]>
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> |
ab51fbab | 67 | #include <linux/fault-inject.h> |
b488893a | 68 | |
4732efbe | 69 | #include <asm/futex.h> |
1da177e4 | 70 | |
1696a8be | 71 | #include "locking/rtmutex_common.h" |
c87e2837 | 72 | |
99b60ce6 | 73 | /* |
d7e8af1a DB |
74 | * READ this before attempting to hack on futexes! |
75 | * | |
76 | * Basic futex operation and ordering guarantees | |
77 | * ============================================= | |
99b60ce6 TG |
78 | * |
79 | * The waiter reads the futex value in user space and calls | |
80 | * futex_wait(). This function computes the hash bucket and acquires | |
81 | * the hash bucket lock. After that it reads the futex user space value | |
b0c29f79 DB |
82 | * again and verifies that the data has not changed. If it has not changed |
83 | * it enqueues itself into the hash bucket, releases the hash bucket lock | |
84 | * and schedules. | |
99b60ce6 TG |
85 | * |
86 | * The waker side modifies the user space value of the futex and calls | |
b0c29f79 DB |
87 | * futex_wake(). This function computes the hash bucket and acquires the |
88 | * hash bucket lock. Then it looks for waiters on that futex in the hash | |
89 | * bucket and wakes them. | |
99b60ce6 | 90 | * |
b0c29f79 DB |
91 | * In futex wake up scenarios where no tasks are blocked on a futex, taking |
92 | * the hb spinlock can be avoided and simply return. In order for this | |
93 | * optimization to work, ordering guarantees must exist so that the waiter | |
94 | * being added to the list is acknowledged when the list is concurrently being | |
95 | * checked by the waker, avoiding scenarios like the following: | |
99b60ce6 TG |
96 | * |
97 | * CPU 0 CPU 1 | |
98 | * val = *futex; | |
99 | * sys_futex(WAIT, futex, val); | |
100 | * futex_wait(futex, val); | |
101 | * uval = *futex; | |
102 | * *futex = newval; | |
103 | * sys_futex(WAKE, futex); | |
104 | * futex_wake(futex); | |
105 | * if (queue_empty()) | |
106 | * return; | |
107 | * if (uval == val) | |
108 | * lock(hash_bucket(futex)); | |
109 | * queue(); | |
110 | * unlock(hash_bucket(futex)); | |
111 | * schedule(); | |
112 | * | |
113 | * This would cause the waiter on CPU 0 to wait forever because it | |
114 | * missed the transition of the user space value from val to newval | |
115 | * and the waker did not find the waiter in the hash bucket queue. | |
99b60ce6 | 116 | * |
b0c29f79 DB |
117 | * The correct serialization ensures that a waiter either observes |
118 | * the changed user space value before blocking or is woken by a | |
119 | * concurrent waker: | |
120 | * | |
121 | * CPU 0 CPU 1 | |
99b60ce6 TG |
122 | * val = *futex; |
123 | * sys_futex(WAIT, futex, val); | |
124 | * futex_wait(futex, val); | |
b0c29f79 | 125 | * |
d7e8af1a | 126 | * waiters++; (a) |
8ad7b378 DB |
127 | * smp_mb(); (A) <-- paired with -. |
128 | * | | |
129 | * lock(hash_bucket(futex)); | | |
130 | * | | |
131 | * uval = *futex; | | |
132 | * | *futex = newval; | |
133 | * | sys_futex(WAKE, futex); | |
134 | * | futex_wake(futex); | |
135 | * | | |
136 | * `--------> smp_mb(); (B) | |
99b60ce6 | 137 | * if (uval == val) |
b0c29f79 | 138 | * queue(); |
99b60ce6 | 139 | * unlock(hash_bucket(futex)); |
b0c29f79 DB |
140 | * schedule(); if (waiters) |
141 | * lock(hash_bucket(futex)); | |
d7e8af1a DB |
142 | * else wake_waiters(futex); |
143 | * waiters--; (b) unlock(hash_bucket(futex)); | |
b0c29f79 | 144 | * |
d7e8af1a DB |
145 | * Where (A) orders the waiters increment and the futex value read through |
146 | * atomic operations (see hb_waiters_inc) and where (B) orders the write | |
993b2ff2 DB |
147 | * to futex and the waiters read -- this is done by the barriers for both |
148 | * shared and private futexes in get_futex_key_refs(). | |
b0c29f79 DB |
149 | * |
150 | * This yields the following case (where X:=waiters, Y:=futex): | |
151 | * | |
152 | * X = Y = 0 | |
153 | * | |
154 | * w[X]=1 w[Y]=1 | |
155 | * MB MB | |
156 | * r[Y]=y r[X]=x | |
157 | * | |
158 | * Which guarantees that x==0 && y==0 is impossible; which translates back into | |
159 | * the guarantee that we cannot both miss the futex variable change and the | |
160 | * enqueue. | |
d7e8af1a DB |
161 | * |
162 | * Note that a new waiter is accounted for in (a) even when it is possible that | |
163 | * the wait call can return error, in which case we backtrack from it in (b). | |
164 | * Refer to the comment in queue_lock(). | |
165 | * | |
166 | * Similarly, in order to account for waiters being requeued on another | |
167 | * address we always increment the waiters for the destination bucket before | |
168 | * acquiring the lock. It then decrements them again after releasing it - | |
169 | * the code that actually moves the futex(es) between hash buckets (requeue_futex) | |
170 | * will do the additional required waiter count housekeeping. This is done for | |
171 | * double_lock_hb() and double_unlock_hb(), respectively. | |
99b60ce6 TG |
172 | */ |
173 | ||
03b8c7b6 | 174 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 175 | int __read_mostly futex_cmpxchg_enabled; |
03b8c7b6 | 176 | #endif |
a0c1e907 | 177 | |
b41277dc DH |
178 | /* |
179 | * Futex flags used to encode options to functions and preserve them across | |
180 | * restarts. | |
181 | */ | |
182 | #define FLAGS_SHARED 0x01 | |
183 | #define FLAGS_CLOCKRT 0x02 | |
184 | #define FLAGS_HAS_TIMEOUT 0x04 | |
185 | ||
c87e2837 IM |
186 | /* |
187 | * Priority Inheritance state: | |
188 | */ | |
189 | struct futex_pi_state { | |
190 | /* | |
191 | * list of 'owned' pi_state instances - these have to be | |
192 | * cleaned up in do_exit() if the task exits prematurely: | |
193 | */ | |
194 | struct list_head list; | |
195 | ||
196 | /* | |
197 | * The PI object: | |
198 | */ | |
199 | struct rt_mutex pi_mutex; | |
200 | ||
201 | struct task_struct *owner; | |
202 | atomic_t refcount; | |
203 | ||
204 | union futex_key key; | |
205 | }; | |
206 | ||
d8d88fbb DH |
207 | /** |
208 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 209 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
210 | * @task: the task waiting on the futex |
211 | * @lock_ptr: the hash bucket lock | |
212 | * @key: the key the futex is hashed on | |
213 | * @pi_state: optional priority inheritance state | |
214 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
215 | * @requeue_pi_key: the requeue_pi target futex key | |
216 | * @bitset: bitset for the optional bitmasked wakeup | |
217 | * | |
218 | * We use this hashed waitqueue, instead of a normal wait_queue_t, so | |
1da177e4 LT |
219 | * we can wake only the relevant ones (hashed queues may be shared). |
220 | * | |
221 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 222 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 223 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
224 | * the second. |
225 | * | |
226 | * PI futexes are typically woken before they are removed from the hash list via | |
227 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
228 | */ |
229 | struct futex_q { | |
ec92d082 | 230 | struct plist_node list; |
1da177e4 | 231 | |
d8d88fbb | 232 | struct task_struct *task; |
1da177e4 | 233 | spinlock_t *lock_ptr; |
1da177e4 | 234 | union futex_key key; |
c87e2837 | 235 | struct futex_pi_state *pi_state; |
52400ba9 | 236 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 237 | union futex_key *requeue_pi_key; |
cd689985 | 238 | u32 bitset; |
1da177e4 LT |
239 | }; |
240 | ||
5bdb05f9 DH |
241 | static const struct futex_q futex_q_init = { |
242 | /* list gets initialized in queue_me()*/ | |
243 | .key = FUTEX_KEY_INIT, | |
244 | .bitset = FUTEX_BITSET_MATCH_ANY | |
245 | }; | |
246 | ||
1da177e4 | 247 | /* |
b2d0994b DH |
248 | * Hash buckets are shared by all the futex_keys that hash to the same |
249 | * location. Each key may have multiple futex_q structures, one for each task | |
250 | * waiting on a futex. | |
1da177e4 LT |
251 | */ |
252 | struct futex_hash_bucket { | |
11d4616b | 253 | atomic_t waiters; |
ec92d082 PP |
254 | spinlock_t lock; |
255 | struct plist_head chain; | |
a52b89eb | 256 | } ____cacheline_aligned_in_smp; |
1da177e4 | 257 | |
ac742d37 RV |
258 | /* |
259 | * The base of the bucket array and its size are always used together | |
260 | * (after initialization only in hash_futex()), so ensure that they | |
261 | * reside in the same cacheline. | |
262 | */ | |
263 | static struct { | |
264 | struct futex_hash_bucket *queues; | |
265 | unsigned long hashsize; | |
266 | } __futex_data __read_mostly __aligned(2*sizeof(long)); | |
267 | #define futex_queues (__futex_data.queues) | |
268 | #define futex_hashsize (__futex_data.hashsize) | |
a52b89eb | 269 | |
1da177e4 | 270 | |
ab51fbab DB |
271 | /* |
272 | * Fault injections for futexes. | |
273 | */ | |
274 | #ifdef CONFIG_FAIL_FUTEX | |
275 | ||
276 | static struct { | |
277 | struct fault_attr attr; | |
278 | ||
621a5f7a | 279 | bool ignore_private; |
ab51fbab DB |
280 | } fail_futex = { |
281 | .attr = FAULT_ATTR_INITIALIZER, | |
621a5f7a | 282 | .ignore_private = false, |
ab51fbab DB |
283 | }; |
284 | ||
285 | static int __init setup_fail_futex(char *str) | |
286 | { | |
287 | return setup_fault_attr(&fail_futex.attr, str); | |
288 | } | |
289 | __setup("fail_futex=", setup_fail_futex); | |
290 | ||
5d285a7f | 291 | static bool should_fail_futex(bool fshared) |
ab51fbab DB |
292 | { |
293 | if (fail_futex.ignore_private && !fshared) | |
294 | return false; | |
295 | ||
296 | return should_fail(&fail_futex.attr, 1); | |
297 | } | |
298 | ||
299 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
300 | ||
301 | static int __init fail_futex_debugfs(void) | |
302 | { | |
303 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
304 | struct dentry *dir; | |
305 | ||
306 | dir = fault_create_debugfs_attr("fail_futex", NULL, | |
307 | &fail_futex.attr); | |
308 | if (IS_ERR(dir)) | |
309 | return PTR_ERR(dir); | |
310 | ||
311 | if (!debugfs_create_bool("ignore-private", mode, dir, | |
312 | &fail_futex.ignore_private)) { | |
313 | debugfs_remove_recursive(dir); | |
314 | return -ENOMEM; | |
315 | } | |
316 | ||
317 | return 0; | |
318 | } | |
319 | ||
320 | late_initcall(fail_futex_debugfs); | |
321 | ||
322 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
323 | ||
324 | #else | |
325 | static inline bool should_fail_futex(bool fshared) | |
326 | { | |
327 | return false; | |
328 | } | |
329 | #endif /* CONFIG_FAIL_FUTEX */ | |
330 | ||
b0c29f79 DB |
331 | static inline void futex_get_mm(union futex_key *key) |
332 | { | |
333 | atomic_inc(&key->private.mm->mm_count); | |
334 | /* | |
335 | * Ensure futex_get_mm() implies a full barrier such that | |
336 | * get_futex_key() implies a full barrier. This is relied upon | |
8ad7b378 | 337 | * as smp_mb(); (B), see the ordering comment above. |
b0c29f79 | 338 | */ |
4e857c58 | 339 | smp_mb__after_atomic(); |
b0c29f79 DB |
340 | } |
341 | ||
11d4616b LT |
342 | /* |
343 | * Reflects a new waiter being added to the waitqueue. | |
344 | */ | |
345 | static inline void hb_waiters_inc(struct futex_hash_bucket *hb) | |
b0c29f79 DB |
346 | { |
347 | #ifdef CONFIG_SMP | |
11d4616b | 348 | atomic_inc(&hb->waiters); |
b0c29f79 | 349 | /* |
11d4616b | 350 | * Full barrier (A), see the ordering comment above. |
b0c29f79 | 351 | */ |
4e857c58 | 352 | smp_mb__after_atomic(); |
11d4616b LT |
353 | #endif |
354 | } | |
355 | ||
356 | /* | |
357 | * Reflects a waiter being removed from the waitqueue by wakeup | |
358 | * paths. | |
359 | */ | |
360 | static inline void hb_waiters_dec(struct futex_hash_bucket *hb) | |
361 | { | |
362 | #ifdef CONFIG_SMP | |
363 | atomic_dec(&hb->waiters); | |
364 | #endif | |
365 | } | |
b0c29f79 | 366 | |
11d4616b LT |
367 | static inline int hb_waiters_pending(struct futex_hash_bucket *hb) |
368 | { | |
369 | #ifdef CONFIG_SMP | |
370 | return atomic_read(&hb->waiters); | |
b0c29f79 | 371 | #else |
11d4616b | 372 | return 1; |
b0c29f79 DB |
373 | #endif |
374 | } | |
375 | ||
1da177e4 LT |
376 | /* |
377 | * We hash on the keys returned from get_futex_key (see below). | |
378 | */ | |
379 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
380 | { | |
381 | u32 hash = jhash2((u32*)&key->both.word, | |
382 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
383 | key->both.offset); | |
a52b89eb | 384 | return &futex_queues[hash & (futex_hashsize - 1)]; |
1da177e4 LT |
385 | } |
386 | ||
387 | /* | |
388 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
389 | */ | |
390 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
391 | { | |
2bc87203 DH |
392 | return (key1 && key2 |
393 | && key1->both.word == key2->both.word | |
1da177e4 LT |
394 | && key1->both.ptr == key2->both.ptr |
395 | && key1->both.offset == key2->both.offset); | |
396 | } | |
397 | ||
38d47c1b PZ |
398 | /* |
399 | * Take a reference to the resource addressed by a key. | |
400 | * Can be called while holding spinlocks. | |
401 | * | |
402 | */ | |
403 | static void get_futex_key_refs(union futex_key *key) | |
404 | { | |
405 | if (!key->both.ptr) | |
406 | return; | |
407 | ||
408 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
409 | case FUT_OFF_INODE: | |
8ad7b378 | 410 | ihold(key->shared.inode); /* implies smp_mb(); (B) */ |
38d47c1b PZ |
411 | break; |
412 | case FUT_OFF_MMSHARED: | |
8ad7b378 | 413 | futex_get_mm(key); /* implies smp_mb(); (B) */ |
38d47c1b | 414 | break; |
76835b0e | 415 | default: |
993b2ff2 DB |
416 | /* |
417 | * Private futexes do not hold reference on an inode or | |
418 | * mm, therefore the only purpose of calling get_futex_key_refs | |
419 | * is because we need the barrier for the lockless waiter check. | |
420 | */ | |
8ad7b378 | 421 | smp_mb(); /* explicit smp_mb(); (B) */ |
38d47c1b PZ |
422 | } |
423 | } | |
424 | ||
425 | /* | |
426 | * Drop a reference to the resource addressed by a key. | |
993b2ff2 DB |
427 | * The hash bucket spinlock must not be held. This is |
428 | * a no-op for private futexes, see comment in the get | |
429 | * counterpart. | |
38d47c1b PZ |
430 | */ |
431 | static void drop_futex_key_refs(union futex_key *key) | |
432 | { | |
90621c40 DH |
433 | if (!key->both.ptr) { |
434 | /* If we're here then we tried to put a key we failed to get */ | |
435 | WARN_ON_ONCE(1); | |
38d47c1b | 436 | return; |
90621c40 | 437 | } |
38d47c1b PZ |
438 | |
439 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
440 | case FUT_OFF_INODE: | |
441 | iput(key->shared.inode); | |
442 | break; | |
443 | case FUT_OFF_MMSHARED: | |
444 | mmdrop(key->private.mm); | |
445 | break; | |
446 | } | |
447 | } | |
448 | ||
34f01cc1 | 449 | /** |
d96ee56c DH |
450 | * get_futex_key() - Get parameters which are the keys for a futex |
451 | * @uaddr: virtual address of the futex | |
452 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED | |
453 | * @key: address where result is stored. | |
9ea71503 SB |
454 | * @rw: mapping needs to be read/write (values: VERIFY_READ, |
455 | * VERIFY_WRITE) | |
34f01cc1 | 456 | * |
6c23cbbd RD |
457 | * Return: a negative error code or 0 |
458 | * | |
34f01cc1 | 459 | * The key words are stored in *key on success. |
1da177e4 | 460 | * |
6131ffaa | 461 | * For shared mappings, it's (page->index, file_inode(vma->vm_file), |
1da177e4 LT |
462 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
463 | * We can usually work out the index without swapping in the page. | |
464 | * | |
b2d0994b | 465 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 466 | */ |
64d1304a | 467 | static int |
9ea71503 | 468 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) |
1da177e4 | 469 | { |
e2970f2f | 470 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 471 | struct mm_struct *mm = current->mm; |
14d27abd KS |
472 | struct page *page; |
473 | struct address_space *mapping; | |
9ea71503 | 474 | int err, ro = 0; |
1da177e4 LT |
475 | |
476 | /* | |
477 | * The futex address must be "naturally" aligned. | |
478 | */ | |
e2970f2f | 479 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 480 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 481 | return -EINVAL; |
e2970f2f | 482 | address -= key->both.offset; |
1da177e4 | 483 | |
5cdec2d8 LT |
484 | if (unlikely(!access_ok(rw, uaddr, sizeof(u32)))) |
485 | return -EFAULT; | |
486 | ||
ab51fbab DB |
487 | if (unlikely(should_fail_futex(fshared))) |
488 | return -EFAULT; | |
489 | ||
34f01cc1 ED |
490 | /* |
491 | * PROCESS_PRIVATE futexes are fast. | |
492 | * As the mm cannot disappear under us and the 'key' only needs | |
493 | * virtual address, we dont even have to find the underlying vma. | |
494 | * Note : We do have to check 'uaddr' is a valid user address, | |
495 | * but access_ok() should be faster than find_vma() | |
496 | */ | |
497 | if (!fshared) { | |
34f01cc1 ED |
498 | key->private.mm = mm; |
499 | key->private.address = address; | |
8ad7b378 | 500 | get_futex_key_refs(key); /* implies smp_mb(); (B) */ |
34f01cc1 ED |
501 | return 0; |
502 | } | |
1da177e4 | 503 | |
38d47c1b | 504 | again: |
ab51fbab DB |
505 | /* Ignore any VERIFY_READ mapping (futex common case) */ |
506 | if (unlikely(should_fail_futex(fshared))) | |
507 | return -EFAULT; | |
508 | ||
7485d0d3 | 509 | err = get_user_pages_fast(address, 1, 1, &page); |
9ea71503 SB |
510 | /* |
511 | * If write access is not required (eg. FUTEX_WAIT), try | |
512 | * and get read-only access. | |
513 | */ | |
514 | if (err == -EFAULT && rw == VERIFY_READ) { | |
515 | err = get_user_pages_fast(address, 1, 0, &page); | |
516 | ro = 1; | |
517 | } | |
38d47c1b PZ |
518 | if (err < 0) |
519 | return err; | |
9ea71503 SB |
520 | else |
521 | err = 0; | |
38d47c1b | 522 | |
65d8fc77 MG |
523 | /* |
524 | * The treatment of mapping from this point on is critical. The page | |
525 | * lock protects many things but in this context the page lock | |
526 | * stabilizes mapping, prevents inode freeing in the shared | |
527 | * file-backed region case and guards against movement to swap cache. | |
528 | * | |
529 | * Strictly speaking the page lock is not needed in all cases being | |
530 | * considered here and page lock forces unnecessarily serialization | |
531 | * From this point on, mapping will be re-verified if necessary and | |
532 | * page lock will be acquired only if it is unavoidable | |
533 | */ | |
534 | page = compound_head(page); | |
535 | mapping = READ_ONCE(page->mapping); | |
536 | ||
e6780f72 | 537 | /* |
14d27abd | 538 | * If page->mapping is NULL, then it cannot be a PageAnon |
e6780f72 HD |
539 | * page; but it might be the ZERO_PAGE or in the gate area or |
540 | * in a special mapping (all cases which we are happy to fail); | |
541 | * or it may have been a good file page when get_user_pages_fast | |
542 | * found it, but truncated or holepunched or subjected to | |
543 | * invalidate_complete_page2 before we got the page lock (also | |
544 | * cases which we are happy to fail). And we hold a reference, | |
545 | * so refcount care in invalidate_complete_page's remove_mapping | |
546 | * prevents drop_caches from setting mapping to NULL beneath us. | |
547 | * | |
548 | * The case we do have to guard against is when memory pressure made | |
549 | * shmem_writepage move it from filecache to swapcache beneath us: | |
14d27abd | 550 | * an unlikely race, but we do need to retry for page->mapping. |
e6780f72 | 551 | */ |
65d8fc77 MG |
552 | if (unlikely(!mapping)) { |
553 | int shmem_swizzled; | |
554 | ||
555 | /* | |
556 | * Page lock is required to identify which special case above | |
557 | * applies. If this is really a shmem page then the page lock | |
558 | * will prevent unexpected transitions. | |
559 | */ | |
560 | lock_page(page); | |
561 | shmem_swizzled = PageSwapCache(page) || page->mapping; | |
14d27abd KS |
562 | unlock_page(page); |
563 | put_page(page); | |
65d8fc77 | 564 | |
e6780f72 HD |
565 | if (shmem_swizzled) |
566 | goto again; | |
65d8fc77 | 567 | |
e6780f72 | 568 | return -EFAULT; |
38d47c1b | 569 | } |
1da177e4 LT |
570 | |
571 | /* | |
572 | * Private mappings are handled in a simple way. | |
573 | * | |
65d8fc77 MG |
574 | * If the futex key is stored on an anonymous page, then the associated |
575 | * object is the mm which is implicitly pinned by the calling process. | |
576 | * | |
1da177e4 LT |
577 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if |
578 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 579 | * the object not the particular process. |
1da177e4 | 580 | */ |
14d27abd | 581 | if (PageAnon(page)) { |
9ea71503 SB |
582 | /* |
583 | * A RO anonymous page will never change and thus doesn't make | |
584 | * sense for futex operations. | |
585 | */ | |
ab51fbab | 586 | if (unlikely(should_fail_futex(fshared)) || ro) { |
9ea71503 SB |
587 | err = -EFAULT; |
588 | goto out; | |
589 | } | |
590 | ||
38d47c1b | 591 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 592 | key->private.mm = mm; |
e2970f2f | 593 | key->private.address = address; |
65d8fc77 MG |
594 | |
595 | get_futex_key_refs(key); /* implies smp_mb(); (B) */ | |
596 | ||
38d47c1b | 597 | } else { |
65d8fc77 MG |
598 | struct inode *inode; |
599 | ||
600 | /* | |
601 | * The associated futex object in this case is the inode and | |
602 | * the page->mapping must be traversed. Ordinarily this should | |
603 | * be stabilised under page lock but it's not strictly | |
604 | * necessary in this case as we just want to pin the inode, not | |
605 | * update the radix tree or anything like that. | |
606 | * | |
607 | * The RCU read lock is taken as the inode is finally freed | |
608 | * under RCU. If the mapping still matches expectations then the | |
609 | * mapping->host can be safely accessed as being a valid inode. | |
610 | */ | |
611 | rcu_read_lock(); | |
612 | ||
613 | if (READ_ONCE(page->mapping) != mapping) { | |
614 | rcu_read_unlock(); | |
615 | put_page(page); | |
616 | ||
617 | goto again; | |
618 | } | |
619 | ||
620 | inode = READ_ONCE(mapping->host); | |
621 | if (!inode) { | |
622 | rcu_read_unlock(); | |
623 | put_page(page); | |
624 | ||
625 | goto again; | |
626 | } | |
627 | ||
628 | /* | |
629 | * Take a reference unless it is about to be freed. Previously | |
630 | * this reference was taken by ihold under the page lock | |
631 | * pinning the inode in place so i_lock was unnecessary. The | |
632 | * only way for this check to fail is if the inode was | |
633 | * truncated in parallel so warn for now if this happens. | |
634 | * | |
635 | * We are not calling into get_futex_key_refs() in file-backed | |
636 | * cases, therefore a successful atomic_inc return below will | |
637 | * guarantee that get_futex_key() will still imply smp_mb(); (B). | |
638 | */ | |
639 | if (WARN_ON_ONCE(!atomic_inc_not_zero(&inode->i_count))) { | |
640 | rcu_read_unlock(); | |
641 | put_page(page); | |
642 | ||
643 | goto again; | |
644 | } | |
645 | ||
646 | /* Should be impossible but lets be paranoid for now */ | |
647 | if (WARN_ON_ONCE(inode->i_mapping != mapping)) { | |
648 | err = -EFAULT; | |
649 | rcu_read_unlock(); | |
650 | iput(inode); | |
651 | ||
652 | goto out; | |
653 | } | |
654 | ||
38d47c1b | 655 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ |
65d8fc77 | 656 | key->shared.inode = inode; |
13d60f4b | 657 | key->shared.pgoff = basepage_index(page); |
65d8fc77 | 658 | rcu_read_unlock(); |
1da177e4 LT |
659 | } |
660 | ||
9ea71503 | 661 | out: |
14d27abd | 662 | put_page(page); |
9ea71503 | 663 | return err; |
1da177e4 LT |
664 | } |
665 | ||
ae791a2d | 666 | static inline void put_futex_key(union futex_key *key) |
1da177e4 | 667 | { |
38d47c1b | 668 | drop_futex_key_refs(key); |
1da177e4 LT |
669 | } |
670 | ||
d96ee56c DH |
671 | /** |
672 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
673 | * @uaddr: pointer to faulting user space address |
674 | * | |
675 | * Slow path to fixup the fault we just took in the atomic write | |
676 | * access to @uaddr. | |
677 | * | |
fb62db2b | 678 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
679 | * user address. We know that we faulted in the atomic pagefault |
680 | * disabled section so we can as well avoid the #PF overhead by | |
681 | * calling get_user_pages() right away. | |
682 | */ | |
683 | static int fault_in_user_writeable(u32 __user *uaddr) | |
684 | { | |
722d0172 AK |
685 | struct mm_struct *mm = current->mm; |
686 | int ret; | |
687 | ||
688 | down_read(&mm->mmap_sem); | |
2efaca92 | 689 | ret = fixup_user_fault(current, mm, (unsigned long)uaddr, |
4a9e1cda | 690 | FAULT_FLAG_WRITE, NULL); |
722d0172 AK |
691 | up_read(&mm->mmap_sem); |
692 | ||
d0725992 TG |
693 | return ret < 0 ? ret : 0; |
694 | } | |
695 | ||
4b1c486b DH |
696 | /** |
697 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
698 | * @hb: the hash bucket the futex_q's reside in |
699 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
700 | * |
701 | * Must be called with the hb lock held. | |
702 | */ | |
703 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
704 | union futex_key *key) | |
705 | { | |
706 | struct futex_q *this; | |
707 | ||
708 | plist_for_each_entry(this, &hb->chain, list) { | |
709 | if (match_futex(&this->key, key)) | |
710 | return this; | |
711 | } | |
712 | return NULL; | |
713 | } | |
714 | ||
37a9d912 ML |
715 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
716 | u32 uval, u32 newval) | |
36cf3b5c | 717 | { |
37a9d912 | 718 | int ret; |
36cf3b5c TG |
719 | |
720 | pagefault_disable(); | |
37a9d912 | 721 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
722 | pagefault_enable(); |
723 | ||
37a9d912 | 724 | return ret; |
36cf3b5c TG |
725 | } |
726 | ||
727 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
728 | { |
729 | int ret; | |
730 | ||
a866374a | 731 | pagefault_disable(); |
e2970f2f | 732 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 733 | pagefault_enable(); |
1da177e4 LT |
734 | |
735 | return ret ? -EFAULT : 0; | |
736 | } | |
737 | ||
c87e2837 IM |
738 | |
739 | /* | |
740 | * PI code: | |
741 | */ | |
742 | static int refill_pi_state_cache(void) | |
743 | { | |
744 | struct futex_pi_state *pi_state; | |
745 | ||
746 | if (likely(current->pi_state_cache)) | |
747 | return 0; | |
748 | ||
4668edc3 | 749 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
750 | |
751 | if (!pi_state) | |
752 | return -ENOMEM; | |
753 | ||
c87e2837 IM |
754 | INIT_LIST_HEAD(&pi_state->list); |
755 | /* pi_mutex gets initialized later */ | |
756 | pi_state->owner = NULL; | |
757 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 758 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
759 | |
760 | current->pi_state_cache = pi_state; | |
761 | ||
762 | return 0; | |
763 | } | |
764 | ||
765 | static struct futex_pi_state * alloc_pi_state(void) | |
766 | { | |
767 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
768 | ||
769 | WARN_ON(!pi_state); | |
770 | current->pi_state_cache = NULL; | |
771 | ||
772 | return pi_state; | |
773 | } | |
774 | ||
30a6b803 | 775 | /* |
29e9ee5d TG |
776 | * Drops a reference to the pi_state object and frees or caches it |
777 | * when the last reference is gone. | |
778 | * | |
30a6b803 BS |
779 | * Must be called with the hb lock held. |
780 | */ | |
29e9ee5d | 781 | static void put_pi_state(struct futex_pi_state *pi_state) |
c87e2837 | 782 | { |
30a6b803 BS |
783 | if (!pi_state) |
784 | return; | |
785 | ||
c87e2837 IM |
786 | if (!atomic_dec_and_test(&pi_state->refcount)) |
787 | return; | |
788 | ||
789 | /* | |
790 | * If pi_state->owner is NULL, the owner is most probably dying | |
791 | * and has cleaned up the pi_state already | |
792 | */ | |
793 | if (pi_state->owner) { | |
1d615482 | 794 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
c87e2837 | 795 | list_del_init(&pi_state->list); |
1d615482 | 796 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
c87e2837 IM |
797 | |
798 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
799 | } | |
800 | ||
801 | if (current->pi_state_cache) | |
802 | kfree(pi_state); | |
803 | else { | |
804 | /* | |
805 | * pi_state->list is already empty. | |
806 | * clear pi_state->owner. | |
807 | * refcount is at 0 - put it back to 1. | |
808 | */ | |
809 | pi_state->owner = NULL; | |
810 | atomic_set(&pi_state->refcount, 1); | |
811 | current->pi_state_cache = pi_state; | |
812 | } | |
813 | } | |
814 | ||
815 | /* | |
816 | * Look up the task based on what TID userspace gave us. | |
817 | * We dont trust it. | |
818 | */ | |
819 | static struct task_struct * futex_find_get_task(pid_t pid) | |
820 | { | |
821 | struct task_struct *p; | |
822 | ||
d359b549 | 823 | rcu_read_lock(); |
228ebcbe | 824 | p = find_task_by_vpid(pid); |
7a0ea09a MH |
825 | if (p) |
826 | get_task_struct(p); | |
a06381fe | 827 | |
d359b549 | 828 | rcu_read_unlock(); |
c87e2837 IM |
829 | |
830 | return p; | |
831 | } | |
832 | ||
833 | /* | |
834 | * This task is holding PI mutexes at exit time => bad. | |
835 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
836 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
837 | */ | |
838 | void exit_pi_state_list(struct task_struct *curr) | |
839 | { | |
c87e2837 IM |
840 | struct list_head *next, *head = &curr->pi_state_list; |
841 | struct futex_pi_state *pi_state; | |
627371d7 | 842 | struct futex_hash_bucket *hb; |
38d47c1b | 843 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 844 | |
a0c1e907 TG |
845 | if (!futex_cmpxchg_enabled) |
846 | return; | |
c87e2837 IM |
847 | /* |
848 | * We are a ZOMBIE and nobody can enqueue itself on | |
849 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 850 | * versus waiters unqueueing themselves: |
c87e2837 | 851 | */ |
1d615482 | 852 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 IM |
853 | while (!list_empty(head)) { |
854 | ||
855 | next = head->next; | |
856 | pi_state = list_entry(next, struct futex_pi_state, list); | |
857 | key = pi_state->key; | |
627371d7 | 858 | hb = hash_futex(&key); |
1d615482 | 859 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 860 | |
c87e2837 IM |
861 | spin_lock(&hb->lock); |
862 | ||
1d615482 | 863 | raw_spin_lock_irq(&curr->pi_lock); |
627371d7 IM |
864 | /* |
865 | * We dropped the pi-lock, so re-check whether this | |
866 | * task still owns the PI-state: | |
867 | */ | |
c87e2837 IM |
868 | if (head->next != next) { |
869 | spin_unlock(&hb->lock); | |
870 | continue; | |
871 | } | |
872 | ||
c87e2837 | 873 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
874 | WARN_ON(list_empty(&pi_state->list)); |
875 | list_del_init(&pi_state->list); | |
c87e2837 | 876 | pi_state->owner = NULL; |
1d615482 | 877 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
878 | |
879 | rt_mutex_unlock(&pi_state->pi_mutex); | |
880 | ||
881 | spin_unlock(&hb->lock); | |
882 | ||
1d615482 | 883 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 884 | } |
1d615482 | 885 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
886 | } |
887 | ||
54a21788 TG |
888 | /* |
889 | * We need to check the following states: | |
890 | * | |
891 | * Waiter | pi_state | pi->owner | uTID | uODIED | ? | |
892 | * | |
893 | * [1] NULL | --- | --- | 0 | 0/1 | Valid | |
894 | * [2] NULL | --- | --- | >0 | 0/1 | Valid | |
895 | * | |
896 | * [3] Found | NULL | -- | Any | 0/1 | Invalid | |
897 | * | |
898 | * [4] Found | Found | NULL | 0 | 1 | Valid | |
899 | * [5] Found | Found | NULL | >0 | 1 | Invalid | |
900 | * | |
901 | * [6] Found | Found | task | 0 | 1 | Valid | |
902 | * | |
903 | * [7] Found | Found | NULL | Any | 0 | Invalid | |
904 | * | |
905 | * [8] Found | Found | task | ==taskTID | 0/1 | Valid | |
906 | * [9] Found | Found | task | 0 | 0 | Invalid | |
907 | * [10] Found | Found | task | !=taskTID | 0/1 | Invalid | |
908 | * | |
909 | * [1] Indicates that the kernel can acquire the futex atomically. We | |
910 | * came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. | |
911 | * | |
912 | * [2] Valid, if TID does not belong to a kernel thread. If no matching | |
913 | * thread is found then it indicates that the owner TID has died. | |
914 | * | |
915 | * [3] Invalid. The waiter is queued on a non PI futex | |
916 | * | |
917 | * [4] Valid state after exit_robust_list(), which sets the user space | |
918 | * value to FUTEX_WAITERS | FUTEX_OWNER_DIED. | |
919 | * | |
920 | * [5] The user space value got manipulated between exit_robust_list() | |
921 | * and exit_pi_state_list() | |
922 | * | |
923 | * [6] Valid state after exit_pi_state_list() which sets the new owner in | |
924 | * the pi_state but cannot access the user space value. | |
925 | * | |
926 | * [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. | |
927 | * | |
928 | * [8] Owner and user space value match | |
929 | * | |
930 | * [9] There is no transient state which sets the user space TID to 0 | |
931 | * except exit_robust_list(), but this is indicated by the | |
932 | * FUTEX_OWNER_DIED bit. See [4] | |
933 | * | |
934 | * [10] There is no transient state which leaves owner and user space | |
935 | * TID out of sync. | |
936 | */ | |
e60cbc5c TG |
937 | |
938 | /* | |
939 | * Validate that the existing waiter has a pi_state and sanity check | |
940 | * the pi_state against the user space value. If correct, attach to | |
941 | * it. | |
942 | */ | |
943 | static int attach_to_pi_state(u32 uval, struct futex_pi_state *pi_state, | |
944 | struct futex_pi_state **ps) | |
c87e2837 | 945 | { |
778e9a9c | 946 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 | 947 | |
e60cbc5c TG |
948 | /* |
949 | * Userspace might have messed up non-PI and PI futexes [3] | |
950 | */ | |
951 | if (unlikely(!pi_state)) | |
952 | return -EINVAL; | |
06a9ec29 | 953 | |
e60cbc5c | 954 | WARN_ON(!atomic_read(&pi_state->refcount)); |
59647b6a | 955 | |
e60cbc5c TG |
956 | /* |
957 | * Handle the owner died case: | |
958 | */ | |
959 | if (uval & FUTEX_OWNER_DIED) { | |
bd1dbcc6 | 960 | /* |
e60cbc5c TG |
961 | * exit_pi_state_list sets owner to NULL and wakes the |
962 | * topmost waiter. The task which acquires the | |
963 | * pi_state->rt_mutex will fixup owner. | |
bd1dbcc6 | 964 | */ |
e60cbc5c | 965 | if (!pi_state->owner) { |
59647b6a | 966 | /* |
e60cbc5c TG |
967 | * No pi state owner, but the user space TID |
968 | * is not 0. Inconsistent state. [5] | |
59647b6a | 969 | */ |
e60cbc5c TG |
970 | if (pid) |
971 | return -EINVAL; | |
bd1dbcc6 | 972 | /* |
e60cbc5c | 973 | * Take a ref on the state and return success. [4] |
866293ee | 974 | */ |
e60cbc5c | 975 | goto out_state; |
c87e2837 | 976 | } |
bd1dbcc6 TG |
977 | |
978 | /* | |
e60cbc5c TG |
979 | * If TID is 0, then either the dying owner has not |
980 | * yet executed exit_pi_state_list() or some waiter | |
981 | * acquired the rtmutex in the pi state, but did not | |
982 | * yet fixup the TID in user space. | |
983 | * | |
984 | * Take a ref on the state and return success. [6] | |
985 | */ | |
986 | if (!pid) | |
987 | goto out_state; | |
988 | } else { | |
989 | /* | |
990 | * If the owner died bit is not set, then the pi_state | |
991 | * must have an owner. [7] | |
bd1dbcc6 | 992 | */ |
e60cbc5c | 993 | if (!pi_state->owner) |
bd1dbcc6 | 994 | return -EINVAL; |
c87e2837 IM |
995 | } |
996 | ||
e60cbc5c TG |
997 | /* |
998 | * Bail out if user space manipulated the futex value. If pi | |
999 | * state exists then the owner TID must be the same as the | |
1000 | * user space TID. [9/10] | |
1001 | */ | |
1002 | if (pid != task_pid_vnr(pi_state->owner)) | |
1003 | return -EINVAL; | |
1004 | out_state: | |
1005 | atomic_inc(&pi_state->refcount); | |
1006 | *ps = pi_state; | |
1007 | return 0; | |
1008 | } | |
1009 | ||
04e1b2e5 TG |
1010 | /* |
1011 | * Lookup the task for the TID provided from user space and attach to | |
1012 | * it after doing proper sanity checks. | |
1013 | */ | |
1014 | static int attach_to_pi_owner(u32 uval, union futex_key *key, | |
1015 | struct futex_pi_state **ps) | |
e60cbc5c | 1016 | { |
e60cbc5c | 1017 | pid_t pid = uval & FUTEX_TID_MASK; |
04e1b2e5 TG |
1018 | struct futex_pi_state *pi_state; |
1019 | struct task_struct *p; | |
e60cbc5c | 1020 | |
c87e2837 | 1021 | /* |
e3f2ddea | 1022 | * We are the first waiter - try to look up the real owner and attach |
54a21788 | 1023 | * the new pi_state to it, but bail out when TID = 0 [1] |
c87e2837 | 1024 | */ |
778e9a9c | 1025 | if (!pid) |
e3f2ddea | 1026 | return -ESRCH; |
c87e2837 | 1027 | p = futex_find_get_task(pid); |
7a0ea09a MH |
1028 | if (!p) |
1029 | return -ESRCH; | |
778e9a9c | 1030 | |
a2129464 | 1031 | if (unlikely(p->flags & PF_KTHREAD)) { |
f0d71b3d TG |
1032 | put_task_struct(p); |
1033 | return -EPERM; | |
1034 | } | |
1035 | ||
778e9a9c AK |
1036 | /* |
1037 | * We need to look at the task state flags to figure out, | |
1038 | * whether the task is exiting. To protect against the do_exit | |
1039 | * change of the task flags, we do this protected by | |
1040 | * p->pi_lock: | |
1041 | */ | |
1d615482 | 1042 | raw_spin_lock_irq(&p->pi_lock); |
778e9a9c AK |
1043 | if (unlikely(p->flags & PF_EXITING)) { |
1044 | /* | |
1045 | * The task is on the way out. When PF_EXITPIDONE is | |
1046 | * set, we know that the task has finished the | |
1047 | * cleanup: | |
1048 | */ | |
1049 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
1050 | ||
1d615482 | 1051 | raw_spin_unlock_irq(&p->pi_lock); |
778e9a9c AK |
1052 | put_task_struct(p); |
1053 | return ret; | |
1054 | } | |
c87e2837 | 1055 | |
54a21788 TG |
1056 | /* |
1057 | * No existing pi state. First waiter. [2] | |
1058 | */ | |
c87e2837 IM |
1059 | pi_state = alloc_pi_state(); |
1060 | ||
1061 | /* | |
04e1b2e5 | 1062 | * Initialize the pi_mutex in locked state and make @p |
c87e2837 IM |
1063 | * the owner of it: |
1064 | */ | |
1065 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
1066 | ||
1067 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 1068 | pi_state->key = *key; |
c87e2837 | 1069 | |
627371d7 | 1070 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
1071 | list_add(&pi_state->list, &p->pi_state_list); |
1072 | pi_state->owner = p; | |
1d615482 | 1073 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
1074 | |
1075 | put_task_struct(p); | |
1076 | ||
d0aa7a70 | 1077 | *ps = pi_state; |
c87e2837 IM |
1078 | |
1079 | return 0; | |
1080 | } | |
1081 | ||
04e1b2e5 TG |
1082 | static int lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
1083 | union futex_key *key, struct futex_pi_state **ps) | |
1084 | { | |
1085 | struct futex_q *match = futex_top_waiter(hb, key); | |
1086 | ||
1087 | /* | |
1088 | * If there is a waiter on that futex, validate it and | |
1089 | * attach to the pi_state when the validation succeeds. | |
1090 | */ | |
1091 | if (match) | |
1092 | return attach_to_pi_state(uval, match->pi_state, ps); | |
1093 | ||
1094 | /* | |
1095 | * We are the first waiter - try to look up the owner based on | |
1096 | * @uval and attach to it. | |
1097 | */ | |
1098 | return attach_to_pi_owner(uval, key, ps); | |
1099 | } | |
1100 | ||
af54d6a1 TG |
1101 | static int lock_pi_update_atomic(u32 __user *uaddr, u32 uval, u32 newval) |
1102 | { | |
1103 | u32 uninitialized_var(curval); | |
1104 | ||
ab51fbab DB |
1105 | if (unlikely(should_fail_futex(true))) |
1106 | return -EFAULT; | |
1107 | ||
af54d6a1 TG |
1108 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))) |
1109 | return -EFAULT; | |
1110 | ||
1111 | /*If user space value changed, let the caller retry */ | |
1112 | return curval != uval ? -EAGAIN : 0; | |
1113 | } | |
1114 | ||
1a52084d | 1115 | /** |
d96ee56c | 1116 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
1117 | * @uaddr: the pi futex user address |
1118 | * @hb: the pi futex hash bucket | |
1119 | * @key: the futex key associated with uaddr and hb | |
1120 | * @ps: the pi_state pointer where we store the result of the | |
1121 | * lookup | |
1122 | * @task: the task to perform the atomic lock work for. This will | |
1123 | * be "current" except in the case of requeue pi. | |
1124 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d | 1125 | * |
6c23cbbd RD |
1126 | * Return: |
1127 | * 0 - ready to wait; | |
1128 | * 1 - acquired the lock; | |
1a52084d DH |
1129 | * <0 - error |
1130 | * | |
1131 | * The hb->lock and futex_key refs shall be held by the caller. | |
1132 | */ | |
1133 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
1134 | union futex_key *key, | |
1135 | struct futex_pi_state **ps, | |
bab5bc9e | 1136 | struct task_struct *task, int set_waiters) |
1a52084d | 1137 | { |
af54d6a1 TG |
1138 | u32 uval, newval, vpid = task_pid_vnr(task); |
1139 | struct futex_q *match; | |
1140 | int ret; | |
1a52084d DH |
1141 | |
1142 | /* | |
af54d6a1 TG |
1143 | * Read the user space value first so we can validate a few |
1144 | * things before proceeding further. | |
1a52084d | 1145 | */ |
af54d6a1 | 1146 | if (get_futex_value_locked(&uval, uaddr)) |
1a52084d DH |
1147 | return -EFAULT; |
1148 | ||
ab51fbab DB |
1149 | if (unlikely(should_fail_futex(true))) |
1150 | return -EFAULT; | |
1151 | ||
1a52084d DH |
1152 | /* |
1153 | * Detect deadlocks. | |
1154 | */ | |
af54d6a1 | 1155 | if ((unlikely((uval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
1156 | return -EDEADLK; |
1157 | ||
ab51fbab DB |
1158 | if ((unlikely(should_fail_futex(true)))) |
1159 | return -EDEADLK; | |
1160 | ||
1a52084d | 1161 | /* |
af54d6a1 TG |
1162 | * Lookup existing state first. If it exists, try to attach to |
1163 | * its pi_state. | |
1a52084d | 1164 | */ |
af54d6a1 TG |
1165 | match = futex_top_waiter(hb, key); |
1166 | if (match) | |
1167 | return attach_to_pi_state(uval, match->pi_state, ps); | |
1a52084d DH |
1168 | |
1169 | /* | |
af54d6a1 TG |
1170 | * No waiter and user TID is 0. We are here because the |
1171 | * waiters or the owner died bit is set or called from | |
1172 | * requeue_cmp_pi or for whatever reason something took the | |
1173 | * syscall. | |
1a52084d | 1174 | */ |
af54d6a1 | 1175 | if (!(uval & FUTEX_TID_MASK)) { |
59fa6245 | 1176 | /* |
af54d6a1 TG |
1177 | * We take over the futex. No other waiters and the user space |
1178 | * TID is 0. We preserve the owner died bit. | |
59fa6245 | 1179 | */ |
af54d6a1 TG |
1180 | newval = uval & FUTEX_OWNER_DIED; |
1181 | newval |= vpid; | |
1a52084d | 1182 | |
af54d6a1 TG |
1183 | /* The futex requeue_pi code can enforce the waiters bit */ |
1184 | if (set_waiters) | |
1185 | newval |= FUTEX_WAITERS; | |
1186 | ||
1187 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1188 | /* If the take over worked, return 1 */ | |
1189 | return ret < 0 ? ret : 1; | |
1190 | } | |
1a52084d DH |
1191 | |
1192 | /* | |
af54d6a1 TG |
1193 | * First waiter. Set the waiters bit before attaching ourself to |
1194 | * the owner. If owner tries to unlock, it will be forced into | |
1195 | * the kernel and blocked on hb->lock. | |
1a52084d | 1196 | */ |
af54d6a1 TG |
1197 | newval = uval | FUTEX_WAITERS; |
1198 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1199 | if (ret) | |
1200 | return ret; | |
1a52084d | 1201 | /* |
af54d6a1 TG |
1202 | * If the update of the user space value succeeded, we try to |
1203 | * attach to the owner. If that fails, no harm done, we only | |
1204 | * set the FUTEX_WAITERS bit in the user space variable. | |
1a52084d | 1205 | */ |
af54d6a1 | 1206 | return attach_to_pi_owner(uval, key, ps); |
1a52084d DH |
1207 | } |
1208 | ||
2e12978a LJ |
1209 | /** |
1210 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
1211 | * @q: The futex_q to unqueue | |
1212 | * | |
1213 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
1214 | */ | |
1215 | static void __unqueue_futex(struct futex_q *q) | |
1216 | { | |
1217 | struct futex_hash_bucket *hb; | |
1218 | ||
29096202 SR |
1219 | if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr)) |
1220 | || WARN_ON(plist_node_empty(&q->list))) | |
2e12978a LJ |
1221 | return; |
1222 | ||
1223 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
1224 | plist_del(&q->list, &hb->chain); | |
11d4616b | 1225 | hb_waiters_dec(hb); |
2e12978a LJ |
1226 | } |
1227 | ||
1da177e4 LT |
1228 | /* |
1229 | * The hash bucket lock must be held when this is called. | |
1d0dcb3a DB |
1230 | * Afterwards, the futex_q must not be accessed. Callers |
1231 | * must ensure to later call wake_up_q() for the actual | |
1232 | * wakeups to occur. | |
1da177e4 | 1233 | */ |
1d0dcb3a | 1234 | static void mark_wake_futex(struct wake_q_head *wake_q, struct futex_q *q) |
1da177e4 | 1235 | { |
f1a11e05 TG |
1236 | struct task_struct *p = q->task; |
1237 | ||
aa10990e DH |
1238 | if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n")) |
1239 | return; | |
1240 | ||
1da177e4 | 1241 | /* |
1d0dcb3a DB |
1242 | * Queue the task for later wakeup for after we've released |
1243 | * the hb->lock. wake_q_add() grabs reference to p. | |
1da177e4 | 1244 | */ |
1d0dcb3a | 1245 | wake_q_add(wake_q, p); |
2e12978a | 1246 | __unqueue_futex(q); |
1da177e4 | 1247 | /* |
f1a11e05 TG |
1248 | * The waiting task can free the futex_q as soon as |
1249 | * q->lock_ptr = NULL is written, without taking any locks. A | |
1250 | * memory barrier is required here to prevent the following | |
1251 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 1252 | */ |
ccdea2f8 | 1253 | smp_wmb(); |
1da177e4 LT |
1254 | q->lock_ptr = NULL; |
1255 | } | |
1256 | ||
802ab58d SAS |
1257 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this, |
1258 | struct futex_hash_bucket *hb) | |
c87e2837 IM |
1259 | { |
1260 | struct task_struct *new_owner; | |
1261 | struct futex_pi_state *pi_state = this->pi_state; | |
7cfdaf38 | 1262 | u32 uninitialized_var(curval), newval; |
802ab58d SAS |
1263 | WAKE_Q(wake_q); |
1264 | bool deboost; | |
13fbca4c | 1265 | int ret = 0; |
c87e2837 IM |
1266 | |
1267 | if (!pi_state) | |
1268 | return -EINVAL; | |
1269 | ||
51246bfd TG |
1270 | /* |
1271 | * If current does not own the pi_state then the futex is | |
1272 | * inconsistent and user space fiddled with the futex value. | |
1273 | */ | |
1274 | if (pi_state->owner != current) | |
1275 | return -EINVAL; | |
1276 | ||
b4abf910 | 1277 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
1278 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
1279 | ||
1280 | /* | |
f123c98e SR |
1281 | * It is possible that the next waiter (the one that brought |
1282 | * this owner to the kernel) timed out and is no longer | |
1283 | * waiting on the lock. | |
c87e2837 IM |
1284 | */ |
1285 | if (!new_owner) | |
1286 | new_owner = this->task; | |
1287 | ||
1288 | /* | |
13fbca4c TG |
1289 | * We pass it to the next owner. The WAITERS bit is always |
1290 | * kept enabled while there is PI state around. We cleanup the | |
1291 | * owner died bit, because we are the owner. | |
c87e2837 | 1292 | */ |
13fbca4c | 1293 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 1294 | |
ab51fbab DB |
1295 | if (unlikely(should_fail_futex(true))) |
1296 | ret = -EFAULT; | |
1297 | ||
89e9e66b | 1298 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) { |
13fbca4c | 1299 | ret = -EFAULT; |
89e9e66b SAS |
1300 | } else if (curval != uval) { |
1301 | /* | |
1302 | * If a unconditional UNLOCK_PI operation (user space did not | |
1303 | * try the TID->0 transition) raced with a waiter setting the | |
1304 | * FUTEX_WAITERS flag between get_user() and locking the hash | |
1305 | * bucket lock, retry the operation. | |
1306 | */ | |
1307 | if ((FUTEX_TID_MASK & curval) == uval) | |
1308 | ret = -EAGAIN; | |
1309 | else | |
1310 | ret = -EINVAL; | |
1311 | } | |
13fbca4c | 1312 | if (ret) { |
b4abf910 | 1313 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
13fbca4c | 1314 | return ret; |
e3f2ddea | 1315 | } |
c87e2837 | 1316 | |
b4abf910 | 1317 | raw_spin_lock(&pi_state->owner->pi_lock); |
627371d7 IM |
1318 | WARN_ON(list_empty(&pi_state->list)); |
1319 | list_del_init(&pi_state->list); | |
b4abf910 | 1320 | raw_spin_unlock(&pi_state->owner->pi_lock); |
627371d7 | 1321 | |
b4abf910 | 1322 | raw_spin_lock(&new_owner->pi_lock); |
627371d7 | 1323 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
1324 | list_add(&pi_state->list, &new_owner->pi_state_list); |
1325 | pi_state->owner = new_owner; | |
b4abf910 | 1326 | raw_spin_unlock(&new_owner->pi_lock); |
627371d7 | 1327 | |
b4abf910 | 1328 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
802ab58d SAS |
1329 | |
1330 | deboost = rt_mutex_futex_unlock(&pi_state->pi_mutex, &wake_q); | |
1331 | ||
1332 | /* | |
1333 | * First unlock HB so the waiter does not spin on it once he got woken | |
1334 | * up. Second wake up the waiter before the priority is adjusted. If we | |
1335 | * deboost first (and lose our higher priority), then the task might get | |
1336 | * scheduled away before the wake up can take place. | |
1337 | */ | |
1338 | spin_unlock(&hb->lock); | |
1339 | wake_up_q(&wake_q); | |
1340 | if (deboost) | |
1341 | rt_mutex_adjust_prio(current); | |
c87e2837 IM |
1342 | |
1343 | return 0; | |
1344 | } | |
1345 | ||
8b8f319f IM |
1346 | /* |
1347 | * Express the locking dependencies for lockdep: | |
1348 | */ | |
1349 | static inline void | |
1350 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1351 | { | |
1352 | if (hb1 <= hb2) { | |
1353 | spin_lock(&hb1->lock); | |
1354 | if (hb1 < hb2) | |
1355 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
1356 | } else { /* hb1 > hb2 */ | |
1357 | spin_lock(&hb2->lock); | |
1358 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
1359 | } | |
1360 | } | |
1361 | ||
5eb3dc62 DH |
1362 | static inline void |
1363 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1364 | { | |
f061d351 | 1365 | spin_unlock(&hb1->lock); |
88f502fe IM |
1366 | if (hb1 != hb2) |
1367 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
1368 | } |
1369 | ||
1da177e4 | 1370 | /* |
b2d0994b | 1371 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 1372 | */ |
b41277dc DH |
1373 | static int |
1374 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 1375 | { |
e2970f2f | 1376 | struct futex_hash_bucket *hb; |
1da177e4 | 1377 | struct futex_q *this, *next; |
38d47c1b | 1378 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 | 1379 | int ret; |
1d0dcb3a | 1380 | WAKE_Q(wake_q); |
1da177e4 | 1381 | |
cd689985 TG |
1382 | if (!bitset) |
1383 | return -EINVAL; | |
1384 | ||
9ea71503 | 1385 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ); |
1da177e4 LT |
1386 | if (unlikely(ret != 0)) |
1387 | goto out; | |
1388 | ||
e2970f2f | 1389 | hb = hash_futex(&key); |
b0c29f79 DB |
1390 | |
1391 | /* Make sure we really have tasks to wakeup */ | |
1392 | if (!hb_waiters_pending(hb)) | |
1393 | goto out_put_key; | |
1394 | ||
e2970f2f | 1395 | spin_lock(&hb->lock); |
1da177e4 | 1396 | |
0d00c7b2 | 1397 | plist_for_each_entry_safe(this, next, &hb->chain, list) { |
1da177e4 | 1398 | if (match_futex (&this->key, &key)) { |
52400ba9 | 1399 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
1400 | ret = -EINVAL; |
1401 | break; | |
1402 | } | |
cd689985 TG |
1403 | |
1404 | /* Check if one of the bits is set in both bitsets */ | |
1405 | if (!(this->bitset & bitset)) | |
1406 | continue; | |
1407 | ||
1d0dcb3a | 1408 | mark_wake_futex(&wake_q, this); |
1da177e4 LT |
1409 | if (++ret >= nr_wake) |
1410 | break; | |
1411 | } | |
1412 | } | |
1413 | ||
e2970f2f | 1414 | spin_unlock(&hb->lock); |
1d0dcb3a | 1415 | wake_up_q(&wake_q); |
b0c29f79 | 1416 | out_put_key: |
ae791a2d | 1417 | put_futex_key(&key); |
42d35d48 | 1418 | out: |
1da177e4 LT |
1419 | return ret; |
1420 | } | |
1421 | ||
4732efbe JJ |
1422 | /* |
1423 | * Wake up all waiters hashed on the physical page that is mapped | |
1424 | * to this virtual address: | |
1425 | */ | |
e2970f2f | 1426 | static int |
b41277dc | 1427 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 1428 | int nr_wake, int nr_wake2, int op) |
4732efbe | 1429 | { |
38d47c1b | 1430 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 1431 | struct futex_hash_bucket *hb1, *hb2; |
4732efbe | 1432 | struct futex_q *this, *next; |
e4dc5b7a | 1433 | int ret, op_ret; |
1d0dcb3a | 1434 | WAKE_Q(wake_q); |
4732efbe | 1435 | |
e4dc5b7a | 1436 | retry: |
9ea71503 | 1437 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
4732efbe JJ |
1438 | if (unlikely(ret != 0)) |
1439 | goto out; | |
9ea71503 | 1440 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
4732efbe | 1441 | if (unlikely(ret != 0)) |
42d35d48 | 1442 | goto out_put_key1; |
4732efbe | 1443 | |
e2970f2f IM |
1444 | hb1 = hash_futex(&key1); |
1445 | hb2 = hash_futex(&key2); | |
4732efbe | 1446 | |
e4dc5b7a | 1447 | retry_private: |
eaaea803 | 1448 | double_lock_hb(hb1, hb2); |
e2970f2f | 1449 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1450 | if (unlikely(op_ret < 0)) { |
4732efbe | 1451 | |
5eb3dc62 | 1452 | double_unlock_hb(hb1, hb2); |
4732efbe | 1453 | |
7ee1dd3f | 1454 | #ifndef CONFIG_MMU |
e2970f2f IM |
1455 | /* |
1456 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
1457 | * but we might get them from range checking | |
1458 | */ | |
7ee1dd3f | 1459 | ret = op_ret; |
42d35d48 | 1460 | goto out_put_keys; |
7ee1dd3f DH |
1461 | #endif |
1462 | ||
796f8d9b DG |
1463 | if (unlikely(op_ret != -EFAULT)) { |
1464 | ret = op_ret; | |
42d35d48 | 1465 | goto out_put_keys; |
796f8d9b DG |
1466 | } |
1467 | ||
d0725992 | 1468 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 1469 | if (ret) |
de87fcc1 | 1470 | goto out_put_keys; |
4732efbe | 1471 | |
b41277dc | 1472 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1473 | goto retry_private; |
1474 | ||
ae791a2d TG |
1475 | put_futex_key(&key2); |
1476 | put_futex_key(&key1); | |
e4dc5b7a | 1477 | goto retry; |
4732efbe JJ |
1478 | } |
1479 | ||
0d00c7b2 | 1480 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
4732efbe | 1481 | if (match_futex (&this->key, &key1)) { |
aa10990e DH |
1482 | if (this->pi_state || this->rt_waiter) { |
1483 | ret = -EINVAL; | |
1484 | goto out_unlock; | |
1485 | } | |
1d0dcb3a | 1486 | mark_wake_futex(&wake_q, this); |
4732efbe JJ |
1487 | if (++ret >= nr_wake) |
1488 | break; | |
1489 | } | |
1490 | } | |
1491 | ||
1492 | if (op_ret > 0) { | |
4732efbe | 1493 | op_ret = 0; |
0d00c7b2 | 1494 | plist_for_each_entry_safe(this, next, &hb2->chain, list) { |
4732efbe | 1495 | if (match_futex (&this->key, &key2)) { |
aa10990e DH |
1496 | if (this->pi_state || this->rt_waiter) { |
1497 | ret = -EINVAL; | |
1498 | goto out_unlock; | |
1499 | } | |
1d0dcb3a | 1500 | mark_wake_futex(&wake_q, this); |
4732efbe JJ |
1501 | if (++op_ret >= nr_wake2) |
1502 | break; | |
1503 | } | |
1504 | } | |
1505 | ret += op_ret; | |
1506 | } | |
1507 | ||
aa10990e | 1508 | out_unlock: |
5eb3dc62 | 1509 | double_unlock_hb(hb1, hb2); |
1d0dcb3a | 1510 | wake_up_q(&wake_q); |
42d35d48 | 1511 | out_put_keys: |
ae791a2d | 1512 | put_futex_key(&key2); |
42d35d48 | 1513 | out_put_key1: |
ae791a2d | 1514 | put_futex_key(&key1); |
42d35d48 | 1515 | out: |
4732efbe JJ |
1516 | return ret; |
1517 | } | |
1518 | ||
9121e478 DH |
1519 | /** |
1520 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1521 | * @q: the futex_q to requeue | |
1522 | * @hb1: the source hash_bucket | |
1523 | * @hb2: the target hash_bucket | |
1524 | * @key2: the new key for the requeued futex_q | |
1525 | */ | |
1526 | static inline | |
1527 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1528 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1529 | { | |
1530 | ||
1531 | /* | |
1532 | * If key1 and key2 hash to the same bucket, no need to | |
1533 | * requeue. | |
1534 | */ | |
1535 | if (likely(&hb1->chain != &hb2->chain)) { | |
1536 | plist_del(&q->list, &hb1->chain); | |
11d4616b | 1537 | hb_waiters_dec(hb1); |
11d4616b | 1538 | hb_waiters_inc(hb2); |
fe1bce9e | 1539 | plist_add(&q->list, &hb2->chain); |
9121e478 | 1540 | q->lock_ptr = &hb2->lock; |
9121e478 DH |
1541 | } |
1542 | get_futex_key_refs(key2); | |
1543 | q->key = *key2; | |
1544 | } | |
1545 | ||
52400ba9 DH |
1546 | /** |
1547 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1548 | * @q: the futex_q |
1549 | * @key: the key of the requeue target futex | |
1550 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1551 | * |
1552 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1553 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1554 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1555 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1556 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1557 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1558 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1559 | */ |
1560 | static inline | |
beda2c7e DH |
1561 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1562 | struct futex_hash_bucket *hb) | |
52400ba9 | 1563 | { |
52400ba9 DH |
1564 | get_futex_key_refs(key); |
1565 | q->key = *key; | |
1566 | ||
2e12978a | 1567 | __unqueue_futex(q); |
52400ba9 DH |
1568 | |
1569 | WARN_ON(!q->rt_waiter); | |
1570 | q->rt_waiter = NULL; | |
1571 | ||
beda2c7e | 1572 | q->lock_ptr = &hb->lock; |
beda2c7e | 1573 | |
f1a11e05 | 1574 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1575 | } |
1576 | ||
1577 | /** | |
1578 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1579 | * @pifutex: the user address of the to futex |
1580 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1581 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1582 | * @key1: the from futex key | |
1583 | * @key2: the to futex key | |
1584 | * @ps: address to store the pi_state pointer | |
1585 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1586 | * |
1587 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1588 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1589 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1590 | * hb1 and hb2 must be held by the caller. | |
52400ba9 | 1591 | * |
6c23cbbd RD |
1592 | * Return: |
1593 | * 0 - failed to acquire the lock atomically; | |
866293ee | 1594 | * >0 - acquired the lock, return value is vpid of the top_waiter |
52400ba9 DH |
1595 | * <0 - error |
1596 | */ | |
1597 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1598 | struct futex_hash_bucket *hb1, | |
1599 | struct futex_hash_bucket *hb2, | |
1600 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1601 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1602 | { |
bab5bc9e | 1603 | struct futex_q *top_waiter = NULL; |
52400ba9 | 1604 | u32 curval; |
866293ee | 1605 | int ret, vpid; |
52400ba9 DH |
1606 | |
1607 | if (get_futex_value_locked(&curval, pifutex)) | |
1608 | return -EFAULT; | |
1609 | ||
ab51fbab DB |
1610 | if (unlikely(should_fail_futex(true))) |
1611 | return -EFAULT; | |
1612 | ||
bab5bc9e DH |
1613 | /* |
1614 | * Find the top_waiter and determine if there are additional waiters. | |
1615 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1616 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1617 | * as we have means to handle the possible fault. If not, don't set | |
1618 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1619 | * the kernel. | |
1620 | */ | |
52400ba9 DH |
1621 | top_waiter = futex_top_waiter(hb1, key1); |
1622 | ||
1623 | /* There are no waiters, nothing for us to do. */ | |
1624 | if (!top_waiter) | |
1625 | return 0; | |
1626 | ||
84bc4af5 DH |
1627 | /* Ensure we requeue to the expected futex. */ |
1628 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1629 | return -EINVAL; | |
1630 | ||
52400ba9 | 1631 | /* |
bab5bc9e DH |
1632 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1633 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1634 | * in ps in contended cases. | |
52400ba9 | 1635 | */ |
866293ee | 1636 | vpid = task_pid_vnr(top_waiter->task); |
bab5bc9e DH |
1637 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1638 | set_waiters); | |
866293ee | 1639 | if (ret == 1) { |
beda2c7e | 1640 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
866293ee TG |
1641 | return vpid; |
1642 | } | |
52400ba9 DH |
1643 | return ret; |
1644 | } | |
1645 | ||
1646 | /** | |
1647 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1648 | * @uaddr1: source futex user address |
b41277dc | 1649 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1650 | * @uaddr2: target futex user address |
1651 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1652 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1653 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1654 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1655 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1656 | * |
1657 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1658 | * uaddr2 atomically on behalf of the top waiter. | |
1659 | * | |
6c23cbbd RD |
1660 | * Return: |
1661 | * >=0 - on success, the number of tasks requeued or woken; | |
52400ba9 | 1662 | * <0 - on error |
1da177e4 | 1663 | */ |
b41277dc DH |
1664 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
1665 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
1666 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 1667 | { |
38d47c1b | 1668 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1669 | int drop_count = 0, task_count = 0, ret; |
1670 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1671 | struct futex_hash_bucket *hb1, *hb2; |
1da177e4 | 1672 | struct futex_q *this, *next; |
1d0dcb3a | 1673 | WAKE_Q(wake_q); |
52400ba9 DH |
1674 | |
1675 | if (requeue_pi) { | |
e9c243a5 TG |
1676 | /* |
1677 | * Requeue PI only works on two distinct uaddrs. This | |
1678 | * check is only valid for private futexes. See below. | |
1679 | */ | |
1680 | if (uaddr1 == uaddr2) | |
1681 | return -EINVAL; | |
1682 | ||
52400ba9 DH |
1683 | /* |
1684 | * requeue_pi requires a pi_state, try to allocate it now | |
1685 | * without any locks in case it fails. | |
1686 | */ | |
1687 | if (refill_pi_state_cache()) | |
1688 | return -ENOMEM; | |
1689 | /* | |
1690 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1691 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1692 | * returning to userspace, so as to not leave the rt_mutex with | |
1693 | * waiters and no owner. However, second and third wake-ups | |
1694 | * cannot be predicted as they involve race conditions with the | |
1695 | * first wake and a fault while looking up the pi_state. Both | |
1696 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1697 | * use nr_wake=1. | |
1698 | */ | |
1699 | if (nr_wake != 1) | |
1700 | return -EINVAL; | |
1701 | } | |
1da177e4 | 1702 | |
42d35d48 | 1703 | retry: |
9ea71503 | 1704 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
1da177e4 LT |
1705 | if (unlikely(ret != 0)) |
1706 | goto out; | |
9ea71503 SB |
1707 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, |
1708 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1709 | if (unlikely(ret != 0)) |
42d35d48 | 1710 | goto out_put_key1; |
1da177e4 | 1711 | |
e9c243a5 TG |
1712 | /* |
1713 | * The check above which compares uaddrs is not sufficient for | |
1714 | * shared futexes. We need to compare the keys: | |
1715 | */ | |
1716 | if (requeue_pi && match_futex(&key1, &key2)) { | |
1717 | ret = -EINVAL; | |
1718 | goto out_put_keys; | |
1719 | } | |
1720 | ||
e2970f2f IM |
1721 | hb1 = hash_futex(&key1); |
1722 | hb2 = hash_futex(&key2); | |
1da177e4 | 1723 | |
e4dc5b7a | 1724 | retry_private: |
69cd9eba | 1725 | hb_waiters_inc(hb2); |
8b8f319f | 1726 | double_lock_hb(hb1, hb2); |
1da177e4 | 1727 | |
e2970f2f IM |
1728 | if (likely(cmpval != NULL)) { |
1729 | u32 curval; | |
1da177e4 | 1730 | |
e2970f2f | 1731 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1732 | |
1733 | if (unlikely(ret)) { | |
5eb3dc62 | 1734 | double_unlock_hb(hb1, hb2); |
69cd9eba | 1735 | hb_waiters_dec(hb2); |
1da177e4 | 1736 | |
e2970f2f | 1737 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1738 | if (ret) |
1739 | goto out_put_keys; | |
1da177e4 | 1740 | |
b41277dc | 1741 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 1742 | goto retry_private; |
1da177e4 | 1743 | |
ae791a2d TG |
1744 | put_futex_key(&key2); |
1745 | put_futex_key(&key1); | |
e4dc5b7a | 1746 | goto retry; |
1da177e4 | 1747 | } |
e2970f2f | 1748 | if (curval != *cmpval) { |
1da177e4 LT |
1749 | ret = -EAGAIN; |
1750 | goto out_unlock; | |
1751 | } | |
1752 | } | |
1753 | ||
52400ba9 | 1754 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1755 | /* |
1756 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1757 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1758 | * bit. We force this here where we are able to easily handle | |
1759 | * faults rather in the requeue loop below. | |
1760 | */ | |
52400ba9 | 1761 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1762 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1763 | |
1764 | /* | |
1765 | * At this point the top_waiter has either taken uaddr2 or is | |
1766 | * waiting on it. If the former, then the pi_state will not | |
1767 | * exist yet, look it up one more time to ensure we have a | |
866293ee TG |
1768 | * reference to it. If the lock was taken, ret contains the |
1769 | * vpid of the top waiter task. | |
ecb38b78 TG |
1770 | * If the lock was not taken, we have pi_state and an initial |
1771 | * refcount on it. In case of an error we have nothing. | |
52400ba9 | 1772 | */ |
866293ee | 1773 | if (ret > 0) { |
52400ba9 | 1774 | WARN_ON(pi_state); |
89061d3d | 1775 | drop_count++; |
52400ba9 | 1776 | task_count++; |
866293ee | 1777 | /* |
ecb38b78 TG |
1778 | * If we acquired the lock, then the user space value |
1779 | * of uaddr2 should be vpid. It cannot be changed by | |
1780 | * the top waiter as it is blocked on hb2 lock if it | |
1781 | * tries to do so. If something fiddled with it behind | |
1782 | * our back the pi state lookup might unearth it. So | |
1783 | * we rather use the known value than rereading and | |
1784 | * handing potential crap to lookup_pi_state. | |
1785 | * | |
1786 | * If that call succeeds then we have pi_state and an | |
1787 | * initial refcount on it. | |
866293ee | 1788 | */ |
54a21788 | 1789 | ret = lookup_pi_state(ret, hb2, &key2, &pi_state); |
52400ba9 DH |
1790 | } |
1791 | ||
1792 | switch (ret) { | |
1793 | case 0: | |
ecb38b78 | 1794 | /* We hold a reference on the pi state. */ |
52400ba9 | 1795 | break; |
4959f2de TG |
1796 | |
1797 | /* If the above failed, then pi_state is NULL */ | |
52400ba9 DH |
1798 | case -EFAULT: |
1799 | double_unlock_hb(hb1, hb2); | |
69cd9eba | 1800 | hb_waiters_dec(hb2); |
ae791a2d TG |
1801 | put_futex_key(&key2); |
1802 | put_futex_key(&key1); | |
d0725992 | 1803 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
1804 | if (!ret) |
1805 | goto retry; | |
1806 | goto out; | |
1807 | case -EAGAIN: | |
af54d6a1 TG |
1808 | /* |
1809 | * Two reasons for this: | |
1810 | * - Owner is exiting and we just wait for the | |
1811 | * exit to complete. | |
1812 | * - The user space value changed. | |
1813 | */ | |
52400ba9 | 1814 | double_unlock_hb(hb1, hb2); |
69cd9eba | 1815 | hb_waiters_dec(hb2); |
ae791a2d TG |
1816 | put_futex_key(&key2); |
1817 | put_futex_key(&key1); | |
52400ba9 DH |
1818 | cond_resched(); |
1819 | goto retry; | |
1820 | default: | |
1821 | goto out_unlock; | |
1822 | } | |
1823 | } | |
1824 | ||
0d00c7b2 | 1825 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
52400ba9 DH |
1826 | if (task_count - nr_wake >= nr_requeue) |
1827 | break; | |
1828 | ||
1829 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1830 | continue; |
52400ba9 | 1831 | |
392741e0 DH |
1832 | /* |
1833 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
1834 | * be paired with each other and no other futex ops. | |
aa10990e DH |
1835 | * |
1836 | * We should never be requeueing a futex_q with a pi_state, | |
1837 | * which is awaiting a futex_unlock_pi(). | |
392741e0 DH |
1838 | */ |
1839 | if ((requeue_pi && !this->rt_waiter) || | |
aa10990e DH |
1840 | (!requeue_pi && this->rt_waiter) || |
1841 | this->pi_state) { | |
392741e0 DH |
1842 | ret = -EINVAL; |
1843 | break; | |
1844 | } | |
52400ba9 DH |
1845 | |
1846 | /* | |
1847 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1848 | * lock, we already woke the top_waiter. If not, it will be | |
1849 | * woken by futex_unlock_pi(). | |
1850 | */ | |
1851 | if (++task_count <= nr_wake && !requeue_pi) { | |
1d0dcb3a | 1852 | mark_wake_futex(&wake_q, this); |
52400ba9 DH |
1853 | continue; |
1854 | } | |
1da177e4 | 1855 | |
84bc4af5 DH |
1856 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
1857 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
1858 | ret = -EINVAL; | |
1859 | break; | |
1860 | } | |
1861 | ||
52400ba9 DH |
1862 | /* |
1863 | * Requeue nr_requeue waiters and possibly one more in the case | |
1864 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1865 | */ | |
1866 | if (requeue_pi) { | |
ecb38b78 TG |
1867 | /* |
1868 | * Prepare the waiter to take the rt_mutex. Take a | |
1869 | * refcount on the pi_state and store the pointer in | |
1870 | * the futex_q object of the waiter. | |
1871 | */ | |
52400ba9 DH |
1872 | atomic_inc(&pi_state->refcount); |
1873 | this->pi_state = pi_state; | |
1874 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1875 | this->rt_waiter, | |
c051b21f | 1876 | this->task); |
52400ba9 | 1877 | if (ret == 1) { |
ecb38b78 TG |
1878 | /* |
1879 | * We got the lock. We do neither drop the | |
1880 | * refcount on pi_state nor clear | |
1881 | * this->pi_state because the waiter needs the | |
1882 | * pi_state for cleaning up the user space | |
1883 | * value. It will drop the refcount after | |
1884 | * doing so. | |
1885 | */ | |
beda2c7e | 1886 | requeue_pi_wake_futex(this, &key2, hb2); |
89061d3d | 1887 | drop_count++; |
52400ba9 DH |
1888 | continue; |
1889 | } else if (ret) { | |
ecb38b78 TG |
1890 | /* |
1891 | * rt_mutex_start_proxy_lock() detected a | |
1892 | * potential deadlock when we tried to queue | |
1893 | * that waiter. Drop the pi_state reference | |
1894 | * which we took above and remove the pointer | |
1895 | * to the state from the waiters futex_q | |
1896 | * object. | |
1897 | */ | |
52400ba9 | 1898 | this->pi_state = NULL; |
29e9ee5d | 1899 | put_pi_state(pi_state); |
885c2cb7 TG |
1900 | /* |
1901 | * We stop queueing more waiters and let user | |
1902 | * space deal with the mess. | |
1903 | */ | |
1904 | break; | |
52400ba9 | 1905 | } |
1da177e4 | 1906 | } |
52400ba9 DH |
1907 | requeue_futex(this, hb1, hb2, &key2); |
1908 | drop_count++; | |
1da177e4 LT |
1909 | } |
1910 | ||
ecb38b78 TG |
1911 | /* |
1912 | * We took an extra initial reference to the pi_state either | |
1913 | * in futex_proxy_trylock_atomic() or in lookup_pi_state(). We | |
1914 | * need to drop it here again. | |
1915 | */ | |
29e9ee5d | 1916 | put_pi_state(pi_state); |
885c2cb7 TG |
1917 | |
1918 | out_unlock: | |
5eb3dc62 | 1919 | double_unlock_hb(hb1, hb2); |
1d0dcb3a | 1920 | wake_up_q(&wake_q); |
69cd9eba | 1921 | hb_waiters_dec(hb2); |
1da177e4 | 1922 | |
cd84a42f DH |
1923 | /* |
1924 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1925 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1926 | * one at key2 and updated their key pointer. We no longer need to | |
1927 | * hold the references to key1. | |
1928 | */ | |
1da177e4 | 1929 | while (--drop_count >= 0) |
9adef58b | 1930 | drop_futex_key_refs(&key1); |
1da177e4 | 1931 | |
42d35d48 | 1932 | out_put_keys: |
ae791a2d | 1933 | put_futex_key(&key2); |
42d35d48 | 1934 | out_put_key1: |
ae791a2d | 1935 | put_futex_key(&key1); |
42d35d48 | 1936 | out: |
52400ba9 | 1937 | return ret ? ret : task_count; |
1da177e4 LT |
1938 | } |
1939 | ||
1940 | /* The key must be already stored in q->key. */ | |
82af7aca | 1941 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 1942 | __acquires(&hb->lock) |
1da177e4 | 1943 | { |
e2970f2f | 1944 | struct futex_hash_bucket *hb; |
1da177e4 | 1945 | |
e2970f2f | 1946 | hb = hash_futex(&q->key); |
11d4616b LT |
1947 | |
1948 | /* | |
1949 | * Increment the counter before taking the lock so that | |
1950 | * a potential waker won't miss a to-be-slept task that is | |
1951 | * waiting for the spinlock. This is safe as all queue_lock() | |
1952 | * users end up calling queue_me(). Similarly, for housekeeping, | |
1953 | * decrement the counter at queue_unlock() when some error has | |
1954 | * occurred and we don't end up adding the task to the list. | |
1955 | */ | |
1956 | hb_waiters_inc(hb); | |
1957 | ||
e2970f2f | 1958 | q->lock_ptr = &hb->lock; |
1da177e4 | 1959 | |
8ad7b378 | 1960 | spin_lock(&hb->lock); /* implies smp_mb(); (A) */ |
e2970f2f | 1961 | return hb; |
1da177e4 LT |
1962 | } |
1963 | ||
d40d65c8 | 1964 | static inline void |
0d00c7b2 | 1965 | queue_unlock(struct futex_hash_bucket *hb) |
15e408cd | 1966 | __releases(&hb->lock) |
d40d65c8 DH |
1967 | { |
1968 | spin_unlock(&hb->lock); | |
11d4616b | 1969 | hb_waiters_dec(hb); |
d40d65c8 DH |
1970 | } |
1971 | ||
1972 | /** | |
1973 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
1974 | * @q: The futex_q to enqueue | |
1975 | * @hb: The destination hash bucket | |
1976 | * | |
1977 | * The hb->lock must be held by the caller, and is released here. A call to | |
1978 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
1979 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
1980 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
1981 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
1982 | * an example). | |
1983 | */ | |
82af7aca | 1984 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
15e408cd | 1985 | __releases(&hb->lock) |
1da177e4 | 1986 | { |
ec92d082 PP |
1987 | int prio; |
1988 | ||
1989 | /* | |
1990 | * The priority used to register this element is | |
1991 | * - either the real thread-priority for the real-time threads | |
1992 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
1993 | * - or MAX_RT_PRIO for non-RT threads. | |
1994 | * Thus, all RT-threads are woken first in priority order, and | |
1995 | * the others are woken last, in FIFO order. | |
1996 | */ | |
1997 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1998 | ||
1999 | plist_node_init(&q->list, prio); | |
ec92d082 | 2000 | plist_add(&q->list, &hb->chain); |
c87e2837 | 2001 | q->task = current; |
e2970f2f | 2002 | spin_unlock(&hb->lock); |
1da177e4 LT |
2003 | } |
2004 | ||
d40d65c8 DH |
2005 | /** |
2006 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
2007 | * @q: The futex_q to unqueue | |
2008 | * | |
2009 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
2010 | * be paired with exactly one earlier call to queue_me(). | |
2011 | * | |
6c23cbbd RD |
2012 | * Return: |
2013 | * 1 - if the futex_q was still queued (and we removed unqueued it); | |
d40d65c8 | 2014 | * 0 - if the futex_q was already removed by the waking thread |
1da177e4 | 2015 | */ |
1da177e4 LT |
2016 | static int unqueue_me(struct futex_q *q) |
2017 | { | |
1da177e4 | 2018 | spinlock_t *lock_ptr; |
e2970f2f | 2019 | int ret = 0; |
1da177e4 LT |
2020 | |
2021 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 2022 | retry: |
29b75eb2 JZ |
2023 | /* |
2024 | * q->lock_ptr can change between this read and the following spin_lock. | |
2025 | * Use READ_ONCE to forbid the compiler from reloading q->lock_ptr and | |
2026 | * optimizing lock_ptr out of the logic below. | |
2027 | */ | |
2028 | lock_ptr = READ_ONCE(q->lock_ptr); | |
c80544dc | 2029 | if (lock_ptr != NULL) { |
1da177e4 LT |
2030 | spin_lock(lock_ptr); |
2031 | /* | |
2032 | * q->lock_ptr can change between reading it and | |
2033 | * spin_lock(), causing us to take the wrong lock. This | |
2034 | * corrects the race condition. | |
2035 | * | |
2036 | * Reasoning goes like this: if we have the wrong lock, | |
2037 | * q->lock_ptr must have changed (maybe several times) | |
2038 | * between reading it and the spin_lock(). It can | |
2039 | * change again after the spin_lock() but only if it was | |
2040 | * already changed before the spin_lock(). It cannot, | |
2041 | * however, change back to the original value. Therefore | |
2042 | * we can detect whether we acquired the correct lock. | |
2043 | */ | |
2044 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
2045 | spin_unlock(lock_ptr); | |
2046 | goto retry; | |
2047 | } | |
2e12978a | 2048 | __unqueue_futex(q); |
c87e2837 IM |
2049 | |
2050 | BUG_ON(q->pi_state); | |
2051 | ||
1da177e4 LT |
2052 | spin_unlock(lock_ptr); |
2053 | ret = 1; | |
2054 | } | |
2055 | ||
9adef58b | 2056 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
2057 | return ret; |
2058 | } | |
2059 | ||
c87e2837 IM |
2060 | /* |
2061 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
2062 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
2063 | * and dropped here. | |
c87e2837 | 2064 | */ |
d0aa7a70 | 2065 | static void unqueue_me_pi(struct futex_q *q) |
15e408cd | 2066 | __releases(q->lock_ptr) |
c87e2837 | 2067 | { |
2e12978a | 2068 | __unqueue_futex(q); |
c87e2837 IM |
2069 | |
2070 | BUG_ON(!q->pi_state); | |
29e9ee5d | 2071 | put_pi_state(q->pi_state); |
c87e2837 IM |
2072 | q->pi_state = NULL; |
2073 | ||
d0aa7a70 | 2074 | spin_unlock(q->lock_ptr); |
c87e2837 IM |
2075 | } |
2076 | ||
d0aa7a70 | 2077 | /* |
cdf71a10 | 2078 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 2079 | * |
778e9a9c AK |
2080 | * Must be called with hash bucket lock held and mm->sem held for non |
2081 | * private futexes. | |
d0aa7a70 | 2082 | */ |
778e9a9c | 2083 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
ae791a2d | 2084 | struct task_struct *newowner) |
d0aa7a70 | 2085 | { |
cdf71a10 | 2086 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 2087 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 2088 | struct task_struct *oldowner = pi_state->owner; |
7cfdaf38 | 2089 | u32 uval, uninitialized_var(curval), newval; |
e4dc5b7a | 2090 | int ret; |
d0aa7a70 PP |
2091 | |
2092 | /* Owner died? */ | |
1b7558e4 TG |
2093 | if (!pi_state->owner) |
2094 | newtid |= FUTEX_OWNER_DIED; | |
2095 | ||
2096 | /* | |
2097 | * We are here either because we stole the rtmutex from the | |
8161239a LJ |
2098 | * previous highest priority waiter or we are the highest priority |
2099 | * waiter but failed to get the rtmutex the first time. | |
2100 | * We have to replace the newowner TID in the user space variable. | |
2101 | * This must be atomic as we have to preserve the owner died bit here. | |
1b7558e4 | 2102 | * |
b2d0994b DH |
2103 | * Note: We write the user space value _before_ changing the pi_state |
2104 | * because we can fault here. Imagine swapped out pages or a fork | |
2105 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
2106 | * |
2107 | * Modifying pi_state _before_ the user space value would | |
2108 | * leave the pi_state in an inconsistent state when we fault | |
2109 | * here, because we need to drop the hash bucket lock to | |
2110 | * handle the fault. This might be observed in the PID check | |
2111 | * in lookup_pi_state. | |
2112 | */ | |
2113 | retry: | |
2114 | if (get_futex_value_locked(&uval, uaddr)) | |
2115 | goto handle_fault; | |
2116 | ||
2117 | while (1) { | |
2118 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
2119 | ||
37a9d912 | 2120 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1b7558e4 TG |
2121 | goto handle_fault; |
2122 | if (curval == uval) | |
2123 | break; | |
2124 | uval = curval; | |
2125 | } | |
2126 | ||
2127 | /* | |
2128 | * We fixed up user space. Now we need to fix the pi_state | |
2129 | * itself. | |
2130 | */ | |
d0aa7a70 | 2131 | if (pi_state->owner != NULL) { |
1d615482 | 2132 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
d0aa7a70 PP |
2133 | WARN_ON(list_empty(&pi_state->list)); |
2134 | list_del_init(&pi_state->list); | |
1d615482 | 2135 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
1b7558e4 | 2136 | } |
d0aa7a70 | 2137 | |
cdf71a10 | 2138 | pi_state->owner = newowner; |
d0aa7a70 | 2139 | |
1d615482 | 2140 | raw_spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 2141 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 | 2142 | list_add(&pi_state->list, &newowner->pi_state_list); |
1d615482 | 2143 | raw_spin_unlock_irq(&newowner->pi_lock); |
1b7558e4 | 2144 | return 0; |
d0aa7a70 | 2145 | |
d0aa7a70 | 2146 | /* |
1b7558e4 | 2147 | * To handle the page fault we need to drop the hash bucket |
8161239a LJ |
2148 | * lock here. That gives the other task (either the highest priority |
2149 | * waiter itself or the task which stole the rtmutex) the | |
1b7558e4 TG |
2150 | * chance to try the fixup of the pi_state. So once we are |
2151 | * back from handling the fault we need to check the pi_state | |
2152 | * after reacquiring the hash bucket lock and before trying to | |
2153 | * do another fixup. When the fixup has been done already we | |
2154 | * simply return. | |
d0aa7a70 | 2155 | */ |
1b7558e4 TG |
2156 | handle_fault: |
2157 | spin_unlock(q->lock_ptr); | |
778e9a9c | 2158 | |
d0725992 | 2159 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 2160 | |
1b7558e4 | 2161 | spin_lock(q->lock_ptr); |
778e9a9c | 2162 | |
1b7558e4 TG |
2163 | /* |
2164 | * Check if someone else fixed it for us: | |
2165 | */ | |
2166 | if (pi_state->owner != oldowner) | |
2167 | return 0; | |
2168 | ||
2169 | if (ret) | |
2170 | return ret; | |
2171 | ||
2172 | goto retry; | |
d0aa7a70 PP |
2173 | } |
2174 | ||
72c1bbf3 | 2175 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 2176 | |
dd973998 DH |
2177 | /** |
2178 | * fixup_owner() - Post lock pi_state and corner case management | |
2179 | * @uaddr: user address of the futex | |
dd973998 DH |
2180 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
2181 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
2182 | * | |
2183 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
2184 | * the pi_state owner as well as handle race conditions that may allow us to | |
2185 | * acquire the lock. Must be called with the hb lock held. | |
2186 | * | |
6c23cbbd RD |
2187 | * Return: |
2188 | * 1 - success, lock taken; | |
2189 | * 0 - success, lock not taken; | |
dd973998 DH |
2190 | * <0 - on error (-EFAULT) |
2191 | */ | |
ae791a2d | 2192 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 DH |
2193 | { |
2194 | struct task_struct *owner; | |
2195 | int ret = 0; | |
2196 | ||
2197 | if (locked) { | |
2198 | /* | |
2199 | * Got the lock. We might not be the anticipated owner if we | |
2200 | * did a lock-steal - fix up the PI-state in that case: | |
2201 | */ | |
2202 | if (q->pi_state->owner != current) | |
ae791a2d | 2203 | ret = fixup_pi_state_owner(uaddr, q, current); |
dd973998 DH |
2204 | goto out; |
2205 | } | |
2206 | ||
2207 | /* | |
2208 | * Catch the rare case, where the lock was released when we were on the | |
2209 | * way back before we locked the hash bucket. | |
2210 | */ | |
2211 | if (q->pi_state->owner == current) { | |
2212 | /* | |
2213 | * Try to get the rt_mutex now. This might fail as some other | |
2214 | * task acquired the rt_mutex after we removed ourself from the | |
2215 | * rt_mutex waiters list. | |
2216 | */ | |
2217 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
2218 | locked = 1; | |
2219 | goto out; | |
2220 | } | |
2221 | ||
2222 | /* | |
2223 | * pi_state is incorrect, some other task did a lock steal and | |
2224 | * we returned due to timeout or signal without taking the | |
8161239a | 2225 | * rt_mutex. Too late. |
dd973998 | 2226 | */ |
b4abf910 | 2227 | raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); |
dd973998 | 2228 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); |
8161239a LJ |
2229 | if (!owner) |
2230 | owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); | |
b4abf910 | 2231 | raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); |
ae791a2d | 2232 | ret = fixup_pi_state_owner(uaddr, q, owner); |
dd973998 DH |
2233 | goto out; |
2234 | } | |
2235 | ||
2236 | /* | |
2237 | * Paranoia check. If we did not take the lock, then we should not be | |
8161239a | 2238 | * the owner of the rt_mutex. |
dd973998 DH |
2239 | */ |
2240 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
2241 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
2242 | "pi-state %p\n", ret, | |
2243 | q->pi_state->pi_mutex.owner, | |
2244 | q->pi_state->owner); | |
2245 | ||
2246 | out: | |
2247 | return ret ? ret : locked; | |
2248 | } | |
2249 | ||
ca5f9524 DH |
2250 | /** |
2251 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
2252 | * @hb: the futex hash bucket, must be locked by the caller | |
2253 | * @q: the futex_q to queue up on | |
2254 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
2255 | */ |
2256 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 2257 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 2258 | { |
9beba3c5 DH |
2259 | /* |
2260 | * The task state is guaranteed to be set before another task can | |
b92b8b35 | 2261 | * wake it. set_current_state() is implemented using smp_store_mb() and |
9beba3c5 DH |
2262 | * queue_me() calls spin_unlock() upon completion, both serializing |
2263 | * access to the hash list and forcing another memory barrier. | |
2264 | */ | |
f1a11e05 | 2265 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 2266 | queue_me(q, hb); |
ca5f9524 DH |
2267 | |
2268 | /* Arm the timer */ | |
2e4b0d3f | 2269 | if (timeout) |
ca5f9524 | 2270 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); |
ca5f9524 DH |
2271 | |
2272 | /* | |
0729e196 DH |
2273 | * If we have been removed from the hash list, then another task |
2274 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
2275 | */ |
2276 | if (likely(!plist_node_empty(&q->list))) { | |
2277 | /* | |
2278 | * If the timer has already expired, current will already be | |
2279 | * flagged for rescheduling. Only call schedule if there | |
2280 | * is no timeout, or if it has yet to expire. | |
2281 | */ | |
2282 | if (!timeout || timeout->task) | |
88c8004f | 2283 | freezable_schedule(); |
ca5f9524 DH |
2284 | } |
2285 | __set_current_state(TASK_RUNNING); | |
2286 | } | |
2287 | ||
f801073f DH |
2288 | /** |
2289 | * futex_wait_setup() - Prepare to wait on a futex | |
2290 | * @uaddr: the futex userspace address | |
2291 | * @val: the expected value | |
b41277dc | 2292 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
2293 | * @q: the associated futex_q |
2294 | * @hb: storage for hash_bucket pointer to be returned to caller | |
2295 | * | |
2296 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
2297 | * compare it with the expected value. Handle atomic faults internally. | |
2298 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
2299 | * with no q.key reference on failure. | |
2300 | * | |
6c23cbbd RD |
2301 | * Return: |
2302 | * 0 - uaddr contains val and hb has been locked; | |
ca4a04cf | 2303 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked |
f801073f | 2304 | */ |
b41277dc | 2305 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 2306 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 2307 | { |
e2970f2f IM |
2308 | u32 uval; |
2309 | int ret; | |
1da177e4 | 2310 | |
1da177e4 | 2311 | /* |
b2d0994b | 2312 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
2313 | * Order is important: |
2314 | * | |
2315 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
2316 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
2317 | * | |
2318 | * The basic logical guarantee of a futex is that it blocks ONLY | |
2319 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
2320 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
2321 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
2322 | * cond(var) false, which would violate the guarantee. |
2323 | * | |
8fe8f545 ML |
2324 | * On the other hand, we insert q and release the hash-bucket only |
2325 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
2326 | * absorb a wakeup if *uaddr does not match the desired values | |
2327 | * while the syscall executes. | |
1da177e4 | 2328 | */ |
f801073f | 2329 | retry: |
9ea71503 | 2330 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ); |
f801073f | 2331 | if (unlikely(ret != 0)) |
a5a2a0c7 | 2332 | return ret; |
f801073f DH |
2333 | |
2334 | retry_private: | |
2335 | *hb = queue_lock(q); | |
2336 | ||
e2970f2f | 2337 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 2338 | |
f801073f | 2339 | if (ret) { |
0d00c7b2 | 2340 | queue_unlock(*hb); |
1da177e4 | 2341 | |
e2970f2f | 2342 | ret = get_user(uval, uaddr); |
e4dc5b7a | 2343 | if (ret) |
f801073f | 2344 | goto out; |
1da177e4 | 2345 | |
b41277dc | 2346 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2347 | goto retry_private; |
2348 | ||
ae791a2d | 2349 | put_futex_key(&q->key); |
e4dc5b7a | 2350 | goto retry; |
1da177e4 | 2351 | } |
ca5f9524 | 2352 | |
f801073f | 2353 | if (uval != val) { |
0d00c7b2 | 2354 | queue_unlock(*hb); |
f801073f | 2355 | ret = -EWOULDBLOCK; |
2fff78c7 | 2356 | } |
1da177e4 | 2357 | |
f801073f DH |
2358 | out: |
2359 | if (ret) | |
ae791a2d | 2360 | put_futex_key(&q->key); |
f801073f DH |
2361 | return ret; |
2362 | } | |
2363 | ||
b41277dc DH |
2364 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
2365 | ktime_t *abs_time, u32 bitset) | |
f801073f DH |
2366 | { |
2367 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
2368 | struct restart_block *restart; |
2369 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 2370 | struct futex_q q = futex_q_init; |
f801073f DH |
2371 | int ret; |
2372 | ||
2373 | if (!bitset) | |
2374 | return -EINVAL; | |
f801073f DH |
2375 | q.bitset = bitset; |
2376 | ||
2377 | if (abs_time) { | |
2378 | to = &timeout; | |
2379 | ||
b41277dc DH |
2380 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2381 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2382 | HRTIMER_MODE_ABS); | |
f801073f DH |
2383 | hrtimer_init_sleeper(to, current); |
2384 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2385 | current->timer_slack_ns); | |
2386 | } | |
2387 | ||
d58e6576 | 2388 | retry: |
7ada876a DH |
2389 | /* |
2390 | * Prepare to wait on uaddr. On success, holds hb lock and increments | |
2391 | * q.key refs. | |
2392 | */ | |
b41277dc | 2393 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
2394 | if (ret) |
2395 | goto out; | |
2396 | ||
ca5f9524 | 2397 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 2398 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
2399 | |
2400 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 2401 | ret = 0; |
7ada876a | 2402 | /* unqueue_me() drops q.key ref */ |
1da177e4 | 2403 | if (!unqueue_me(&q)) |
7ada876a | 2404 | goto out; |
2fff78c7 | 2405 | ret = -ETIMEDOUT; |
ca5f9524 | 2406 | if (to && !to->task) |
7ada876a | 2407 | goto out; |
72c1bbf3 | 2408 | |
e2970f2f | 2409 | /* |
d58e6576 TG |
2410 | * We expect signal_pending(current), but we might be the |
2411 | * victim of a spurious wakeup as well. | |
e2970f2f | 2412 | */ |
7ada876a | 2413 | if (!signal_pending(current)) |
d58e6576 | 2414 | goto retry; |
d58e6576 | 2415 | |
2fff78c7 | 2416 | ret = -ERESTARTSYS; |
c19384b5 | 2417 | if (!abs_time) |
7ada876a | 2418 | goto out; |
1da177e4 | 2419 | |
f56141e3 | 2420 | restart = ¤t->restart_block; |
2fff78c7 | 2421 | restart->fn = futex_wait_restart; |
a3c74c52 | 2422 | restart->futex.uaddr = uaddr; |
2fff78c7 PZ |
2423 | restart->futex.val = val; |
2424 | restart->futex.time = abs_time->tv64; | |
2425 | restart->futex.bitset = bitset; | |
0cd9c649 | 2426 | restart->futex.flags = flags | FLAGS_HAS_TIMEOUT; |
42d35d48 | 2427 | |
2fff78c7 PZ |
2428 | ret = -ERESTART_RESTARTBLOCK; |
2429 | ||
42d35d48 | 2430 | out: |
ca5f9524 DH |
2431 | if (to) { |
2432 | hrtimer_cancel(&to->timer); | |
2433 | destroy_hrtimer_on_stack(&to->timer); | |
2434 | } | |
c87e2837 IM |
2435 | return ret; |
2436 | } | |
2437 | ||
72c1bbf3 NP |
2438 | |
2439 | static long futex_wait_restart(struct restart_block *restart) | |
2440 | { | |
a3c74c52 | 2441 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 2442 | ktime_t t, *tp = NULL; |
72c1bbf3 | 2443 | |
a72188d8 DH |
2444 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
2445 | t.tv64 = restart->futex.time; | |
2446 | tp = &t; | |
2447 | } | |
72c1bbf3 | 2448 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
2449 | |
2450 | return (long)futex_wait(uaddr, restart->futex.flags, | |
2451 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
2452 | } |
2453 | ||
2454 | ||
c87e2837 IM |
2455 | /* |
2456 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
2457 | * and failed. The kernel side here does the whole locking operation: | |
767f509c DB |
2458 | * if there are waiters then it will block as a consequence of relying |
2459 | * on rt-mutexes, it does PI, etc. (Due to races the kernel might see | |
2460 | * a 0 value of the futex too.). | |
2461 | * | |
2462 | * Also serves as futex trylock_pi()'ing, and due semantics. | |
c87e2837 | 2463 | */ |
996636dd | 2464 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, |
b41277dc | 2465 | ktime_t *time, int trylock) |
c87e2837 | 2466 | { |
c5780e97 | 2467 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 2468 | struct futex_hash_bucket *hb; |
5bdb05f9 | 2469 | struct futex_q q = futex_q_init; |
dd973998 | 2470 | int res, ret; |
c87e2837 IM |
2471 | |
2472 | if (refill_pi_state_cache()) | |
2473 | return -ENOMEM; | |
2474 | ||
c19384b5 | 2475 | if (time) { |
c5780e97 | 2476 | to = &timeout; |
237fc6e7 TG |
2477 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
2478 | HRTIMER_MODE_ABS); | |
c5780e97 | 2479 | hrtimer_init_sleeper(to, current); |
cc584b21 | 2480 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
2481 | } |
2482 | ||
42d35d48 | 2483 | retry: |
9ea71503 | 2484 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE); |
c87e2837 | 2485 | if (unlikely(ret != 0)) |
42d35d48 | 2486 | goto out; |
c87e2837 | 2487 | |
e4dc5b7a | 2488 | retry_private: |
82af7aca | 2489 | hb = queue_lock(&q); |
c87e2837 | 2490 | |
bab5bc9e | 2491 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 2492 | if (unlikely(ret)) { |
767f509c DB |
2493 | /* |
2494 | * Atomic work succeeded and we got the lock, | |
2495 | * or failed. Either way, we do _not_ block. | |
2496 | */ | |
778e9a9c | 2497 | switch (ret) { |
1a52084d DH |
2498 | case 1: |
2499 | /* We got the lock. */ | |
2500 | ret = 0; | |
2501 | goto out_unlock_put_key; | |
2502 | case -EFAULT: | |
2503 | goto uaddr_faulted; | |
778e9a9c AK |
2504 | case -EAGAIN: |
2505 | /* | |
af54d6a1 TG |
2506 | * Two reasons for this: |
2507 | * - Task is exiting and we just wait for the | |
2508 | * exit to complete. | |
2509 | * - The user space value changed. | |
778e9a9c | 2510 | */ |
0d00c7b2 | 2511 | queue_unlock(hb); |
ae791a2d | 2512 | put_futex_key(&q.key); |
778e9a9c AK |
2513 | cond_resched(); |
2514 | goto retry; | |
778e9a9c | 2515 | default: |
42d35d48 | 2516 | goto out_unlock_put_key; |
c87e2837 | 2517 | } |
c87e2837 IM |
2518 | } |
2519 | ||
2520 | /* | |
2521 | * Only actually queue now that the atomic ops are done: | |
2522 | */ | |
82af7aca | 2523 | queue_me(&q, hb); |
c87e2837 | 2524 | |
c87e2837 IM |
2525 | WARN_ON(!q.pi_state); |
2526 | /* | |
2527 | * Block on the PI mutex: | |
2528 | */ | |
c051b21f TG |
2529 | if (!trylock) { |
2530 | ret = rt_mutex_timed_futex_lock(&q.pi_state->pi_mutex, to); | |
2531 | } else { | |
c87e2837 IM |
2532 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); |
2533 | /* Fixup the trylock return value: */ | |
2534 | ret = ret ? 0 : -EWOULDBLOCK; | |
2535 | } | |
2536 | ||
a99e4e41 | 2537 | spin_lock(q.lock_ptr); |
dd973998 DH |
2538 | /* |
2539 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2540 | * haven't already. | |
2541 | */ | |
ae791a2d | 2542 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 DH |
2543 | /* |
2544 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
2545 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
2546 | */ | |
2547 | if (res) | |
2548 | ret = (res < 0) ? res : 0; | |
c87e2837 | 2549 | |
e8f6386c | 2550 | /* |
dd973998 DH |
2551 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
2552 | * it and return the fault to userspace. | |
e8f6386c DH |
2553 | */ |
2554 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
2555 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
2556 | ||
778e9a9c AK |
2557 | /* Unqueue and drop the lock */ |
2558 | unqueue_me_pi(&q); | |
c87e2837 | 2559 | |
5ecb01cf | 2560 | goto out_put_key; |
c87e2837 | 2561 | |
42d35d48 | 2562 | out_unlock_put_key: |
0d00c7b2 | 2563 | queue_unlock(hb); |
c87e2837 | 2564 | |
42d35d48 | 2565 | out_put_key: |
ae791a2d | 2566 | put_futex_key(&q.key); |
42d35d48 | 2567 | out: |
237fc6e7 TG |
2568 | if (to) |
2569 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 2570 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 2571 | |
42d35d48 | 2572 | uaddr_faulted: |
0d00c7b2 | 2573 | queue_unlock(hb); |
778e9a9c | 2574 | |
d0725992 | 2575 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
2576 | if (ret) |
2577 | goto out_put_key; | |
c87e2837 | 2578 | |
b41277dc | 2579 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2580 | goto retry_private; |
2581 | ||
ae791a2d | 2582 | put_futex_key(&q.key); |
e4dc5b7a | 2583 | goto retry; |
c87e2837 IM |
2584 | } |
2585 | ||
c87e2837 IM |
2586 | /* |
2587 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
2588 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
2589 | * and do the rt-mutex unlock. | |
2590 | */ | |
b41277dc | 2591 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 | 2592 | { |
ccf9e6a8 | 2593 | u32 uninitialized_var(curval), uval, vpid = task_pid_vnr(current); |
38d47c1b | 2594 | union futex_key key = FUTEX_KEY_INIT; |
ccf9e6a8 TG |
2595 | struct futex_hash_bucket *hb; |
2596 | struct futex_q *match; | |
e4dc5b7a | 2597 | int ret; |
c87e2837 IM |
2598 | |
2599 | retry: | |
2600 | if (get_user(uval, uaddr)) | |
2601 | return -EFAULT; | |
2602 | /* | |
2603 | * We release only a lock we actually own: | |
2604 | */ | |
c0c9ed15 | 2605 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 2606 | return -EPERM; |
c87e2837 | 2607 | |
9ea71503 | 2608 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE); |
ccf9e6a8 TG |
2609 | if (ret) |
2610 | return ret; | |
c87e2837 IM |
2611 | |
2612 | hb = hash_futex(&key); | |
2613 | spin_lock(&hb->lock); | |
2614 | ||
c87e2837 | 2615 | /* |
ccf9e6a8 TG |
2616 | * Check waiters first. We do not trust user space values at |
2617 | * all and we at least want to know if user space fiddled | |
2618 | * with the futex value instead of blindly unlocking. | |
c87e2837 | 2619 | */ |
ccf9e6a8 TG |
2620 | match = futex_top_waiter(hb, &key); |
2621 | if (match) { | |
802ab58d SAS |
2622 | ret = wake_futex_pi(uaddr, uval, match, hb); |
2623 | /* | |
2624 | * In case of success wake_futex_pi dropped the hash | |
2625 | * bucket lock. | |
2626 | */ | |
2627 | if (!ret) | |
2628 | goto out_putkey; | |
c87e2837 | 2629 | /* |
ccf9e6a8 TG |
2630 | * The atomic access to the futex value generated a |
2631 | * pagefault, so retry the user-access and the wakeup: | |
c87e2837 IM |
2632 | */ |
2633 | if (ret == -EFAULT) | |
2634 | goto pi_faulted; | |
89e9e66b SAS |
2635 | /* |
2636 | * A unconditional UNLOCK_PI op raced against a waiter | |
2637 | * setting the FUTEX_WAITERS bit. Try again. | |
2638 | */ | |
2639 | if (ret == -EAGAIN) { | |
2640 | spin_unlock(&hb->lock); | |
2641 | put_futex_key(&key); | |
2642 | goto retry; | |
2643 | } | |
802ab58d SAS |
2644 | /* |
2645 | * wake_futex_pi has detected invalid state. Tell user | |
2646 | * space. | |
2647 | */ | |
c87e2837 IM |
2648 | goto out_unlock; |
2649 | } | |
ccf9e6a8 | 2650 | |
c87e2837 | 2651 | /* |
ccf9e6a8 TG |
2652 | * We have no kernel internal state, i.e. no waiters in the |
2653 | * kernel. Waiters which are about to queue themselves are stuck | |
2654 | * on hb->lock. So we can safely ignore them. We do neither | |
2655 | * preserve the WAITERS bit not the OWNER_DIED one. We are the | |
2656 | * owner. | |
c87e2837 | 2657 | */ |
ccf9e6a8 | 2658 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, 0)) |
13fbca4c | 2659 | goto pi_faulted; |
c87e2837 | 2660 | |
ccf9e6a8 TG |
2661 | /* |
2662 | * If uval has changed, let user space handle it. | |
2663 | */ | |
2664 | ret = (curval == uval) ? 0 : -EAGAIN; | |
2665 | ||
c87e2837 IM |
2666 | out_unlock: |
2667 | spin_unlock(&hb->lock); | |
802ab58d | 2668 | out_putkey: |
ae791a2d | 2669 | put_futex_key(&key); |
c87e2837 IM |
2670 | return ret; |
2671 | ||
2672 | pi_faulted: | |
778e9a9c | 2673 | spin_unlock(&hb->lock); |
ae791a2d | 2674 | put_futex_key(&key); |
c87e2837 | 2675 | |
d0725992 | 2676 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2677 | if (!ret) |
c87e2837 IM |
2678 | goto retry; |
2679 | ||
1da177e4 LT |
2680 | return ret; |
2681 | } | |
2682 | ||
52400ba9 DH |
2683 | /** |
2684 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2685 | * @hb: the hash_bucket futex_q was original enqueued on | |
2686 | * @q: the futex_q woken while waiting to be requeued | |
2687 | * @key2: the futex_key of the requeue target futex | |
2688 | * @timeout: the timeout associated with the wait (NULL if none) | |
2689 | * | |
2690 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2691 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2692 | * the wakeup and return the appropriate error code to the caller. Must be | |
2693 | * called with the hb lock held. | |
2694 | * | |
6c23cbbd RD |
2695 | * Return: |
2696 | * 0 = no early wakeup detected; | |
2697 | * <0 = -ETIMEDOUT or -ERESTARTNOINTR | |
52400ba9 DH |
2698 | */ |
2699 | static inline | |
2700 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2701 | struct futex_q *q, union futex_key *key2, | |
2702 | struct hrtimer_sleeper *timeout) | |
2703 | { | |
2704 | int ret = 0; | |
2705 | ||
2706 | /* | |
2707 | * With the hb lock held, we avoid races while we process the wakeup. | |
2708 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2709 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2710 | * It can't be requeued from uaddr2 to something else since we don't | |
2711 | * support a PI aware source futex for requeue. | |
2712 | */ | |
2713 | if (!match_futex(&q->key, key2)) { | |
2714 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2715 | /* | |
2716 | * We were woken prior to requeue by a timeout or a signal. | |
2717 | * Unqueue the futex_q and determine which it was. | |
2718 | */ | |
2e12978a | 2719 | plist_del(&q->list, &hb->chain); |
11d4616b | 2720 | hb_waiters_dec(hb); |
52400ba9 | 2721 | |
d58e6576 | 2722 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 2723 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2724 | if (timeout && !timeout->task) |
2725 | ret = -ETIMEDOUT; | |
d58e6576 | 2726 | else if (signal_pending(current)) |
1c840c14 | 2727 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
2728 | } |
2729 | return ret; | |
2730 | } | |
2731 | ||
2732 | /** | |
2733 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 2734 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 2735 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
ab51fbab | 2736 | * the same type, no requeueing from private to shared, etc. |
52400ba9 DH |
2737 | * @val: the expected value of uaddr |
2738 | * @abs_time: absolute timeout | |
56ec1607 | 2739 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
2740 | * @uaddr2: the pi futex we will take prior to returning to user-space |
2741 | * | |
2742 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
6f7b0a2a DH |
2743 | * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake |
2744 | * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to | |
2745 | * userspace. This ensures the rt_mutex maintains an owner when it has waiters; | |
2746 | * without one, the pi logic would not know which task to boost/deboost, if | |
2747 | * there was a need to. | |
52400ba9 DH |
2748 | * |
2749 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
6c23cbbd | 2750 | * via the following-- |
52400ba9 | 2751 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() |
cc6db4e6 DH |
2752 | * 2) wakeup on uaddr2 after a requeue |
2753 | * 3) signal | |
2754 | * 4) timeout | |
52400ba9 | 2755 | * |
cc6db4e6 | 2756 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
2757 | * |
2758 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2759 | * 5) successful lock | |
2760 | * 6) signal | |
2761 | * 7) timeout | |
2762 | * 8) other lock acquisition failure | |
2763 | * | |
cc6db4e6 | 2764 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
2765 | * |
2766 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2767 | * | |
6c23cbbd RD |
2768 | * Return: |
2769 | * 0 - On success; | |
52400ba9 DH |
2770 | * <0 - On error |
2771 | */ | |
b41277dc | 2772 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 2773 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 2774 | u32 __user *uaddr2) |
52400ba9 DH |
2775 | { |
2776 | struct hrtimer_sleeper timeout, *to = NULL; | |
2777 | struct rt_mutex_waiter rt_waiter; | |
2778 | struct rt_mutex *pi_mutex = NULL; | |
52400ba9 | 2779 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
2780 | union futex_key key2 = FUTEX_KEY_INIT; |
2781 | struct futex_q q = futex_q_init; | |
52400ba9 | 2782 | int res, ret; |
52400ba9 | 2783 | |
6f7b0a2a DH |
2784 | if (uaddr == uaddr2) |
2785 | return -EINVAL; | |
2786 | ||
52400ba9 DH |
2787 | if (!bitset) |
2788 | return -EINVAL; | |
2789 | ||
2790 | if (abs_time) { | |
2791 | to = &timeout; | |
b41277dc DH |
2792 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2793 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2794 | HRTIMER_MODE_ABS); | |
52400ba9 DH |
2795 | hrtimer_init_sleeper(to, current); |
2796 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2797 | current->timer_slack_ns); | |
2798 | } | |
2799 | ||
2800 | /* | |
2801 | * The waiter is allocated on our stack, manipulated by the requeue | |
2802 | * code while we sleep on uaddr. | |
2803 | */ | |
2804 | debug_rt_mutex_init_waiter(&rt_waiter); | |
fb00aca4 PZ |
2805 | RB_CLEAR_NODE(&rt_waiter.pi_tree_entry); |
2806 | RB_CLEAR_NODE(&rt_waiter.tree_entry); | |
52400ba9 DH |
2807 | rt_waiter.task = NULL; |
2808 | ||
9ea71503 | 2809 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
52400ba9 DH |
2810 | if (unlikely(ret != 0)) |
2811 | goto out; | |
2812 | ||
84bc4af5 DH |
2813 | q.bitset = bitset; |
2814 | q.rt_waiter = &rt_waiter; | |
2815 | q.requeue_pi_key = &key2; | |
2816 | ||
7ada876a DH |
2817 | /* |
2818 | * Prepare to wait on uaddr. On success, increments q.key (key1) ref | |
2819 | * count. | |
2820 | */ | |
b41277dc | 2821 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 TG |
2822 | if (ret) |
2823 | goto out_key2; | |
52400ba9 | 2824 | |
e9c243a5 TG |
2825 | /* |
2826 | * The check above which compares uaddrs is not sufficient for | |
2827 | * shared futexes. We need to compare the keys: | |
2828 | */ | |
2829 | if (match_futex(&q.key, &key2)) { | |
13c42c2f | 2830 | queue_unlock(hb); |
e9c243a5 TG |
2831 | ret = -EINVAL; |
2832 | goto out_put_keys; | |
2833 | } | |
2834 | ||
52400ba9 | 2835 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ |
f1a11e05 | 2836 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2837 | |
2838 | spin_lock(&hb->lock); | |
2839 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2840 | spin_unlock(&hb->lock); | |
2841 | if (ret) | |
2842 | goto out_put_keys; | |
2843 | ||
2844 | /* | |
2845 | * In order for us to be here, we know our q.key == key2, and since | |
2846 | * we took the hb->lock above, we also know that futex_requeue() has | |
2847 | * completed and we no longer have to concern ourselves with a wakeup | |
7ada876a DH |
2848 | * race with the atomic proxy lock acquisition by the requeue code. The |
2849 | * futex_requeue dropped our key1 reference and incremented our key2 | |
2850 | * reference count. | |
52400ba9 DH |
2851 | */ |
2852 | ||
2853 | /* Check if the requeue code acquired the second futex for us. */ | |
2854 | if (!q.rt_waiter) { | |
2855 | /* | |
2856 | * Got the lock. We might not be the anticipated owner if we | |
2857 | * did a lock-steal - fix up the PI-state in that case. | |
2858 | */ | |
2859 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2860 | spin_lock(q.lock_ptr); | |
ae791a2d | 2861 | ret = fixup_pi_state_owner(uaddr2, &q, current); |
fb75a428 TG |
2862 | /* |
2863 | * Drop the reference to the pi state which | |
2864 | * the requeue_pi() code acquired for us. | |
2865 | */ | |
29e9ee5d | 2866 | put_pi_state(q.pi_state); |
52400ba9 DH |
2867 | spin_unlock(q.lock_ptr); |
2868 | } | |
2869 | } else { | |
2870 | /* | |
2871 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2872 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2873 | * the pi_state. | |
2874 | */ | |
f27071cb | 2875 | WARN_ON(!q.pi_state); |
52400ba9 | 2876 | pi_mutex = &q.pi_state->pi_mutex; |
c051b21f | 2877 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter); |
52400ba9 DH |
2878 | debug_rt_mutex_free_waiter(&rt_waiter); |
2879 | ||
2880 | spin_lock(q.lock_ptr); | |
2881 | /* | |
2882 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2883 | * haven't already. | |
2884 | */ | |
ae791a2d | 2885 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 DH |
2886 | /* |
2887 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 2888 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
2889 | */ |
2890 | if (res) | |
2891 | ret = (res < 0) ? res : 0; | |
2892 | ||
2893 | /* Unqueue and drop the lock. */ | |
2894 | unqueue_me_pi(&q); | |
2895 | } | |
2896 | ||
2897 | /* | |
2898 | * If fixup_pi_state_owner() faulted and was unable to handle the | |
2899 | * fault, unlock the rt_mutex and return the fault to userspace. | |
2900 | */ | |
2901 | if (ret == -EFAULT) { | |
b6070a8d | 2902 | if (pi_mutex && rt_mutex_owner(pi_mutex) == current) |
52400ba9 DH |
2903 | rt_mutex_unlock(pi_mutex); |
2904 | } else if (ret == -EINTR) { | |
52400ba9 | 2905 | /* |
cc6db4e6 DH |
2906 | * We've already been requeued, but cannot restart by calling |
2907 | * futex_lock_pi() directly. We could restart this syscall, but | |
2908 | * it would detect that the user space "val" changed and return | |
2909 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
2910 | * -EWOULDBLOCK directly. | |
52400ba9 | 2911 | */ |
2070887f | 2912 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2913 | } |
2914 | ||
2915 | out_put_keys: | |
ae791a2d | 2916 | put_futex_key(&q.key); |
c8b15a70 | 2917 | out_key2: |
ae791a2d | 2918 | put_futex_key(&key2); |
52400ba9 DH |
2919 | |
2920 | out: | |
2921 | if (to) { | |
2922 | hrtimer_cancel(&to->timer); | |
2923 | destroy_hrtimer_on_stack(&to->timer); | |
2924 | } | |
2925 | return ret; | |
2926 | } | |
2927 | ||
0771dfef IM |
2928 | /* |
2929 | * Support for robust futexes: the kernel cleans up held futexes at | |
2930 | * thread exit time. | |
2931 | * | |
2932 | * Implementation: user-space maintains a per-thread list of locks it | |
2933 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2934 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2935 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2936 | * always manipulated with the lock held, so the list is private and |
2937 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2938 | * field, to allow the kernel to clean up if the thread dies after | |
2939 | * acquiring the lock, but just before it could have added itself to | |
2940 | * the list. There can only be one such pending lock. | |
2941 | */ | |
2942 | ||
2943 | /** | |
d96ee56c DH |
2944 | * sys_set_robust_list() - Set the robust-futex list head of a task |
2945 | * @head: pointer to the list-head | |
2946 | * @len: length of the list-head, as userspace expects | |
0771dfef | 2947 | */ |
836f92ad HC |
2948 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2949 | size_t, len) | |
0771dfef | 2950 | { |
a0c1e907 TG |
2951 | if (!futex_cmpxchg_enabled) |
2952 | return -ENOSYS; | |
0771dfef IM |
2953 | /* |
2954 | * The kernel knows only one size for now: | |
2955 | */ | |
2956 | if (unlikely(len != sizeof(*head))) | |
2957 | return -EINVAL; | |
2958 | ||
2959 | current->robust_list = head; | |
2960 | ||
2961 | return 0; | |
2962 | } | |
2963 | ||
2964 | /** | |
d96ee56c DH |
2965 | * sys_get_robust_list() - Get the robust-futex list head of a task |
2966 | * @pid: pid of the process [zero for current task] | |
2967 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
2968 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 2969 | */ |
836f92ad HC |
2970 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
2971 | struct robust_list_head __user * __user *, head_ptr, | |
2972 | size_t __user *, len_ptr) | |
0771dfef | 2973 | { |
ba46df98 | 2974 | struct robust_list_head __user *head; |
0771dfef | 2975 | unsigned long ret; |
bdbb776f | 2976 | struct task_struct *p; |
0771dfef | 2977 | |
a0c1e907 TG |
2978 | if (!futex_cmpxchg_enabled) |
2979 | return -ENOSYS; | |
2980 | ||
bdbb776f KC |
2981 | rcu_read_lock(); |
2982 | ||
2983 | ret = -ESRCH; | |
0771dfef | 2984 | if (!pid) |
bdbb776f | 2985 | p = current; |
0771dfef | 2986 | else { |
228ebcbe | 2987 | p = find_task_by_vpid(pid); |
0771dfef IM |
2988 | if (!p) |
2989 | goto err_unlock; | |
0771dfef IM |
2990 | } |
2991 | ||
bdbb776f | 2992 | ret = -EPERM; |
caaee623 | 2993 | if (!ptrace_may_access(p, PTRACE_MODE_READ_REALCREDS)) |
bdbb776f KC |
2994 | goto err_unlock; |
2995 | ||
2996 | head = p->robust_list; | |
2997 | rcu_read_unlock(); | |
2998 | ||
0771dfef IM |
2999 | if (put_user(sizeof(*head), len_ptr)) |
3000 | return -EFAULT; | |
3001 | return put_user(head, head_ptr); | |
3002 | ||
3003 | err_unlock: | |
aaa2a97e | 3004 | rcu_read_unlock(); |
0771dfef IM |
3005 | |
3006 | return ret; | |
3007 | } | |
3008 | ||
3009 | /* | |
3010 | * Process a futex-list entry, check whether it's owned by the | |
3011 | * dying task, and do notification if so: | |
3012 | */ | |
e3f2ddea | 3013 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 3014 | { |
7cfdaf38 | 3015 | u32 uval, uninitialized_var(nval), mval; |
0771dfef | 3016 | |
8f17d3a5 IM |
3017 | retry: |
3018 | if (get_user(uval, uaddr)) | |
0771dfef IM |
3019 | return -1; |
3020 | ||
b488893a | 3021 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
3022 | /* |
3023 | * Ok, this dying thread is truly holding a futex | |
3024 | * of interest. Set the OWNER_DIED bit atomically | |
3025 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
3026 | * set, wake up a waiter (if any). (We have to do a | |
3027 | * futex_wake() even if OWNER_DIED is already set - | |
3028 | * to handle the rare but possible case of recursive | |
3029 | * thread-death.) The rest of the cleanup is done in | |
3030 | * userspace. | |
3031 | */ | |
e3f2ddea | 3032 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
6e0aa9f8 TG |
3033 | /* |
3034 | * We are not holding a lock here, but we want to have | |
3035 | * the pagefault_disable/enable() protection because | |
3036 | * we want to handle the fault gracefully. If the | |
3037 | * access fails we try to fault in the futex with R/W | |
3038 | * verification via get_user_pages. get_user() above | |
3039 | * does not guarantee R/W access. If that fails we | |
3040 | * give up and leave the futex locked. | |
3041 | */ | |
3042 | if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) { | |
3043 | if (fault_in_user_writeable(uaddr)) | |
3044 | return -1; | |
3045 | goto retry; | |
3046 | } | |
c87e2837 | 3047 | if (nval != uval) |
8f17d3a5 | 3048 | goto retry; |
0771dfef | 3049 | |
e3f2ddea IM |
3050 | /* |
3051 | * Wake robust non-PI futexes here. The wakeup of | |
3052 | * PI futexes happens in exit_pi_state(): | |
3053 | */ | |
36cf3b5c | 3054 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 3055 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
3056 | } |
3057 | return 0; | |
3058 | } | |
3059 | ||
e3f2ddea IM |
3060 | /* |
3061 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
3062 | */ | |
3063 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 3064 | struct robust_list __user * __user *head, |
1dcc41bb | 3065 | unsigned int *pi) |
e3f2ddea IM |
3066 | { |
3067 | unsigned long uentry; | |
3068 | ||
ba46df98 | 3069 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
3070 | return -EFAULT; |
3071 | ||
ba46df98 | 3072 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
3073 | *pi = uentry & 1; |
3074 | ||
3075 | return 0; | |
3076 | } | |
3077 | ||
0771dfef IM |
3078 | /* |
3079 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
3080 | * and mark any locks found there dead, and notify any waiters. | |
3081 | * | |
3082 | * We silently return on any sign of list-walking problem. | |
3083 | */ | |
3084 | void exit_robust_list(struct task_struct *curr) | |
3085 | { | |
3086 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 3087 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 DH |
3088 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
3089 | unsigned int uninitialized_var(next_pi); | |
0771dfef | 3090 | unsigned long futex_offset; |
9f96cb1e | 3091 | int rc; |
0771dfef | 3092 | |
a0c1e907 TG |
3093 | if (!futex_cmpxchg_enabled) |
3094 | return; | |
3095 | ||
0771dfef IM |
3096 | /* |
3097 | * Fetch the list head (which was registered earlier, via | |
3098 | * sys_set_robust_list()): | |
3099 | */ | |
e3f2ddea | 3100 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
3101 | return; |
3102 | /* | |
3103 | * Fetch the relative futex offset: | |
3104 | */ | |
3105 | if (get_user(futex_offset, &head->futex_offset)) | |
3106 | return; | |
3107 | /* | |
3108 | * Fetch any possibly pending lock-add first, and handle it | |
3109 | * if it exists: | |
3110 | */ | |
e3f2ddea | 3111 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 3112 | return; |
e3f2ddea | 3113 | |
9f96cb1e | 3114 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 3115 | while (entry != &head->list) { |
9f96cb1e MS |
3116 | /* |
3117 | * Fetch the next entry in the list before calling | |
3118 | * handle_futex_death: | |
3119 | */ | |
3120 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
3121 | /* |
3122 | * A pending lock might already be on the list, so | |
c87e2837 | 3123 | * don't process it twice: |
0771dfef IM |
3124 | */ |
3125 | if (entry != pending) | |
ba46df98 | 3126 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 3127 | curr, pi)) |
0771dfef | 3128 | return; |
9f96cb1e | 3129 | if (rc) |
0771dfef | 3130 | return; |
9f96cb1e MS |
3131 | entry = next_entry; |
3132 | pi = next_pi; | |
0771dfef IM |
3133 | /* |
3134 | * Avoid excessively long or circular lists: | |
3135 | */ | |
3136 | if (!--limit) | |
3137 | break; | |
3138 | ||
3139 | cond_resched(); | |
3140 | } | |
9f96cb1e MS |
3141 | |
3142 | if (pending) | |
3143 | handle_futex_death((void __user *)pending + futex_offset, | |
3144 | curr, pip); | |
0771dfef IM |
3145 | } |
3146 | ||
c19384b5 | 3147 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 3148 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 3149 | { |
81b40539 | 3150 | int cmd = op & FUTEX_CMD_MASK; |
b41277dc | 3151 | unsigned int flags = 0; |
34f01cc1 ED |
3152 | |
3153 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 3154 | flags |= FLAGS_SHARED; |
1da177e4 | 3155 | |
b41277dc DH |
3156 | if (op & FUTEX_CLOCK_REALTIME) { |
3157 | flags |= FLAGS_CLOCKRT; | |
337f1304 DH |
3158 | if (cmd != FUTEX_WAIT && cmd != FUTEX_WAIT_BITSET && \ |
3159 | cmd != FUTEX_WAIT_REQUEUE_PI) | |
b41277dc DH |
3160 | return -ENOSYS; |
3161 | } | |
1da177e4 | 3162 | |
59263b51 TG |
3163 | switch (cmd) { |
3164 | case FUTEX_LOCK_PI: | |
3165 | case FUTEX_UNLOCK_PI: | |
3166 | case FUTEX_TRYLOCK_PI: | |
3167 | case FUTEX_WAIT_REQUEUE_PI: | |
3168 | case FUTEX_CMP_REQUEUE_PI: | |
3169 | if (!futex_cmpxchg_enabled) | |
3170 | return -ENOSYS; | |
3171 | } | |
3172 | ||
34f01cc1 | 3173 | switch (cmd) { |
1da177e4 | 3174 | case FUTEX_WAIT: |
cd689985 TG |
3175 | val3 = FUTEX_BITSET_MATCH_ANY; |
3176 | case FUTEX_WAIT_BITSET: | |
81b40539 | 3177 | return futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 | 3178 | case FUTEX_WAKE: |
cd689985 TG |
3179 | val3 = FUTEX_BITSET_MATCH_ANY; |
3180 | case FUTEX_WAKE_BITSET: | |
81b40539 | 3181 | return futex_wake(uaddr, flags, val, val3); |
1da177e4 | 3182 | case FUTEX_REQUEUE: |
81b40539 | 3183 | return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 | 3184 | case FUTEX_CMP_REQUEUE: |
81b40539 | 3185 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
4732efbe | 3186 | case FUTEX_WAKE_OP: |
81b40539 | 3187 | return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
c87e2837 | 3188 | case FUTEX_LOCK_PI: |
996636dd | 3189 | return futex_lock_pi(uaddr, flags, timeout, 0); |
c87e2837 | 3190 | case FUTEX_UNLOCK_PI: |
81b40539 | 3191 | return futex_unlock_pi(uaddr, flags); |
c87e2837 | 3192 | case FUTEX_TRYLOCK_PI: |
996636dd | 3193 | return futex_lock_pi(uaddr, flags, NULL, 1); |
52400ba9 DH |
3194 | case FUTEX_WAIT_REQUEUE_PI: |
3195 | val3 = FUTEX_BITSET_MATCH_ANY; | |
81b40539 TG |
3196 | return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
3197 | uaddr2); | |
52400ba9 | 3198 | case FUTEX_CMP_REQUEUE_PI: |
81b40539 | 3199 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
1da177e4 | 3200 | } |
81b40539 | 3201 | return -ENOSYS; |
1da177e4 LT |
3202 | } |
3203 | ||
3204 | ||
17da2bd9 HC |
3205 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
3206 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
3207 | u32, val3) | |
1da177e4 | 3208 | { |
c19384b5 PP |
3209 | struct timespec ts; |
3210 | ktime_t t, *tp = NULL; | |
e2970f2f | 3211 | u32 val2 = 0; |
34f01cc1 | 3212 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 3213 | |
cd689985 | 3214 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
3215 | cmd == FUTEX_WAIT_BITSET || |
3216 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
ab51fbab DB |
3217 | if (unlikely(should_fail_futex(!(op & FUTEX_PRIVATE_FLAG)))) |
3218 | return -EFAULT; | |
c19384b5 | 3219 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 3220 | return -EFAULT; |
c19384b5 | 3221 | if (!timespec_valid(&ts)) |
9741ef96 | 3222 | return -EINVAL; |
c19384b5 PP |
3223 | |
3224 | t = timespec_to_ktime(ts); | |
34f01cc1 | 3225 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 3226 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 3227 | tp = &t; |
1da177e4 LT |
3228 | } |
3229 | /* | |
52400ba9 | 3230 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 3231 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 3232 | */ |
f54f0986 | 3233 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 3234 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 3235 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 3236 | |
c19384b5 | 3237 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
3238 | } |
3239 | ||
03b8c7b6 | 3240 | static void __init futex_detect_cmpxchg(void) |
1da177e4 | 3241 | { |
03b8c7b6 | 3242 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 3243 | u32 curval; |
03b8c7b6 HC |
3244 | |
3245 | /* | |
3246 | * This will fail and we want it. Some arch implementations do | |
3247 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
3248 | * functionality. We want to know that before we call in any | |
3249 | * of the complex code paths. Also we want to prevent | |
3250 | * registration of robust lists in that case. NULL is | |
3251 | * guaranteed to fault and we get -EFAULT on functional | |
3252 | * implementation, the non-functional ones will return | |
3253 | * -ENOSYS. | |
3254 | */ | |
3255 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) | |
3256 | futex_cmpxchg_enabled = 1; | |
3257 | #endif | |
3258 | } | |
3259 | ||
3260 | static int __init futex_init(void) | |
3261 | { | |
63b1a816 | 3262 | unsigned int futex_shift; |
a52b89eb DB |
3263 | unsigned long i; |
3264 | ||
3265 | #if CONFIG_BASE_SMALL | |
3266 | futex_hashsize = 16; | |
3267 | #else | |
3268 | futex_hashsize = roundup_pow_of_two(256 * num_possible_cpus()); | |
3269 | #endif | |
3270 | ||
3271 | futex_queues = alloc_large_system_hash("futex", sizeof(*futex_queues), | |
3272 | futex_hashsize, 0, | |
3273 | futex_hashsize < 256 ? HASH_SMALL : 0, | |
63b1a816 HC |
3274 | &futex_shift, NULL, |
3275 | futex_hashsize, futex_hashsize); | |
3276 | futex_hashsize = 1UL << futex_shift; | |
03b8c7b6 HC |
3277 | |
3278 | futex_detect_cmpxchg(); | |
a0c1e907 | 3279 | |
a52b89eb | 3280 | for (i = 0; i < futex_hashsize; i++) { |
11d4616b | 3281 | atomic_set(&futex_queues[i].waiters, 0); |
732375c6 | 3282 | plist_head_init(&futex_queues[i].chain); |
3e4ab747 TG |
3283 | spin_lock_init(&futex_queues[i].lock); |
3284 | } | |
3285 | ||
1da177e4 LT |
3286 | return 0; |
3287 | } | |
f6d107fb | 3288 | __initcall(futex_init); |