]>
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 | */ | |
784bdf3b TG |
182 | #ifdef CONFIG_MMU |
183 | # define FLAGS_SHARED 0x01 | |
184 | #else | |
185 | /* | |
186 | * NOMMU does not have per process address space. Let the compiler optimize | |
187 | * code away. | |
188 | */ | |
189 | # define FLAGS_SHARED 0x00 | |
190 | #endif | |
b41277dc DH |
191 | #define FLAGS_CLOCKRT 0x02 |
192 | #define FLAGS_HAS_TIMEOUT 0x04 | |
193 | ||
c87e2837 IM |
194 | /* |
195 | * Priority Inheritance state: | |
196 | */ | |
197 | struct futex_pi_state { | |
198 | /* | |
199 | * list of 'owned' pi_state instances - these have to be | |
200 | * cleaned up in do_exit() if the task exits prematurely: | |
201 | */ | |
202 | struct list_head list; | |
203 | ||
204 | /* | |
205 | * The PI object: | |
206 | */ | |
207 | struct rt_mutex pi_mutex; | |
208 | ||
209 | struct task_struct *owner; | |
210 | atomic_t refcount; | |
211 | ||
212 | union futex_key key; | |
213 | }; | |
214 | ||
d8d88fbb DH |
215 | /** |
216 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 217 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
218 | * @task: the task waiting on the futex |
219 | * @lock_ptr: the hash bucket lock | |
220 | * @key: the key the futex is hashed on | |
221 | * @pi_state: optional priority inheritance state | |
222 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
223 | * @requeue_pi_key: the requeue_pi target futex key | |
224 | * @bitset: bitset for the optional bitmasked wakeup | |
225 | * | |
226 | * We use this hashed waitqueue, instead of a normal wait_queue_t, so | |
1da177e4 LT |
227 | * we can wake only the relevant ones (hashed queues may be shared). |
228 | * | |
229 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 230 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 231 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
232 | * the second. |
233 | * | |
234 | * PI futexes are typically woken before they are removed from the hash list via | |
235 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
236 | */ |
237 | struct futex_q { | |
ec92d082 | 238 | struct plist_node list; |
1da177e4 | 239 | |
d8d88fbb | 240 | struct task_struct *task; |
1da177e4 | 241 | spinlock_t *lock_ptr; |
1da177e4 | 242 | union futex_key key; |
c87e2837 | 243 | struct futex_pi_state *pi_state; |
52400ba9 | 244 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 245 | union futex_key *requeue_pi_key; |
cd689985 | 246 | u32 bitset; |
1da177e4 LT |
247 | }; |
248 | ||
5bdb05f9 DH |
249 | static const struct futex_q futex_q_init = { |
250 | /* list gets initialized in queue_me()*/ | |
251 | .key = FUTEX_KEY_INIT, | |
252 | .bitset = FUTEX_BITSET_MATCH_ANY | |
253 | }; | |
254 | ||
1da177e4 | 255 | /* |
b2d0994b DH |
256 | * Hash buckets are shared by all the futex_keys that hash to the same |
257 | * location. Each key may have multiple futex_q structures, one for each task | |
258 | * waiting on a futex. | |
1da177e4 LT |
259 | */ |
260 | struct futex_hash_bucket { | |
11d4616b | 261 | atomic_t waiters; |
ec92d082 PP |
262 | spinlock_t lock; |
263 | struct plist_head chain; | |
a52b89eb | 264 | } ____cacheline_aligned_in_smp; |
1da177e4 | 265 | |
ac742d37 RV |
266 | /* |
267 | * The base of the bucket array and its size are always used together | |
268 | * (after initialization only in hash_futex()), so ensure that they | |
269 | * reside in the same cacheline. | |
270 | */ | |
271 | static struct { | |
272 | struct futex_hash_bucket *queues; | |
273 | unsigned long hashsize; | |
274 | } __futex_data __read_mostly __aligned(2*sizeof(long)); | |
275 | #define futex_queues (__futex_data.queues) | |
276 | #define futex_hashsize (__futex_data.hashsize) | |
a52b89eb | 277 | |
1da177e4 | 278 | |
ab51fbab DB |
279 | /* |
280 | * Fault injections for futexes. | |
281 | */ | |
282 | #ifdef CONFIG_FAIL_FUTEX | |
283 | ||
284 | static struct { | |
285 | struct fault_attr attr; | |
286 | ||
621a5f7a | 287 | bool ignore_private; |
ab51fbab DB |
288 | } fail_futex = { |
289 | .attr = FAULT_ATTR_INITIALIZER, | |
621a5f7a | 290 | .ignore_private = false, |
ab51fbab DB |
291 | }; |
292 | ||
293 | static int __init setup_fail_futex(char *str) | |
294 | { | |
295 | return setup_fault_attr(&fail_futex.attr, str); | |
296 | } | |
297 | __setup("fail_futex=", setup_fail_futex); | |
298 | ||
5d285a7f | 299 | static bool should_fail_futex(bool fshared) |
ab51fbab DB |
300 | { |
301 | if (fail_futex.ignore_private && !fshared) | |
302 | return false; | |
303 | ||
304 | return should_fail(&fail_futex.attr, 1); | |
305 | } | |
306 | ||
307 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
308 | ||
309 | static int __init fail_futex_debugfs(void) | |
310 | { | |
311 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
312 | struct dentry *dir; | |
313 | ||
314 | dir = fault_create_debugfs_attr("fail_futex", NULL, | |
315 | &fail_futex.attr); | |
316 | if (IS_ERR(dir)) | |
317 | return PTR_ERR(dir); | |
318 | ||
319 | if (!debugfs_create_bool("ignore-private", mode, dir, | |
320 | &fail_futex.ignore_private)) { | |
321 | debugfs_remove_recursive(dir); | |
322 | return -ENOMEM; | |
323 | } | |
324 | ||
325 | return 0; | |
326 | } | |
327 | ||
328 | late_initcall(fail_futex_debugfs); | |
329 | ||
330 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
331 | ||
332 | #else | |
333 | static inline bool should_fail_futex(bool fshared) | |
334 | { | |
335 | return false; | |
336 | } | |
337 | #endif /* CONFIG_FAIL_FUTEX */ | |
338 | ||
b0c29f79 DB |
339 | static inline void futex_get_mm(union futex_key *key) |
340 | { | |
341 | atomic_inc(&key->private.mm->mm_count); | |
342 | /* | |
343 | * Ensure futex_get_mm() implies a full barrier such that | |
344 | * get_futex_key() implies a full barrier. This is relied upon | |
8ad7b378 | 345 | * as smp_mb(); (B), see the ordering comment above. |
b0c29f79 | 346 | */ |
4e857c58 | 347 | smp_mb__after_atomic(); |
b0c29f79 DB |
348 | } |
349 | ||
11d4616b LT |
350 | /* |
351 | * Reflects a new waiter being added to the waitqueue. | |
352 | */ | |
353 | static inline void hb_waiters_inc(struct futex_hash_bucket *hb) | |
b0c29f79 DB |
354 | { |
355 | #ifdef CONFIG_SMP | |
11d4616b | 356 | atomic_inc(&hb->waiters); |
b0c29f79 | 357 | /* |
11d4616b | 358 | * Full barrier (A), see the ordering comment above. |
b0c29f79 | 359 | */ |
4e857c58 | 360 | smp_mb__after_atomic(); |
11d4616b LT |
361 | #endif |
362 | } | |
363 | ||
364 | /* | |
365 | * Reflects a waiter being removed from the waitqueue by wakeup | |
366 | * paths. | |
367 | */ | |
368 | static inline void hb_waiters_dec(struct futex_hash_bucket *hb) | |
369 | { | |
370 | #ifdef CONFIG_SMP | |
371 | atomic_dec(&hb->waiters); | |
372 | #endif | |
373 | } | |
b0c29f79 | 374 | |
11d4616b LT |
375 | static inline int hb_waiters_pending(struct futex_hash_bucket *hb) |
376 | { | |
377 | #ifdef CONFIG_SMP | |
378 | return atomic_read(&hb->waiters); | |
b0c29f79 | 379 | #else |
11d4616b | 380 | return 1; |
b0c29f79 DB |
381 | #endif |
382 | } | |
383 | ||
e8b61b3f TG |
384 | /** |
385 | * hash_futex - Return the hash bucket in the global hash | |
386 | * @key: Pointer to the futex key for which the hash is calculated | |
387 | * | |
388 | * We hash on the keys returned from get_futex_key (see below) and return the | |
389 | * corresponding hash bucket in the global hash. | |
1da177e4 LT |
390 | */ |
391 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
392 | { | |
393 | u32 hash = jhash2((u32*)&key->both.word, | |
394 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
395 | key->both.offset); | |
a52b89eb | 396 | return &futex_queues[hash & (futex_hashsize - 1)]; |
1da177e4 LT |
397 | } |
398 | ||
e8b61b3f TG |
399 | |
400 | /** | |
401 | * match_futex - Check whether two futex keys are equal | |
402 | * @key1: Pointer to key1 | |
403 | * @key2: Pointer to key2 | |
404 | * | |
1da177e4 LT |
405 | * Return 1 if two futex_keys are equal, 0 otherwise. |
406 | */ | |
407 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
408 | { | |
2bc87203 DH |
409 | return (key1 && key2 |
410 | && key1->both.word == key2->both.word | |
1da177e4 LT |
411 | && key1->both.ptr == key2->both.ptr |
412 | && key1->both.offset == key2->both.offset); | |
413 | } | |
414 | ||
38d47c1b PZ |
415 | /* |
416 | * Take a reference to the resource addressed by a key. | |
417 | * Can be called while holding spinlocks. | |
418 | * | |
419 | */ | |
420 | static void get_futex_key_refs(union futex_key *key) | |
421 | { | |
422 | if (!key->both.ptr) | |
423 | return; | |
424 | ||
784bdf3b TG |
425 | /* |
426 | * On MMU less systems futexes are always "private" as there is no per | |
427 | * process address space. We need the smp wmb nevertheless - yes, | |
428 | * arch/blackfin has MMU less SMP ... | |
429 | */ | |
430 | if (!IS_ENABLED(CONFIG_MMU)) { | |
431 | smp_mb(); /* explicit smp_mb(); (B) */ | |
432 | return; | |
433 | } | |
434 | ||
38d47c1b PZ |
435 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { |
436 | case FUT_OFF_INODE: | |
8ad7b378 | 437 | ihold(key->shared.inode); /* implies smp_mb(); (B) */ |
38d47c1b PZ |
438 | break; |
439 | case FUT_OFF_MMSHARED: | |
8ad7b378 | 440 | futex_get_mm(key); /* implies smp_mb(); (B) */ |
38d47c1b | 441 | break; |
76835b0e | 442 | default: |
993b2ff2 DB |
443 | /* |
444 | * Private futexes do not hold reference on an inode or | |
445 | * mm, therefore the only purpose of calling get_futex_key_refs | |
446 | * is because we need the barrier for the lockless waiter check. | |
447 | */ | |
8ad7b378 | 448 | smp_mb(); /* explicit smp_mb(); (B) */ |
38d47c1b PZ |
449 | } |
450 | } | |
451 | ||
452 | /* | |
453 | * Drop a reference to the resource addressed by a key. | |
993b2ff2 DB |
454 | * The hash bucket spinlock must not be held. This is |
455 | * a no-op for private futexes, see comment in the get | |
456 | * counterpart. | |
38d47c1b PZ |
457 | */ |
458 | static void drop_futex_key_refs(union futex_key *key) | |
459 | { | |
90621c40 DH |
460 | if (!key->both.ptr) { |
461 | /* If we're here then we tried to put a key we failed to get */ | |
462 | WARN_ON_ONCE(1); | |
38d47c1b | 463 | return; |
90621c40 | 464 | } |
38d47c1b | 465 | |
784bdf3b TG |
466 | if (!IS_ENABLED(CONFIG_MMU)) |
467 | return; | |
468 | ||
38d47c1b PZ |
469 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { |
470 | case FUT_OFF_INODE: | |
471 | iput(key->shared.inode); | |
472 | break; | |
473 | case FUT_OFF_MMSHARED: | |
474 | mmdrop(key->private.mm); | |
475 | break; | |
476 | } | |
477 | } | |
478 | ||
34f01cc1 | 479 | /** |
d96ee56c DH |
480 | * get_futex_key() - Get parameters which are the keys for a futex |
481 | * @uaddr: virtual address of the futex | |
482 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED | |
483 | * @key: address where result is stored. | |
9ea71503 SB |
484 | * @rw: mapping needs to be read/write (values: VERIFY_READ, |
485 | * VERIFY_WRITE) | |
34f01cc1 | 486 | * |
6c23cbbd RD |
487 | * Return: a negative error code or 0 |
488 | * | |
34f01cc1 | 489 | * The key words are stored in *key on success. |
1da177e4 | 490 | * |
6131ffaa | 491 | * For shared mappings, it's (page->index, file_inode(vma->vm_file), |
1da177e4 LT |
492 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
493 | * We can usually work out the index without swapping in the page. | |
494 | * | |
b2d0994b | 495 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 496 | */ |
64d1304a | 497 | static int |
9ea71503 | 498 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) |
1da177e4 | 499 | { |
e2970f2f | 500 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 501 | struct mm_struct *mm = current->mm; |
077fa7ae | 502 | struct page *page, *tail; |
14d27abd | 503 | struct address_space *mapping; |
9ea71503 | 504 | int err, ro = 0; |
1da177e4 LT |
505 | |
506 | /* | |
507 | * The futex address must be "naturally" aligned. | |
508 | */ | |
e2970f2f | 509 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 510 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 511 | return -EINVAL; |
e2970f2f | 512 | address -= key->both.offset; |
1da177e4 | 513 | |
5cdec2d8 LT |
514 | if (unlikely(!access_ok(rw, uaddr, sizeof(u32)))) |
515 | return -EFAULT; | |
516 | ||
ab51fbab DB |
517 | if (unlikely(should_fail_futex(fshared))) |
518 | return -EFAULT; | |
519 | ||
34f01cc1 ED |
520 | /* |
521 | * PROCESS_PRIVATE futexes are fast. | |
522 | * As the mm cannot disappear under us and the 'key' only needs | |
523 | * virtual address, we dont even have to find the underlying vma. | |
524 | * Note : We do have to check 'uaddr' is a valid user address, | |
525 | * but access_ok() should be faster than find_vma() | |
526 | */ | |
527 | if (!fshared) { | |
34f01cc1 ED |
528 | key->private.mm = mm; |
529 | key->private.address = address; | |
8ad7b378 | 530 | get_futex_key_refs(key); /* implies smp_mb(); (B) */ |
34f01cc1 ED |
531 | return 0; |
532 | } | |
1da177e4 | 533 | |
38d47c1b | 534 | again: |
ab51fbab DB |
535 | /* Ignore any VERIFY_READ mapping (futex common case) */ |
536 | if (unlikely(should_fail_futex(fshared))) | |
537 | return -EFAULT; | |
538 | ||
7485d0d3 | 539 | err = get_user_pages_fast(address, 1, 1, &page); |
9ea71503 SB |
540 | /* |
541 | * If write access is not required (eg. FUTEX_WAIT), try | |
542 | * and get read-only access. | |
543 | */ | |
544 | if (err == -EFAULT && rw == VERIFY_READ) { | |
545 | err = get_user_pages_fast(address, 1, 0, &page); | |
546 | ro = 1; | |
547 | } | |
38d47c1b PZ |
548 | if (err < 0) |
549 | return err; | |
9ea71503 SB |
550 | else |
551 | err = 0; | |
38d47c1b | 552 | |
65d8fc77 MG |
553 | /* |
554 | * The treatment of mapping from this point on is critical. The page | |
555 | * lock protects many things but in this context the page lock | |
556 | * stabilizes mapping, prevents inode freeing in the shared | |
557 | * file-backed region case and guards against movement to swap cache. | |
558 | * | |
559 | * Strictly speaking the page lock is not needed in all cases being | |
560 | * considered here and page lock forces unnecessarily serialization | |
561 | * From this point on, mapping will be re-verified if necessary and | |
562 | * page lock will be acquired only if it is unavoidable | |
077fa7ae MG |
563 | * |
564 | * Mapping checks require the head page for any compound page so the | |
565 | * head page and mapping is looked up now. For anonymous pages, it | |
566 | * does not matter if the page splits in the future as the key is | |
567 | * based on the address. For filesystem-backed pages, the tail is | |
568 | * required as the index of the page determines the key. For | |
569 | * base pages, there is no tail page and tail == page. | |
65d8fc77 | 570 | */ |
077fa7ae | 571 | tail = page; |
65d8fc77 MG |
572 | page = compound_head(page); |
573 | mapping = READ_ONCE(page->mapping); | |
574 | ||
e6780f72 | 575 | /* |
14d27abd | 576 | * If page->mapping is NULL, then it cannot be a PageAnon |
e6780f72 HD |
577 | * page; but it might be the ZERO_PAGE or in the gate area or |
578 | * in a special mapping (all cases which we are happy to fail); | |
579 | * or it may have been a good file page when get_user_pages_fast | |
580 | * found it, but truncated or holepunched or subjected to | |
581 | * invalidate_complete_page2 before we got the page lock (also | |
582 | * cases which we are happy to fail). And we hold a reference, | |
583 | * so refcount care in invalidate_complete_page's remove_mapping | |
584 | * prevents drop_caches from setting mapping to NULL beneath us. | |
585 | * | |
586 | * The case we do have to guard against is when memory pressure made | |
587 | * shmem_writepage move it from filecache to swapcache beneath us: | |
14d27abd | 588 | * an unlikely race, but we do need to retry for page->mapping. |
e6780f72 | 589 | */ |
65d8fc77 MG |
590 | if (unlikely(!mapping)) { |
591 | int shmem_swizzled; | |
592 | ||
593 | /* | |
594 | * Page lock is required to identify which special case above | |
595 | * applies. If this is really a shmem page then the page lock | |
596 | * will prevent unexpected transitions. | |
597 | */ | |
598 | lock_page(page); | |
599 | shmem_swizzled = PageSwapCache(page) || page->mapping; | |
14d27abd KS |
600 | unlock_page(page); |
601 | put_page(page); | |
65d8fc77 | 602 | |
e6780f72 HD |
603 | if (shmem_swizzled) |
604 | goto again; | |
65d8fc77 | 605 | |
e6780f72 | 606 | return -EFAULT; |
38d47c1b | 607 | } |
1da177e4 LT |
608 | |
609 | /* | |
610 | * Private mappings are handled in a simple way. | |
611 | * | |
65d8fc77 MG |
612 | * If the futex key is stored on an anonymous page, then the associated |
613 | * object is the mm which is implicitly pinned by the calling process. | |
614 | * | |
1da177e4 LT |
615 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if |
616 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 617 | * the object not the particular process. |
1da177e4 | 618 | */ |
14d27abd | 619 | if (PageAnon(page)) { |
9ea71503 SB |
620 | /* |
621 | * A RO anonymous page will never change and thus doesn't make | |
622 | * sense for futex operations. | |
623 | */ | |
ab51fbab | 624 | if (unlikely(should_fail_futex(fshared)) || ro) { |
9ea71503 SB |
625 | err = -EFAULT; |
626 | goto out; | |
627 | } | |
628 | ||
38d47c1b | 629 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 630 | key->private.mm = mm; |
e2970f2f | 631 | key->private.address = address; |
65d8fc77 MG |
632 | |
633 | get_futex_key_refs(key); /* implies smp_mb(); (B) */ | |
634 | ||
38d47c1b | 635 | } else { |
65d8fc77 MG |
636 | struct inode *inode; |
637 | ||
638 | /* | |
639 | * The associated futex object in this case is the inode and | |
640 | * the page->mapping must be traversed. Ordinarily this should | |
641 | * be stabilised under page lock but it's not strictly | |
642 | * necessary in this case as we just want to pin the inode, not | |
643 | * update the radix tree or anything like that. | |
644 | * | |
645 | * The RCU read lock is taken as the inode is finally freed | |
646 | * under RCU. If the mapping still matches expectations then the | |
647 | * mapping->host can be safely accessed as being a valid inode. | |
648 | */ | |
649 | rcu_read_lock(); | |
650 | ||
651 | if (READ_ONCE(page->mapping) != mapping) { | |
652 | rcu_read_unlock(); | |
653 | put_page(page); | |
654 | ||
655 | goto again; | |
656 | } | |
657 | ||
658 | inode = READ_ONCE(mapping->host); | |
659 | if (!inode) { | |
660 | rcu_read_unlock(); | |
661 | put_page(page); | |
662 | ||
663 | goto again; | |
664 | } | |
665 | ||
666 | /* | |
667 | * Take a reference unless it is about to be freed. Previously | |
668 | * this reference was taken by ihold under the page lock | |
669 | * pinning the inode in place so i_lock was unnecessary. The | |
670 | * only way for this check to fail is if the inode was | |
671 | * truncated in parallel so warn for now if this happens. | |
672 | * | |
673 | * We are not calling into get_futex_key_refs() in file-backed | |
674 | * cases, therefore a successful atomic_inc return below will | |
675 | * guarantee that get_futex_key() will still imply smp_mb(); (B). | |
676 | */ | |
677 | if (WARN_ON_ONCE(!atomic_inc_not_zero(&inode->i_count))) { | |
678 | rcu_read_unlock(); | |
679 | put_page(page); | |
680 | ||
681 | goto again; | |
682 | } | |
683 | ||
684 | /* Should be impossible but lets be paranoid for now */ | |
685 | if (WARN_ON_ONCE(inode->i_mapping != mapping)) { | |
686 | err = -EFAULT; | |
687 | rcu_read_unlock(); | |
688 | iput(inode); | |
689 | ||
690 | goto out; | |
691 | } | |
692 | ||
38d47c1b | 693 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ |
65d8fc77 | 694 | key->shared.inode = inode; |
077fa7ae | 695 | key->shared.pgoff = basepage_index(tail); |
65d8fc77 | 696 | rcu_read_unlock(); |
1da177e4 LT |
697 | } |
698 | ||
9ea71503 | 699 | out: |
14d27abd | 700 | put_page(page); |
9ea71503 | 701 | return err; |
1da177e4 LT |
702 | } |
703 | ||
ae791a2d | 704 | static inline void put_futex_key(union futex_key *key) |
1da177e4 | 705 | { |
38d47c1b | 706 | drop_futex_key_refs(key); |
1da177e4 LT |
707 | } |
708 | ||
d96ee56c DH |
709 | /** |
710 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
711 | * @uaddr: pointer to faulting user space address |
712 | * | |
713 | * Slow path to fixup the fault we just took in the atomic write | |
714 | * access to @uaddr. | |
715 | * | |
fb62db2b | 716 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
717 | * user address. We know that we faulted in the atomic pagefault |
718 | * disabled section so we can as well avoid the #PF overhead by | |
719 | * calling get_user_pages() right away. | |
720 | */ | |
721 | static int fault_in_user_writeable(u32 __user *uaddr) | |
722 | { | |
722d0172 AK |
723 | struct mm_struct *mm = current->mm; |
724 | int ret; | |
725 | ||
726 | down_read(&mm->mmap_sem); | |
2efaca92 | 727 | ret = fixup_user_fault(current, mm, (unsigned long)uaddr, |
4a9e1cda | 728 | FAULT_FLAG_WRITE, NULL); |
722d0172 AK |
729 | up_read(&mm->mmap_sem); |
730 | ||
d0725992 TG |
731 | return ret < 0 ? ret : 0; |
732 | } | |
733 | ||
4b1c486b DH |
734 | /** |
735 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
736 | * @hb: the hash bucket the futex_q's reside in |
737 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
738 | * |
739 | * Must be called with the hb lock held. | |
740 | */ | |
741 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
742 | union futex_key *key) | |
743 | { | |
744 | struct futex_q *this; | |
745 | ||
746 | plist_for_each_entry(this, &hb->chain, list) { | |
747 | if (match_futex(&this->key, key)) | |
748 | return this; | |
749 | } | |
750 | return NULL; | |
751 | } | |
752 | ||
37a9d912 ML |
753 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
754 | u32 uval, u32 newval) | |
36cf3b5c | 755 | { |
37a9d912 | 756 | int ret; |
36cf3b5c TG |
757 | |
758 | pagefault_disable(); | |
37a9d912 | 759 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
760 | pagefault_enable(); |
761 | ||
37a9d912 | 762 | return ret; |
36cf3b5c TG |
763 | } |
764 | ||
765 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
766 | { |
767 | int ret; | |
768 | ||
a866374a | 769 | pagefault_disable(); |
bd28b145 | 770 | ret = __get_user(*dest, from); |
a866374a | 771 | pagefault_enable(); |
1da177e4 LT |
772 | |
773 | return ret ? -EFAULT : 0; | |
774 | } | |
775 | ||
c87e2837 IM |
776 | |
777 | /* | |
778 | * PI code: | |
779 | */ | |
780 | static int refill_pi_state_cache(void) | |
781 | { | |
782 | struct futex_pi_state *pi_state; | |
783 | ||
784 | if (likely(current->pi_state_cache)) | |
785 | return 0; | |
786 | ||
4668edc3 | 787 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
788 | |
789 | if (!pi_state) | |
790 | return -ENOMEM; | |
791 | ||
c87e2837 IM |
792 | INIT_LIST_HEAD(&pi_state->list); |
793 | /* pi_mutex gets initialized later */ | |
794 | pi_state->owner = NULL; | |
795 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 796 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
797 | |
798 | current->pi_state_cache = pi_state; | |
799 | ||
800 | return 0; | |
801 | } | |
802 | ||
803 | static struct futex_pi_state * alloc_pi_state(void) | |
804 | { | |
805 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
806 | ||
807 | WARN_ON(!pi_state); | |
808 | current->pi_state_cache = NULL; | |
809 | ||
810 | return pi_state; | |
811 | } | |
812 | ||
30a6b803 | 813 | /* |
29e9ee5d TG |
814 | * Drops a reference to the pi_state object and frees or caches it |
815 | * when the last reference is gone. | |
816 | * | |
30a6b803 BS |
817 | * Must be called with the hb lock held. |
818 | */ | |
29e9ee5d | 819 | static void put_pi_state(struct futex_pi_state *pi_state) |
c87e2837 | 820 | { |
30a6b803 BS |
821 | if (!pi_state) |
822 | return; | |
823 | ||
c87e2837 IM |
824 | if (!atomic_dec_and_test(&pi_state->refcount)) |
825 | return; | |
826 | ||
827 | /* | |
828 | * If pi_state->owner is NULL, the owner is most probably dying | |
829 | * and has cleaned up the pi_state already | |
830 | */ | |
831 | if (pi_state->owner) { | |
1d615482 | 832 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
c87e2837 | 833 | list_del_init(&pi_state->list); |
1d615482 | 834 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
c87e2837 IM |
835 | |
836 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
837 | } | |
838 | ||
839 | if (current->pi_state_cache) | |
840 | kfree(pi_state); | |
841 | else { | |
842 | /* | |
843 | * pi_state->list is already empty. | |
844 | * clear pi_state->owner. | |
845 | * refcount is at 0 - put it back to 1. | |
846 | */ | |
847 | pi_state->owner = NULL; | |
848 | atomic_set(&pi_state->refcount, 1); | |
849 | current->pi_state_cache = pi_state; | |
850 | } | |
851 | } | |
852 | ||
853 | /* | |
854 | * Look up the task based on what TID userspace gave us. | |
855 | * We dont trust it. | |
856 | */ | |
857 | static struct task_struct * futex_find_get_task(pid_t pid) | |
858 | { | |
859 | struct task_struct *p; | |
860 | ||
d359b549 | 861 | rcu_read_lock(); |
228ebcbe | 862 | p = find_task_by_vpid(pid); |
7a0ea09a MH |
863 | if (p) |
864 | get_task_struct(p); | |
a06381fe | 865 | |
d359b549 | 866 | rcu_read_unlock(); |
c87e2837 IM |
867 | |
868 | return p; | |
869 | } | |
870 | ||
871 | /* | |
872 | * This task is holding PI mutexes at exit time => bad. | |
873 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
874 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
875 | */ | |
876 | void exit_pi_state_list(struct task_struct *curr) | |
877 | { | |
c87e2837 IM |
878 | struct list_head *next, *head = &curr->pi_state_list; |
879 | struct futex_pi_state *pi_state; | |
627371d7 | 880 | struct futex_hash_bucket *hb; |
38d47c1b | 881 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 882 | |
a0c1e907 TG |
883 | if (!futex_cmpxchg_enabled) |
884 | return; | |
c87e2837 IM |
885 | /* |
886 | * We are a ZOMBIE and nobody can enqueue itself on | |
887 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 888 | * versus waiters unqueueing themselves: |
c87e2837 | 889 | */ |
1d615482 | 890 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 IM |
891 | while (!list_empty(head)) { |
892 | ||
893 | next = head->next; | |
894 | pi_state = list_entry(next, struct futex_pi_state, list); | |
895 | key = pi_state->key; | |
627371d7 | 896 | hb = hash_futex(&key); |
1d615482 | 897 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 898 | |
c87e2837 IM |
899 | spin_lock(&hb->lock); |
900 | ||
1d615482 | 901 | raw_spin_lock_irq(&curr->pi_lock); |
627371d7 IM |
902 | /* |
903 | * We dropped the pi-lock, so re-check whether this | |
904 | * task still owns the PI-state: | |
905 | */ | |
c87e2837 IM |
906 | if (head->next != next) { |
907 | spin_unlock(&hb->lock); | |
908 | continue; | |
909 | } | |
910 | ||
c87e2837 | 911 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
912 | WARN_ON(list_empty(&pi_state->list)); |
913 | list_del_init(&pi_state->list); | |
c87e2837 | 914 | pi_state->owner = NULL; |
1d615482 | 915 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
916 | |
917 | rt_mutex_unlock(&pi_state->pi_mutex); | |
918 | ||
919 | spin_unlock(&hb->lock); | |
920 | ||
1d615482 | 921 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 922 | } |
1d615482 | 923 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
924 | } |
925 | ||
54a21788 TG |
926 | /* |
927 | * We need to check the following states: | |
928 | * | |
929 | * Waiter | pi_state | pi->owner | uTID | uODIED | ? | |
930 | * | |
931 | * [1] NULL | --- | --- | 0 | 0/1 | Valid | |
932 | * [2] NULL | --- | --- | >0 | 0/1 | Valid | |
933 | * | |
934 | * [3] Found | NULL | -- | Any | 0/1 | Invalid | |
935 | * | |
936 | * [4] Found | Found | NULL | 0 | 1 | Valid | |
937 | * [5] Found | Found | NULL | >0 | 1 | Invalid | |
938 | * | |
939 | * [6] Found | Found | task | 0 | 1 | Valid | |
940 | * | |
941 | * [7] Found | Found | NULL | Any | 0 | Invalid | |
942 | * | |
943 | * [8] Found | Found | task | ==taskTID | 0/1 | Valid | |
944 | * [9] Found | Found | task | 0 | 0 | Invalid | |
945 | * [10] Found | Found | task | !=taskTID | 0/1 | Invalid | |
946 | * | |
947 | * [1] Indicates that the kernel can acquire the futex atomically. We | |
948 | * came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. | |
949 | * | |
950 | * [2] Valid, if TID does not belong to a kernel thread. If no matching | |
951 | * thread is found then it indicates that the owner TID has died. | |
952 | * | |
953 | * [3] Invalid. The waiter is queued on a non PI futex | |
954 | * | |
955 | * [4] Valid state after exit_robust_list(), which sets the user space | |
956 | * value to FUTEX_WAITERS | FUTEX_OWNER_DIED. | |
957 | * | |
958 | * [5] The user space value got manipulated between exit_robust_list() | |
959 | * and exit_pi_state_list() | |
960 | * | |
961 | * [6] Valid state after exit_pi_state_list() which sets the new owner in | |
962 | * the pi_state but cannot access the user space value. | |
963 | * | |
964 | * [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. | |
965 | * | |
966 | * [8] Owner and user space value match | |
967 | * | |
968 | * [9] There is no transient state which sets the user space TID to 0 | |
969 | * except exit_robust_list(), but this is indicated by the | |
970 | * FUTEX_OWNER_DIED bit. See [4] | |
971 | * | |
972 | * [10] There is no transient state which leaves owner and user space | |
973 | * TID out of sync. | |
974 | */ | |
e60cbc5c TG |
975 | |
976 | /* | |
977 | * Validate that the existing waiter has a pi_state and sanity check | |
978 | * the pi_state against the user space value. If correct, attach to | |
979 | * it. | |
980 | */ | |
981 | static int attach_to_pi_state(u32 uval, struct futex_pi_state *pi_state, | |
982 | struct futex_pi_state **ps) | |
c87e2837 | 983 | { |
778e9a9c | 984 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 | 985 | |
e60cbc5c TG |
986 | /* |
987 | * Userspace might have messed up non-PI and PI futexes [3] | |
988 | */ | |
989 | if (unlikely(!pi_state)) | |
990 | return -EINVAL; | |
06a9ec29 | 991 | |
e60cbc5c | 992 | WARN_ON(!atomic_read(&pi_state->refcount)); |
59647b6a | 993 | |
e60cbc5c TG |
994 | /* |
995 | * Handle the owner died case: | |
996 | */ | |
997 | if (uval & FUTEX_OWNER_DIED) { | |
bd1dbcc6 | 998 | /* |
e60cbc5c TG |
999 | * exit_pi_state_list sets owner to NULL and wakes the |
1000 | * topmost waiter. The task which acquires the | |
1001 | * pi_state->rt_mutex will fixup owner. | |
bd1dbcc6 | 1002 | */ |
e60cbc5c | 1003 | if (!pi_state->owner) { |
59647b6a | 1004 | /* |
e60cbc5c TG |
1005 | * No pi state owner, but the user space TID |
1006 | * is not 0. Inconsistent state. [5] | |
59647b6a | 1007 | */ |
e60cbc5c TG |
1008 | if (pid) |
1009 | return -EINVAL; | |
bd1dbcc6 | 1010 | /* |
e60cbc5c | 1011 | * Take a ref on the state and return success. [4] |
866293ee | 1012 | */ |
e60cbc5c | 1013 | goto out_state; |
c87e2837 | 1014 | } |
bd1dbcc6 TG |
1015 | |
1016 | /* | |
e60cbc5c TG |
1017 | * If TID is 0, then either the dying owner has not |
1018 | * yet executed exit_pi_state_list() or some waiter | |
1019 | * acquired the rtmutex in the pi state, but did not | |
1020 | * yet fixup the TID in user space. | |
1021 | * | |
1022 | * Take a ref on the state and return success. [6] | |
1023 | */ | |
1024 | if (!pid) | |
1025 | goto out_state; | |
1026 | } else { | |
1027 | /* | |
1028 | * If the owner died bit is not set, then the pi_state | |
1029 | * must have an owner. [7] | |
bd1dbcc6 | 1030 | */ |
e60cbc5c | 1031 | if (!pi_state->owner) |
bd1dbcc6 | 1032 | return -EINVAL; |
c87e2837 IM |
1033 | } |
1034 | ||
e60cbc5c TG |
1035 | /* |
1036 | * Bail out if user space manipulated the futex value. If pi | |
1037 | * state exists then the owner TID must be the same as the | |
1038 | * user space TID. [9/10] | |
1039 | */ | |
1040 | if (pid != task_pid_vnr(pi_state->owner)) | |
1041 | return -EINVAL; | |
1042 | out_state: | |
1043 | atomic_inc(&pi_state->refcount); | |
1044 | *ps = pi_state; | |
1045 | return 0; | |
1046 | } | |
1047 | ||
04e1b2e5 TG |
1048 | /* |
1049 | * Lookup the task for the TID provided from user space and attach to | |
1050 | * it after doing proper sanity checks. | |
1051 | */ | |
1052 | static int attach_to_pi_owner(u32 uval, union futex_key *key, | |
1053 | struct futex_pi_state **ps) | |
e60cbc5c | 1054 | { |
e60cbc5c | 1055 | pid_t pid = uval & FUTEX_TID_MASK; |
04e1b2e5 TG |
1056 | struct futex_pi_state *pi_state; |
1057 | struct task_struct *p; | |
e60cbc5c | 1058 | |
c87e2837 | 1059 | /* |
e3f2ddea | 1060 | * We are the first waiter - try to look up the real owner and attach |
54a21788 | 1061 | * the new pi_state to it, but bail out when TID = 0 [1] |
c87e2837 | 1062 | */ |
778e9a9c | 1063 | if (!pid) |
e3f2ddea | 1064 | return -ESRCH; |
c87e2837 | 1065 | p = futex_find_get_task(pid); |
7a0ea09a MH |
1066 | if (!p) |
1067 | return -ESRCH; | |
778e9a9c | 1068 | |
a2129464 | 1069 | if (unlikely(p->flags & PF_KTHREAD)) { |
f0d71b3d TG |
1070 | put_task_struct(p); |
1071 | return -EPERM; | |
1072 | } | |
1073 | ||
778e9a9c AK |
1074 | /* |
1075 | * We need to look at the task state flags to figure out, | |
1076 | * whether the task is exiting. To protect against the do_exit | |
1077 | * change of the task flags, we do this protected by | |
1078 | * p->pi_lock: | |
1079 | */ | |
1d615482 | 1080 | raw_spin_lock_irq(&p->pi_lock); |
778e9a9c AK |
1081 | if (unlikely(p->flags & PF_EXITING)) { |
1082 | /* | |
1083 | * The task is on the way out. When PF_EXITPIDONE is | |
1084 | * set, we know that the task has finished the | |
1085 | * cleanup: | |
1086 | */ | |
1087 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
1088 | ||
1d615482 | 1089 | raw_spin_unlock_irq(&p->pi_lock); |
778e9a9c AK |
1090 | put_task_struct(p); |
1091 | return ret; | |
1092 | } | |
c87e2837 | 1093 | |
54a21788 TG |
1094 | /* |
1095 | * No existing pi state. First waiter. [2] | |
1096 | */ | |
c87e2837 IM |
1097 | pi_state = alloc_pi_state(); |
1098 | ||
1099 | /* | |
04e1b2e5 | 1100 | * Initialize the pi_mutex in locked state and make @p |
c87e2837 IM |
1101 | * the owner of it: |
1102 | */ | |
1103 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
1104 | ||
1105 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 1106 | pi_state->key = *key; |
c87e2837 | 1107 | |
627371d7 | 1108 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
1109 | list_add(&pi_state->list, &p->pi_state_list); |
1110 | pi_state->owner = p; | |
1d615482 | 1111 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
1112 | |
1113 | put_task_struct(p); | |
1114 | ||
d0aa7a70 | 1115 | *ps = pi_state; |
c87e2837 IM |
1116 | |
1117 | return 0; | |
1118 | } | |
1119 | ||
04e1b2e5 TG |
1120 | static int lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
1121 | union futex_key *key, struct futex_pi_state **ps) | |
1122 | { | |
1123 | struct futex_q *match = futex_top_waiter(hb, key); | |
1124 | ||
1125 | /* | |
1126 | * If there is a waiter on that futex, validate it and | |
1127 | * attach to the pi_state when the validation succeeds. | |
1128 | */ | |
1129 | if (match) | |
1130 | return attach_to_pi_state(uval, match->pi_state, ps); | |
1131 | ||
1132 | /* | |
1133 | * We are the first waiter - try to look up the owner based on | |
1134 | * @uval and attach to it. | |
1135 | */ | |
1136 | return attach_to_pi_owner(uval, key, ps); | |
1137 | } | |
1138 | ||
af54d6a1 TG |
1139 | static int lock_pi_update_atomic(u32 __user *uaddr, u32 uval, u32 newval) |
1140 | { | |
1141 | u32 uninitialized_var(curval); | |
1142 | ||
ab51fbab DB |
1143 | if (unlikely(should_fail_futex(true))) |
1144 | return -EFAULT; | |
1145 | ||
af54d6a1 TG |
1146 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))) |
1147 | return -EFAULT; | |
1148 | ||
1149 | /*If user space value changed, let the caller retry */ | |
1150 | return curval != uval ? -EAGAIN : 0; | |
1151 | } | |
1152 | ||
1a52084d | 1153 | /** |
d96ee56c | 1154 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
1155 | * @uaddr: the pi futex user address |
1156 | * @hb: the pi futex hash bucket | |
1157 | * @key: the futex key associated with uaddr and hb | |
1158 | * @ps: the pi_state pointer where we store the result of the | |
1159 | * lookup | |
1160 | * @task: the task to perform the atomic lock work for. This will | |
1161 | * be "current" except in the case of requeue pi. | |
1162 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d | 1163 | * |
6c23cbbd RD |
1164 | * Return: |
1165 | * 0 - ready to wait; | |
1166 | * 1 - acquired the lock; | |
1a52084d DH |
1167 | * <0 - error |
1168 | * | |
1169 | * The hb->lock and futex_key refs shall be held by the caller. | |
1170 | */ | |
1171 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
1172 | union futex_key *key, | |
1173 | struct futex_pi_state **ps, | |
bab5bc9e | 1174 | struct task_struct *task, int set_waiters) |
1a52084d | 1175 | { |
af54d6a1 TG |
1176 | u32 uval, newval, vpid = task_pid_vnr(task); |
1177 | struct futex_q *match; | |
1178 | int ret; | |
1a52084d DH |
1179 | |
1180 | /* | |
af54d6a1 TG |
1181 | * Read the user space value first so we can validate a few |
1182 | * things before proceeding further. | |
1a52084d | 1183 | */ |
af54d6a1 | 1184 | if (get_futex_value_locked(&uval, uaddr)) |
1a52084d DH |
1185 | return -EFAULT; |
1186 | ||
ab51fbab DB |
1187 | if (unlikely(should_fail_futex(true))) |
1188 | return -EFAULT; | |
1189 | ||
1a52084d DH |
1190 | /* |
1191 | * Detect deadlocks. | |
1192 | */ | |
af54d6a1 | 1193 | if ((unlikely((uval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
1194 | return -EDEADLK; |
1195 | ||
ab51fbab DB |
1196 | if ((unlikely(should_fail_futex(true)))) |
1197 | return -EDEADLK; | |
1198 | ||
1a52084d | 1199 | /* |
af54d6a1 TG |
1200 | * Lookup existing state first. If it exists, try to attach to |
1201 | * its pi_state. | |
1a52084d | 1202 | */ |
af54d6a1 TG |
1203 | match = futex_top_waiter(hb, key); |
1204 | if (match) | |
1205 | return attach_to_pi_state(uval, match->pi_state, ps); | |
1a52084d DH |
1206 | |
1207 | /* | |
af54d6a1 TG |
1208 | * No waiter and user TID is 0. We are here because the |
1209 | * waiters or the owner died bit is set or called from | |
1210 | * requeue_cmp_pi or for whatever reason something took the | |
1211 | * syscall. | |
1a52084d | 1212 | */ |
af54d6a1 | 1213 | if (!(uval & FUTEX_TID_MASK)) { |
59fa6245 | 1214 | /* |
af54d6a1 TG |
1215 | * We take over the futex. No other waiters and the user space |
1216 | * TID is 0. We preserve the owner died bit. | |
59fa6245 | 1217 | */ |
af54d6a1 TG |
1218 | newval = uval & FUTEX_OWNER_DIED; |
1219 | newval |= vpid; | |
1a52084d | 1220 | |
af54d6a1 TG |
1221 | /* The futex requeue_pi code can enforce the waiters bit */ |
1222 | if (set_waiters) | |
1223 | newval |= FUTEX_WAITERS; | |
1224 | ||
1225 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1226 | /* If the take over worked, return 1 */ | |
1227 | return ret < 0 ? ret : 1; | |
1228 | } | |
1a52084d DH |
1229 | |
1230 | /* | |
af54d6a1 TG |
1231 | * First waiter. Set the waiters bit before attaching ourself to |
1232 | * the owner. If owner tries to unlock, it will be forced into | |
1233 | * the kernel and blocked on hb->lock. | |
1a52084d | 1234 | */ |
af54d6a1 TG |
1235 | newval = uval | FUTEX_WAITERS; |
1236 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1237 | if (ret) | |
1238 | return ret; | |
1a52084d | 1239 | /* |
af54d6a1 TG |
1240 | * If the update of the user space value succeeded, we try to |
1241 | * attach to the owner. If that fails, no harm done, we only | |
1242 | * set the FUTEX_WAITERS bit in the user space variable. | |
1a52084d | 1243 | */ |
af54d6a1 | 1244 | return attach_to_pi_owner(uval, key, ps); |
1a52084d DH |
1245 | } |
1246 | ||
2e12978a LJ |
1247 | /** |
1248 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
1249 | * @q: The futex_q to unqueue | |
1250 | * | |
1251 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
1252 | */ | |
1253 | static void __unqueue_futex(struct futex_q *q) | |
1254 | { | |
1255 | struct futex_hash_bucket *hb; | |
1256 | ||
29096202 SR |
1257 | if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr)) |
1258 | || WARN_ON(plist_node_empty(&q->list))) | |
2e12978a LJ |
1259 | return; |
1260 | ||
1261 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
1262 | plist_del(&q->list, &hb->chain); | |
11d4616b | 1263 | hb_waiters_dec(hb); |
2e12978a LJ |
1264 | } |
1265 | ||
1da177e4 LT |
1266 | /* |
1267 | * The hash bucket lock must be held when this is called. | |
1d0dcb3a DB |
1268 | * Afterwards, the futex_q must not be accessed. Callers |
1269 | * must ensure to later call wake_up_q() for the actual | |
1270 | * wakeups to occur. | |
1da177e4 | 1271 | */ |
1d0dcb3a | 1272 | static void mark_wake_futex(struct wake_q_head *wake_q, struct futex_q *q) |
1da177e4 | 1273 | { |
f1a11e05 TG |
1274 | struct task_struct *p = q->task; |
1275 | ||
aa10990e DH |
1276 | if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n")) |
1277 | return; | |
1278 | ||
1da177e4 | 1279 | /* |
1d0dcb3a DB |
1280 | * Queue the task for later wakeup for after we've released |
1281 | * the hb->lock. wake_q_add() grabs reference to p. | |
1da177e4 | 1282 | */ |
1d0dcb3a | 1283 | wake_q_add(wake_q, p); |
2e12978a | 1284 | __unqueue_futex(q); |
1da177e4 | 1285 | /* |
f1a11e05 TG |
1286 | * The waiting task can free the futex_q as soon as |
1287 | * q->lock_ptr = NULL is written, without taking any locks. A | |
1288 | * memory barrier is required here to prevent the following | |
1289 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 1290 | */ |
ccdea2f8 | 1291 | smp_wmb(); |
1da177e4 LT |
1292 | q->lock_ptr = NULL; |
1293 | } | |
1294 | ||
802ab58d SAS |
1295 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this, |
1296 | struct futex_hash_bucket *hb) | |
c87e2837 IM |
1297 | { |
1298 | struct task_struct *new_owner; | |
1299 | struct futex_pi_state *pi_state = this->pi_state; | |
7cfdaf38 | 1300 | u32 uninitialized_var(curval), newval; |
194a6b5b | 1301 | DEFINE_WAKE_Q(wake_q); |
802ab58d | 1302 | bool deboost; |
13fbca4c | 1303 | int ret = 0; |
c87e2837 IM |
1304 | |
1305 | if (!pi_state) | |
1306 | return -EINVAL; | |
1307 | ||
51246bfd TG |
1308 | /* |
1309 | * If current does not own the pi_state then the futex is | |
1310 | * inconsistent and user space fiddled with the futex value. | |
1311 | */ | |
1312 | if (pi_state->owner != current) | |
1313 | return -EINVAL; | |
1314 | ||
b4abf910 | 1315 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
1316 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
1317 | ||
1318 | /* | |
f123c98e SR |
1319 | * It is possible that the next waiter (the one that brought |
1320 | * this owner to the kernel) timed out and is no longer | |
1321 | * waiting on the lock. | |
c87e2837 IM |
1322 | */ |
1323 | if (!new_owner) | |
1324 | new_owner = this->task; | |
1325 | ||
1326 | /* | |
13fbca4c TG |
1327 | * We pass it to the next owner. The WAITERS bit is always |
1328 | * kept enabled while there is PI state around. We cleanup the | |
1329 | * owner died bit, because we are the owner. | |
c87e2837 | 1330 | */ |
13fbca4c | 1331 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 1332 | |
ab51fbab DB |
1333 | if (unlikely(should_fail_futex(true))) |
1334 | ret = -EFAULT; | |
1335 | ||
89e9e66b | 1336 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) { |
13fbca4c | 1337 | ret = -EFAULT; |
89e9e66b SAS |
1338 | } else if (curval != uval) { |
1339 | /* | |
1340 | * If a unconditional UNLOCK_PI operation (user space did not | |
1341 | * try the TID->0 transition) raced with a waiter setting the | |
1342 | * FUTEX_WAITERS flag between get_user() and locking the hash | |
1343 | * bucket lock, retry the operation. | |
1344 | */ | |
1345 | if ((FUTEX_TID_MASK & curval) == uval) | |
1346 | ret = -EAGAIN; | |
1347 | else | |
1348 | ret = -EINVAL; | |
1349 | } | |
13fbca4c | 1350 | if (ret) { |
b4abf910 | 1351 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
13fbca4c | 1352 | return ret; |
e3f2ddea | 1353 | } |
c87e2837 | 1354 | |
b4abf910 | 1355 | raw_spin_lock(&pi_state->owner->pi_lock); |
627371d7 IM |
1356 | WARN_ON(list_empty(&pi_state->list)); |
1357 | list_del_init(&pi_state->list); | |
b4abf910 | 1358 | raw_spin_unlock(&pi_state->owner->pi_lock); |
627371d7 | 1359 | |
b4abf910 | 1360 | raw_spin_lock(&new_owner->pi_lock); |
627371d7 | 1361 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
1362 | list_add(&pi_state->list, &new_owner->pi_state_list); |
1363 | pi_state->owner = new_owner; | |
b4abf910 | 1364 | raw_spin_unlock(&new_owner->pi_lock); |
627371d7 | 1365 | |
b4abf910 | 1366 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
802ab58d SAS |
1367 | |
1368 | deboost = rt_mutex_futex_unlock(&pi_state->pi_mutex, &wake_q); | |
1369 | ||
1370 | /* | |
1371 | * First unlock HB so the waiter does not spin on it once he got woken | |
1372 | * up. Second wake up the waiter before the priority is adjusted. If we | |
1373 | * deboost first (and lose our higher priority), then the task might get | |
1374 | * scheduled away before the wake up can take place. | |
1375 | */ | |
1376 | spin_unlock(&hb->lock); | |
1377 | wake_up_q(&wake_q); | |
1378 | if (deboost) | |
1379 | rt_mutex_adjust_prio(current); | |
c87e2837 IM |
1380 | |
1381 | return 0; | |
1382 | } | |
1383 | ||
8b8f319f IM |
1384 | /* |
1385 | * Express the locking dependencies for lockdep: | |
1386 | */ | |
1387 | static inline void | |
1388 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1389 | { | |
1390 | if (hb1 <= hb2) { | |
1391 | spin_lock(&hb1->lock); | |
1392 | if (hb1 < hb2) | |
1393 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
1394 | } else { /* hb1 > hb2 */ | |
1395 | spin_lock(&hb2->lock); | |
1396 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
1397 | } | |
1398 | } | |
1399 | ||
5eb3dc62 DH |
1400 | static inline void |
1401 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1402 | { | |
f061d351 | 1403 | spin_unlock(&hb1->lock); |
88f502fe IM |
1404 | if (hb1 != hb2) |
1405 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
1406 | } |
1407 | ||
1da177e4 | 1408 | /* |
b2d0994b | 1409 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 1410 | */ |
b41277dc DH |
1411 | static int |
1412 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 1413 | { |
e2970f2f | 1414 | struct futex_hash_bucket *hb; |
1da177e4 | 1415 | struct futex_q *this, *next; |
38d47c1b | 1416 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 | 1417 | int ret; |
194a6b5b | 1418 | DEFINE_WAKE_Q(wake_q); |
1da177e4 | 1419 | |
cd689985 TG |
1420 | if (!bitset) |
1421 | return -EINVAL; | |
1422 | ||
9ea71503 | 1423 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ); |
1da177e4 LT |
1424 | if (unlikely(ret != 0)) |
1425 | goto out; | |
1426 | ||
e2970f2f | 1427 | hb = hash_futex(&key); |
b0c29f79 DB |
1428 | |
1429 | /* Make sure we really have tasks to wakeup */ | |
1430 | if (!hb_waiters_pending(hb)) | |
1431 | goto out_put_key; | |
1432 | ||
e2970f2f | 1433 | spin_lock(&hb->lock); |
1da177e4 | 1434 | |
0d00c7b2 | 1435 | plist_for_each_entry_safe(this, next, &hb->chain, list) { |
1da177e4 | 1436 | if (match_futex (&this->key, &key)) { |
52400ba9 | 1437 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
1438 | ret = -EINVAL; |
1439 | break; | |
1440 | } | |
cd689985 TG |
1441 | |
1442 | /* Check if one of the bits is set in both bitsets */ | |
1443 | if (!(this->bitset & bitset)) | |
1444 | continue; | |
1445 | ||
1d0dcb3a | 1446 | mark_wake_futex(&wake_q, this); |
1da177e4 LT |
1447 | if (++ret >= nr_wake) |
1448 | break; | |
1449 | } | |
1450 | } | |
1451 | ||
e2970f2f | 1452 | spin_unlock(&hb->lock); |
1d0dcb3a | 1453 | wake_up_q(&wake_q); |
b0c29f79 | 1454 | out_put_key: |
ae791a2d | 1455 | put_futex_key(&key); |
42d35d48 | 1456 | out: |
1da177e4 LT |
1457 | return ret; |
1458 | } | |
1459 | ||
4732efbe JJ |
1460 | /* |
1461 | * Wake up all waiters hashed on the physical page that is mapped | |
1462 | * to this virtual address: | |
1463 | */ | |
e2970f2f | 1464 | static int |
b41277dc | 1465 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 1466 | int nr_wake, int nr_wake2, int op) |
4732efbe | 1467 | { |
38d47c1b | 1468 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 1469 | struct futex_hash_bucket *hb1, *hb2; |
4732efbe | 1470 | struct futex_q *this, *next; |
e4dc5b7a | 1471 | int ret, op_ret; |
194a6b5b | 1472 | DEFINE_WAKE_Q(wake_q); |
4732efbe | 1473 | |
e4dc5b7a | 1474 | retry: |
9ea71503 | 1475 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
4732efbe JJ |
1476 | if (unlikely(ret != 0)) |
1477 | goto out; | |
9ea71503 | 1478 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
4732efbe | 1479 | if (unlikely(ret != 0)) |
42d35d48 | 1480 | goto out_put_key1; |
4732efbe | 1481 | |
e2970f2f IM |
1482 | hb1 = hash_futex(&key1); |
1483 | hb2 = hash_futex(&key2); | |
4732efbe | 1484 | |
e4dc5b7a | 1485 | retry_private: |
eaaea803 | 1486 | double_lock_hb(hb1, hb2); |
e2970f2f | 1487 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1488 | if (unlikely(op_ret < 0)) { |
4732efbe | 1489 | |
5eb3dc62 | 1490 | double_unlock_hb(hb1, hb2); |
4732efbe | 1491 | |
7ee1dd3f | 1492 | #ifndef CONFIG_MMU |
e2970f2f IM |
1493 | /* |
1494 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
1495 | * but we might get them from range checking | |
1496 | */ | |
7ee1dd3f | 1497 | ret = op_ret; |
42d35d48 | 1498 | goto out_put_keys; |
7ee1dd3f DH |
1499 | #endif |
1500 | ||
796f8d9b DG |
1501 | if (unlikely(op_ret != -EFAULT)) { |
1502 | ret = op_ret; | |
42d35d48 | 1503 | goto out_put_keys; |
796f8d9b DG |
1504 | } |
1505 | ||
d0725992 | 1506 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 1507 | if (ret) |
de87fcc1 | 1508 | goto out_put_keys; |
4732efbe | 1509 | |
b41277dc | 1510 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1511 | goto retry_private; |
1512 | ||
ae791a2d TG |
1513 | put_futex_key(&key2); |
1514 | put_futex_key(&key1); | |
e4dc5b7a | 1515 | goto retry; |
4732efbe JJ |
1516 | } |
1517 | ||
0d00c7b2 | 1518 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
4732efbe | 1519 | if (match_futex (&this->key, &key1)) { |
aa10990e DH |
1520 | if (this->pi_state || this->rt_waiter) { |
1521 | ret = -EINVAL; | |
1522 | goto out_unlock; | |
1523 | } | |
1d0dcb3a | 1524 | mark_wake_futex(&wake_q, this); |
4732efbe JJ |
1525 | if (++ret >= nr_wake) |
1526 | break; | |
1527 | } | |
1528 | } | |
1529 | ||
1530 | if (op_ret > 0) { | |
4732efbe | 1531 | op_ret = 0; |
0d00c7b2 | 1532 | plist_for_each_entry_safe(this, next, &hb2->chain, list) { |
4732efbe | 1533 | if (match_futex (&this->key, &key2)) { |
aa10990e DH |
1534 | if (this->pi_state || this->rt_waiter) { |
1535 | ret = -EINVAL; | |
1536 | goto out_unlock; | |
1537 | } | |
1d0dcb3a | 1538 | mark_wake_futex(&wake_q, this); |
4732efbe JJ |
1539 | if (++op_ret >= nr_wake2) |
1540 | break; | |
1541 | } | |
1542 | } | |
1543 | ret += op_ret; | |
1544 | } | |
1545 | ||
aa10990e | 1546 | out_unlock: |
5eb3dc62 | 1547 | double_unlock_hb(hb1, hb2); |
1d0dcb3a | 1548 | wake_up_q(&wake_q); |
42d35d48 | 1549 | out_put_keys: |
ae791a2d | 1550 | put_futex_key(&key2); |
42d35d48 | 1551 | out_put_key1: |
ae791a2d | 1552 | put_futex_key(&key1); |
42d35d48 | 1553 | out: |
4732efbe JJ |
1554 | return ret; |
1555 | } | |
1556 | ||
9121e478 DH |
1557 | /** |
1558 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1559 | * @q: the futex_q to requeue | |
1560 | * @hb1: the source hash_bucket | |
1561 | * @hb2: the target hash_bucket | |
1562 | * @key2: the new key for the requeued futex_q | |
1563 | */ | |
1564 | static inline | |
1565 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1566 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1567 | { | |
1568 | ||
1569 | /* | |
1570 | * If key1 and key2 hash to the same bucket, no need to | |
1571 | * requeue. | |
1572 | */ | |
1573 | if (likely(&hb1->chain != &hb2->chain)) { | |
1574 | plist_del(&q->list, &hb1->chain); | |
11d4616b | 1575 | hb_waiters_dec(hb1); |
11d4616b | 1576 | hb_waiters_inc(hb2); |
fe1bce9e | 1577 | plist_add(&q->list, &hb2->chain); |
9121e478 | 1578 | q->lock_ptr = &hb2->lock; |
9121e478 DH |
1579 | } |
1580 | get_futex_key_refs(key2); | |
1581 | q->key = *key2; | |
1582 | } | |
1583 | ||
52400ba9 DH |
1584 | /** |
1585 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1586 | * @q: the futex_q |
1587 | * @key: the key of the requeue target futex | |
1588 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1589 | * |
1590 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1591 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1592 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1593 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1594 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1595 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1596 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1597 | */ |
1598 | static inline | |
beda2c7e DH |
1599 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1600 | struct futex_hash_bucket *hb) | |
52400ba9 | 1601 | { |
52400ba9 DH |
1602 | get_futex_key_refs(key); |
1603 | q->key = *key; | |
1604 | ||
2e12978a | 1605 | __unqueue_futex(q); |
52400ba9 DH |
1606 | |
1607 | WARN_ON(!q->rt_waiter); | |
1608 | q->rt_waiter = NULL; | |
1609 | ||
beda2c7e | 1610 | q->lock_ptr = &hb->lock; |
beda2c7e | 1611 | |
f1a11e05 | 1612 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1613 | } |
1614 | ||
1615 | /** | |
1616 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1617 | * @pifutex: the user address of the to futex |
1618 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1619 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1620 | * @key1: the from futex key | |
1621 | * @key2: the to futex key | |
1622 | * @ps: address to store the pi_state pointer | |
1623 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1624 | * |
1625 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1626 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1627 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1628 | * hb1 and hb2 must be held by the caller. | |
52400ba9 | 1629 | * |
6c23cbbd RD |
1630 | * Return: |
1631 | * 0 - failed to acquire the lock atomically; | |
866293ee | 1632 | * >0 - acquired the lock, return value is vpid of the top_waiter |
52400ba9 DH |
1633 | * <0 - error |
1634 | */ | |
1635 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1636 | struct futex_hash_bucket *hb1, | |
1637 | struct futex_hash_bucket *hb2, | |
1638 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1639 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1640 | { |
bab5bc9e | 1641 | struct futex_q *top_waiter = NULL; |
52400ba9 | 1642 | u32 curval; |
866293ee | 1643 | int ret, vpid; |
52400ba9 DH |
1644 | |
1645 | if (get_futex_value_locked(&curval, pifutex)) | |
1646 | return -EFAULT; | |
1647 | ||
ab51fbab DB |
1648 | if (unlikely(should_fail_futex(true))) |
1649 | return -EFAULT; | |
1650 | ||
bab5bc9e DH |
1651 | /* |
1652 | * Find the top_waiter and determine if there are additional waiters. | |
1653 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1654 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1655 | * as we have means to handle the possible fault. If not, don't set | |
1656 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1657 | * the kernel. | |
1658 | */ | |
52400ba9 DH |
1659 | top_waiter = futex_top_waiter(hb1, key1); |
1660 | ||
1661 | /* There are no waiters, nothing for us to do. */ | |
1662 | if (!top_waiter) | |
1663 | return 0; | |
1664 | ||
84bc4af5 DH |
1665 | /* Ensure we requeue to the expected futex. */ |
1666 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1667 | return -EINVAL; | |
1668 | ||
52400ba9 | 1669 | /* |
bab5bc9e DH |
1670 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1671 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1672 | * in ps in contended cases. | |
52400ba9 | 1673 | */ |
866293ee | 1674 | vpid = task_pid_vnr(top_waiter->task); |
bab5bc9e DH |
1675 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1676 | set_waiters); | |
866293ee | 1677 | if (ret == 1) { |
beda2c7e | 1678 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
866293ee TG |
1679 | return vpid; |
1680 | } | |
52400ba9 DH |
1681 | return ret; |
1682 | } | |
1683 | ||
1684 | /** | |
1685 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1686 | * @uaddr1: source futex user address |
b41277dc | 1687 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1688 | * @uaddr2: target futex user address |
1689 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1690 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1691 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1692 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1693 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1694 | * |
1695 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1696 | * uaddr2 atomically on behalf of the top waiter. | |
1697 | * | |
6c23cbbd RD |
1698 | * Return: |
1699 | * >=0 - on success, the number of tasks requeued or woken; | |
52400ba9 | 1700 | * <0 - on error |
1da177e4 | 1701 | */ |
b41277dc DH |
1702 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
1703 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
1704 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 1705 | { |
38d47c1b | 1706 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1707 | int drop_count = 0, task_count = 0, ret; |
1708 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1709 | struct futex_hash_bucket *hb1, *hb2; |
1da177e4 | 1710 | struct futex_q *this, *next; |
194a6b5b | 1711 | DEFINE_WAKE_Q(wake_q); |
52400ba9 DH |
1712 | |
1713 | if (requeue_pi) { | |
e9c243a5 TG |
1714 | /* |
1715 | * Requeue PI only works on two distinct uaddrs. This | |
1716 | * check is only valid for private futexes. See below. | |
1717 | */ | |
1718 | if (uaddr1 == uaddr2) | |
1719 | return -EINVAL; | |
1720 | ||
52400ba9 DH |
1721 | /* |
1722 | * requeue_pi requires a pi_state, try to allocate it now | |
1723 | * without any locks in case it fails. | |
1724 | */ | |
1725 | if (refill_pi_state_cache()) | |
1726 | return -ENOMEM; | |
1727 | /* | |
1728 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1729 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1730 | * returning to userspace, so as to not leave the rt_mutex with | |
1731 | * waiters and no owner. However, second and third wake-ups | |
1732 | * cannot be predicted as they involve race conditions with the | |
1733 | * first wake and a fault while looking up the pi_state. Both | |
1734 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1735 | * use nr_wake=1. | |
1736 | */ | |
1737 | if (nr_wake != 1) | |
1738 | return -EINVAL; | |
1739 | } | |
1da177e4 | 1740 | |
42d35d48 | 1741 | retry: |
9ea71503 | 1742 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
1da177e4 LT |
1743 | if (unlikely(ret != 0)) |
1744 | goto out; | |
9ea71503 SB |
1745 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, |
1746 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1747 | if (unlikely(ret != 0)) |
42d35d48 | 1748 | goto out_put_key1; |
1da177e4 | 1749 | |
e9c243a5 TG |
1750 | /* |
1751 | * The check above which compares uaddrs is not sufficient for | |
1752 | * shared futexes. We need to compare the keys: | |
1753 | */ | |
1754 | if (requeue_pi && match_futex(&key1, &key2)) { | |
1755 | ret = -EINVAL; | |
1756 | goto out_put_keys; | |
1757 | } | |
1758 | ||
e2970f2f IM |
1759 | hb1 = hash_futex(&key1); |
1760 | hb2 = hash_futex(&key2); | |
1da177e4 | 1761 | |
e4dc5b7a | 1762 | retry_private: |
69cd9eba | 1763 | hb_waiters_inc(hb2); |
8b8f319f | 1764 | double_lock_hb(hb1, hb2); |
1da177e4 | 1765 | |
e2970f2f IM |
1766 | if (likely(cmpval != NULL)) { |
1767 | u32 curval; | |
1da177e4 | 1768 | |
e2970f2f | 1769 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1770 | |
1771 | if (unlikely(ret)) { | |
5eb3dc62 | 1772 | double_unlock_hb(hb1, hb2); |
69cd9eba | 1773 | hb_waiters_dec(hb2); |
1da177e4 | 1774 | |
e2970f2f | 1775 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1776 | if (ret) |
1777 | goto out_put_keys; | |
1da177e4 | 1778 | |
b41277dc | 1779 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 1780 | goto retry_private; |
1da177e4 | 1781 | |
ae791a2d TG |
1782 | put_futex_key(&key2); |
1783 | put_futex_key(&key1); | |
e4dc5b7a | 1784 | goto retry; |
1da177e4 | 1785 | } |
e2970f2f | 1786 | if (curval != *cmpval) { |
1da177e4 LT |
1787 | ret = -EAGAIN; |
1788 | goto out_unlock; | |
1789 | } | |
1790 | } | |
1791 | ||
52400ba9 | 1792 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1793 | /* |
1794 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1795 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1796 | * bit. We force this here where we are able to easily handle | |
1797 | * faults rather in the requeue loop below. | |
1798 | */ | |
52400ba9 | 1799 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1800 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1801 | |
1802 | /* | |
1803 | * At this point the top_waiter has either taken uaddr2 or is | |
1804 | * waiting on it. If the former, then the pi_state will not | |
1805 | * exist yet, look it up one more time to ensure we have a | |
866293ee TG |
1806 | * reference to it. If the lock was taken, ret contains the |
1807 | * vpid of the top waiter task. | |
ecb38b78 TG |
1808 | * If the lock was not taken, we have pi_state and an initial |
1809 | * refcount on it. In case of an error we have nothing. | |
52400ba9 | 1810 | */ |
866293ee | 1811 | if (ret > 0) { |
52400ba9 | 1812 | WARN_ON(pi_state); |
89061d3d | 1813 | drop_count++; |
52400ba9 | 1814 | task_count++; |
866293ee | 1815 | /* |
ecb38b78 TG |
1816 | * If we acquired the lock, then the user space value |
1817 | * of uaddr2 should be vpid. It cannot be changed by | |
1818 | * the top waiter as it is blocked on hb2 lock if it | |
1819 | * tries to do so. If something fiddled with it behind | |
1820 | * our back the pi state lookup might unearth it. So | |
1821 | * we rather use the known value than rereading and | |
1822 | * handing potential crap to lookup_pi_state. | |
1823 | * | |
1824 | * If that call succeeds then we have pi_state and an | |
1825 | * initial refcount on it. | |
866293ee | 1826 | */ |
54a21788 | 1827 | ret = lookup_pi_state(ret, hb2, &key2, &pi_state); |
52400ba9 DH |
1828 | } |
1829 | ||
1830 | switch (ret) { | |
1831 | case 0: | |
ecb38b78 | 1832 | /* We hold a reference on the pi state. */ |
52400ba9 | 1833 | break; |
4959f2de TG |
1834 | |
1835 | /* If the above failed, then pi_state is NULL */ | |
52400ba9 DH |
1836 | case -EFAULT: |
1837 | double_unlock_hb(hb1, hb2); | |
69cd9eba | 1838 | hb_waiters_dec(hb2); |
ae791a2d TG |
1839 | put_futex_key(&key2); |
1840 | put_futex_key(&key1); | |
d0725992 | 1841 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
1842 | if (!ret) |
1843 | goto retry; | |
1844 | goto out; | |
1845 | case -EAGAIN: | |
af54d6a1 TG |
1846 | /* |
1847 | * Two reasons for this: | |
1848 | * - Owner is exiting and we just wait for the | |
1849 | * exit to complete. | |
1850 | * - The user space value changed. | |
1851 | */ | |
52400ba9 | 1852 | double_unlock_hb(hb1, hb2); |
69cd9eba | 1853 | hb_waiters_dec(hb2); |
ae791a2d TG |
1854 | put_futex_key(&key2); |
1855 | put_futex_key(&key1); | |
52400ba9 DH |
1856 | cond_resched(); |
1857 | goto retry; | |
1858 | default: | |
1859 | goto out_unlock; | |
1860 | } | |
1861 | } | |
1862 | ||
0d00c7b2 | 1863 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
52400ba9 DH |
1864 | if (task_count - nr_wake >= nr_requeue) |
1865 | break; | |
1866 | ||
1867 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1868 | continue; |
52400ba9 | 1869 | |
392741e0 DH |
1870 | /* |
1871 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
1872 | * be paired with each other and no other futex ops. | |
aa10990e DH |
1873 | * |
1874 | * We should never be requeueing a futex_q with a pi_state, | |
1875 | * which is awaiting a futex_unlock_pi(). | |
392741e0 DH |
1876 | */ |
1877 | if ((requeue_pi && !this->rt_waiter) || | |
aa10990e DH |
1878 | (!requeue_pi && this->rt_waiter) || |
1879 | this->pi_state) { | |
392741e0 DH |
1880 | ret = -EINVAL; |
1881 | break; | |
1882 | } | |
52400ba9 DH |
1883 | |
1884 | /* | |
1885 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1886 | * lock, we already woke the top_waiter. If not, it will be | |
1887 | * woken by futex_unlock_pi(). | |
1888 | */ | |
1889 | if (++task_count <= nr_wake && !requeue_pi) { | |
1d0dcb3a | 1890 | mark_wake_futex(&wake_q, this); |
52400ba9 DH |
1891 | continue; |
1892 | } | |
1da177e4 | 1893 | |
84bc4af5 DH |
1894 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
1895 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
1896 | ret = -EINVAL; | |
1897 | break; | |
1898 | } | |
1899 | ||
52400ba9 DH |
1900 | /* |
1901 | * Requeue nr_requeue waiters and possibly one more in the case | |
1902 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1903 | */ | |
1904 | if (requeue_pi) { | |
ecb38b78 TG |
1905 | /* |
1906 | * Prepare the waiter to take the rt_mutex. Take a | |
1907 | * refcount on the pi_state and store the pointer in | |
1908 | * the futex_q object of the waiter. | |
1909 | */ | |
52400ba9 DH |
1910 | atomic_inc(&pi_state->refcount); |
1911 | this->pi_state = pi_state; | |
1912 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1913 | this->rt_waiter, | |
c051b21f | 1914 | this->task); |
52400ba9 | 1915 | if (ret == 1) { |
ecb38b78 TG |
1916 | /* |
1917 | * We got the lock. We do neither drop the | |
1918 | * refcount on pi_state nor clear | |
1919 | * this->pi_state because the waiter needs the | |
1920 | * pi_state for cleaning up the user space | |
1921 | * value. It will drop the refcount after | |
1922 | * doing so. | |
1923 | */ | |
beda2c7e | 1924 | requeue_pi_wake_futex(this, &key2, hb2); |
89061d3d | 1925 | drop_count++; |
52400ba9 DH |
1926 | continue; |
1927 | } else if (ret) { | |
ecb38b78 TG |
1928 | /* |
1929 | * rt_mutex_start_proxy_lock() detected a | |
1930 | * potential deadlock when we tried to queue | |
1931 | * that waiter. Drop the pi_state reference | |
1932 | * which we took above and remove the pointer | |
1933 | * to the state from the waiters futex_q | |
1934 | * object. | |
1935 | */ | |
52400ba9 | 1936 | this->pi_state = NULL; |
29e9ee5d | 1937 | put_pi_state(pi_state); |
885c2cb7 TG |
1938 | /* |
1939 | * We stop queueing more waiters and let user | |
1940 | * space deal with the mess. | |
1941 | */ | |
1942 | break; | |
52400ba9 | 1943 | } |
1da177e4 | 1944 | } |
52400ba9 DH |
1945 | requeue_futex(this, hb1, hb2, &key2); |
1946 | drop_count++; | |
1da177e4 LT |
1947 | } |
1948 | ||
ecb38b78 TG |
1949 | /* |
1950 | * We took an extra initial reference to the pi_state either | |
1951 | * in futex_proxy_trylock_atomic() or in lookup_pi_state(). We | |
1952 | * need to drop it here again. | |
1953 | */ | |
29e9ee5d | 1954 | put_pi_state(pi_state); |
885c2cb7 TG |
1955 | |
1956 | out_unlock: | |
5eb3dc62 | 1957 | double_unlock_hb(hb1, hb2); |
1d0dcb3a | 1958 | wake_up_q(&wake_q); |
69cd9eba | 1959 | hb_waiters_dec(hb2); |
1da177e4 | 1960 | |
cd84a42f DH |
1961 | /* |
1962 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1963 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1964 | * one at key2 and updated their key pointer. We no longer need to | |
1965 | * hold the references to key1. | |
1966 | */ | |
1da177e4 | 1967 | while (--drop_count >= 0) |
9adef58b | 1968 | drop_futex_key_refs(&key1); |
1da177e4 | 1969 | |
42d35d48 | 1970 | out_put_keys: |
ae791a2d | 1971 | put_futex_key(&key2); |
42d35d48 | 1972 | out_put_key1: |
ae791a2d | 1973 | put_futex_key(&key1); |
42d35d48 | 1974 | out: |
52400ba9 | 1975 | return ret ? ret : task_count; |
1da177e4 LT |
1976 | } |
1977 | ||
1978 | /* The key must be already stored in q->key. */ | |
82af7aca | 1979 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 1980 | __acquires(&hb->lock) |
1da177e4 | 1981 | { |
e2970f2f | 1982 | struct futex_hash_bucket *hb; |
1da177e4 | 1983 | |
e2970f2f | 1984 | hb = hash_futex(&q->key); |
11d4616b LT |
1985 | |
1986 | /* | |
1987 | * Increment the counter before taking the lock so that | |
1988 | * a potential waker won't miss a to-be-slept task that is | |
1989 | * waiting for the spinlock. This is safe as all queue_lock() | |
1990 | * users end up calling queue_me(). Similarly, for housekeeping, | |
1991 | * decrement the counter at queue_unlock() when some error has | |
1992 | * occurred and we don't end up adding the task to the list. | |
1993 | */ | |
1994 | hb_waiters_inc(hb); | |
1995 | ||
e2970f2f | 1996 | q->lock_ptr = &hb->lock; |
1da177e4 | 1997 | |
8ad7b378 | 1998 | spin_lock(&hb->lock); /* implies smp_mb(); (A) */ |
e2970f2f | 1999 | return hb; |
1da177e4 LT |
2000 | } |
2001 | ||
d40d65c8 | 2002 | static inline void |
0d00c7b2 | 2003 | queue_unlock(struct futex_hash_bucket *hb) |
15e408cd | 2004 | __releases(&hb->lock) |
d40d65c8 DH |
2005 | { |
2006 | spin_unlock(&hb->lock); | |
11d4616b | 2007 | hb_waiters_dec(hb); |
d40d65c8 DH |
2008 | } |
2009 | ||
2010 | /** | |
2011 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
2012 | * @q: The futex_q to enqueue | |
2013 | * @hb: The destination hash bucket | |
2014 | * | |
2015 | * The hb->lock must be held by the caller, and is released here. A call to | |
2016 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
2017 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
2018 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
2019 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
2020 | * an example). | |
2021 | */ | |
82af7aca | 2022 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
15e408cd | 2023 | __releases(&hb->lock) |
1da177e4 | 2024 | { |
ec92d082 PP |
2025 | int prio; |
2026 | ||
2027 | /* | |
2028 | * The priority used to register this element is | |
2029 | * - either the real thread-priority for the real-time threads | |
2030 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
2031 | * - or MAX_RT_PRIO for non-RT threads. | |
2032 | * Thus, all RT-threads are woken first in priority order, and | |
2033 | * the others are woken last, in FIFO order. | |
2034 | */ | |
2035 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
2036 | ||
2037 | plist_node_init(&q->list, prio); | |
ec92d082 | 2038 | plist_add(&q->list, &hb->chain); |
c87e2837 | 2039 | q->task = current; |
e2970f2f | 2040 | spin_unlock(&hb->lock); |
1da177e4 LT |
2041 | } |
2042 | ||
d40d65c8 DH |
2043 | /** |
2044 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
2045 | * @q: The futex_q to unqueue | |
2046 | * | |
2047 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
2048 | * be paired with exactly one earlier call to queue_me(). | |
2049 | * | |
6c23cbbd RD |
2050 | * Return: |
2051 | * 1 - if the futex_q was still queued (and we removed unqueued it); | |
d40d65c8 | 2052 | * 0 - if the futex_q was already removed by the waking thread |
1da177e4 | 2053 | */ |
1da177e4 LT |
2054 | static int unqueue_me(struct futex_q *q) |
2055 | { | |
1da177e4 | 2056 | spinlock_t *lock_ptr; |
e2970f2f | 2057 | int ret = 0; |
1da177e4 LT |
2058 | |
2059 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 2060 | retry: |
29b75eb2 JZ |
2061 | /* |
2062 | * q->lock_ptr can change between this read and the following spin_lock. | |
2063 | * Use READ_ONCE to forbid the compiler from reloading q->lock_ptr and | |
2064 | * optimizing lock_ptr out of the logic below. | |
2065 | */ | |
2066 | lock_ptr = READ_ONCE(q->lock_ptr); | |
c80544dc | 2067 | if (lock_ptr != NULL) { |
1da177e4 LT |
2068 | spin_lock(lock_ptr); |
2069 | /* | |
2070 | * q->lock_ptr can change between reading it and | |
2071 | * spin_lock(), causing us to take the wrong lock. This | |
2072 | * corrects the race condition. | |
2073 | * | |
2074 | * Reasoning goes like this: if we have the wrong lock, | |
2075 | * q->lock_ptr must have changed (maybe several times) | |
2076 | * between reading it and the spin_lock(). It can | |
2077 | * change again after the spin_lock() but only if it was | |
2078 | * already changed before the spin_lock(). It cannot, | |
2079 | * however, change back to the original value. Therefore | |
2080 | * we can detect whether we acquired the correct lock. | |
2081 | */ | |
2082 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
2083 | spin_unlock(lock_ptr); | |
2084 | goto retry; | |
2085 | } | |
2e12978a | 2086 | __unqueue_futex(q); |
c87e2837 IM |
2087 | |
2088 | BUG_ON(q->pi_state); | |
2089 | ||
1da177e4 LT |
2090 | spin_unlock(lock_ptr); |
2091 | ret = 1; | |
2092 | } | |
2093 | ||
9adef58b | 2094 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
2095 | return ret; |
2096 | } | |
2097 | ||
c87e2837 IM |
2098 | /* |
2099 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
2100 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
2101 | * and dropped here. | |
c87e2837 | 2102 | */ |
d0aa7a70 | 2103 | static void unqueue_me_pi(struct futex_q *q) |
15e408cd | 2104 | __releases(q->lock_ptr) |
c87e2837 | 2105 | { |
2e12978a | 2106 | __unqueue_futex(q); |
c87e2837 IM |
2107 | |
2108 | BUG_ON(!q->pi_state); | |
29e9ee5d | 2109 | put_pi_state(q->pi_state); |
c87e2837 IM |
2110 | q->pi_state = NULL; |
2111 | ||
d0aa7a70 | 2112 | spin_unlock(q->lock_ptr); |
c87e2837 IM |
2113 | } |
2114 | ||
d0aa7a70 | 2115 | /* |
cdf71a10 | 2116 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 2117 | * |
778e9a9c AK |
2118 | * Must be called with hash bucket lock held and mm->sem held for non |
2119 | * private futexes. | |
d0aa7a70 | 2120 | */ |
778e9a9c | 2121 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
ae791a2d | 2122 | struct task_struct *newowner) |
d0aa7a70 | 2123 | { |
cdf71a10 | 2124 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 2125 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 2126 | struct task_struct *oldowner = pi_state->owner; |
7cfdaf38 | 2127 | u32 uval, uninitialized_var(curval), newval; |
e4dc5b7a | 2128 | int ret; |
d0aa7a70 PP |
2129 | |
2130 | /* Owner died? */ | |
1b7558e4 TG |
2131 | if (!pi_state->owner) |
2132 | newtid |= FUTEX_OWNER_DIED; | |
2133 | ||
2134 | /* | |
2135 | * We are here either because we stole the rtmutex from the | |
8161239a LJ |
2136 | * previous highest priority waiter or we are the highest priority |
2137 | * waiter but failed to get the rtmutex the first time. | |
2138 | * We have to replace the newowner TID in the user space variable. | |
2139 | * This must be atomic as we have to preserve the owner died bit here. | |
1b7558e4 | 2140 | * |
b2d0994b DH |
2141 | * Note: We write the user space value _before_ changing the pi_state |
2142 | * because we can fault here. Imagine swapped out pages or a fork | |
2143 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
2144 | * |
2145 | * Modifying pi_state _before_ the user space value would | |
2146 | * leave the pi_state in an inconsistent state when we fault | |
2147 | * here, because we need to drop the hash bucket lock to | |
2148 | * handle the fault. This might be observed in the PID check | |
2149 | * in lookup_pi_state. | |
2150 | */ | |
2151 | retry: | |
2152 | if (get_futex_value_locked(&uval, uaddr)) | |
2153 | goto handle_fault; | |
2154 | ||
2155 | while (1) { | |
2156 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
2157 | ||
37a9d912 | 2158 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1b7558e4 TG |
2159 | goto handle_fault; |
2160 | if (curval == uval) | |
2161 | break; | |
2162 | uval = curval; | |
2163 | } | |
2164 | ||
2165 | /* | |
2166 | * We fixed up user space. Now we need to fix the pi_state | |
2167 | * itself. | |
2168 | */ | |
d0aa7a70 | 2169 | if (pi_state->owner != NULL) { |
1d615482 | 2170 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
d0aa7a70 PP |
2171 | WARN_ON(list_empty(&pi_state->list)); |
2172 | list_del_init(&pi_state->list); | |
1d615482 | 2173 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
1b7558e4 | 2174 | } |
d0aa7a70 | 2175 | |
cdf71a10 | 2176 | pi_state->owner = newowner; |
d0aa7a70 | 2177 | |
1d615482 | 2178 | raw_spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 2179 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 | 2180 | list_add(&pi_state->list, &newowner->pi_state_list); |
1d615482 | 2181 | raw_spin_unlock_irq(&newowner->pi_lock); |
1b7558e4 | 2182 | return 0; |
d0aa7a70 | 2183 | |
d0aa7a70 | 2184 | /* |
1b7558e4 | 2185 | * To handle the page fault we need to drop the hash bucket |
8161239a LJ |
2186 | * lock here. That gives the other task (either the highest priority |
2187 | * waiter itself or the task which stole the rtmutex) the | |
1b7558e4 TG |
2188 | * chance to try the fixup of the pi_state. So once we are |
2189 | * back from handling the fault we need to check the pi_state | |
2190 | * after reacquiring the hash bucket lock and before trying to | |
2191 | * do another fixup. When the fixup has been done already we | |
2192 | * simply return. | |
d0aa7a70 | 2193 | */ |
1b7558e4 TG |
2194 | handle_fault: |
2195 | spin_unlock(q->lock_ptr); | |
778e9a9c | 2196 | |
d0725992 | 2197 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 2198 | |
1b7558e4 | 2199 | spin_lock(q->lock_ptr); |
778e9a9c | 2200 | |
1b7558e4 TG |
2201 | /* |
2202 | * Check if someone else fixed it for us: | |
2203 | */ | |
2204 | if (pi_state->owner != oldowner) | |
2205 | return 0; | |
2206 | ||
2207 | if (ret) | |
2208 | return ret; | |
2209 | ||
2210 | goto retry; | |
d0aa7a70 PP |
2211 | } |
2212 | ||
72c1bbf3 | 2213 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 2214 | |
dd973998 DH |
2215 | /** |
2216 | * fixup_owner() - Post lock pi_state and corner case management | |
2217 | * @uaddr: user address of the futex | |
dd973998 DH |
2218 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
2219 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
2220 | * | |
2221 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
2222 | * the pi_state owner as well as handle race conditions that may allow us to | |
2223 | * acquire the lock. Must be called with the hb lock held. | |
2224 | * | |
6c23cbbd RD |
2225 | * Return: |
2226 | * 1 - success, lock taken; | |
2227 | * 0 - success, lock not taken; | |
dd973998 DH |
2228 | * <0 - on error (-EFAULT) |
2229 | */ | |
ae791a2d | 2230 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 DH |
2231 | { |
2232 | struct task_struct *owner; | |
2233 | int ret = 0; | |
2234 | ||
2235 | if (locked) { | |
2236 | /* | |
2237 | * Got the lock. We might not be the anticipated owner if we | |
2238 | * did a lock-steal - fix up the PI-state in that case: | |
2239 | */ | |
2240 | if (q->pi_state->owner != current) | |
ae791a2d | 2241 | ret = fixup_pi_state_owner(uaddr, q, current); |
dd973998 DH |
2242 | goto out; |
2243 | } | |
2244 | ||
2245 | /* | |
2246 | * Catch the rare case, where the lock was released when we were on the | |
2247 | * way back before we locked the hash bucket. | |
2248 | */ | |
2249 | if (q->pi_state->owner == current) { | |
2250 | /* | |
2251 | * Try to get the rt_mutex now. This might fail as some other | |
2252 | * task acquired the rt_mutex after we removed ourself from the | |
2253 | * rt_mutex waiters list. | |
2254 | */ | |
2255 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
2256 | locked = 1; | |
2257 | goto out; | |
2258 | } | |
2259 | ||
2260 | /* | |
2261 | * pi_state is incorrect, some other task did a lock steal and | |
2262 | * we returned due to timeout or signal without taking the | |
8161239a | 2263 | * rt_mutex. Too late. |
dd973998 | 2264 | */ |
b4abf910 | 2265 | raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); |
dd973998 | 2266 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); |
8161239a LJ |
2267 | if (!owner) |
2268 | owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); | |
b4abf910 | 2269 | raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); |
ae791a2d | 2270 | ret = fixup_pi_state_owner(uaddr, q, owner); |
dd973998 DH |
2271 | goto out; |
2272 | } | |
2273 | ||
2274 | /* | |
2275 | * Paranoia check. If we did not take the lock, then we should not be | |
8161239a | 2276 | * the owner of the rt_mutex. |
dd973998 DH |
2277 | */ |
2278 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
2279 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
2280 | "pi-state %p\n", ret, | |
2281 | q->pi_state->pi_mutex.owner, | |
2282 | q->pi_state->owner); | |
2283 | ||
2284 | out: | |
2285 | return ret ? ret : locked; | |
2286 | } | |
2287 | ||
ca5f9524 DH |
2288 | /** |
2289 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
2290 | * @hb: the futex hash bucket, must be locked by the caller | |
2291 | * @q: the futex_q to queue up on | |
2292 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
2293 | */ |
2294 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 2295 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 2296 | { |
9beba3c5 DH |
2297 | /* |
2298 | * The task state is guaranteed to be set before another task can | |
b92b8b35 | 2299 | * wake it. set_current_state() is implemented using smp_store_mb() and |
9beba3c5 DH |
2300 | * queue_me() calls spin_unlock() upon completion, both serializing |
2301 | * access to the hash list and forcing another memory barrier. | |
2302 | */ | |
f1a11e05 | 2303 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 2304 | queue_me(q, hb); |
ca5f9524 DH |
2305 | |
2306 | /* Arm the timer */ | |
2e4b0d3f | 2307 | if (timeout) |
ca5f9524 | 2308 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); |
ca5f9524 DH |
2309 | |
2310 | /* | |
0729e196 DH |
2311 | * If we have been removed from the hash list, then another task |
2312 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
2313 | */ |
2314 | if (likely(!plist_node_empty(&q->list))) { | |
2315 | /* | |
2316 | * If the timer has already expired, current will already be | |
2317 | * flagged for rescheduling. Only call schedule if there | |
2318 | * is no timeout, or if it has yet to expire. | |
2319 | */ | |
2320 | if (!timeout || timeout->task) | |
88c8004f | 2321 | freezable_schedule(); |
ca5f9524 DH |
2322 | } |
2323 | __set_current_state(TASK_RUNNING); | |
2324 | } | |
2325 | ||
f801073f DH |
2326 | /** |
2327 | * futex_wait_setup() - Prepare to wait on a futex | |
2328 | * @uaddr: the futex userspace address | |
2329 | * @val: the expected value | |
b41277dc | 2330 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
2331 | * @q: the associated futex_q |
2332 | * @hb: storage for hash_bucket pointer to be returned to caller | |
2333 | * | |
2334 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
2335 | * compare it with the expected value. Handle atomic faults internally. | |
2336 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
2337 | * with no q.key reference on failure. | |
2338 | * | |
6c23cbbd RD |
2339 | * Return: |
2340 | * 0 - uaddr contains val and hb has been locked; | |
ca4a04cf | 2341 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked |
f801073f | 2342 | */ |
b41277dc | 2343 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 2344 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 2345 | { |
e2970f2f IM |
2346 | u32 uval; |
2347 | int ret; | |
1da177e4 | 2348 | |
1da177e4 | 2349 | /* |
b2d0994b | 2350 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
2351 | * Order is important: |
2352 | * | |
2353 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
2354 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
2355 | * | |
2356 | * The basic logical guarantee of a futex is that it blocks ONLY | |
2357 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
2358 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
2359 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
2360 | * cond(var) false, which would violate the guarantee. |
2361 | * | |
8fe8f545 ML |
2362 | * On the other hand, we insert q and release the hash-bucket only |
2363 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
2364 | * absorb a wakeup if *uaddr does not match the desired values | |
2365 | * while the syscall executes. | |
1da177e4 | 2366 | */ |
f801073f | 2367 | retry: |
9ea71503 | 2368 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ); |
f801073f | 2369 | if (unlikely(ret != 0)) |
a5a2a0c7 | 2370 | return ret; |
f801073f DH |
2371 | |
2372 | retry_private: | |
2373 | *hb = queue_lock(q); | |
2374 | ||
e2970f2f | 2375 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 2376 | |
f801073f | 2377 | if (ret) { |
0d00c7b2 | 2378 | queue_unlock(*hb); |
1da177e4 | 2379 | |
e2970f2f | 2380 | ret = get_user(uval, uaddr); |
e4dc5b7a | 2381 | if (ret) |
f801073f | 2382 | goto out; |
1da177e4 | 2383 | |
b41277dc | 2384 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2385 | goto retry_private; |
2386 | ||
ae791a2d | 2387 | put_futex_key(&q->key); |
e4dc5b7a | 2388 | goto retry; |
1da177e4 | 2389 | } |
ca5f9524 | 2390 | |
f801073f | 2391 | if (uval != val) { |
0d00c7b2 | 2392 | queue_unlock(*hb); |
f801073f | 2393 | ret = -EWOULDBLOCK; |
2fff78c7 | 2394 | } |
1da177e4 | 2395 | |
f801073f DH |
2396 | out: |
2397 | if (ret) | |
ae791a2d | 2398 | put_futex_key(&q->key); |
f801073f DH |
2399 | return ret; |
2400 | } | |
2401 | ||
b41277dc DH |
2402 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
2403 | ktime_t *abs_time, u32 bitset) | |
f801073f DH |
2404 | { |
2405 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
2406 | struct restart_block *restart; |
2407 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 2408 | struct futex_q q = futex_q_init; |
f801073f DH |
2409 | int ret; |
2410 | ||
2411 | if (!bitset) | |
2412 | return -EINVAL; | |
f801073f DH |
2413 | q.bitset = bitset; |
2414 | ||
2415 | if (abs_time) { | |
2416 | to = &timeout; | |
2417 | ||
b41277dc DH |
2418 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2419 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2420 | HRTIMER_MODE_ABS); | |
f801073f DH |
2421 | hrtimer_init_sleeper(to, current); |
2422 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2423 | current->timer_slack_ns); | |
2424 | } | |
2425 | ||
d58e6576 | 2426 | retry: |
7ada876a DH |
2427 | /* |
2428 | * Prepare to wait on uaddr. On success, holds hb lock and increments | |
2429 | * q.key refs. | |
2430 | */ | |
b41277dc | 2431 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
2432 | if (ret) |
2433 | goto out; | |
2434 | ||
ca5f9524 | 2435 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 2436 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
2437 | |
2438 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 2439 | ret = 0; |
7ada876a | 2440 | /* unqueue_me() drops q.key ref */ |
1da177e4 | 2441 | if (!unqueue_me(&q)) |
7ada876a | 2442 | goto out; |
2fff78c7 | 2443 | ret = -ETIMEDOUT; |
ca5f9524 | 2444 | if (to && !to->task) |
7ada876a | 2445 | goto out; |
72c1bbf3 | 2446 | |
e2970f2f | 2447 | /* |
d58e6576 TG |
2448 | * We expect signal_pending(current), but we might be the |
2449 | * victim of a spurious wakeup as well. | |
e2970f2f | 2450 | */ |
7ada876a | 2451 | if (!signal_pending(current)) |
d58e6576 | 2452 | goto retry; |
d58e6576 | 2453 | |
2fff78c7 | 2454 | ret = -ERESTARTSYS; |
c19384b5 | 2455 | if (!abs_time) |
7ada876a | 2456 | goto out; |
1da177e4 | 2457 | |
f56141e3 | 2458 | restart = ¤t->restart_block; |
2fff78c7 | 2459 | restart->fn = futex_wait_restart; |
a3c74c52 | 2460 | restart->futex.uaddr = uaddr; |
2fff78c7 | 2461 | restart->futex.val = val; |
2456e855 | 2462 | restart->futex.time = *abs_time; |
2fff78c7 | 2463 | restart->futex.bitset = bitset; |
0cd9c649 | 2464 | restart->futex.flags = flags | FLAGS_HAS_TIMEOUT; |
42d35d48 | 2465 | |
2fff78c7 PZ |
2466 | ret = -ERESTART_RESTARTBLOCK; |
2467 | ||
42d35d48 | 2468 | out: |
ca5f9524 DH |
2469 | if (to) { |
2470 | hrtimer_cancel(&to->timer); | |
2471 | destroy_hrtimer_on_stack(&to->timer); | |
2472 | } | |
c87e2837 IM |
2473 | return ret; |
2474 | } | |
2475 | ||
72c1bbf3 NP |
2476 | |
2477 | static long futex_wait_restart(struct restart_block *restart) | |
2478 | { | |
a3c74c52 | 2479 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 2480 | ktime_t t, *tp = NULL; |
72c1bbf3 | 2481 | |
a72188d8 | 2482 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
2456e855 | 2483 | t = restart->futex.time; |
a72188d8 DH |
2484 | tp = &t; |
2485 | } | |
72c1bbf3 | 2486 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
2487 | |
2488 | return (long)futex_wait(uaddr, restart->futex.flags, | |
2489 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
2490 | } |
2491 | ||
2492 | ||
c87e2837 IM |
2493 | /* |
2494 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
2495 | * and failed. The kernel side here does the whole locking operation: | |
767f509c DB |
2496 | * if there are waiters then it will block as a consequence of relying |
2497 | * on rt-mutexes, it does PI, etc. (Due to races the kernel might see | |
2498 | * a 0 value of the futex too.). | |
2499 | * | |
2500 | * Also serves as futex trylock_pi()'ing, and due semantics. | |
c87e2837 | 2501 | */ |
996636dd | 2502 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, |
b41277dc | 2503 | ktime_t *time, int trylock) |
c87e2837 | 2504 | { |
c5780e97 | 2505 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 2506 | struct futex_hash_bucket *hb; |
5bdb05f9 | 2507 | struct futex_q q = futex_q_init; |
dd973998 | 2508 | int res, ret; |
c87e2837 IM |
2509 | |
2510 | if (refill_pi_state_cache()) | |
2511 | return -ENOMEM; | |
2512 | ||
c19384b5 | 2513 | if (time) { |
c5780e97 | 2514 | to = &timeout; |
237fc6e7 TG |
2515 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
2516 | HRTIMER_MODE_ABS); | |
c5780e97 | 2517 | hrtimer_init_sleeper(to, current); |
cc584b21 | 2518 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
2519 | } |
2520 | ||
42d35d48 | 2521 | retry: |
9ea71503 | 2522 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE); |
c87e2837 | 2523 | if (unlikely(ret != 0)) |
42d35d48 | 2524 | goto out; |
c87e2837 | 2525 | |
e4dc5b7a | 2526 | retry_private: |
82af7aca | 2527 | hb = queue_lock(&q); |
c87e2837 | 2528 | |
bab5bc9e | 2529 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 2530 | if (unlikely(ret)) { |
767f509c DB |
2531 | /* |
2532 | * Atomic work succeeded and we got the lock, | |
2533 | * or failed. Either way, we do _not_ block. | |
2534 | */ | |
778e9a9c | 2535 | switch (ret) { |
1a52084d DH |
2536 | case 1: |
2537 | /* We got the lock. */ | |
2538 | ret = 0; | |
2539 | goto out_unlock_put_key; | |
2540 | case -EFAULT: | |
2541 | goto uaddr_faulted; | |
778e9a9c AK |
2542 | case -EAGAIN: |
2543 | /* | |
af54d6a1 TG |
2544 | * Two reasons for this: |
2545 | * - Task is exiting and we just wait for the | |
2546 | * exit to complete. | |
2547 | * - The user space value changed. | |
778e9a9c | 2548 | */ |
0d00c7b2 | 2549 | queue_unlock(hb); |
ae791a2d | 2550 | put_futex_key(&q.key); |
778e9a9c AK |
2551 | cond_resched(); |
2552 | goto retry; | |
778e9a9c | 2553 | default: |
42d35d48 | 2554 | goto out_unlock_put_key; |
c87e2837 | 2555 | } |
c87e2837 IM |
2556 | } |
2557 | ||
2558 | /* | |
2559 | * Only actually queue now that the atomic ops are done: | |
2560 | */ | |
82af7aca | 2561 | queue_me(&q, hb); |
c87e2837 | 2562 | |
c87e2837 IM |
2563 | WARN_ON(!q.pi_state); |
2564 | /* | |
2565 | * Block on the PI mutex: | |
2566 | */ | |
c051b21f TG |
2567 | if (!trylock) { |
2568 | ret = rt_mutex_timed_futex_lock(&q.pi_state->pi_mutex, to); | |
2569 | } else { | |
c87e2837 IM |
2570 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); |
2571 | /* Fixup the trylock return value: */ | |
2572 | ret = ret ? 0 : -EWOULDBLOCK; | |
2573 | } | |
2574 | ||
a99e4e41 | 2575 | spin_lock(q.lock_ptr); |
dd973998 DH |
2576 | /* |
2577 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2578 | * haven't already. | |
2579 | */ | |
ae791a2d | 2580 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 DH |
2581 | /* |
2582 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
2583 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
2584 | */ | |
2585 | if (res) | |
2586 | ret = (res < 0) ? res : 0; | |
c87e2837 | 2587 | |
e8f6386c | 2588 | /* |
dd973998 DH |
2589 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
2590 | * it and return the fault to userspace. | |
e8f6386c DH |
2591 | */ |
2592 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
2593 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
2594 | ||
778e9a9c AK |
2595 | /* Unqueue and drop the lock */ |
2596 | unqueue_me_pi(&q); | |
c87e2837 | 2597 | |
5ecb01cf | 2598 | goto out_put_key; |
c87e2837 | 2599 | |
42d35d48 | 2600 | out_unlock_put_key: |
0d00c7b2 | 2601 | queue_unlock(hb); |
c87e2837 | 2602 | |
42d35d48 | 2603 | out_put_key: |
ae791a2d | 2604 | put_futex_key(&q.key); |
42d35d48 | 2605 | out: |
237fc6e7 TG |
2606 | if (to) |
2607 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 2608 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 2609 | |
42d35d48 | 2610 | uaddr_faulted: |
0d00c7b2 | 2611 | queue_unlock(hb); |
778e9a9c | 2612 | |
d0725992 | 2613 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
2614 | if (ret) |
2615 | goto out_put_key; | |
c87e2837 | 2616 | |
b41277dc | 2617 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2618 | goto retry_private; |
2619 | ||
ae791a2d | 2620 | put_futex_key(&q.key); |
e4dc5b7a | 2621 | goto retry; |
c87e2837 IM |
2622 | } |
2623 | ||
c87e2837 IM |
2624 | /* |
2625 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
2626 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
2627 | * and do the rt-mutex unlock. | |
2628 | */ | |
b41277dc | 2629 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 | 2630 | { |
ccf9e6a8 | 2631 | u32 uninitialized_var(curval), uval, vpid = task_pid_vnr(current); |
38d47c1b | 2632 | union futex_key key = FUTEX_KEY_INIT; |
ccf9e6a8 TG |
2633 | struct futex_hash_bucket *hb; |
2634 | struct futex_q *match; | |
e4dc5b7a | 2635 | int ret; |
c87e2837 IM |
2636 | |
2637 | retry: | |
2638 | if (get_user(uval, uaddr)) | |
2639 | return -EFAULT; | |
2640 | /* | |
2641 | * We release only a lock we actually own: | |
2642 | */ | |
c0c9ed15 | 2643 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 2644 | return -EPERM; |
c87e2837 | 2645 | |
9ea71503 | 2646 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE); |
ccf9e6a8 TG |
2647 | if (ret) |
2648 | return ret; | |
c87e2837 IM |
2649 | |
2650 | hb = hash_futex(&key); | |
2651 | spin_lock(&hb->lock); | |
2652 | ||
c87e2837 | 2653 | /* |
ccf9e6a8 TG |
2654 | * Check waiters first. We do not trust user space values at |
2655 | * all and we at least want to know if user space fiddled | |
2656 | * with the futex value instead of blindly unlocking. | |
c87e2837 | 2657 | */ |
ccf9e6a8 TG |
2658 | match = futex_top_waiter(hb, &key); |
2659 | if (match) { | |
802ab58d SAS |
2660 | ret = wake_futex_pi(uaddr, uval, match, hb); |
2661 | /* | |
2662 | * In case of success wake_futex_pi dropped the hash | |
2663 | * bucket lock. | |
2664 | */ | |
2665 | if (!ret) | |
2666 | goto out_putkey; | |
c87e2837 | 2667 | /* |
ccf9e6a8 TG |
2668 | * The atomic access to the futex value generated a |
2669 | * pagefault, so retry the user-access and the wakeup: | |
c87e2837 IM |
2670 | */ |
2671 | if (ret == -EFAULT) | |
2672 | goto pi_faulted; | |
89e9e66b SAS |
2673 | /* |
2674 | * A unconditional UNLOCK_PI op raced against a waiter | |
2675 | * setting the FUTEX_WAITERS bit. Try again. | |
2676 | */ | |
2677 | if (ret == -EAGAIN) { | |
2678 | spin_unlock(&hb->lock); | |
2679 | put_futex_key(&key); | |
2680 | goto retry; | |
2681 | } | |
802ab58d SAS |
2682 | /* |
2683 | * wake_futex_pi has detected invalid state. Tell user | |
2684 | * space. | |
2685 | */ | |
c87e2837 IM |
2686 | goto out_unlock; |
2687 | } | |
ccf9e6a8 | 2688 | |
c87e2837 | 2689 | /* |
ccf9e6a8 TG |
2690 | * We have no kernel internal state, i.e. no waiters in the |
2691 | * kernel. Waiters which are about to queue themselves are stuck | |
2692 | * on hb->lock. So we can safely ignore them. We do neither | |
2693 | * preserve the WAITERS bit not the OWNER_DIED one. We are the | |
2694 | * owner. | |
c87e2837 | 2695 | */ |
ccf9e6a8 | 2696 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, 0)) |
13fbca4c | 2697 | goto pi_faulted; |
c87e2837 | 2698 | |
ccf9e6a8 TG |
2699 | /* |
2700 | * If uval has changed, let user space handle it. | |
2701 | */ | |
2702 | ret = (curval == uval) ? 0 : -EAGAIN; | |
2703 | ||
c87e2837 IM |
2704 | out_unlock: |
2705 | spin_unlock(&hb->lock); | |
802ab58d | 2706 | out_putkey: |
ae791a2d | 2707 | put_futex_key(&key); |
c87e2837 IM |
2708 | return ret; |
2709 | ||
2710 | pi_faulted: | |
778e9a9c | 2711 | spin_unlock(&hb->lock); |
ae791a2d | 2712 | put_futex_key(&key); |
c87e2837 | 2713 | |
d0725992 | 2714 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2715 | if (!ret) |
c87e2837 IM |
2716 | goto retry; |
2717 | ||
1da177e4 LT |
2718 | return ret; |
2719 | } | |
2720 | ||
52400ba9 DH |
2721 | /** |
2722 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2723 | * @hb: the hash_bucket futex_q was original enqueued on | |
2724 | * @q: the futex_q woken while waiting to be requeued | |
2725 | * @key2: the futex_key of the requeue target futex | |
2726 | * @timeout: the timeout associated with the wait (NULL if none) | |
2727 | * | |
2728 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2729 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2730 | * the wakeup and return the appropriate error code to the caller. Must be | |
2731 | * called with the hb lock held. | |
2732 | * | |
6c23cbbd RD |
2733 | * Return: |
2734 | * 0 = no early wakeup detected; | |
2735 | * <0 = -ETIMEDOUT or -ERESTARTNOINTR | |
52400ba9 DH |
2736 | */ |
2737 | static inline | |
2738 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2739 | struct futex_q *q, union futex_key *key2, | |
2740 | struct hrtimer_sleeper *timeout) | |
2741 | { | |
2742 | int ret = 0; | |
2743 | ||
2744 | /* | |
2745 | * With the hb lock held, we avoid races while we process the wakeup. | |
2746 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2747 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2748 | * It can't be requeued from uaddr2 to something else since we don't | |
2749 | * support a PI aware source futex for requeue. | |
2750 | */ | |
2751 | if (!match_futex(&q->key, key2)) { | |
2752 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2753 | /* | |
2754 | * We were woken prior to requeue by a timeout or a signal. | |
2755 | * Unqueue the futex_q and determine which it was. | |
2756 | */ | |
2e12978a | 2757 | plist_del(&q->list, &hb->chain); |
11d4616b | 2758 | hb_waiters_dec(hb); |
52400ba9 | 2759 | |
d58e6576 | 2760 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 2761 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2762 | if (timeout && !timeout->task) |
2763 | ret = -ETIMEDOUT; | |
d58e6576 | 2764 | else if (signal_pending(current)) |
1c840c14 | 2765 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
2766 | } |
2767 | return ret; | |
2768 | } | |
2769 | ||
2770 | /** | |
2771 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 2772 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 2773 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
ab51fbab | 2774 | * the same type, no requeueing from private to shared, etc. |
52400ba9 DH |
2775 | * @val: the expected value of uaddr |
2776 | * @abs_time: absolute timeout | |
56ec1607 | 2777 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
2778 | * @uaddr2: the pi futex we will take prior to returning to user-space |
2779 | * | |
2780 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
6f7b0a2a DH |
2781 | * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake |
2782 | * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to | |
2783 | * userspace. This ensures the rt_mutex maintains an owner when it has waiters; | |
2784 | * without one, the pi logic would not know which task to boost/deboost, if | |
2785 | * there was a need to. | |
52400ba9 DH |
2786 | * |
2787 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
6c23cbbd | 2788 | * via the following-- |
52400ba9 | 2789 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() |
cc6db4e6 DH |
2790 | * 2) wakeup on uaddr2 after a requeue |
2791 | * 3) signal | |
2792 | * 4) timeout | |
52400ba9 | 2793 | * |
cc6db4e6 | 2794 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
2795 | * |
2796 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2797 | * 5) successful lock | |
2798 | * 6) signal | |
2799 | * 7) timeout | |
2800 | * 8) other lock acquisition failure | |
2801 | * | |
cc6db4e6 | 2802 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
2803 | * |
2804 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2805 | * | |
6c23cbbd RD |
2806 | * Return: |
2807 | * 0 - On success; | |
52400ba9 DH |
2808 | * <0 - On error |
2809 | */ | |
b41277dc | 2810 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 2811 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 2812 | u32 __user *uaddr2) |
52400ba9 DH |
2813 | { |
2814 | struct hrtimer_sleeper timeout, *to = NULL; | |
2815 | struct rt_mutex_waiter rt_waiter; | |
2816 | struct rt_mutex *pi_mutex = NULL; | |
52400ba9 | 2817 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
2818 | union futex_key key2 = FUTEX_KEY_INIT; |
2819 | struct futex_q q = futex_q_init; | |
52400ba9 | 2820 | int res, ret; |
52400ba9 | 2821 | |
6f7b0a2a DH |
2822 | if (uaddr == uaddr2) |
2823 | return -EINVAL; | |
2824 | ||
52400ba9 DH |
2825 | if (!bitset) |
2826 | return -EINVAL; | |
2827 | ||
2828 | if (abs_time) { | |
2829 | to = &timeout; | |
b41277dc DH |
2830 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2831 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2832 | HRTIMER_MODE_ABS); | |
52400ba9 DH |
2833 | hrtimer_init_sleeper(to, current); |
2834 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2835 | current->timer_slack_ns); | |
2836 | } | |
2837 | ||
2838 | /* | |
2839 | * The waiter is allocated on our stack, manipulated by the requeue | |
2840 | * code while we sleep on uaddr. | |
2841 | */ | |
2842 | debug_rt_mutex_init_waiter(&rt_waiter); | |
fb00aca4 PZ |
2843 | RB_CLEAR_NODE(&rt_waiter.pi_tree_entry); |
2844 | RB_CLEAR_NODE(&rt_waiter.tree_entry); | |
52400ba9 DH |
2845 | rt_waiter.task = NULL; |
2846 | ||
9ea71503 | 2847 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
52400ba9 DH |
2848 | if (unlikely(ret != 0)) |
2849 | goto out; | |
2850 | ||
84bc4af5 DH |
2851 | q.bitset = bitset; |
2852 | q.rt_waiter = &rt_waiter; | |
2853 | q.requeue_pi_key = &key2; | |
2854 | ||
7ada876a DH |
2855 | /* |
2856 | * Prepare to wait on uaddr. On success, increments q.key (key1) ref | |
2857 | * count. | |
2858 | */ | |
b41277dc | 2859 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 TG |
2860 | if (ret) |
2861 | goto out_key2; | |
52400ba9 | 2862 | |
e9c243a5 TG |
2863 | /* |
2864 | * The check above which compares uaddrs is not sufficient for | |
2865 | * shared futexes. We need to compare the keys: | |
2866 | */ | |
2867 | if (match_futex(&q.key, &key2)) { | |
13c42c2f | 2868 | queue_unlock(hb); |
e9c243a5 TG |
2869 | ret = -EINVAL; |
2870 | goto out_put_keys; | |
2871 | } | |
2872 | ||
52400ba9 | 2873 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ |
f1a11e05 | 2874 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2875 | |
2876 | spin_lock(&hb->lock); | |
2877 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2878 | spin_unlock(&hb->lock); | |
2879 | if (ret) | |
2880 | goto out_put_keys; | |
2881 | ||
2882 | /* | |
2883 | * In order for us to be here, we know our q.key == key2, and since | |
2884 | * we took the hb->lock above, we also know that futex_requeue() has | |
2885 | * completed and we no longer have to concern ourselves with a wakeup | |
7ada876a DH |
2886 | * race with the atomic proxy lock acquisition by the requeue code. The |
2887 | * futex_requeue dropped our key1 reference and incremented our key2 | |
2888 | * reference count. | |
52400ba9 DH |
2889 | */ |
2890 | ||
2891 | /* Check if the requeue code acquired the second futex for us. */ | |
2892 | if (!q.rt_waiter) { | |
2893 | /* | |
2894 | * Got the lock. We might not be the anticipated owner if we | |
2895 | * did a lock-steal - fix up the PI-state in that case. | |
2896 | */ | |
2897 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2898 | spin_lock(q.lock_ptr); | |
ae791a2d | 2899 | ret = fixup_pi_state_owner(uaddr2, &q, current); |
fb75a428 TG |
2900 | /* |
2901 | * Drop the reference to the pi state which | |
2902 | * the requeue_pi() code acquired for us. | |
2903 | */ | |
29e9ee5d | 2904 | put_pi_state(q.pi_state); |
52400ba9 DH |
2905 | spin_unlock(q.lock_ptr); |
2906 | } | |
2907 | } else { | |
2908 | /* | |
2909 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2910 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2911 | * the pi_state. | |
2912 | */ | |
f27071cb | 2913 | WARN_ON(!q.pi_state); |
52400ba9 | 2914 | pi_mutex = &q.pi_state->pi_mutex; |
c051b21f | 2915 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter); |
52400ba9 DH |
2916 | debug_rt_mutex_free_waiter(&rt_waiter); |
2917 | ||
2918 | spin_lock(q.lock_ptr); | |
2919 | /* | |
2920 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2921 | * haven't already. | |
2922 | */ | |
ae791a2d | 2923 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 DH |
2924 | /* |
2925 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 2926 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
2927 | */ |
2928 | if (res) | |
2929 | ret = (res < 0) ? res : 0; | |
2930 | ||
2931 | /* Unqueue and drop the lock. */ | |
2932 | unqueue_me_pi(&q); | |
2933 | } | |
2934 | ||
2935 | /* | |
2936 | * If fixup_pi_state_owner() faulted and was unable to handle the | |
2937 | * fault, unlock the rt_mutex and return the fault to userspace. | |
2938 | */ | |
2939 | if (ret == -EFAULT) { | |
b6070a8d | 2940 | if (pi_mutex && rt_mutex_owner(pi_mutex) == current) |
52400ba9 DH |
2941 | rt_mutex_unlock(pi_mutex); |
2942 | } else if (ret == -EINTR) { | |
52400ba9 | 2943 | /* |
cc6db4e6 DH |
2944 | * We've already been requeued, but cannot restart by calling |
2945 | * futex_lock_pi() directly. We could restart this syscall, but | |
2946 | * it would detect that the user space "val" changed and return | |
2947 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
2948 | * -EWOULDBLOCK directly. | |
52400ba9 | 2949 | */ |
2070887f | 2950 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2951 | } |
2952 | ||
2953 | out_put_keys: | |
ae791a2d | 2954 | put_futex_key(&q.key); |
c8b15a70 | 2955 | out_key2: |
ae791a2d | 2956 | put_futex_key(&key2); |
52400ba9 DH |
2957 | |
2958 | out: | |
2959 | if (to) { | |
2960 | hrtimer_cancel(&to->timer); | |
2961 | destroy_hrtimer_on_stack(&to->timer); | |
2962 | } | |
2963 | return ret; | |
2964 | } | |
2965 | ||
0771dfef IM |
2966 | /* |
2967 | * Support for robust futexes: the kernel cleans up held futexes at | |
2968 | * thread exit time. | |
2969 | * | |
2970 | * Implementation: user-space maintains a per-thread list of locks it | |
2971 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2972 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2973 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2974 | * always manipulated with the lock held, so the list is private and |
2975 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2976 | * field, to allow the kernel to clean up if the thread dies after | |
2977 | * acquiring the lock, but just before it could have added itself to | |
2978 | * the list. There can only be one such pending lock. | |
2979 | */ | |
2980 | ||
2981 | /** | |
d96ee56c DH |
2982 | * sys_set_robust_list() - Set the robust-futex list head of a task |
2983 | * @head: pointer to the list-head | |
2984 | * @len: length of the list-head, as userspace expects | |
0771dfef | 2985 | */ |
836f92ad HC |
2986 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2987 | size_t, len) | |
0771dfef | 2988 | { |
a0c1e907 TG |
2989 | if (!futex_cmpxchg_enabled) |
2990 | return -ENOSYS; | |
0771dfef IM |
2991 | /* |
2992 | * The kernel knows only one size for now: | |
2993 | */ | |
2994 | if (unlikely(len != sizeof(*head))) | |
2995 | return -EINVAL; | |
2996 | ||
2997 | current->robust_list = head; | |
2998 | ||
2999 | return 0; | |
3000 | } | |
3001 | ||
3002 | /** | |
d96ee56c DH |
3003 | * sys_get_robust_list() - Get the robust-futex list head of a task |
3004 | * @pid: pid of the process [zero for current task] | |
3005 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
3006 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 3007 | */ |
836f92ad HC |
3008 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
3009 | struct robust_list_head __user * __user *, head_ptr, | |
3010 | size_t __user *, len_ptr) | |
0771dfef | 3011 | { |
ba46df98 | 3012 | struct robust_list_head __user *head; |
0771dfef | 3013 | unsigned long ret; |
bdbb776f | 3014 | struct task_struct *p; |
0771dfef | 3015 | |
a0c1e907 TG |
3016 | if (!futex_cmpxchg_enabled) |
3017 | return -ENOSYS; | |
3018 | ||
bdbb776f KC |
3019 | rcu_read_lock(); |
3020 | ||
3021 | ret = -ESRCH; | |
0771dfef | 3022 | if (!pid) |
bdbb776f | 3023 | p = current; |
0771dfef | 3024 | else { |
228ebcbe | 3025 | p = find_task_by_vpid(pid); |
0771dfef IM |
3026 | if (!p) |
3027 | goto err_unlock; | |
0771dfef IM |
3028 | } |
3029 | ||
bdbb776f | 3030 | ret = -EPERM; |
caaee623 | 3031 | if (!ptrace_may_access(p, PTRACE_MODE_READ_REALCREDS)) |
bdbb776f KC |
3032 | goto err_unlock; |
3033 | ||
3034 | head = p->robust_list; | |
3035 | rcu_read_unlock(); | |
3036 | ||
0771dfef IM |
3037 | if (put_user(sizeof(*head), len_ptr)) |
3038 | return -EFAULT; | |
3039 | return put_user(head, head_ptr); | |
3040 | ||
3041 | err_unlock: | |
aaa2a97e | 3042 | rcu_read_unlock(); |
0771dfef IM |
3043 | |
3044 | return ret; | |
3045 | } | |
3046 | ||
3047 | /* | |
3048 | * Process a futex-list entry, check whether it's owned by the | |
3049 | * dying task, and do notification if so: | |
3050 | */ | |
e3f2ddea | 3051 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 3052 | { |
7cfdaf38 | 3053 | u32 uval, uninitialized_var(nval), mval; |
0771dfef | 3054 | |
8f17d3a5 IM |
3055 | retry: |
3056 | if (get_user(uval, uaddr)) | |
0771dfef IM |
3057 | return -1; |
3058 | ||
b488893a | 3059 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
3060 | /* |
3061 | * Ok, this dying thread is truly holding a futex | |
3062 | * of interest. Set the OWNER_DIED bit atomically | |
3063 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
3064 | * set, wake up a waiter (if any). (We have to do a | |
3065 | * futex_wake() even if OWNER_DIED is already set - | |
3066 | * to handle the rare but possible case of recursive | |
3067 | * thread-death.) The rest of the cleanup is done in | |
3068 | * userspace. | |
3069 | */ | |
e3f2ddea | 3070 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
6e0aa9f8 TG |
3071 | /* |
3072 | * We are not holding a lock here, but we want to have | |
3073 | * the pagefault_disable/enable() protection because | |
3074 | * we want to handle the fault gracefully. If the | |
3075 | * access fails we try to fault in the futex with R/W | |
3076 | * verification via get_user_pages. get_user() above | |
3077 | * does not guarantee R/W access. If that fails we | |
3078 | * give up and leave the futex locked. | |
3079 | */ | |
3080 | if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) { | |
3081 | if (fault_in_user_writeable(uaddr)) | |
3082 | return -1; | |
3083 | goto retry; | |
3084 | } | |
c87e2837 | 3085 | if (nval != uval) |
8f17d3a5 | 3086 | goto retry; |
0771dfef | 3087 | |
e3f2ddea IM |
3088 | /* |
3089 | * Wake robust non-PI futexes here. The wakeup of | |
3090 | * PI futexes happens in exit_pi_state(): | |
3091 | */ | |
36cf3b5c | 3092 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 3093 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
3094 | } |
3095 | return 0; | |
3096 | } | |
3097 | ||
e3f2ddea IM |
3098 | /* |
3099 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
3100 | */ | |
3101 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 3102 | struct robust_list __user * __user *head, |
1dcc41bb | 3103 | unsigned int *pi) |
e3f2ddea IM |
3104 | { |
3105 | unsigned long uentry; | |
3106 | ||
ba46df98 | 3107 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
3108 | return -EFAULT; |
3109 | ||
ba46df98 | 3110 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
3111 | *pi = uentry & 1; |
3112 | ||
3113 | return 0; | |
3114 | } | |
3115 | ||
0771dfef IM |
3116 | /* |
3117 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
3118 | * and mark any locks found there dead, and notify any waiters. | |
3119 | * | |
3120 | * We silently return on any sign of list-walking problem. | |
3121 | */ | |
3122 | void exit_robust_list(struct task_struct *curr) | |
3123 | { | |
3124 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 3125 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 DH |
3126 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
3127 | unsigned int uninitialized_var(next_pi); | |
0771dfef | 3128 | unsigned long futex_offset; |
9f96cb1e | 3129 | int rc; |
0771dfef | 3130 | |
a0c1e907 TG |
3131 | if (!futex_cmpxchg_enabled) |
3132 | return; | |
3133 | ||
0771dfef IM |
3134 | /* |
3135 | * Fetch the list head (which was registered earlier, via | |
3136 | * sys_set_robust_list()): | |
3137 | */ | |
e3f2ddea | 3138 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
3139 | return; |
3140 | /* | |
3141 | * Fetch the relative futex offset: | |
3142 | */ | |
3143 | if (get_user(futex_offset, &head->futex_offset)) | |
3144 | return; | |
3145 | /* | |
3146 | * Fetch any possibly pending lock-add first, and handle it | |
3147 | * if it exists: | |
3148 | */ | |
e3f2ddea | 3149 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 3150 | return; |
e3f2ddea | 3151 | |
9f96cb1e | 3152 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 3153 | while (entry != &head->list) { |
9f96cb1e MS |
3154 | /* |
3155 | * Fetch the next entry in the list before calling | |
3156 | * handle_futex_death: | |
3157 | */ | |
3158 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
3159 | /* |
3160 | * A pending lock might already be on the list, so | |
c87e2837 | 3161 | * don't process it twice: |
0771dfef IM |
3162 | */ |
3163 | if (entry != pending) | |
ba46df98 | 3164 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 3165 | curr, pi)) |
0771dfef | 3166 | return; |
9f96cb1e | 3167 | if (rc) |
0771dfef | 3168 | return; |
9f96cb1e MS |
3169 | entry = next_entry; |
3170 | pi = next_pi; | |
0771dfef IM |
3171 | /* |
3172 | * Avoid excessively long or circular lists: | |
3173 | */ | |
3174 | if (!--limit) | |
3175 | break; | |
3176 | ||
3177 | cond_resched(); | |
3178 | } | |
9f96cb1e MS |
3179 | |
3180 | if (pending) | |
3181 | handle_futex_death((void __user *)pending + futex_offset, | |
3182 | curr, pip); | |
0771dfef IM |
3183 | } |
3184 | ||
c19384b5 | 3185 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 3186 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 3187 | { |
81b40539 | 3188 | int cmd = op & FUTEX_CMD_MASK; |
b41277dc | 3189 | unsigned int flags = 0; |
34f01cc1 ED |
3190 | |
3191 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 3192 | flags |= FLAGS_SHARED; |
1da177e4 | 3193 | |
b41277dc DH |
3194 | if (op & FUTEX_CLOCK_REALTIME) { |
3195 | flags |= FLAGS_CLOCKRT; | |
337f1304 DH |
3196 | if (cmd != FUTEX_WAIT && cmd != FUTEX_WAIT_BITSET && \ |
3197 | cmd != FUTEX_WAIT_REQUEUE_PI) | |
b41277dc DH |
3198 | return -ENOSYS; |
3199 | } | |
1da177e4 | 3200 | |
59263b51 TG |
3201 | switch (cmd) { |
3202 | case FUTEX_LOCK_PI: | |
3203 | case FUTEX_UNLOCK_PI: | |
3204 | case FUTEX_TRYLOCK_PI: | |
3205 | case FUTEX_WAIT_REQUEUE_PI: | |
3206 | case FUTEX_CMP_REQUEUE_PI: | |
3207 | if (!futex_cmpxchg_enabled) | |
3208 | return -ENOSYS; | |
3209 | } | |
3210 | ||
34f01cc1 | 3211 | switch (cmd) { |
1da177e4 | 3212 | case FUTEX_WAIT: |
cd689985 TG |
3213 | val3 = FUTEX_BITSET_MATCH_ANY; |
3214 | case FUTEX_WAIT_BITSET: | |
81b40539 | 3215 | return futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 | 3216 | case FUTEX_WAKE: |
cd689985 TG |
3217 | val3 = FUTEX_BITSET_MATCH_ANY; |
3218 | case FUTEX_WAKE_BITSET: | |
81b40539 | 3219 | return futex_wake(uaddr, flags, val, val3); |
1da177e4 | 3220 | case FUTEX_REQUEUE: |
81b40539 | 3221 | return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 | 3222 | case FUTEX_CMP_REQUEUE: |
81b40539 | 3223 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
4732efbe | 3224 | case FUTEX_WAKE_OP: |
81b40539 | 3225 | return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
c87e2837 | 3226 | case FUTEX_LOCK_PI: |
996636dd | 3227 | return futex_lock_pi(uaddr, flags, timeout, 0); |
c87e2837 | 3228 | case FUTEX_UNLOCK_PI: |
81b40539 | 3229 | return futex_unlock_pi(uaddr, flags); |
c87e2837 | 3230 | case FUTEX_TRYLOCK_PI: |
996636dd | 3231 | return futex_lock_pi(uaddr, flags, NULL, 1); |
52400ba9 DH |
3232 | case FUTEX_WAIT_REQUEUE_PI: |
3233 | val3 = FUTEX_BITSET_MATCH_ANY; | |
81b40539 TG |
3234 | return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
3235 | uaddr2); | |
52400ba9 | 3236 | case FUTEX_CMP_REQUEUE_PI: |
81b40539 | 3237 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
1da177e4 | 3238 | } |
81b40539 | 3239 | return -ENOSYS; |
1da177e4 LT |
3240 | } |
3241 | ||
3242 | ||
17da2bd9 HC |
3243 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
3244 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
3245 | u32, val3) | |
1da177e4 | 3246 | { |
c19384b5 PP |
3247 | struct timespec ts; |
3248 | ktime_t t, *tp = NULL; | |
e2970f2f | 3249 | u32 val2 = 0; |
34f01cc1 | 3250 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 3251 | |
cd689985 | 3252 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
3253 | cmd == FUTEX_WAIT_BITSET || |
3254 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
ab51fbab DB |
3255 | if (unlikely(should_fail_futex(!(op & FUTEX_PRIVATE_FLAG)))) |
3256 | return -EFAULT; | |
c19384b5 | 3257 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 3258 | return -EFAULT; |
c19384b5 | 3259 | if (!timespec_valid(&ts)) |
9741ef96 | 3260 | return -EINVAL; |
c19384b5 PP |
3261 | |
3262 | t = timespec_to_ktime(ts); | |
34f01cc1 | 3263 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 3264 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 3265 | tp = &t; |
1da177e4 LT |
3266 | } |
3267 | /* | |
52400ba9 | 3268 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 3269 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 3270 | */ |
f54f0986 | 3271 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 3272 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 3273 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 3274 | |
c19384b5 | 3275 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
3276 | } |
3277 | ||
03b8c7b6 | 3278 | static void __init futex_detect_cmpxchg(void) |
1da177e4 | 3279 | { |
03b8c7b6 | 3280 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 3281 | u32 curval; |
03b8c7b6 HC |
3282 | |
3283 | /* | |
3284 | * This will fail and we want it. Some arch implementations do | |
3285 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
3286 | * functionality. We want to know that before we call in any | |
3287 | * of the complex code paths. Also we want to prevent | |
3288 | * registration of robust lists in that case. NULL is | |
3289 | * guaranteed to fault and we get -EFAULT on functional | |
3290 | * implementation, the non-functional ones will return | |
3291 | * -ENOSYS. | |
3292 | */ | |
3293 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) | |
3294 | futex_cmpxchg_enabled = 1; | |
3295 | #endif | |
3296 | } | |
3297 | ||
3298 | static int __init futex_init(void) | |
3299 | { | |
63b1a816 | 3300 | unsigned int futex_shift; |
a52b89eb DB |
3301 | unsigned long i; |
3302 | ||
3303 | #if CONFIG_BASE_SMALL | |
3304 | futex_hashsize = 16; | |
3305 | #else | |
3306 | futex_hashsize = roundup_pow_of_two(256 * num_possible_cpus()); | |
3307 | #endif | |
3308 | ||
3309 | futex_queues = alloc_large_system_hash("futex", sizeof(*futex_queues), | |
3310 | futex_hashsize, 0, | |
3311 | futex_hashsize < 256 ? HASH_SMALL : 0, | |
63b1a816 HC |
3312 | &futex_shift, NULL, |
3313 | futex_hashsize, futex_hashsize); | |
3314 | futex_hashsize = 1UL << futex_shift; | |
03b8c7b6 HC |
3315 | |
3316 | futex_detect_cmpxchg(); | |
a0c1e907 | 3317 | |
a52b89eb | 3318 | for (i = 0; i < futex_hashsize; i++) { |
11d4616b | 3319 | atomic_set(&futex_queues[i].waiters, 0); |
732375c6 | 3320 | plist_head_init(&futex_queues[i].chain); |
3e4ab747 TG |
3321 | spin_lock_init(&futex_queues[i].lock); |
3322 | } | |
3323 | ||
1da177e4 LT |
3324 | return 0; |
3325 | } | |
f6d107fb | 3326 | __initcall(futex_init); |