | |
1020 | ||
51246bfd TG |
1021 | /* |
1022 | * If current does not own the pi_state then the futex is | |
1023 | * inconsistent and user space fiddled with the futex value. | |
1024 | */ | |
1025 | if (pi_state->owner != current) | |
1026 | return -EINVAL; | |
1027 | ||
d209d74d | 1028 | raw_spin_lock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
1029 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
1030 | ||
1031 | /* | |
f123c98e SR |
1032 | * It is possible that the next waiter (the one that brought |
1033 | * this owner to the kernel) timed out and is no longer | |
1034 | * waiting on the lock. | |
c87e2837 IM |
1035 | */ |
1036 | if (!new_owner) | |
1037 | new_owner = this->task; | |
1038 | ||
1039 | /* | |
1040 | * We pass it to the next owner. (The WAITERS bit is always | |
1041 | * kept enabled while there is PI state around. We must also | |
1042 | * preserve the owner died bit.) | |
1043 | */ | |
e3f2ddea | 1044 | if (!(uval & FUTEX_OWNER_DIED)) { |
778e9a9c AK |
1045 | int ret = 0; |
1046 | ||
b488893a | 1047 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 1048 | |
37a9d912 | 1049 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
778e9a9c | 1050 | ret = -EFAULT; |
cde898fa | 1051 | else if (curval != uval) |
778e9a9c AK |
1052 | ret = -EINVAL; |
1053 | if (ret) { | |
d209d74d | 1054 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
778e9a9c AK |
1055 | return ret; |
1056 | } | |
e3f2ddea | 1057 | } |
c87e2837 | 1058 | |
1d615482 | 1059 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
627371d7 IM |
1060 | WARN_ON(list_empty(&pi_state->list)); |
1061 | list_del_init(&pi_state->list); | |
1d615482 | 1062 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
627371d7 | 1063 | |
1d615482 | 1064 | raw_spin_lock_irq(&new_owner->pi_lock); |
627371d7 | 1065 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
1066 | list_add(&pi_state->list, &new_owner->pi_state_list); |
1067 | pi_state->owner = new_owner; | |
1d615482 | 1068 | raw_spin_unlock_irq(&new_owner->pi_lock); |
627371d7 | 1069 | |
d209d74d | 1070 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
1071 | rt_mutex_unlock(&pi_state->pi_mutex); |
1072 | ||
1073 | return 0; | |
1074 | } | |
1075 | ||
1076 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) | |
1077 | { | |
7cfdaf38 | 1078 | u32 uninitialized_var(oldval); |
c87e2837 IM |
1079 | |
1080 | /* | |
1081 | * There is no waiter, so we unlock the futex. The owner died | |
1082 | * bit has not to be preserved here. We are the owner: | |
1083 | */ | |
37a9d912 ML |
1084 | if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0)) |
1085 | return -EFAULT; | |
c87e2837 IM |
1086 | if (oldval != uval) |
1087 | return -EAGAIN; | |
1088 | ||
1089 | return 0; | |
1090 | } | |
1091 | ||
8b8f319f IM |
1092 | /* |
1093 | * Express the locking dependencies for lockdep: | |
1094 | */ | |
1095 | static inline void | |
1096 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1097 | { | |
1098 | if (hb1 <= hb2) { | |
1099 | spin_lock(&hb1->lock); | |
1100 | if (hb1 < hb2) | |
1101 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
1102 | } else { /* hb1 > hb2 */ | |
1103 | spin_lock(&hb2->lock); | |
1104 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
1105 | } | |
1106 | } | |
1107 | ||
5eb3dc62 DH |
1108 | static inline void |
1109 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1110 | { | |
f061d351 | 1111 | spin_unlock(&hb1->lock); |
88f502fe IM |
1112 | if (hb1 != hb2) |
1113 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
1114 | } |
1115 | ||
1da177e4 | 1116 | /* |
b2d0994b | 1117 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 1118 | */ |
b41277dc DH |
1119 | static int |
1120 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 1121 | { |
e2970f2f | 1122 | struct futex_hash_bucket *hb; |
1da177e4 | 1123 | struct futex_q *this, *next; |
38d47c1b | 1124 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 LT |
1125 | int ret; |
1126 | ||
cd689985 TG |
1127 | if (!bitset) |
1128 | return -EINVAL; | |
1129 | ||
9ea71503 | 1130 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ); |
1da177e4 LT |
1131 | if (unlikely(ret != 0)) |
1132 | goto out; | |
1133 | ||
e2970f2f | 1134 | hb = hash_futex(&key); |
b0c29f79 DB |
1135 | |
1136 | /* Make sure we really have tasks to wakeup */ | |
1137 | if (!hb_waiters_pending(hb)) | |
1138 | goto out_put_key; | |
1139 | ||
e2970f2f | 1140 | spin_lock(&hb->lock); |
1da177e4 | 1141 | |
0d00c7b2 | 1142 | plist_for_each_entry_safe(this, next, &hb->chain, list) { |
1da177e4 | 1143 | if (match_futex (&this->key, &key)) { |
52400ba9 | 1144 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
1145 | ret = -EINVAL; |
1146 | break; | |
1147 | } | |
cd689985 TG |
1148 | |
1149 | /* Check if one of the bits is set in both bitsets */ | |
1150 | if (!(this->bitset & bitset)) | |
1151 | continue; | |
1152 | ||
1da177e4 LT |
1153 | wake_futex(this); |
1154 | if (++ret >= nr_wake) | |
1155 | break; | |
1156 | } | |
1157 | } | |
1158 | ||
e2970f2f | 1159 | spin_unlock(&hb->lock); |
b0c29f79 | 1160 | out_put_key: |
ae791a2d | 1161 | put_futex_key(&key); |
42d35d48 | 1162 | out: |
1da177e4 LT |
1163 | return ret; |
1164 | } | |
1165 | ||
4732efbe JJ |
1166 | /* |
1167 | * Wake up all waiters hashed on the physical page that is mapped | |
1168 | * to this virtual address: | |
1169 | */ | |
e2970f2f | 1170 | static int |
b41277dc | 1171 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 1172 | int nr_wake, int nr_wake2, int op) |
4732efbe | 1173 | { |
38d47c1b | 1174 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 1175 | struct futex_hash_bucket *hb1, *hb2; |
4732efbe | 1176 | struct futex_q *this, *next; |
e4dc5b7a | 1177 | int ret, op_ret; |
4732efbe | 1178 | |
e4dc5b7a | 1179 | retry: |
9ea71503 | 1180 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
4732efbe JJ |
1181 | if (unlikely(ret != 0)) |
1182 | goto out; | |
9ea71503 | 1183 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
4732efbe | 1184 | if (unlikely(ret != 0)) |
42d35d48 | 1185 | goto out_put_key1; |
4732efbe | 1186 | |
e2970f2f IM |
1187 | hb1 = hash_futex(&key1); |
1188 | hb2 = hash_futex(&key2); | |
4732efbe | 1189 | |
e4dc5b7a | 1190 | retry_private: |
eaaea803 | 1191 | double_lock_hb(hb1, hb2); |
e2970f2f | 1192 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1193 | if (unlikely(op_ret < 0)) { |
4732efbe | 1194 | |
5eb3dc62 | 1195 | double_unlock_hb(hb1, hb2); |
4732efbe | 1196 | |
7ee1dd3f | 1197 | #ifndef CONFIG_MMU |
e2970f2f IM |
1198 | /* |
1199 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
1200 | * but we might get them from range checking | |
1201 | */ | |
7ee1dd3f | 1202 | ret = op_ret; |
42d35d48 | 1203 | goto out_put_keys; |
7ee1dd3f DH |
1204 | #endif |
1205 | ||
796f8d9b DG |
1206 | if (unlikely(op_ret != -EFAULT)) { |
1207 | ret = op_ret; | |
42d35d48 | 1208 | goto out_put_keys; |
796f8d9b DG |
1209 | } |
1210 | ||
d0725992 | 1211 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 1212 | if (ret) |
de87fcc1 | 1213 | goto out_put_keys; |
4732efbe | 1214 | |
b41277dc | 1215 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1216 | goto retry_private; |
1217 | ||
ae791a2d TG |
1218 | put_futex_key(&key2); |
1219 | put_futex_key(&key1); | |
e4dc5b7a | 1220 | goto retry; |
4732efbe JJ |
1221 | } |
1222 | ||
0d00c7b2 | 1223 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
4732efbe | 1224 | if (match_futex (&this->key, &key1)) { |
aa10990e DH |
1225 | if (this->pi_state || this->rt_waiter) { |
1226 | ret = -EINVAL; | |
1227 | goto out_unlock; | |
1228 | } | |
4732efbe JJ |
1229 | wake_futex(this); |
1230 | if (++ret >= nr_wake) | |
1231 | break; | |
1232 | } | |
1233 | } | |
1234 | ||
1235 | if (op_ret > 0) { | |
4732efbe | 1236 | op_ret = 0; |
0d00c7b2 | 1237 | plist_for_each_entry_safe(this, next, &hb2->chain, list) { |
4732efbe | 1238 | if (match_futex (&this->key, &key2)) { |
aa10990e DH |
1239 | if (this->pi_state || this->rt_waiter) { |
1240 | ret = -EINVAL; | |
1241 | goto out_unlock; | |
1242 | } | |
4732efbe JJ |
1243 | wake_futex(this); |
1244 | if (++op_ret >= nr_wake2) | |
1245 | break; | |
1246 | } | |
1247 | } | |
1248 | ret += op_ret; | |
1249 | } | |
1250 | ||
aa10990e | 1251 | out_unlock: |
5eb3dc62 | 1252 | double_unlock_hb(hb1, hb2); |
42d35d48 | 1253 | out_put_keys: |
ae791a2d | 1254 | put_futex_key(&key2); |
42d35d48 | 1255 | out_put_key1: |
ae791a2d | 1256 | put_futex_key(&key1); |
42d35d48 | 1257 | out: |
4732efbe JJ |
1258 | return ret; |
1259 | } | |
1260 | ||
9121e478 DH |
1261 | /** |
1262 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1263 | * @q: the futex_q to requeue | |
1264 | * @hb1: the source hash_bucket | |
1265 | * @hb2: the target hash_bucket | |
1266 | * @key2: the new key for the requeued futex_q | |
1267 | */ | |
1268 | static inline | |
1269 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1270 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1271 | { | |
1272 | ||
1273 | /* | |
1274 | * If key1 and key2 hash to the same bucket, no need to | |
1275 | * requeue. | |
1276 | */ | |
1277 | if (likely(&hb1->chain != &hb2->chain)) { | |
1278 | plist_del(&q->list, &hb1->chain); | |
11d4616b | 1279 | hb_waiters_dec(hb1); |
9121e478 | 1280 | plist_add(&q->list, &hb2->chain); |
11d4616b | 1281 | hb_waiters_inc(hb2); |
9121e478 | 1282 | q->lock_ptr = &hb2->lock; |
9121e478 DH |
1283 | } |
1284 | get_futex_key_refs(key2); | |
1285 | q->key = *key2; | |
1286 | } | |
1287 | ||
52400ba9 DH |
1288 | /** |
1289 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1290 | * @q: the futex_q |
1291 | * @key: the key of the requeue target futex | |
1292 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1293 | * |
1294 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1295 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1296 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1297 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1298 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1299 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1300 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1301 | */ |
1302 | static inline | |
beda2c7e DH |
1303 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1304 | struct futex_hash_bucket *hb) | |
52400ba9 | 1305 | { |
52400ba9 DH |
1306 | get_futex_key_refs(key); |
1307 | q->key = *key; | |
1308 | ||
2e12978a | 1309 | __unqueue_futex(q); |
52400ba9 DH |
1310 | |
1311 | WARN_ON(!q->rt_waiter); | |
1312 | q->rt_waiter = NULL; | |
1313 | ||
beda2c7e | 1314 | q->lock_ptr = &hb->lock; |
beda2c7e | 1315 | |
f1a11e05 | 1316 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1317 | } |
1318 | ||
1319 | /** | |
1320 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1321 | * @pifutex: the user address of the to futex |
1322 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1323 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1324 | * @key1: the from futex key | |
1325 | * @key2: the to futex key | |
1326 | * @ps: address to store the pi_state pointer | |
1327 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1328 | * |
1329 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1330 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1331 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1332 | * hb1 and hb2 must be held by the caller. | |
52400ba9 | 1333 | * |
6c23cbbd RD |
1334 | * Return: |
1335 | * 0 - failed to acquire the lock atomically; | |
1336 | * 1 - acquired the lock; | |
52400ba9 DH |
1337 | * <0 - error |
1338 | */ | |
1339 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1340 | struct futex_hash_bucket *hb1, | |
1341 | struct futex_hash_bucket *hb2, | |
1342 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1343 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1344 | { |
bab5bc9e | 1345 | struct futex_q *top_waiter = NULL; |
52400ba9 DH |
1346 | u32 curval; |
1347 | int ret; | |
1348 | ||
1349 | if (get_futex_value_locked(&curval, pifutex)) | |
1350 | return -EFAULT; | |
1351 | ||
bab5bc9e DH |
1352 | /* |
1353 | * Find the top_waiter and determine if there are additional waiters. | |
1354 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1355 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1356 | * as we have means to handle the possible fault. If not, don't set | |
1357 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1358 | * the kernel. | |
1359 | */ | |
52400ba9 DH |
1360 | top_waiter = futex_top_waiter(hb1, key1); |
1361 | ||
1362 | /* There are no waiters, nothing for us to do. */ | |
1363 | if (!top_waiter) | |
1364 | return 0; | |
1365 | ||
84bc4af5 DH |
1366 | /* Ensure we requeue to the expected futex. */ |
1367 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1368 | return -EINVAL; | |
1369 | ||
52400ba9 | 1370 | /* |
bab5bc9e DH |
1371 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1372 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1373 | * in ps in contended cases. | |
52400ba9 | 1374 | */ |
bab5bc9e DH |
1375 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1376 | set_waiters); | |
52400ba9 | 1377 | if (ret == 1) |
beda2c7e | 1378 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
52400ba9 DH |
1379 | |
1380 | return ret; | |
1381 | } | |
1382 | ||
1383 | /** | |
1384 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1385 | * @uaddr1: source futex user address |
b41277dc | 1386 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1387 | * @uaddr2: target futex user address |
1388 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1389 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1390 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1391 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1392 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1393 | * |
1394 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1395 | * uaddr2 atomically on behalf of the top waiter. | |
1396 | * | |
6c23cbbd RD |
1397 | * Return: |
1398 | * >=0 - on success, the number of tasks requeued or woken; | |
52400ba9 | 1399 | * <0 - on error |
1da177e4 | 1400 | */ |
b41277dc DH |
1401 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
1402 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
1403 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 1404 | { |
38d47c1b | 1405 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1406 | int drop_count = 0, task_count = 0, ret; |
1407 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1408 | struct futex_hash_bucket *hb1, *hb2; |
1da177e4 | 1409 | struct futex_q *this, *next; |
52400ba9 DH |
1410 | u32 curval2; |
1411 | ||
1412 | if (requeue_pi) { | |
1413 | /* | |
1414 | * requeue_pi requires a pi_state, try to allocate it now | |
1415 | * without any locks in case it fails. | |
1416 | */ | |
1417 | if (refill_pi_state_cache()) | |
1418 | return -ENOMEM; | |
1419 | /* | |
1420 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1421 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1422 | * returning to userspace, so as to not leave the rt_mutex with | |
1423 | * waiters and no owner. However, second and third wake-ups | |
1424 | * cannot be predicted as they involve race conditions with the | |
1425 | * first wake and a fault while looking up the pi_state. Both | |
1426 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1427 | * use nr_wake=1. | |
1428 | */ | |
1429 | if (nr_wake != 1) | |
1430 | return -EINVAL; | |
1431 | } | |
1da177e4 | 1432 | |
42d35d48 | 1433 | retry: |
52400ba9 DH |
1434 | if (pi_state != NULL) { |
1435 | /* | |
1436 | * We will have to lookup the pi_state again, so free this one | |
1437 | * to keep the accounting correct. | |
1438 | */ | |
1439 | free_pi_state(pi_state); | |
1440 | pi_state = NULL; | |
1441 | } | |
1442 | ||
9ea71503 | 1443 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
1da177e4 LT |
1444 | if (unlikely(ret != 0)) |
1445 | goto out; | |
9ea71503 SB |
1446 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, |
1447 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1448 | if (unlikely(ret != 0)) |
42d35d48 | 1449 | goto out_put_key1; |
1da177e4 | 1450 | |
e2970f2f IM |
1451 | hb1 = hash_futex(&key1); |
1452 | hb2 = hash_futex(&key2); | |
1da177e4 | 1453 | |
e4dc5b7a | 1454 | retry_private: |
8b8f319f | 1455 | double_lock_hb(hb1, hb2); |
1da177e4 | 1456 | |
e2970f2f IM |
1457 | if (likely(cmpval != NULL)) { |
1458 | u32 curval; | |
1da177e4 | 1459 | |
e2970f2f | 1460 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1461 | |
1462 | if (unlikely(ret)) { | |
5eb3dc62 | 1463 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1464 | |
e2970f2f | 1465 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1466 | if (ret) |
1467 | goto out_put_keys; | |
1da177e4 | 1468 | |
b41277dc | 1469 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 1470 | goto retry_private; |
1da177e4 | 1471 | |
ae791a2d TG |
1472 | put_futex_key(&key2); |
1473 | put_futex_key(&key1); | |
e4dc5b7a | 1474 | goto retry; |
1da177e4 | 1475 | } |
e2970f2f | 1476 | if (curval != *cmpval) { |
1da177e4 LT |
1477 | ret = -EAGAIN; |
1478 | goto out_unlock; | |
1479 | } | |
1480 | } | |
1481 | ||
52400ba9 | 1482 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1483 | /* |
1484 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1485 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1486 | * bit. We force this here where we are able to easily handle | |
1487 | * faults rather in the requeue loop below. | |
1488 | */ | |
52400ba9 | 1489 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1490 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1491 | |
1492 | /* | |
1493 | * At this point the top_waiter has either taken uaddr2 or is | |
1494 | * waiting on it. If the former, then the pi_state will not | |
1495 | * exist yet, look it up one more time to ensure we have a | |
1496 | * reference to it. | |
1497 | */ | |
1498 | if (ret == 1) { | |
1499 | WARN_ON(pi_state); | |
89061d3d | 1500 | drop_count++; |
52400ba9 DH |
1501 | task_count++; |
1502 | ret = get_futex_value_locked(&curval2, uaddr2); | |
1503 | if (!ret) | |
1504 | ret = lookup_pi_state(curval2, hb2, &key2, | |
1505 | &pi_state); | |
1506 | } | |
1507 | ||
1508 | switch (ret) { | |
1509 | case 0: | |
1510 | break; | |
1511 | case -EFAULT: | |
1512 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1513 | put_futex_key(&key2); |
1514 | put_futex_key(&key1); | |
d0725992 | 1515 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
1516 | if (!ret) |
1517 | goto retry; | |
1518 | goto out; | |
1519 | case -EAGAIN: | |
1520 | /* The owner was exiting, try again. */ | |
1521 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1522 | put_futex_key(&key2); |
1523 | put_futex_key(&key1); | |
52400ba9 DH |
1524 | cond_resched(); |
1525 | goto retry; | |
1526 | default: | |
1527 | goto out_unlock; | |
1528 | } | |
1529 | } | |
1530 | ||
0d00c7b2 | 1531 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
52400ba9 DH |
1532 | if (task_count - nr_wake >= nr_requeue) |
1533 | break; | |
1534 | ||
1535 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1536 | continue; |
52400ba9 | 1537 | |
392741e0 DH |
1538 | /* |
1539 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
1540 | * be paired with each other and no other futex ops. | |
aa10990e DH |
1541 | * |
1542 | * We should never be requeueing a futex_q with a pi_state, | |
1543 | * which is awaiting a futex_unlock_pi(). | |
392741e0 DH |
1544 | */ |
1545 | if ((requeue_pi && !this->rt_waiter) || | |
aa10990e DH |
1546 | (!requeue_pi && this->rt_waiter) || |
1547 | this->pi_state) { | |
392741e0 DH |
1548 | ret = -EINVAL; |
1549 | break; | |
1550 | } | |
52400ba9 DH |
1551 | |
1552 | /* | |
1553 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1554 | * lock, we already woke the top_waiter. If not, it will be | |
1555 | * woken by futex_unlock_pi(). | |
1556 | */ | |
1557 | if (++task_count <= nr_wake && !requeue_pi) { | |
1da177e4 | 1558 | wake_futex(this); |
52400ba9 DH |
1559 | continue; |
1560 | } | |
1da177e4 | 1561 | |
84bc4af5 DH |
1562 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
1563 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
1564 | ret = -EINVAL; | |
1565 | break; | |
1566 | } | |
1567 | ||
52400ba9 DH |
1568 | /* |
1569 | * Requeue nr_requeue waiters and possibly one more in the case | |
1570 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1571 | */ | |
1572 | if (requeue_pi) { | |
1573 | /* Prepare the waiter to take the rt_mutex. */ | |
1574 | atomic_inc(&pi_state->refcount); | |
1575 | this->pi_state = pi_state; | |
1576 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1577 | this->rt_waiter, | |
1578 | this->task, 1); | |
1579 | if (ret == 1) { | |
1580 | /* We got the lock. */ | |
beda2c7e | 1581 | requeue_pi_wake_futex(this, &key2, hb2); |
89061d3d | 1582 | drop_count++; |
52400ba9 DH |
1583 | continue; |
1584 | } else if (ret) { | |
1585 | /* -EDEADLK */ | |
1586 | this->pi_state = NULL; | |
1587 | free_pi_state(pi_state); | |
1588 | goto out_unlock; | |
1589 | } | |
1da177e4 | 1590 | } |
52400ba9 DH |
1591 | requeue_futex(this, hb1, hb2, &key2); |
1592 | drop_count++; | |
1da177e4 LT |
1593 | } |
1594 | ||
1595 | out_unlock: | |
5eb3dc62 | 1596 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1597 | |
cd84a42f DH |
1598 | /* |
1599 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1600 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1601 | * one at key2 and updated their key pointer. We no longer need to | |
1602 | * hold the references to key1. | |
1603 | */ | |
1da177e4 | 1604 | while (--drop_count >= 0) |
9adef58b | 1605 | drop_futex_key_refs(&key1); |
1da177e4 | 1606 | |
42d35d48 | 1607 | out_put_keys: |
ae791a2d | 1608 | put_futex_key(&key2); |
42d35d48 | 1609 | out_put_key1: |
ae791a2d | 1610 | put_futex_key(&key1); |
42d35d48 | 1611 | out: |
52400ba9 DH |
1612 | if (pi_state != NULL) |
1613 | free_pi_state(pi_state); | |
1614 | return ret ? ret : task_count; | |
1da177e4 LT |
1615 | } |
1616 | ||
1617 | /* The key must be already stored in q->key. */ | |
82af7aca | 1618 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 1619 | __acquires(&hb->lock) |
1da177e4 | 1620 | { |
e2970f2f | 1621 | struct futex_hash_bucket *hb; |
1da177e4 | 1622 | |
e2970f2f | 1623 | hb = hash_futex(&q->key); |
11d4616b LT |
1624 | |
1625 | /* | |
1626 | * Increment the counter before taking the lock so that | |
1627 | * a potential waker won't miss a to-be-slept task that is | |
1628 | * waiting for the spinlock. This is safe as all queue_lock() | |
1629 | * users end up calling queue_me(). Similarly, for housekeeping, | |
1630 | * decrement the counter at queue_unlock() when some error has | |
1631 | * occurred and we don't end up adding the task to the list. | |
1632 | */ | |
1633 | hb_waiters_inc(hb); | |
1634 | ||
e2970f2f | 1635 | q->lock_ptr = &hb->lock; |
1da177e4 | 1636 | |
b0c29f79 | 1637 | spin_lock(&hb->lock); /* implies MB (A) */ |
e2970f2f | 1638 | return hb; |
1da177e4 LT |
1639 | } |
1640 | ||
d40d65c8 | 1641 | static inline void |
0d00c7b2 | 1642 | queue_unlock(struct futex_hash_bucket *hb) |
15e408cd | 1643 | __releases(&hb->lock) |
d40d65c8 DH |
1644 | { |
1645 | spin_unlock(&hb->lock); | |
11d4616b | 1646 | hb_waiters_dec(hb); |
d40d65c8 DH |
1647 | } |
1648 | ||
1649 | /** | |
1650 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
1651 | * @q: The futex_q to enqueue | |
1652 | * @hb: The destination hash bucket | |
1653 | * | |
1654 | * The hb->lock must be held by the caller, and is released here. A call to | |
1655 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
1656 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
1657 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
1658 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
1659 | * an example). | |
1660 | */ | |
82af7aca | 1661 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
15e408cd | 1662 | __releases(&hb->lock) |
1da177e4 | 1663 | { |
ec92d082 PP |
1664 | int prio; |
1665 | ||
1666 | /* | |
1667 | * The priority used to register this element is | |
1668 | * - either the real thread-priority for the real-time threads | |
1669 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
1670 | * - or MAX_RT_PRIO for non-RT threads. | |
1671 | * Thus, all RT-threads are woken first in priority order, and | |
1672 | * the others are woken last, in FIFO order. | |
1673 | */ | |
1674 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1675 | ||
1676 | plist_node_init(&q->list, prio); | |
ec92d082 | 1677 | plist_add(&q->list, &hb->chain); |
c87e2837 | 1678 | q->task = current; |
e2970f2f | 1679 | spin_unlock(&hb->lock); |
1da177e4 LT |
1680 | } |
1681 | ||
d40d65c8 DH |
1682 | /** |
1683 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
1684 | * @q: The futex_q to unqueue | |
1685 | * | |
1686 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
1687 | * be paired with exactly one earlier call to queue_me(). | |
1688 | * | |
6c23cbbd RD |
1689 | * Return: |
1690 | * 1 - if the futex_q was still queued (and we removed unqueued it); | |
d40d65c8 | 1691 | * 0 - if the futex_q was already removed by the waking thread |
1da177e4 | 1692 | */ |
1da177e4 LT |
1693 | static int unqueue_me(struct futex_q *q) |
1694 | { | |
1da177e4 | 1695 | spinlock_t *lock_ptr; |
e2970f2f | 1696 | int ret = 0; |
1da177e4 LT |
1697 | |
1698 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 1699 | retry: |
1da177e4 | 1700 | lock_ptr = q->lock_ptr; |
e91467ec | 1701 | barrier(); |
c80544dc | 1702 | if (lock_ptr != NULL) { |
1da177e4 LT |
1703 | spin_lock(lock_ptr); |
1704 | /* | |
1705 | * q->lock_ptr can change between reading it and | |
1706 | * spin_lock(), causing us to take the wrong lock. This | |
1707 | * corrects the race condition. | |
1708 | * | |
1709 | * Reasoning goes like this: if we have the wrong lock, | |
1710 | * q->lock_ptr must have changed (maybe several times) | |
1711 | * between reading it and the spin_lock(). It can | |
1712 | * change again after the spin_lock() but only if it was | |
1713 | * already changed before the spin_lock(). It cannot, | |
1714 | * however, change back to the original value. Therefore | |
1715 | * we can detect whether we acquired the correct lock. | |
1716 | */ | |
1717 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
1718 | spin_unlock(lock_ptr); | |
1719 | goto retry; | |
1720 | } | |
2e12978a | 1721 | __unqueue_futex(q); |
c87e2837 IM |
1722 | |
1723 | BUG_ON(q->pi_state); | |
1724 | ||
1da177e4 LT |
1725 | spin_unlock(lock_ptr); |
1726 | ret = 1; | |
1727 | } | |
1728 | ||
9adef58b | 1729 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1730 | return ret; |
1731 | } | |
1732 | ||
c87e2837 IM |
1733 | /* |
1734 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
1735 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
1736 | * and dropped here. | |
c87e2837 | 1737 | */ |
d0aa7a70 | 1738 | static void unqueue_me_pi(struct futex_q *q) |
15e408cd | 1739 | __releases(q->lock_ptr) |
c87e2837 | 1740 | { |
2e12978a | 1741 | __unqueue_futex(q); |
c87e2837 IM |
1742 | |
1743 | BUG_ON(!q->pi_state); | |
1744 | free_pi_state(q->pi_state); | |
1745 | q->pi_state = NULL; | |
1746 | ||
d0aa7a70 | 1747 | spin_unlock(q->lock_ptr); |
c87e2837 IM |
1748 | } |
1749 | ||
d0aa7a70 | 1750 | /* |
cdf71a10 | 1751 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 1752 | * |
778e9a9c AK |
1753 | * Must be called with hash bucket lock held and mm->sem held for non |
1754 | * private futexes. | |
d0aa7a70 | 1755 | */ |
778e9a9c | 1756 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
ae791a2d | 1757 | struct task_struct *newowner) |
d0aa7a70 | 1758 | { |
cdf71a10 | 1759 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 1760 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 1761 | struct task_struct *oldowner = pi_state->owner; |
7cfdaf38 | 1762 | u32 uval, uninitialized_var(curval), newval; |
e4dc5b7a | 1763 | int ret; |
d0aa7a70 PP |
1764 | |
1765 | /* Owner died? */ | |
1b7558e4 TG |
1766 | if (!pi_state->owner) |
1767 | newtid |= FUTEX_OWNER_DIED; | |
1768 | ||
1769 | /* | |
1770 | * We are here either because we stole the rtmutex from the | |
8161239a LJ |
1771 | * previous highest priority waiter or we are the highest priority |
1772 | * waiter but failed to get the rtmutex the first time. | |
1773 | * We have to replace the newowner TID in the user space variable. | |
1774 | * This must be atomic as we have to preserve the owner died bit here. | |
1b7558e4 | 1775 | * |
b2d0994b DH |
1776 | * Note: We write the user space value _before_ changing the pi_state |
1777 | * because we can fault here. Imagine swapped out pages or a fork | |
1778 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
1779 | * |
1780 | * Modifying pi_state _before_ the user space value would | |
1781 | * leave the pi_state in an inconsistent state when we fault | |
1782 | * here, because we need to drop the hash bucket lock to | |
1783 | * handle the fault. This might be observed in the PID check | |
1784 | * in lookup_pi_state. | |
1785 | */ | |
1786 | retry: | |
1787 | if (get_futex_value_locked(&uval, uaddr)) | |
1788 | goto handle_fault; | |
1789 | ||
1790 | while (1) { | |
1791 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
1792 | ||
37a9d912 | 1793 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1b7558e4 TG |
1794 | goto handle_fault; |
1795 | if (curval == uval) | |
1796 | break; | |
1797 | uval = curval; | |
1798 | } | |
1799 | ||
1800 | /* | |
1801 | * We fixed up user space. Now we need to fix the pi_state | |
1802 | * itself. | |
1803 | */ | |
d0aa7a70 | 1804 | if (pi_state->owner != NULL) { |
1d615482 | 1805 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
d0aa7a70 PP |
1806 | WARN_ON(list_empty(&pi_state->list)); |
1807 | list_del_init(&pi_state->list); | |
1d615482 | 1808 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
1b7558e4 | 1809 | } |
d0aa7a70 | 1810 | |
cdf71a10 | 1811 | pi_state->owner = newowner; |
d0aa7a70 | 1812 | |
1d615482 | 1813 | raw_spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 1814 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 | 1815 | list_add(&pi_state->list, &newowner->pi_state_list); |
1d615482 | 1816 | raw_spin_unlock_irq(&newowner->pi_lock); |
1b7558e4 | 1817 | return 0; |
d0aa7a70 | 1818 | |
d0aa7a70 | 1819 | /* |
1b7558e4 | 1820 | * To handle the page fault we need to drop the hash bucket |
8161239a LJ |
1821 | * lock here. That gives the other task (either the highest priority |
1822 | * waiter itself or the task which stole the rtmutex) the | |
1b7558e4 TG |
1823 | * chance to try the fixup of the pi_state. So once we are |
1824 | * back from handling the fault we need to check the pi_state | |
1825 | * after reacquiring the hash bucket lock and before trying to | |
1826 | * do another fixup. When the fixup has been done already we | |
1827 | * simply return. | |
d0aa7a70 | 1828 | */ |
1b7558e4 TG |
1829 | handle_fault: |
1830 | spin_unlock(q->lock_ptr); | |
778e9a9c | 1831 | |
d0725992 | 1832 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 1833 | |
1b7558e4 | 1834 | spin_lock(q->lock_ptr); |
778e9a9c | 1835 | |
1b7558e4 TG |
1836 | /* |
1837 | * Check if someone else fixed it for us: | |
1838 | */ | |
1839 | if (pi_state->owner != oldowner) | |
1840 | return 0; | |
1841 | ||
1842 | if (ret) | |
1843 | return ret; | |
1844 | ||
1845 | goto retry; | |
d0aa7a70 PP |
1846 | } |
1847 | ||
72c1bbf3 | 1848 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 1849 | |
dd973998 DH |
1850 | /** |
1851 | * fixup_owner() - Post lock pi_state and corner case management | |
1852 | * @uaddr: user address of the futex | |
dd973998 DH |
1853 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
1854 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
1855 | * | |
1856 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
1857 | * the pi_state owner as well as handle race conditions that may allow us to | |
1858 | * acquire the lock. Must be called with the hb lock held. | |
1859 | * | |
6c23cbbd RD |
1860 | * Return: |
1861 | * 1 - success, lock taken; | |
1862 | * 0 - success, lock not taken; | |
dd973998 DH |
1863 | * <0 - on error (-EFAULT) |
1864 | */ | |
ae791a2d | 1865 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 DH |
1866 | { |
1867 | struct task_struct *owner; | |
1868 | int ret = 0; | |
1869 | ||
1870 | if (locked) { | |
1871 | /* | |
1872 | * Got the lock. We might not be the anticipated owner if we | |
1873 | * did a lock-steal - fix up the PI-state in that case: | |
1874 | */ | |
1875 | if (q->pi_state->owner != current) | |
ae791a2d | 1876 | ret = fixup_pi_state_owner(uaddr, q, current); |
dd973998 DH |
1877 | goto out; |
1878 | } | |
1879 | ||
1880 | /* | |
1881 | * Catch the rare case, where the lock was released when we were on the | |
1882 | * way back before we locked the hash bucket. | |
1883 | */ | |
1884 | if (q->pi_state->owner == current) { | |
1885 | /* | |
1886 | * Try to get the rt_mutex now. This might fail as some other | |
1887 | * task acquired the rt_mutex after we removed ourself from the | |
1888 | * rt_mutex waiters list. | |
1889 | */ | |
1890 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
1891 | locked = 1; | |
1892 | goto out; | |
1893 | } | |
1894 | ||
1895 | /* | |
1896 | * pi_state is incorrect, some other task did a lock steal and | |
1897 | * we returned due to timeout or signal without taking the | |
8161239a | 1898 | * rt_mutex. Too late. |
dd973998 | 1899 | */ |
8161239a | 1900 | raw_spin_lock(&q->pi_state->pi_mutex.wait_lock); |
dd973998 | 1901 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); |
8161239a LJ |
1902 | if (!owner) |
1903 | owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); | |
1904 | raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock); | |
ae791a2d | 1905 | ret = fixup_pi_state_owner(uaddr, q, owner); |
dd973998 DH |
1906 | goto out; |
1907 | } | |
1908 | ||
1909 | /* | |
1910 | * Paranoia check. If we did not take the lock, then we should not be | |
8161239a | 1911 | * the owner of the rt_mutex. |
dd973998 DH |
1912 | */ |
1913 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
1914 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
1915 | "pi-state %p\n", ret, | |
1916 | q->pi_state->pi_mutex.owner, | |
1917 | q->pi_state->owner); | |
1918 | ||
1919 | out: | |
1920 | return ret ? ret : locked; | |
1921 | } | |
1922 | ||
ca5f9524 DH |
1923 | /** |
1924 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
1925 | * @hb: the futex hash bucket, must be locked by the caller | |
1926 | * @q: the futex_q to queue up on | |
1927 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
1928 | */ |
1929 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 1930 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 1931 | { |
9beba3c5 DH |
1932 | /* |
1933 | * The task state is guaranteed to be set before another task can | |
1934 | * wake it. set_current_state() is implemented using set_mb() and | |
1935 | * queue_me() calls spin_unlock() upon completion, both serializing | |
1936 | * access to the hash list and forcing another memory barrier. | |
1937 | */ | |
f1a11e05 | 1938 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 1939 | queue_me(q, hb); |
ca5f9524 DH |
1940 | |
1941 | /* Arm the timer */ | |
1942 | if (timeout) { | |
1943 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); | |
1944 | if (!hrtimer_active(&timeout->timer)) | |
1945 | timeout->task = NULL; | |
1946 | } | |
1947 | ||
1948 | /* | |
0729e196 DH |
1949 | * If we have been removed from the hash list, then another task |
1950 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
1951 | */ |
1952 | if (likely(!plist_node_empty(&q->list))) { | |
1953 | /* | |
1954 | * If the timer has already expired, current will already be | |
1955 | * flagged for rescheduling. Only call schedule if there | |
1956 | * is no timeout, or if it has yet to expire. | |
1957 | */ | |
1958 | if (!timeout || timeout->task) | |
88c8004f | 1959 | freezable_schedule(); |
ca5f9524 DH |
1960 | } |
1961 | __set_current_state(TASK_RUNNING); | |
1962 | } | |
1963 | ||
f801073f DH |
1964 | /** |
1965 | * futex_wait_setup() - Prepare to wait on a futex | |
1966 | * @uaddr: the futex userspace address | |
1967 | * @val: the expected value | |
b41277dc | 1968 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
1969 | * @q: the associated futex_q |
1970 | * @hb: storage for hash_bucket pointer to be returned to caller | |
1971 | * | |
1972 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
1973 | * compare it with the expected value. Handle atomic faults internally. | |
1974 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
1975 | * with no q.key reference on failure. | |
1976 | * | |
6c23cbbd RD |
1977 | * Return: |
1978 | * 0 - uaddr contains val and hb has been locked; | |
ca4a04cf | 1979 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked |
f801073f | 1980 | */ |
b41277dc | 1981 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 1982 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 1983 | { |
e2970f2f IM |
1984 | u32 uval; |
1985 | int ret; | |
1da177e4 | 1986 | |
1da177e4 | 1987 | /* |
b2d0994b | 1988 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
1989 | * Order is important: |
1990 | * | |
1991 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
1992 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
1993 | * | |
1994 | * The basic logical guarantee of a futex is that it blocks ONLY | |
1995 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
1996 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
1997 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
1998 | * cond(var) false, which would violate the guarantee. |
1999 | * | |
8fe8f545 ML |
2000 | * On the other hand, we insert q and release the hash-bucket only |
2001 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
2002 | * absorb a wakeup if *uaddr does not match the desired values | |
2003 | * while the syscall executes. | |
1da177e4 | 2004 | */ |
f801073f | 2005 | retry: |
9ea71503 | 2006 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ); |
f801073f | 2007 | if (unlikely(ret != 0)) |
a5a2a0c7 | 2008 | return ret; |
f801073f DH |
2009 | |
2010 | retry_private: | |
2011 | *hb = queue_lock(q); | |
2012 | ||
e2970f2f | 2013 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 2014 | |
f801073f | 2015 | if (ret) { |
0d00c7b2 | 2016 | queue_unlock(*hb); |
1da177e4 | 2017 | |
e2970f2f | 2018 | ret = get_user(uval, uaddr); |
e4dc5b7a | 2019 | if (ret) |
f801073f | 2020 | goto out; |
1da177e4 | 2021 | |
b41277dc | 2022 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2023 | goto retry_private; |
2024 | ||
ae791a2d | 2025 | put_futex_key(&q->key); |
e4dc5b7a | 2026 | goto retry; |
1da177e4 | 2027 | } |
ca5f9524 | 2028 | |
f801073f | 2029 | if (uval != val) { |
0d00c7b2 | 2030 | queue_unlock(*hb); |
f801073f | 2031 | ret = -EWOULDBLOCK; |
2fff78c7 | 2032 | } |
1da177e4 | 2033 | |
f801073f DH |
2034 | out: |
2035 | if (ret) | |
ae791a2d | 2036 | put_futex_key(&q->key); |
f801073f DH |
2037 | return ret; |
2038 | } | |
2039 | ||
b41277dc DH |
2040 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
2041 | ktime_t *abs_time, u32 bitset) | |
f801073f DH |
2042 | { |
2043 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
2044 | struct restart_block *restart; |
2045 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 2046 | struct futex_q q = futex_q_init; |
f801073f DH |
2047 | int ret; |
2048 | ||
2049 | if (!bitset) | |
2050 | return -EINVAL; | |
f801073f DH |
2051 | q.bitset = bitset; |
2052 | ||
2053 | if (abs_time) { | |
2054 | to = &timeout; | |
2055 | ||
b41277dc DH |
2056 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2057 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2058 | HRTIMER_MODE_ABS); | |
f801073f DH |
2059 | hrtimer_init_sleeper(to, current); |
2060 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2061 | current->timer_slack_ns); | |
2062 | } | |
2063 | ||
d58e6576 | 2064 | retry: |
7ada876a DH |
2065 | /* |
2066 | * Prepare to wait on uaddr. On success, holds hb lock and increments | |
2067 | * q.key refs. | |
2068 | */ | |
b41277dc | 2069 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
2070 | if (ret) |
2071 | goto out; | |
2072 | ||
ca5f9524 | 2073 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 2074 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
2075 | |
2076 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 2077 | ret = 0; |
7ada876a | 2078 | /* unqueue_me() drops q.key ref */ |
1da177e4 | 2079 | if (!unqueue_me(&q)) |
7ada876a | 2080 | goto out; |
2fff78c7 | 2081 | ret = -ETIMEDOUT; |
ca5f9524 | 2082 | if (to && !to->task) |
7ada876a | 2083 | goto out; |
72c1bbf3 | 2084 | |
e2970f2f | 2085 | /* |
d58e6576 TG |
2086 | * We expect signal_pending(current), but we might be the |
2087 | * victim of a spurious wakeup as well. | |
e2970f2f | 2088 | */ |
7ada876a | 2089 | if (!signal_pending(current)) |
d58e6576 | 2090 | goto retry; |
d58e6576 | 2091 | |
2fff78c7 | 2092 | ret = -ERESTARTSYS; |
c19384b5 | 2093 | if (!abs_time) |
7ada876a | 2094 | goto out; |
1da177e4 | 2095 | |
2fff78c7 PZ |
2096 | restart = ¤t_thread_info()->restart_block; |
2097 | restart->fn = futex_wait_restart; | |
a3c74c52 | 2098 | restart->futex.uaddr = uaddr; |
2fff78c7 PZ |
2099 | restart->futex.val = val; |
2100 | restart->futex.time = abs_time->tv64; | |
2101 | restart->futex.bitset = bitset; | |
0cd9c649 | 2102 | restart->futex.flags = flags | FLAGS_HAS_TIMEOUT; |
42d35d48 | 2103 | |
2fff78c7 PZ |
2104 | ret = -ERESTART_RESTARTBLOCK; |
2105 | ||
42d35d48 | 2106 | out: |
ca5f9524 DH |
2107 | if (to) { |
2108 | hrtimer_cancel(&to->timer); | |
2109 | destroy_hrtimer_on_stack(&to->timer); | |
2110 | } | |
c87e2837 IM |
2111 | return ret; |
2112 | } | |
2113 | ||
72c1bbf3 NP |
2114 | |
2115 | static long futex_wait_restart(struct restart_block *restart) | |
2116 | { | |
a3c74c52 | 2117 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 2118 | ktime_t t, *tp = NULL; |
72c1bbf3 | 2119 | |
a72188d8 DH |
2120 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
2121 | t.tv64 = restart->futex.time; | |
2122 | tp = &t; | |
2123 | } | |
72c1bbf3 | 2124 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
2125 | |
2126 | return (long)futex_wait(uaddr, restart->futex.flags, | |
2127 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
2128 | } |
2129 | ||
2130 | ||
c87e2837 IM |
2131 | /* |
2132 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
2133 | * and failed. The kernel side here does the whole locking operation: | |
2134 | * if there are waiters then it will block, it does PI, etc. (Due to | |
2135 | * races the kernel might see a 0 value of the futex too.) | |
2136 | */ | |
b41277dc DH |
2137 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect, |
2138 | ktime_t *time, int trylock) | |
c87e2837 | 2139 | { |
c5780e97 | 2140 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 2141 | struct futex_hash_bucket *hb; |
5bdb05f9 | 2142 | struct futex_q q = futex_q_init; |
dd973998 | 2143 | int res, ret; |
c87e2837 IM |
2144 | |
2145 | if (refill_pi_state_cache()) | |
2146 | return -ENOMEM; | |
2147 | ||
c19384b5 | 2148 | if (time) { |
c5780e97 | 2149 | to = &timeout; |
237fc6e7 TG |
2150 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
2151 | HRTIMER_MODE_ABS); | |
c5780e97 | 2152 | hrtimer_init_sleeper(to, current); |
cc584b21 | 2153 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
2154 | } |
2155 | ||
42d35d48 | 2156 | retry: |
9ea71503 | 2157 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE); |
c87e2837 | 2158 | if (unlikely(ret != 0)) |
42d35d48 | 2159 | goto out; |
c87e2837 | 2160 | |
e4dc5b7a | 2161 | retry_private: |
82af7aca | 2162 | hb = queue_lock(&q); |
c87e2837 | 2163 | |
bab5bc9e | 2164 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 2165 | if (unlikely(ret)) { |
778e9a9c | 2166 | switch (ret) { |
1a52084d DH |
2167 | case 1: |
2168 | /* We got the lock. */ | |
2169 | ret = 0; | |
2170 | goto out_unlock_put_key; | |
2171 | case -EFAULT: | |
2172 | goto uaddr_faulted; | |
778e9a9c AK |
2173 | case -EAGAIN: |
2174 | /* | |
2175 | * Task is exiting and we just wait for the | |
2176 | * exit to complete. | |
2177 | */ | |
0d00c7b2 | 2178 | queue_unlock(hb); |
ae791a2d | 2179 | put_futex_key(&q.key); |
778e9a9c AK |
2180 | cond_resched(); |
2181 | goto retry; | |
778e9a9c | 2182 | default: |
42d35d48 | 2183 | goto out_unlock_put_key; |
c87e2837 | 2184 | } |
c87e2837 IM |
2185 | } |
2186 | ||
2187 | /* | |
2188 | * Only actually queue now that the atomic ops are done: | |
2189 | */ | |
82af7aca | 2190 | queue_me(&q, hb); |
c87e2837 | 2191 | |
c87e2837 IM |
2192 | WARN_ON(!q.pi_state); |
2193 | /* | |
2194 | * Block on the PI mutex: | |
2195 | */ | |
2196 | if (!trylock) | |
2197 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); | |
2198 | else { | |
2199 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); | |
2200 | /* Fixup the trylock return value: */ | |
2201 | ret = ret ? 0 : -EWOULDBLOCK; | |
2202 | } | |
2203 | ||
a99e4e41 | 2204 | spin_lock(q.lock_ptr); |
dd973998 DH |
2205 | /* |
2206 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2207 | * haven't already. | |
2208 | */ | |
ae791a2d | 2209 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 DH |
2210 | /* |
2211 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
2212 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
2213 | */ | |
2214 | if (res) | |
2215 | ret = (res < 0) ? res : 0; | |
c87e2837 | 2216 | |
e8f6386c | 2217 | /* |
dd973998 DH |
2218 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
2219 | * it and return the fault to userspace. | |
e8f6386c DH |
2220 | */ |
2221 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
2222 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
2223 | ||
778e9a9c AK |
2224 | /* Unqueue and drop the lock */ |
2225 | unqueue_me_pi(&q); | |
c87e2837 | 2226 | |
5ecb01cf | 2227 | goto out_put_key; |
c87e2837 | 2228 | |
42d35d48 | 2229 | out_unlock_put_key: |
0d00c7b2 | 2230 | queue_unlock(hb); |
c87e2837 | 2231 | |
42d35d48 | 2232 | out_put_key: |
ae791a2d | 2233 | put_futex_key(&q.key); |
42d35d48 | 2234 | out: |
237fc6e7 TG |
2235 | if (to) |
2236 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 2237 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 2238 | |
42d35d48 | 2239 | uaddr_faulted: |
0d00c7b2 | 2240 | queue_unlock(hb); |
778e9a9c | 2241 | |
d0725992 | 2242 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
2243 | if (ret) |
2244 | goto out_put_key; | |
c87e2837 | 2245 | |
b41277dc | 2246 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2247 | goto retry_private; |
2248 | ||
ae791a2d | 2249 | put_futex_key(&q.key); |
e4dc5b7a | 2250 | goto retry; |
c87e2837 IM |
2251 | } |
2252 | ||
c87e2837 IM |
2253 | /* |
2254 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
2255 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
2256 | * and do the rt-mutex unlock. | |
2257 | */ | |
b41277dc | 2258 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 IM |
2259 | { |
2260 | struct futex_hash_bucket *hb; | |
2261 | struct futex_q *this, *next; | |
38d47c1b | 2262 | union futex_key key = FUTEX_KEY_INIT; |
c0c9ed15 | 2263 | u32 uval, vpid = task_pid_vnr(current); |
e4dc5b7a | 2264 | int ret; |
c87e2837 IM |
2265 | |
2266 | retry: | |
2267 | if (get_user(uval, uaddr)) | |
2268 | return -EFAULT; | |
2269 | /* | |
2270 | * We release only a lock we actually own: | |
2271 | */ | |
c0c9ed15 | 2272 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 2273 | return -EPERM; |
c87e2837 | 2274 | |
9ea71503 | 2275 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE); |
c87e2837 IM |
2276 | if (unlikely(ret != 0)) |
2277 | goto out; | |
2278 | ||
2279 | hb = hash_futex(&key); | |
2280 | spin_lock(&hb->lock); | |
2281 | ||
c87e2837 IM |
2282 | /* |
2283 | * To avoid races, try to do the TID -> 0 atomic transition | |
2284 | * again. If it succeeds then we can return without waking | |
2285 | * anyone else up: | |
2286 | */ | |
37a9d912 ML |
2287 | if (!(uval & FUTEX_OWNER_DIED) && |
2288 | cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0)) | |
c87e2837 IM |
2289 | goto pi_faulted; |
2290 | /* | |
2291 | * Rare case: we managed to release the lock atomically, | |
2292 | * no need to wake anyone else up: | |
2293 | */ | |
c0c9ed15 | 2294 | if (unlikely(uval == vpid)) |
c87e2837 IM |
2295 | goto out_unlock; |
2296 | ||
2297 | /* | |
2298 | * Ok, other tasks may need to be woken up - check waiters | |
2299 | * and do the wakeup if necessary: | |
2300 | */ | |
0d00c7b2 | 2301 | plist_for_each_entry_safe(this, next, &hb->chain, list) { |
c87e2837 IM |
2302 | if (!match_futex (&this->key, &key)) |
2303 | continue; | |
2304 | ret = wake_futex_pi(uaddr, uval, this); | |
2305 | /* | |
2306 | * The atomic access to the futex value | |
2307 | * generated a pagefault, so retry the | |
2308 | * user-access and the wakeup: | |
2309 | */ | |
2310 | if (ret == -EFAULT) | |
2311 | goto pi_faulted; | |
2312 | goto out_unlock; | |
2313 | } | |
2314 | /* | |
2315 | * No waiters - kernel unlocks the futex: | |
2316 | */ | |
e3f2ddea IM |
2317 | if (!(uval & FUTEX_OWNER_DIED)) { |
2318 | ret = unlock_futex_pi(uaddr, uval); | |
2319 | if (ret == -EFAULT) | |
2320 | goto pi_faulted; | |
2321 | } | |
c87e2837 IM |
2322 | |
2323 | out_unlock: | |
2324 | spin_unlock(&hb->lock); | |
ae791a2d | 2325 | put_futex_key(&key); |
c87e2837 | 2326 | |
42d35d48 | 2327 | out: |
c87e2837 IM |
2328 | return ret; |
2329 | ||
2330 | pi_faulted: | |
778e9a9c | 2331 | spin_unlock(&hb->lock); |
ae791a2d | 2332 | put_futex_key(&key); |
c87e2837 | 2333 | |
d0725992 | 2334 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2335 | if (!ret) |
c87e2837 IM |
2336 | goto retry; |
2337 | ||
1da177e4 LT |
2338 | return ret; |
2339 | } | |
2340 | ||
52400ba9 DH |
2341 | /** |
2342 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2343 | * @hb: the hash_bucket futex_q was original enqueued on | |
2344 | * @q: the futex_q woken while waiting to be requeued | |
2345 | * @key2: the futex_key of the requeue target futex | |
2346 | * @timeout: the timeout associated with the wait (NULL if none) | |
2347 | * | |
2348 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2349 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2350 | * the wakeup and return the appropriate error code to the caller. Must be | |
2351 | * called with the hb lock held. | |
2352 | * | |
6c23cbbd RD |
2353 | * Return: |
2354 | * 0 = no early wakeup detected; | |
2355 | * <0 = -ETIMEDOUT or -ERESTARTNOINTR | |
52400ba9 DH |
2356 | */ |
2357 | static inline | |
2358 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2359 | struct futex_q *q, union futex_key *key2, | |
2360 | struct hrtimer_sleeper *timeout) | |
2361 | { | |
2362 | int ret = 0; | |
2363 | ||
2364 | /* | |
2365 | * With the hb lock held, we avoid races while we process the wakeup. | |
2366 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2367 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2368 | * It can't be requeued from uaddr2 to something else since we don't | |
2369 | * support a PI aware source futex for requeue. | |
2370 | */ | |
2371 | if (!match_futex(&q->key, key2)) { | |
2372 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2373 | /* | |
2374 | * We were woken prior to requeue by a timeout or a signal. | |
2375 | * Unqueue the futex_q and determine which it was. | |
2376 | */ | |
2e12978a | 2377 | plist_del(&q->list, &hb->chain); |
11d4616b | 2378 | hb_waiters_dec(hb); |
52400ba9 | 2379 | |
d58e6576 | 2380 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 2381 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2382 | if (timeout && !timeout->task) |
2383 | ret = -ETIMEDOUT; | |
d58e6576 | 2384 | else if (signal_pending(current)) |
1c840c14 | 2385 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
2386 | } |
2387 | return ret; | |
2388 | } | |
2389 | ||
2390 | /** | |
2391 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 2392 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 2393 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
52400ba9 DH |
2394 | * the same type, no requeueing from private to shared, etc. |
2395 | * @val: the expected value of uaddr | |
2396 | * @abs_time: absolute timeout | |
56ec1607 | 2397 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
2398 | * @uaddr2: the pi futex we will take prior to returning to user-space |
2399 | * | |
2400 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
6f7b0a2a DH |
2401 | * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake |
2402 | * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to | |
2403 | * userspace. This ensures the rt_mutex maintains an owner when it has waiters; | |
2404 | * without one, the pi logic would not know which task to boost/deboost, if | |
2405 | * there was a need to. | |
52400ba9 DH |
2406 | * |
2407 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
6c23cbbd | 2408 | * via the following-- |
52400ba9 | 2409 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() |
cc6db4e6 DH |
2410 | * 2) wakeup on uaddr2 after a requeue |
2411 | * 3) signal | |
2412 | * 4) timeout | |
52400ba9 | 2413 | * |
cc6db4e6 | 2414 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
2415 | * |
2416 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2417 | * 5) successful lock | |
2418 | * 6) signal | |
2419 | * 7) timeout | |
2420 | * 8) other lock acquisition failure | |
2421 | * | |
cc6db4e6 | 2422 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
2423 | * |
2424 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2425 | * | |
6c23cbbd RD |
2426 | * Return: |
2427 | * 0 - On success; | |
52400ba9 DH |
2428 | * <0 - On error |
2429 | */ | |
b41277dc | 2430 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 2431 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 2432 | u32 __user *uaddr2) |
52400ba9 DH |
2433 | { |
2434 | struct hrtimer_sleeper timeout, *to = NULL; | |
2435 | struct rt_mutex_waiter rt_waiter; | |
2436 | struct rt_mutex *pi_mutex = NULL; | |
52400ba9 | 2437 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
2438 | union futex_key key2 = FUTEX_KEY_INIT; |
2439 | struct futex_q q = futex_q_init; | |
52400ba9 | 2440 | int res, ret; |
52400ba9 | 2441 | |
6f7b0a2a DH |
2442 | if (uaddr == uaddr2) |
2443 | return -EINVAL; | |
2444 | ||
52400ba9 DH |
2445 | if (!bitset) |
2446 | return -EINVAL; | |
2447 | ||
2448 | if (abs_time) { | |
2449 | to = &timeout; | |
b41277dc DH |
2450 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2451 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2452 | HRTIMER_MODE_ABS); | |
52400ba9 DH |
2453 | hrtimer_init_sleeper(to, current); |
2454 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2455 | current->timer_slack_ns); | |
2456 | } | |
2457 | ||
2458 | /* | |
2459 | * The waiter is allocated on our stack, manipulated by the requeue | |
2460 | * code while we sleep on uaddr. | |
2461 | */ | |
2462 | debug_rt_mutex_init_waiter(&rt_waiter); | |
fb00aca4 PZ |
2463 | RB_CLEAR_NODE(&rt_waiter.pi_tree_entry); |
2464 | RB_CLEAR_NODE(&rt_waiter.tree_entry); | |
52400ba9 DH |
2465 | rt_waiter.task = NULL; |
2466 | ||
9ea71503 | 2467 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
52400ba9 DH |
2468 | if (unlikely(ret != 0)) |
2469 | goto out; | |
2470 | ||
84bc4af5 DH |
2471 | q.bitset = bitset; |
2472 | q.rt_waiter = &rt_waiter; | |
2473 | q.requeue_pi_key = &key2; | |
2474 | ||
7ada876a DH |
2475 | /* |
2476 | * Prepare to wait on uaddr. On success, increments q.key (key1) ref | |
2477 | * count. | |
2478 | */ | |
b41277dc | 2479 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 TG |
2480 | if (ret) |
2481 | goto out_key2; | |
52400ba9 DH |
2482 | |
2483 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ | |
f1a11e05 | 2484 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2485 | |
2486 | spin_lock(&hb->lock); | |
2487 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2488 | spin_unlock(&hb->lock); | |
2489 | if (ret) | |
2490 | goto out_put_keys; | |
2491 | ||
2492 | /* | |
2493 | * In order for us to be here, we know our q.key == key2, and since | |
2494 | * we took the hb->lock above, we also know that futex_requeue() has | |
2495 | * completed and we no longer have to concern ourselves with a wakeup | |
7ada876a DH |
2496 | * race with the atomic proxy lock acquisition by the requeue code. The |
2497 | * futex_requeue dropped our key1 reference and incremented our key2 | |
2498 | * reference count. | |
52400ba9 DH |
2499 | */ |
2500 | ||
2501 | /* Check if the requeue code acquired the second futex for us. */ | |
2502 | if (!q.rt_waiter) { | |
2503 | /* | |
2504 | * Got the lock. We might not be the anticipated owner if we | |
2505 | * did a lock-steal - fix up the PI-state in that case. | |
2506 | */ | |
2507 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2508 | spin_lock(q.lock_ptr); | |
ae791a2d | 2509 | ret = fixup_pi_state_owner(uaddr2, &q, current); |
52400ba9 DH |
2510 | spin_unlock(q.lock_ptr); |
2511 | } | |
2512 | } else { | |
2513 | /* | |
2514 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2515 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2516 | * the pi_state. | |
2517 | */ | |
f27071cb | 2518 | WARN_ON(!q.pi_state); |
52400ba9 DH |
2519 | pi_mutex = &q.pi_state->pi_mutex; |
2520 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1); | |
2521 | debug_rt_mutex_free_waiter(&rt_waiter); | |
2522 | ||
2523 | spin_lock(q.lock_ptr); | |
2524 | /* | |
2525 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2526 | * haven't already. | |
2527 | */ | |
ae791a2d | 2528 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 DH |
2529 | /* |
2530 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 2531 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
2532 | */ |
2533 | if (res) | |
2534 | ret = (res < 0) ? res : 0; | |
2535 | ||
2536 | /* Unqueue and drop the lock. */ | |
2537 | unqueue_me_pi(&q); | |
2538 | } | |
2539 | ||
2540 | /* | |
2541 | * If fixup_pi_state_owner() faulted and was unable to handle the | |
2542 | * fault, unlock the rt_mutex and return the fault to userspace. | |
2543 | */ | |
2544 | if (ret == -EFAULT) { | |
b6070a8d | 2545 | if (pi_mutex && rt_mutex_owner(pi_mutex) == current) |
52400ba9 DH |
2546 | rt_mutex_unlock(pi_mutex); |
2547 | } else if (ret == -EINTR) { | |
52400ba9 | 2548 | /* |
cc6db4e6 DH |
2549 | * We've already been requeued, but cannot restart by calling |
2550 | * futex_lock_pi() directly. We could restart this syscall, but | |
2551 | * it would detect that the user space "val" changed and return | |
2552 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
2553 | * -EWOULDBLOCK directly. | |
52400ba9 | 2554 | */ |
2070887f | 2555 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2556 | } |
2557 | ||
2558 | out_put_keys: | |
ae791a2d | 2559 | put_futex_key(&q.key); |
c8b15a70 | 2560 | out_key2: |
ae791a2d | 2561 | put_futex_key(&key2); |
52400ba9 DH |
2562 | |
2563 | out: | |
2564 | if (to) { | |
2565 | hrtimer_cancel(&to->timer); | |
2566 | destroy_hrtimer_on_stack(&to->timer); | |
2567 | } | |
2568 | return ret; | |
2569 | } | |
2570 | ||
0771dfef IM |
2571 | /* |
2572 | * Support for robust futexes: the kernel cleans up held futexes at | |
2573 | * thread exit time. | |
2574 | * | |
2575 | * Implementation: user-space maintains a per-thread list of locks it | |
2576 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2577 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2578 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2579 | * always manipulated with the lock held, so the list is private and |
2580 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2581 | * field, to allow the kernel to clean up if the thread dies after | |
2582 | * acquiring the lock, but just before it could have added itself to | |
2583 | * the list. There can only be one such pending lock. | |
2584 | */ | |
2585 | ||
2586 | /** | |
d96ee56c DH |
2587 | * sys_set_robust_list() - Set the robust-futex list head of a task |
2588 | * @head: pointer to the list-head | |
2589 | * @len: length of the list-head, as userspace expects | |
0771dfef | 2590 | */ |
836f92ad HC |
2591 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2592 | size_t, len) | |
0771dfef | 2593 | { |
a0c1e907 TG |
2594 | if (!futex_cmpxchg_enabled) |
2595 | return -ENOSYS; | |
0771dfef IM |
2596 | /* |
2597 | * The kernel knows only one size for now: | |
2598 | */ | |
2599 | if (unlikely(len != sizeof(*head))) | |
2600 | return -EINVAL; | |
2601 | ||
2602 | current->robust_list = head; | |
2603 | ||
2604 | return 0; | |
2605 | } | |
2606 | ||
2607 | /** | |
d96ee56c DH |
2608 | * sys_get_robust_list() - Get the robust-futex list head of a task |
2609 | * @pid: pid of the process [zero for current task] | |
2610 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
2611 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 2612 | */ |
836f92ad HC |
2613 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
2614 | struct robust_list_head __user * __user *, head_ptr, | |
2615 | size_t __user *, len_ptr) | |
0771dfef | 2616 | { |
ba46df98 | 2617 | struct robust_list_head __user *head; |
0771dfef | 2618 | unsigned long ret; |
bdbb776f | 2619 | struct task_struct *p; |
0771dfef | 2620 | |
a0c1e907 TG |
2621 | if (!futex_cmpxchg_enabled) |
2622 | return -ENOSYS; | |
2623 | ||
bdbb776f KC |
2624 | rcu_read_lock(); |
2625 | ||
2626 | ret = -ESRCH; | |
0771dfef | 2627 | if (!pid) |
bdbb776f | 2628 | p = current; |
0771dfef | 2629 | else { |
228ebcbe | 2630 | p = find_task_by_vpid(pid); |
0771dfef IM |
2631 | if (!p) |
2632 | goto err_unlock; | |
0771dfef IM |
2633 | } |
2634 | ||
bdbb776f KC |
2635 | ret = -EPERM; |
2636 | if (!ptrace_may_access(p, PTRACE_MODE_READ)) | |
2637 | goto err_unlock; | |
2638 | ||
2639 | head = p->robust_list; | |
2640 | rcu_read_unlock(); | |
2641 | ||
0771dfef IM |
2642 | if (put_user(sizeof(*head), len_ptr)) |
2643 | return -EFAULT; | |
2644 | return put_user(head, head_ptr); | |
2645 | ||
2646 | err_unlock: | |
aaa2a97e | 2647 | rcu_read_unlock(); |
0771dfef IM |
2648 | |
2649 | return ret; | |
2650 | } | |
2651 | ||
2652 | /* | |
2653 | * Process a futex-list entry, check whether it's owned by the | |
2654 | * dying task, and do notification if so: | |
2655 | */ | |
e3f2ddea | 2656 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 2657 | { |
7cfdaf38 | 2658 | u32 uval, uninitialized_var(nval), mval; |
0771dfef | 2659 | |
8f17d3a5 IM |
2660 | retry: |
2661 | if (get_user(uval, uaddr)) | |
0771dfef IM |
2662 | return -1; |
2663 | ||
b488893a | 2664 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
2665 | /* |
2666 | * Ok, this dying thread is truly holding a futex | |
2667 | * of interest. Set the OWNER_DIED bit atomically | |
2668 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
2669 | * set, wake up a waiter (if any). (We have to do a | |
2670 | * futex_wake() even if OWNER_DIED is already set - | |
2671 | * to handle the rare but possible case of recursive | |
2672 | * thread-death.) The rest of the cleanup is done in | |
2673 | * userspace. | |
2674 | */ | |
e3f2ddea | 2675 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
6e0aa9f8 TG |
2676 | /* |
2677 | * We are not holding a lock here, but we want to have | |
2678 | * the pagefault_disable/enable() protection because | |
2679 | * we want to handle the fault gracefully. If the | |
2680 | * access fails we try to fault in the futex with R/W | |
2681 | * verification via get_user_pages. get_user() above | |
2682 | * does not guarantee R/W access. If that fails we | |
2683 | * give up and leave the futex locked. | |
2684 | */ | |
2685 | if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) { | |
2686 | if (fault_in_user_writeable(uaddr)) | |
2687 | return -1; | |
2688 | goto retry; | |
2689 | } | |
c87e2837 | 2690 | if (nval != uval) |
8f17d3a5 | 2691 | goto retry; |
0771dfef | 2692 | |
e3f2ddea IM |
2693 | /* |
2694 | * Wake robust non-PI futexes here. The wakeup of | |
2695 | * PI futexes happens in exit_pi_state(): | |
2696 | */ | |
36cf3b5c | 2697 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 2698 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
2699 | } |
2700 | return 0; | |
2701 | } | |
2702 | ||
e3f2ddea IM |
2703 | /* |
2704 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
2705 | */ | |
2706 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 2707 | struct robust_list __user * __user *head, |
1dcc41bb | 2708 | unsigned int *pi) |
e3f2ddea IM |
2709 | { |
2710 | unsigned long uentry; | |
2711 | ||
ba46df98 | 2712 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
2713 | return -EFAULT; |
2714 | ||
ba46df98 | 2715 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
2716 | *pi = uentry & 1; |
2717 | ||
2718 | return 0; | |
2719 | } | |
2720 | ||
0771dfef IM |
2721 | /* |
2722 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
2723 | * and mark any locks found there dead, and notify any waiters. | |
2724 | * | |
2725 | * We silently return on any sign of list-walking problem. | |
2726 | */ | |
2727 | void exit_robust_list(struct task_struct *curr) | |
2728 | { | |
2729 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 2730 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 DH |
2731 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
2732 | unsigned int uninitialized_var(next_pi); | |
0771dfef | 2733 | unsigned long futex_offset; |
9f96cb1e | 2734 | int rc; |
0771dfef | 2735 | |
a0c1e907 TG |
2736 | if (!futex_cmpxchg_enabled) |
2737 | return; | |
2738 | ||
0771dfef IM |
2739 | /* |
2740 | * Fetch the list head (which was registered earlier, via | |
2741 | * sys_set_robust_list()): | |
2742 | */ | |
e3f2ddea | 2743 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
2744 | return; |
2745 | /* | |
2746 | * Fetch the relative futex offset: | |
2747 | */ | |
2748 | if (get_user(futex_offset, &head->futex_offset)) | |
2749 | return; | |
2750 | /* | |
2751 | * Fetch any possibly pending lock-add first, and handle it | |
2752 | * if it exists: | |
2753 | */ | |
e3f2ddea | 2754 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 2755 | return; |
e3f2ddea | 2756 | |
9f96cb1e | 2757 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 2758 | while (entry != &head->list) { |
9f96cb1e MS |
2759 | /* |
2760 | * Fetch the next entry in the list before calling | |
2761 | * handle_futex_death: | |
2762 | */ | |
2763 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
2764 | /* |
2765 | * A pending lock might already be on the list, so | |
c87e2837 | 2766 | * don't process it twice: |
0771dfef IM |
2767 | */ |
2768 | if (entry != pending) | |
ba46df98 | 2769 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 2770 | curr, pi)) |
0771dfef | 2771 | return; |
9f96cb1e | 2772 | if (rc) |
0771dfef | 2773 | return; |
9f96cb1e MS |
2774 | entry = next_entry; |
2775 | pi = next_pi; | |
0771dfef IM |
2776 | /* |
2777 | * Avoid excessively long or circular lists: | |
2778 | */ | |
2779 | if (!--limit) | |
2780 | break; | |
2781 | ||
2782 | cond_resched(); | |
2783 | } | |
9f96cb1e MS |
2784 | |
2785 | if (pending) | |
2786 | handle_futex_death((void __user *)pending + futex_offset, | |
2787 | curr, pip); | |
0771dfef IM |
2788 | } |
2789 | ||
c19384b5 | 2790 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 2791 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 2792 | { |
81b40539 | 2793 | int cmd = op & FUTEX_CMD_MASK; |
b41277dc | 2794 | unsigned int flags = 0; |
34f01cc1 ED |
2795 | |
2796 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 2797 | flags |= FLAGS_SHARED; |
1da177e4 | 2798 | |
b41277dc DH |
2799 | if (op & FUTEX_CLOCK_REALTIME) { |
2800 | flags |= FLAGS_CLOCKRT; | |
2801 | if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI) | |
2802 | return -ENOSYS; | |
2803 | } | |
1da177e4 | 2804 | |
59263b51 TG |
2805 | switch (cmd) { |
2806 | case FUTEX_LOCK_PI: | |
2807 | case FUTEX_UNLOCK_PI: | |
2808 | case FUTEX_TRYLOCK_PI: | |
2809 | case FUTEX_WAIT_REQUEUE_PI: | |
2810 | case FUTEX_CMP_REQUEUE_PI: | |
2811 | if (!futex_cmpxchg_enabled) | |
2812 | return -ENOSYS; | |
2813 | } | |
2814 | ||
34f01cc1 | 2815 | switch (cmd) { |
1da177e4 | 2816 | case FUTEX_WAIT: |
cd689985 TG |
2817 | val3 = FUTEX_BITSET_MATCH_ANY; |
2818 | case FUTEX_WAIT_BITSET: | |
81b40539 | 2819 | return futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 | 2820 | case FUTEX_WAKE: |
cd689985 TG |
2821 | val3 = FUTEX_BITSET_MATCH_ANY; |
2822 | case FUTEX_WAKE_BITSET: | |
81b40539 | 2823 | return futex_wake(uaddr, flags, val, val3); |
1da177e4 | 2824 | case FUTEX_REQUEUE: |
81b40539 | 2825 | return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 | 2826 | case FUTEX_CMP_REQUEUE: |
81b40539 | 2827 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
4732efbe | 2828 | case FUTEX_WAKE_OP: |
81b40539 | 2829 | return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
c87e2837 | 2830 | case FUTEX_LOCK_PI: |
81b40539 | 2831 | return futex_lock_pi(uaddr, flags, val, timeout, 0); |
c87e2837 | 2832 | case FUTEX_UNLOCK_PI: |
81b40539 | 2833 | return futex_unlock_pi(uaddr, flags); |
c87e2837 | 2834 | case FUTEX_TRYLOCK_PI: |
81b40539 | 2835 | return futex_lock_pi(uaddr, flags, 0, timeout, 1); |
52400ba9 DH |
2836 | case FUTEX_WAIT_REQUEUE_PI: |
2837 | val3 = FUTEX_BITSET_MATCH_ANY; | |
81b40539 TG |
2838 | return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
2839 | uaddr2); | |
52400ba9 | 2840 | case FUTEX_CMP_REQUEUE_PI: |
81b40539 | 2841 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
1da177e4 | 2842 | } |
81b40539 | 2843 | return -ENOSYS; |
1da177e4 LT |
2844 | } |
2845 | ||
2846 | ||
17da2bd9 HC |
2847 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
2848 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
2849 | u32, val3) | |
1da177e4 | 2850 | { |
c19384b5 PP |
2851 | struct timespec ts; |
2852 | ktime_t t, *tp = NULL; | |
e2970f2f | 2853 | u32 val2 = 0; |
34f01cc1 | 2854 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 2855 | |
cd689985 | 2856 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
2857 | cmd == FUTEX_WAIT_BITSET || |
2858 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
c19384b5 | 2859 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 2860 | return -EFAULT; |
c19384b5 | 2861 | if (!timespec_valid(&ts)) |
9741ef96 | 2862 | return -EINVAL; |
c19384b5 PP |
2863 | |
2864 | t = timespec_to_ktime(ts); | |
34f01cc1 | 2865 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 2866 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 2867 | tp = &t; |
1da177e4 LT |
2868 | } |
2869 | /* | |
52400ba9 | 2870 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 2871 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 2872 | */ |
f54f0986 | 2873 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 2874 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 2875 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 2876 | |
c19384b5 | 2877 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
2878 | } |
2879 | ||
03b8c7b6 | 2880 | static void __init futex_detect_cmpxchg(void) |
1da177e4 | 2881 | { |
03b8c7b6 | 2882 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 2883 | u32 curval; |
03b8c7b6 HC |
2884 | |
2885 | /* | |
2886 | * This will fail and we want it. Some arch implementations do | |
2887 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
2888 | * functionality. We want to know that before we call in any | |
2889 | * of the complex code paths. Also we want to prevent | |
2890 | * registration of robust lists in that case. NULL is | |
2891 | * guaranteed to fault and we get -EFAULT on functional | |
2892 | * implementation, the non-functional ones will return | |
2893 | * -ENOSYS. | |
2894 | */ | |
2895 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) | |
2896 | futex_cmpxchg_enabled = 1; | |
2897 | #endif | |
2898 | } | |
2899 | ||
2900 | static int __init futex_init(void) | |
2901 | { | |
63b1a816 | 2902 | unsigned int futex_shift; |
a52b89eb DB |
2903 | unsigned long i; |
2904 | ||
2905 | #if CONFIG_BASE_SMALL | |
2906 | futex_hashsize = 16; | |
2907 | #else | |
2908 | futex_hashsize = roundup_pow_of_two(256 * num_possible_cpus()); | |
2909 | #endif | |
2910 | ||
2911 | futex_queues = alloc_large_system_hash("futex", sizeof(*futex_queues), | |
2912 | futex_hashsize, 0, | |
2913 | futex_hashsize < 256 ? HASH_SMALL : 0, | |
63b1a816 HC |
2914 | &futex_shift, NULL, |
2915 | futex_hashsize, futex_hashsize); | |
2916 | futex_hashsize = 1UL << futex_shift; | |
03b8c7b6 HC |
2917 | |
2918 | futex_detect_cmpxchg(); | |
a0c1e907 | 2919 | |
a52b89eb | 2920 | for (i = 0; i < futex_hashsize; i++) { |
11d4616b | 2921 | atomic_set(&futex_queues[i].waiters, 0); |
732375c6 | 2922 | plist_head_init(&futex_queues[i].chain); |
3e4ab747 TG |
2923 | spin_lock_init(&futex_queues[i].lock); |
2924 | } | |
2925 | ||
1da177e4 LT |
2926 | return 0; |
2927 | } | |
f6d107fb | 2928 | __initcall(futex_init); |