Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Fast Userspace Mutexes (which I call "Futexes!"). | |
3 | * (C) Rusty Russell, IBM 2002 | |
4 | * | |
5 | * Generalized futexes, futex requeueing, misc fixes by Ingo Molnar | |
6 | * (C) Copyright 2003 Red Hat Inc, All Rights Reserved | |
7 | * | |
8 | * Removed page pinning, fix privately mapped COW pages and other cleanups | |
9 | * (C) Copyright 2003, 2004 Jamie Lokier | |
10 | * | |
0771dfef IM |
11 | * Robust futex support started by Ingo Molnar |
12 | * (C) Copyright 2006 Red Hat Inc, All Rights Reserved | |
13 | * Thanks to Thomas Gleixner for suggestions, analysis and fixes. | |
14 | * | |
c87e2837 IM |
15 | * PI-futex support started by Ingo Molnar and Thomas Gleixner |
16 | * Copyright (C) 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com> | |
17 | * Copyright (C) 2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com> | |
18 | * | |
34f01cc1 ED |
19 | * PRIVATE futexes by Eric Dumazet |
20 | * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com> | |
21 | * | |
52400ba9 DH |
22 | * Requeue-PI support by Darren Hart <dvhltc@us.ibm.com> |
23 | * Copyright (C) IBM Corporation, 2009 | |
24 | * Thanks to Thomas Gleixner for conceptual design and careful reviews. | |
25 | * | |
1da177e4 LT |
26 | * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly |
27 | * enough at me, Linus for the original (flawed) idea, Matthew | |
28 | * Kirkwood for proof-of-concept implementation. | |
29 | * | |
30 | * "The futexes are also cursed." | |
31 | * "But they come in a choice of three flavours!" | |
32 | * | |
33 | * This program is free software; you can redistribute it and/or modify | |
34 | * it under the terms of the GNU General Public License as published by | |
35 | * the Free Software Foundation; either version 2 of the License, or | |
36 | * (at your option) any later version. | |
37 | * | |
38 | * This program is distributed in the hope that it will be useful, | |
39 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
40 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
41 | * GNU General Public License for more details. | |
42 | * | |
43 | * You should have received a copy of the GNU General Public License | |
44 | * along with this program; if not, write to the Free Software | |
45 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
46 | */ | |
47 | #include <linux/slab.h> | |
48 | #include <linux/poll.h> | |
49 | #include <linux/fs.h> | |
50 | #include <linux/file.h> | |
51 | #include <linux/jhash.h> | |
52 | #include <linux/init.h> | |
53 | #include <linux/futex.h> | |
54 | #include <linux/mount.h> | |
55 | #include <linux/pagemap.h> | |
56 | #include <linux/syscalls.h> | |
7ed20e1a | 57 | #include <linux/signal.h> |
9984de1a | 58 | #include <linux/export.h> |
fd5eea42 | 59 | #include <linux/magic.h> |
b488893a PE |
60 | #include <linux/pid.h> |
61 | #include <linux/nsproxy.h> | |
bdbb776f | 62 | #include <linux/ptrace.h> |
8bd75c77 | 63 | #include <linux/sched/rt.h> |
13d60f4b | 64 | #include <linux/hugetlb.h> |
88c8004f | 65 | #include <linux/freezer.h> |
a52b89eb | 66 | #include <linux/bootmem.h> |
ab51fbab | 67 | #include <linux/fault-inject.h> |
b488893a | 68 | |
4732efbe | 69 | #include <asm/futex.h> |
1da177e4 | 70 | |
1696a8be | 71 | #include "locking/rtmutex_common.h" |
c87e2837 | 72 | |
99b60ce6 | 73 | /* |
d7e8af1a DB |
74 | * READ this before attempting to hack on futexes! |
75 | * | |
76 | * Basic futex operation and ordering guarantees | |
77 | * ============================================= | |
99b60ce6 TG |
78 | * |
79 | * The waiter reads the futex value in user space and calls | |
80 | * futex_wait(). This function computes the hash bucket and acquires | |
81 | * the hash bucket lock. After that it reads the futex user space value | |
b0c29f79 DB |
82 | * again and verifies that the data has not changed. If it has not changed |
83 | * it enqueues itself into the hash bucket, releases the hash bucket lock | |
84 | * and schedules. | |
99b60ce6 TG |
85 | * |
86 | * The waker side modifies the user space value of the futex and calls | |
b0c29f79 DB |
87 | * futex_wake(). This function computes the hash bucket and acquires the |
88 | * hash bucket lock. Then it looks for waiters on that futex in the hash | |
89 | * bucket and wakes them. | |
99b60ce6 | 90 | * |
b0c29f79 DB |
91 | * In futex wake up scenarios where no tasks are blocked on a futex, taking |
92 | * the hb spinlock can be avoided and simply return. In order for this | |
93 | * optimization to work, ordering guarantees must exist so that the waiter | |
94 | * being added to the list is acknowledged when the list is concurrently being | |
95 | * checked by the waker, avoiding scenarios like the following: | |
99b60ce6 TG |
96 | * |
97 | * CPU 0 CPU 1 | |
98 | * val = *futex; | |
99 | * sys_futex(WAIT, futex, val); | |
100 | * futex_wait(futex, val); | |
101 | * uval = *futex; | |
102 | * *futex = newval; | |
103 | * sys_futex(WAKE, futex); | |
104 | * futex_wake(futex); | |
105 | * if (queue_empty()) | |
106 | * return; | |
107 | * if (uval == val) | |
108 | * lock(hash_bucket(futex)); | |
109 | * queue(); | |
110 | * unlock(hash_bucket(futex)); | |
111 | * schedule(); | |
112 | * | |
113 | * This would cause the waiter on CPU 0 to wait forever because it | |
114 | * missed the transition of the user space value from val to newval | |
115 | * and the waker did not find the waiter in the hash bucket queue. | |
99b60ce6 | 116 | * |
b0c29f79 DB |
117 | * The correct serialization ensures that a waiter either observes |
118 | * the changed user space value before blocking or is woken by a | |
119 | * concurrent waker: | |
120 | * | |
121 | * CPU 0 CPU 1 | |
99b60ce6 TG |
122 | * val = *futex; |
123 | * sys_futex(WAIT, futex, val); | |
124 | * futex_wait(futex, val); | |
b0c29f79 | 125 | * |
d7e8af1a | 126 | * waiters++; (a) |
8ad7b378 DB |
127 | * smp_mb(); (A) <-- paired with -. |
128 | * | | |
129 | * lock(hash_bucket(futex)); | | |
130 | * | | |
131 | * uval = *futex; | | |
132 | * | *futex = newval; | |
133 | * | sys_futex(WAKE, futex); | |
134 | * | futex_wake(futex); | |
135 | * | | |
136 | * `--------> smp_mb(); (B) | |
99b60ce6 | 137 | * if (uval == val) |
b0c29f79 | 138 | * queue(); |
99b60ce6 | 139 | * unlock(hash_bucket(futex)); |
b0c29f79 DB |
140 | * schedule(); if (waiters) |
141 | * lock(hash_bucket(futex)); | |
d7e8af1a DB |
142 | * else wake_waiters(futex); |
143 | * waiters--; (b) unlock(hash_bucket(futex)); | |
b0c29f79 | 144 | * |
d7e8af1a DB |
145 | * Where (A) orders the waiters increment and the futex value read through |
146 | * atomic operations (see hb_waiters_inc) and where (B) orders the write | |
993b2ff2 DB |
147 | * to futex and the waiters read -- this is done by the barriers for both |
148 | * shared and private futexes in get_futex_key_refs(). | |
b0c29f79 DB |
149 | * |
150 | * This yields the following case (where X:=waiters, Y:=futex): | |
151 | * | |
152 | * X = Y = 0 | |
153 | * | |
154 | * w[X]=1 w[Y]=1 | |
155 | * MB MB | |
156 | * r[Y]=y r[X]=x | |
157 | * | |
158 | * Which guarantees that x==0 && y==0 is impossible; which translates back into | |
159 | * the guarantee that we cannot both miss the futex variable change and the | |
160 | * enqueue. | |
d7e8af1a DB |
161 | * |
162 | * Note that a new waiter is accounted for in (a) even when it is possible that | |
163 | * the wait call can return error, in which case we backtrack from it in (b). | |
164 | * Refer to the comment in queue_lock(). | |
165 | * | |
166 | * Similarly, in order to account for waiters being requeued on another | |
167 | * address we always increment the waiters for the destination bucket before | |
168 | * acquiring the lock. It then decrements them again after releasing it - | |
169 | * the code that actually moves the futex(es) between hash buckets (requeue_futex) | |
170 | * will do the additional required waiter count housekeeping. This is done for | |
171 | * double_lock_hb() and double_unlock_hb(), respectively. | |
99b60ce6 TG |
172 | */ |
173 | ||
03b8c7b6 | 174 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 175 | int __read_mostly futex_cmpxchg_enabled; |
03b8c7b6 | 176 | #endif |
a0c1e907 | 177 | |
b41277dc DH |
178 | /* |
179 | * Futex flags used to encode options to functions and preserve them across | |
180 | * restarts. | |
181 | */ | |
182 | #define FLAGS_SHARED 0x01 | |
183 | #define FLAGS_CLOCKRT 0x02 | |
184 | #define FLAGS_HAS_TIMEOUT 0x04 | |
185 | ||
c87e2837 IM |
186 | /* |
187 | * Priority Inheritance state: | |
188 | */ | |
189 | struct futex_pi_state { | |
190 | /* | |
191 | * list of 'owned' pi_state instances - these have to be | |
192 | * cleaned up in do_exit() if the task exits prematurely: | |
193 | */ | |
194 | struct list_head list; | |
195 | ||
196 | /* | |
197 | * The PI object: | |
198 | */ | |
199 | struct rt_mutex pi_mutex; | |
200 | ||
201 | struct task_struct *owner; | |
202 | atomic_t refcount; | |
203 | ||
204 | union futex_key key; | |
205 | }; | |
206 | ||
d8d88fbb DH |
207 | /** |
208 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 209 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
210 | * @task: the task waiting on the futex |
211 | * @lock_ptr: the hash bucket lock | |
212 | * @key: the key the futex is hashed on | |
213 | * @pi_state: optional priority inheritance state | |
214 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
215 | * @requeue_pi_key: the requeue_pi target futex key | |
216 | * @bitset: bitset for the optional bitmasked wakeup | |
217 | * | |
218 | * We use this hashed waitqueue, instead of a normal wait_queue_t, so | |
1da177e4 LT |
219 | * we can wake only the relevant ones (hashed queues may be shared). |
220 | * | |
221 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 222 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 223 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
224 | * the second. |
225 | * | |
226 | * PI futexes are typically woken before they are removed from the hash list via | |
227 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
228 | */ |
229 | struct futex_q { | |
ec92d082 | 230 | struct plist_node list; |
1da177e4 | 231 | |
d8d88fbb | 232 | struct task_struct *task; |
1da177e4 | 233 | spinlock_t *lock_ptr; |
1da177e4 | 234 | union futex_key key; |
c87e2837 | 235 | struct futex_pi_state *pi_state; |
52400ba9 | 236 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 237 | union futex_key *requeue_pi_key; |
cd689985 | 238 | u32 bitset; |
1da177e4 LT |
239 | }; |
240 | ||
5bdb05f9 DH |
241 | static const struct futex_q futex_q_init = { |
242 | /* list gets initialized in queue_me()*/ | |
243 | .key = FUTEX_KEY_INIT, | |
244 | .bitset = FUTEX_BITSET_MATCH_ANY | |
245 | }; | |
246 | ||
1da177e4 | 247 | /* |
b2d0994b DH |
248 | * Hash buckets are shared by all the futex_keys that hash to the same |
249 | * location. Each key may have multiple futex_q structures, one for each task | |
250 | * waiting on a futex. | |
1da177e4 LT |
251 | */ |
252 | struct futex_hash_bucket { | |
11d4616b | 253 | atomic_t waiters; |
ec92d082 PP |
254 | spinlock_t lock; |
255 | struct plist_head chain; | |
a52b89eb | 256 | } ____cacheline_aligned_in_smp; |
1da177e4 | 257 | |
ac742d37 RV |
258 | /* |
259 | * The base of the bucket array and its size are always used together | |
260 | * (after initialization only in hash_futex()), so ensure that they | |
261 | * reside in the same cacheline. | |
262 | */ | |
263 | static struct { | |
264 | struct futex_hash_bucket *queues; | |
265 | unsigned long hashsize; | |
266 | } __futex_data __read_mostly __aligned(2*sizeof(long)); | |
267 | #define futex_queues (__futex_data.queues) | |
268 | #define futex_hashsize (__futex_data.hashsize) | |
a52b89eb | 269 | |
1da177e4 | 270 | |
ab51fbab DB |
271 | /* |
272 | * Fault injections for futexes. | |
273 | */ | |
274 | #ifdef CONFIG_FAIL_FUTEX | |
275 | ||
276 | static struct { | |
277 | struct fault_attr attr; | |
278 | ||
621a5f7a | 279 | bool ignore_private; |
ab51fbab DB |
280 | } fail_futex = { |
281 | .attr = FAULT_ATTR_INITIALIZER, | |
621a5f7a | 282 | .ignore_private = false, |
ab51fbab DB |
283 | }; |
284 | ||
285 | static int __init setup_fail_futex(char *str) | |
286 | { | |
287 | return setup_fault_attr(&fail_futex.attr, str); | |
288 | } | |
289 | __setup("fail_futex=", setup_fail_futex); | |
290 | ||
5d285a7f | 291 | static bool should_fail_futex(bool fshared) |
ab51fbab DB |
292 | { |
293 | if (fail_futex.ignore_private && !fshared) | |
294 | return false; | |
295 | ||
296 | return should_fail(&fail_futex.attr, 1); | |
297 | } | |
298 | ||
299 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
300 | ||
301 | static int __init fail_futex_debugfs(void) | |
302 | { | |
303 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
304 | struct dentry *dir; | |
305 | ||
306 | dir = fault_create_debugfs_attr("fail_futex", NULL, | |
307 | &fail_futex.attr); | |
308 | if (IS_ERR(dir)) | |
309 | return PTR_ERR(dir); | |
310 | ||
311 | if (!debugfs_create_bool("ignore-private", mode, dir, | |
312 | &fail_futex.ignore_private)) { | |
313 | debugfs_remove_recursive(dir); | |
314 | return -ENOMEM; | |
315 | } | |
316 | ||
317 | return 0; | |
318 | } | |
319 | ||
320 | late_initcall(fail_futex_debugfs); | |
321 | ||
322 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
323 | ||
324 | #else | |
325 | static inline bool should_fail_futex(bool fshared) | |
326 | { | |
327 | return false; | |
328 | } | |
329 | #endif /* CONFIG_FAIL_FUTEX */ | |
330 | ||
b0c29f79 DB |
331 | static inline void futex_get_mm(union futex_key *key) |
332 | { | |
333 | atomic_inc(&key->private.mm->mm_count); | |
334 | /* | |
335 | * Ensure futex_get_mm() implies a full barrier such that | |
336 | * get_futex_key() implies a full barrier. This is relied upon | |
8ad7b378 | 337 | * as smp_mb(); (B), see the ordering comment above. |
b0c29f79 | 338 | */ |
4e857c58 | 339 | smp_mb__after_atomic(); |
b0c29f79 DB |
340 | } |
341 | ||
11d4616b LT |
342 | /* |
343 | * Reflects a new waiter being added to the waitqueue. | |
344 | */ | |
345 | static inline void hb_waiters_inc(struct futex_hash_bucket *hb) | |
b0c29f79 DB |
346 | { |
347 | #ifdef CONFIG_SMP | |
11d4616b | 348 | atomic_inc(&hb->waiters); |
b0c29f79 | 349 | /* |
11d4616b | 350 | * Full barrier (A), see the ordering comment above. |
b0c29f79 | 351 | */ |
4e857c58 | 352 | smp_mb__after_atomic(); |
11d4616b LT |
353 | #endif |
354 | } | |
355 | ||
356 | /* | |
357 | * Reflects a waiter being removed from the waitqueue by wakeup | |
358 | * paths. | |
359 | */ | |
360 | static inline void hb_waiters_dec(struct futex_hash_bucket *hb) | |
361 | { | |
362 | #ifdef CONFIG_SMP | |
363 | atomic_dec(&hb->waiters); | |
364 | #endif | |
365 | } | |
b0c29f79 | 366 | |
11d4616b LT |
367 | static inline int hb_waiters_pending(struct futex_hash_bucket *hb) |
368 | { | |
369 | #ifdef CONFIG_SMP | |
370 | return atomic_read(&hb->waiters); | |
b0c29f79 | 371 | #else |
11d4616b | 372 | return 1; |
b0c29f79 DB |
373 | #endif |
374 | } | |
375 | ||
1da177e4 LT |
376 | /* |
377 | * We hash on the keys returned from get_futex_key (see below). | |
378 | */ | |
379 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
380 | { | |
381 | u32 hash = jhash2((u32*)&key->both.word, | |
382 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
383 | key->both.offset); | |
a52b89eb | 384 | return &futex_queues[hash & (futex_hashsize - 1)]; |
1da177e4 LT |
385 | } |
386 | ||
387 | /* | |
388 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
389 | */ | |
390 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
391 | { | |
2bc87203 DH |
392 | return (key1 && key2 |
393 | && key1->both.word == key2->both.word | |
1da177e4 LT |
394 | && key1->both.ptr == key2->both.ptr |
395 | && key1->both.offset == key2->both.offset); | |
396 | } | |
397 | ||
38d47c1b PZ |
398 | /* |
399 | * Take a reference to the resource addressed by a key. | |
400 | * Can be called while holding spinlocks. | |
401 | * | |
402 | */ | |
403 | static void get_futex_key_refs(union futex_key *key) | |
404 | { | |
405 | if (!key->both.ptr) | |
406 | return; | |
407 | ||
408 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
409 | case FUT_OFF_INODE: | |
8ad7b378 | 410 | ihold(key->shared.inode); /* implies smp_mb(); (B) */ |
38d47c1b PZ |
411 | break; |
412 | case FUT_OFF_MMSHARED: | |
8ad7b378 | 413 | futex_get_mm(key); /* implies smp_mb(); (B) */ |
38d47c1b | 414 | break; |
76835b0e | 415 | default: |
993b2ff2 DB |
416 | /* |
417 | * Private futexes do not hold reference on an inode or | |
418 | * mm, therefore the only purpose of calling get_futex_key_refs | |
419 | * is because we need the barrier for the lockless waiter check. | |
420 | */ | |
8ad7b378 | 421 | smp_mb(); /* explicit smp_mb(); (B) */ |
38d47c1b PZ |
422 | } |
423 | } | |
424 | ||
425 | /* | |
426 | * Drop a reference to the resource addressed by a key. | |
993b2ff2 DB |
427 | * The hash bucket spinlock must not be held. This is |
428 | * a no-op for private futexes, see comment in the get | |
429 | * counterpart. | |
38d47c1b PZ |
430 | */ |
431 | static void drop_futex_key_refs(union futex_key *key) | |
432 | { | |
90621c40 DH |
433 | if (!key->both.ptr) { |
434 | /* If we're here then we tried to put a key we failed to get */ | |
435 | WARN_ON_ONCE(1); | |
38d47c1b | 436 | return; |
90621c40 | 437 | } |
38d47c1b PZ |
438 | |
439 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
440 | case FUT_OFF_INODE: | |
441 | iput(key->shared.inode); | |
442 | break; | |
443 | case FUT_OFF_MMSHARED: | |
444 | mmdrop(key->private.mm); | |
445 | break; | |
446 | } | |
447 | } | |
448 | ||
34f01cc1 | 449 | /** |
d96ee56c DH |
450 | * get_futex_key() - Get parameters which are the keys for a futex |
451 | * @uaddr: virtual address of the futex | |
452 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED | |
453 | * @key: address where result is stored. | |
9ea71503 SB |
454 | * @rw: mapping needs to be read/write (values: VERIFY_READ, |
455 | * VERIFY_WRITE) | |
34f01cc1 | 456 | * |
6c23cbbd RD |
457 | * Return: a negative error code or 0 |
458 | * | |
34f01cc1 | 459 | * The key words are stored in *key on success. |
1da177e4 | 460 | * |
6131ffaa | 461 | * For shared mappings, it's (page->index, file_inode(vma->vm_file), |
1da177e4 LT |
462 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
463 | * We can usually work out the index without swapping in the page. | |
464 | * | |
b2d0994b | 465 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 466 | */ |
64d1304a | 467 | static int |
9ea71503 | 468 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) |
1da177e4 | 469 | { |
e2970f2f | 470 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 471 | struct mm_struct *mm = current->mm; |
14d27abd KS |
472 | struct page *page; |
473 | struct address_space *mapping; | |
9ea71503 | 474 | int err, ro = 0; |
1da177e4 LT |
475 | |
476 | /* | |
477 | * The futex address must be "naturally" aligned. | |
478 | */ | |
e2970f2f | 479 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 480 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 481 | return -EINVAL; |
e2970f2f | 482 | address -= key->both.offset; |
1da177e4 | 483 | |
5cdec2d8 LT |
484 | if (unlikely(!access_ok(rw, uaddr, sizeof(u32)))) |
485 | return -EFAULT; | |
486 | ||
ab51fbab DB |
487 | if (unlikely(should_fail_futex(fshared))) |
488 | return -EFAULT; | |
489 | ||
34f01cc1 ED |
490 | /* |
491 | * PROCESS_PRIVATE futexes are fast. | |
492 | * As the mm cannot disappear under us and the 'key' only needs | |
493 | * virtual address, we dont even have to find the underlying vma. | |
494 | * Note : We do have to check 'uaddr' is a valid user address, | |
495 | * but access_ok() should be faster than find_vma() | |
496 | */ | |
497 | if (!fshared) { | |
34f01cc1 ED |
498 | key->private.mm = mm; |
499 | key->private.address = address; | |
8ad7b378 | 500 | get_futex_key_refs(key); /* implies smp_mb(); (B) */ |
34f01cc1 ED |
501 | return 0; |
502 | } | |
1da177e4 | 503 | |
38d47c1b | 504 | again: |
ab51fbab DB |
505 | /* Ignore any VERIFY_READ mapping (futex common case) */ |
506 | if (unlikely(should_fail_futex(fshared))) | |
507 | return -EFAULT; | |
508 | ||
7485d0d3 | 509 | err = get_user_pages_fast(address, 1, 1, &page); |
9ea71503 SB |
510 | /* |
511 | * If write access is not required (eg. FUTEX_WAIT), try | |
512 | * and get read-only access. | |
513 | */ | |
514 | if (err == -EFAULT && rw == VERIFY_READ) { | |
515 | err = get_user_pages_fast(address, 1, 0, &page); | |
516 | ro = 1; | |
517 | } | |
38d47c1b PZ |
518 | if (err < 0) |
519 | return err; | |
9ea71503 SB |
520 | else |
521 | err = 0; | |
38d47c1b | 522 | |
65d8fc77 MG |
523 | /* |
524 | * The treatment of mapping from this point on is critical. The page | |
525 | * lock protects many things but in this context the page lock | |
526 | * stabilizes mapping, prevents inode freeing in the shared | |
527 | * file-backed region case and guards against movement to swap cache. | |
528 | * | |
529 | * Strictly speaking the page lock is not needed in all cases being | |
530 | * considered here and page lock forces unnecessarily serialization | |
531 | * From this point on, mapping will be re-verified if necessary and | |
532 | * page lock will be acquired only if it is unavoidable | |
533 | */ | |
534 | page = compound_head(page); | |
535 | mapping = READ_ONCE(page->mapping); | |
536 | ||
e6780f72 | 537 | /* |
14d27abd | 538 | * If page->mapping is NULL, then it cannot be a PageAnon |
e6780f72 HD |
539 | * page; but it might be the ZERO_PAGE or in the gate area or |
540 | * in a special mapping (all cases which we are happy to fail); | |
541 | * or it may have been a good file page when get_user_pages_fast | |
542 | * found it, but truncated or holepunched or subjected to | |
543 | * invalidate_complete_page2 before we got the page lock (also | |
544 | * cases which we are happy to fail). And we hold a reference, | |
545 | * so refcount care in invalidate_complete_page's remove_mapping | |
546 | * prevents drop_caches from setting mapping to NULL beneath us. | |
547 | * | |
548 | * The case we do have to guard against is when memory pressure made | |
549 | * shmem_writepage move it from filecache to swapcache beneath us: | |
14d27abd | 550 | * an unlikely race, but we do need to retry for page->mapping. |
e6780f72 | 551 | */ |
65d8fc77 MG |
552 | if (unlikely(!mapping)) { |
553 | int shmem_swizzled; | |
554 | ||
555 | /* | |
556 | * Page lock is required to identify which special case above | |
557 | * applies. If this is really a shmem page then the page lock | |
558 | * will prevent unexpected transitions. | |
559 | */ | |
560 | lock_page(page); | |
561 | shmem_swizzled = PageSwapCache(page) || page->mapping; | |
14d27abd KS |
562 | unlock_page(page); |
563 | put_page(page); | |
65d8fc77 | 564 | |
e6780f72 HD |
565 | if (shmem_swizzled) |
566 | goto again; | |
65d8fc77 | 567 | |
e6780f72 | 568 | return -EFAULT; |
38d47c1b | 569 | } |
1da177e4 LT |
570 | |
571 | /* | |
572 | * Private mappings are handled in a simple way. | |
573 | * | |
65d8fc77 MG |
574 | * If the futex key is stored on an anonymous page, then the associated |
575 | * object is the mm which is implicitly pinned by the calling process. | |
576 | * | |
1da177e4 LT |
577 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if |
578 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 579 | * the object not the particular process. |
1da177e4 | 580 | */ |
14d27abd | 581 | if (PageAnon(page)) { |
9ea71503 SB |
582 | /* |
583 | * A RO anonymous page will never change and thus doesn't make | |
584 | * sense for futex operations. | |
585 | */ | |
ab51fbab | 586 | if (unlikely(should_fail_futex(fshared)) || ro) { |
9ea71503 SB |
587 | err = -EFAULT; |
588 | goto out; | |
589 | } | |
590 | ||
38d47c1b | 591 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 592 | key->private.mm = mm; |
e2970f2f | 593 | key->private.address = address; |
65d8fc77 MG |
594 | |
595 | get_futex_key_refs(key); /* implies smp_mb(); (B) */ | |
596 | ||
38d47c1b | 597 | } else { |
65d8fc77 MG |
598 | struct inode *inode; |
599 | ||
600 | /* | |
601 | * The associated futex object in this case is the inode and | |
602 | * the page->mapping must be traversed. Ordinarily this should | |
603 | * be stabilised under page lock but it's not strictly | |
604 | * necessary in this case as we just want to pin the inode, not | |
605 | * update the radix tree or anything like that. | |
606 | * | |
607 | * The RCU read lock is taken as the inode is finally freed | |
608 | * under RCU. If the mapping still matches expectations then the | |
609 | * mapping->host can be safely accessed as being a valid inode. | |
610 | */ | |
611 | rcu_read_lock(); | |
612 | ||
613 | if (READ_ONCE(page->mapping) != mapping) { | |
614 | rcu_read_unlock(); | |
615 | put_page(page); | |
616 | ||
617 | goto again; | |
618 | } | |
619 | ||
620 | inode = READ_ONCE(mapping->host); | |
621 | if (!inode) { | |
622 | rcu_read_unlock(); | |
623 | put_page(page); | |
624 | ||
625 | goto again; | |
626 | } | |
627 | ||
628 | /* | |
629 | * Take a reference unless it is about to be freed. Previously | |
630 | * this reference was taken by ihold under the page lock | |
631 | * pinning the inode in place so i_lock was unnecessary. The | |
632 | * only way for this check to fail is if the inode was | |
633 | * truncated in parallel so warn for now if this happens. | |
634 | * | |
635 | * We are not calling into get_futex_key_refs() in file-backed | |
636 | * cases, therefore a successful atomic_inc return below will | |
637 | * guarantee that get_futex_key() will still imply smp_mb(); (B). | |
638 | */ | |
639 | if (WARN_ON_ONCE(!atomic_inc_not_zero(&inode->i_count))) { | |
640 | rcu_read_unlock(); | |
641 | put_page(page); | |
642 | ||
643 | goto again; | |
644 | } | |
645 | ||
646 | /* Should be impossible but lets be paranoid for now */ | |
647 | if (WARN_ON_ONCE(inode->i_mapping != mapping)) { | |
648 | err = -EFAULT; | |
649 | rcu_read_unlock(); | |
650 | iput(inode); | |
651 | ||
652 | goto out; | |
653 | } | |
654 | ||
38d47c1b | 655 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ |
65d8fc77 | 656 | key->shared.inode = inode; |
13d60f4b | 657 | key->shared.pgoff = basepage_index(page); |
65d8fc77 | 658 | rcu_read_unlock(); |
1da177e4 LT |
659 | } |
660 | ||
9ea71503 | 661 | out: |
14d27abd | 662 | put_page(page); |
9ea71503 | 663 | return err; |
1da177e4 LT |
664 | } |
665 | ||
ae791a2d | 666 | static inline void put_futex_key(union futex_key *key) |
1da177e4 | 667 | { |
38d47c1b | 668 | drop_futex_key_refs(key); |
1da177e4 LT |
669 | } |
670 | ||
d96ee56c DH |
671 | /** |
672 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
673 | * @uaddr: pointer to faulting user space address |
674 | * | |
675 | * Slow path to fixup the fault we just took in the atomic write | |
676 | * access to @uaddr. | |
677 | * | |
fb62db2b | 678 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
679 | * user address. We know that we faulted in the atomic pagefault |
680 | * disabled section so we can as well avoid the #PF overhead by | |
681 | * calling get_user_pages() right away. | |
682 | */ | |
683 | static int fault_in_user_writeable(u32 __user *uaddr) | |
684 | { | |
722d0172 AK |
685 | struct mm_struct *mm = current->mm; |
686 | int ret; | |
687 | ||
688 | down_read(&mm->mmap_sem); | |
2efaca92 | 689 | ret = fixup_user_fault(current, mm, (unsigned long)uaddr, |
4a9e1cda | 690 | FAULT_FLAG_WRITE, NULL); |
722d0172 AK |
691 | up_read(&mm->mmap_sem); |
692 | ||
d0725992 TG |
693 | return ret < 0 ? ret : 0; |
694 | } | |
695 | ||
4b1c486b DH |
696 | /** |
697 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
698 | * @hb: the hash bucket the futex_q's reside in |
699 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
700 | * |
701 | * Must be called with the hb lock held. | |
702 | */ | |
703 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
704 | union futex_key *key) | |
705 | { | |
706 | struct futex_q *this; | |
707 | ||
708 | plist_for_each_entry(this, &hb->chain, list) { | |
709 | if (match_futex(&this->key, key)) | |
710 | return this; | |
711 | } | |
712 | return NULL; | |
713 | } | |
714 | ||
37a9d912 ML |
715 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
716 | u32 uval, u32 newval) | |
36cf3b5c | 717 | { |
37a9d912 | 718 | int ret; |
36cf3b5c TG |
719 | |
720 | pagefault_disable(); | |
37a9d912 | 721 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
722 | pagefault_enable(); |
723 | ||
37a9d912 | 724 | return ret; |
36cf3b5c TG |
725 | } |
726 | ||
727 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
728 | { |
729 | int ret; | |
730 | ||
a866374a | 731 | pagefault_disable(); |
e2970f2f | 732 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 733 | pagefault_enable(); |
1da177e4 LT |
734 | |
735 | return ret ? -EFAULT : 0; | |
736 | } | |
737 | ||
c87e2837 IM |
738 | |
739 | /* | |
740 | * PI code: | |
741 | */ | |
742 | static int refill_pi_state_cache(void) | |
743 | { | |
744 | struct futex_pi_state *pi_state; | |
745 | ||
746 | if (likely(current->pi_state_cache)) | |
747 | return 0; | |
748 | ||
4668edc3 | 749 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
750 | |
751 | if (!pi_state) | |
752 | return -ENOMEM; | |
753 | ||
c87e2837 IM |
754 | INIT_LIST_HEAD(&pi_state->list); |
755 | /* pi_mutex gets initialized later */ | |
756 | pi_state->owner = NULL; | |
757 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 758 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
759 | |
760 | current->pi_state_cache = pi_state; | |
761 | ||
762 | return 0; | |
763 | } | |
764 | ||
765 | static struct futex_pi_state * alloc_pi_state(void) | |
766 | { | |
767 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
768 | ||
769 | WARN_ON(!pi_state); | |
770 | current->pi_state_cache = NULL; | |
771 | ||
772 | return pi_state; | |
773 | } | |
774 | ||
30a6b803 | 775 | /* |
29e9ee5d TG |
776 | * Drops a reference to the pi_state object and frees or caches it |
777 | * when the last reference is gone. | |
778 | * | |
30a6b803 BS |
779 | * Must be called with the hb lock held. |
780 | */ | |
29e9ee5d | 781 | static void put_pi_state(struct futex_pi_state *pi_state) |
c87e2837 | 782 | { |
30a6b803 BS |
783 | if (!pi_state) |
784 | return; | |
785 | ||
c87e2837 IM |
786 | if (!atomic_dec_and_test(&pi_state->refcount)) |
787 | return; | |
788 | ||
789 | /* | |
790 | * If pi_state->owner is NULL, the owner is most probably dying | |
791 | * and has cleaned up the pi_state already | |
792 | */ | |
793 | if (pi_state->owner) { | |
1d615482 | 794 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
c87e2837 | 795 | list_del_init(&pi_state->list); |
1d615482 | 796 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
c87e2837 IM |
797 | |
798 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
799 | } | |
800 | ||
801 | if (current->pi_state_cache) | |
802 | kfree(pi_state); | |
803 | else { | |
804 | /* | |
805 | * pi_state->list is already empty. | |
806 | * clear pi_state->owner. | |
807 | * refcount is at 0 - put it back to 1. | |
808 | */ | |
809 | pi_state->owner = NULL; | |
810 | atomic_set(&pi_state->refcount, 1); | |
811 | current->pi_state_cache = pi_state; | |
812 | } | |
813 | } | |
814 | ||
815 | /* | |
816 | * Look up the task based on what TID userspace gave us. | |
817 | * We dont trust it. | |
818 | */ | |
819 | static struct task_struct * futex_find_get_task(pid_t pid) | |
820 | { | |
821 | struct task_struct *p; | |
822 | ||
d359b549 | 823 | rcu_read_lock(); |
228ebcbe | 824 | p = find_task_by_vpid(pid); |
7a0ea09a MH |
825 | if (p) |
826 | get_task_struct(p); | |
a06381fe | 827 | |
d359b549 | 828 | rcu_read_unlock(); |
c87e2837 IM |
829 | |
830 | return p; | |
831 | } | |
832 | ||
833 | /* | |
834 | * This task is holding PI mutexes at exit time => bad. | |
835 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
836 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
837 | */ | |
838 | void exit_pi_state_list(struct task_struct *curr) | |
839 | { | |
c87e2837 IM |
840 | struct list_head *next, *head = &curr->pi_state_list; |
841 | struct futex_pi_state *pi_state; | |
627371d7 | 842 | struct futex_hash_bucket *hb; |
38d47c1b | 843 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 844 | |
a0c1e907 TG |
845 | if (!futex_cmpxchg_enabled) |
846 | return; | |
c87e2837 IM |
847 | /* |
848 | * We are a ZOMBIE and nobody can enqueue itself on | |
849 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 850 | * versus waiters unqueueing themselves: |
c87e2837 | 851 | */ |
1d615482 | 852 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 IM |
853 | while (!list_empty(head)) { |
854 | ||
855 | next = head->next; | |
856 | pi_state = list_entry(next, struct futex_pi_state, list); | |
857 | key = pi_state->key; | |
627371d7 | 858 | hb = hash_futex(&key); |
1d615482 | 859 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 860 | |
c87e2837 IM |
861 | spin_lock(&hb->lock); |
862 | ||
1d615482 | 863 | raw_spin_lock_irq(&curr->pi_lock); |
627371d7 IM |
864 | /* |
865 | * We dropped the pi-lock, so re-check whether this | |
866 | * task still owns the PI-state: | |
867 | */ | |
c87e2837 IM |
868 | if (head->next != next) { |
869 | spin_unlock(&hb->lock); | |
870 | continue; | |
871 | } | |
872 | ||
c87e2837 | 873 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
874 | WARN_ON(list_empty(&pi_state->list)); |
875 | list_del_init(&pi_state->list); | |
c87e2837 | 876 | pi_state->owner = NULL; |
1d615482 | 877 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
878 | |
879 | rt_mutex_unlock(&pi_state->pi_mutex); | |
880 | ||
881 | spin_unlock(&hb->lock); | |
882 | ||
1d615482 | 883 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 884 | } |
1d615482 | 885 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
886 | } |
887 | ||
54a21788 TG |
888 | /* |
889 | * We need to check the following states: | |
890 | * | |
891 | * Waiter | pi_state | pi->owner | uTID | uODIED | ? | |
892 | * | |
893 | * [1] NULL | --- | --- | 0 | 0/1 | Valid | |
894 | * [2] NULL | --- | --- | >0 | 0/1 | Valid | |
895 | * | |
896 | * [3] Found | NULL | -- | Any | 0/1 | Invalid | |
897 | * | |
898 | * [4] Found | Found | NULL | 0 | 1 | Valid | |
899 | * [5] Found | Found | NULL | >0 | 1 | Invalid | |
900 | * | |
901 | * [6] Found | Found | task | 0 | 1 | Valid | |
902 | * | |
903 | * [7] Found | Found | NULL | Any | 0 | Invalid | |
904 | * | |
905 | * [8] Found | Found | task | ==taskTID | 0/1 | Valid | |
906 | * [9] Found | Found | task | 0 | 0 | Invalid | |
907 | * [10] Found | Found | task | !=taskTID | 0/1 | Invalid | |
908 | * | |
909 | * [1] Indicates that the kernel can acquire the futex atomically. We | |
910 | * came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. | |
911 | * | |
912 | * [2] Valid, if TID does not belong to a kernel thread. If no matching | |
913 | * thread is found then it indicates that the owner TID has died. | |
914 | * | |
915 | * [3] Invalid. The waiter is queued on a non PI futex | |
916 | * | |
917 | * [4] Valid state after exit_robust_list(), which sets the user space | |
918 | * value to FUTEX_WAITERS | FUTEX_OWNER_DIED. | |
919 | * | |
920 | * [5] The user space value got manipulated between exit_robust_list() | |
921 | * and exit_pi_state_list() | |
922 | * | |
923 | * [6] Valid state after exit_pi_state_list() which sets the new owner in | |
924 | * the pi_state but cannot access the user space value. | |
925 | * | |
926 | * [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. | |
927 | * | |
928 | * [8] Owner and user space value match | |
929 | * | |
930 | * [9] There is no transient state which sets the user space TID to 0 | |
931 | * except exit_robust_list(), but this is indicated by the | |
932 | * FUTEX_OWNER_DIED bit. See [4] | |
933 | * | |
934 | * [10] There is no transient state which leaves owner and user space | |
935 | * TID out of sync. | |
936 | */ | |
e60cbc5c TG |
937 | |
938 | /* | |
939 | * Validate that the existing waiter has a pi_state and sanity check | |
940 | * the pi_state against the user space value. If correct, attach to | |
941 | * it. | |
942 | */ | |
943 | static int attach_to_pi_state(u32 uval, struct futex_pi_state *pi_state, | |
944 | struct futex_pi_state **ps) | |
c87e2837 | 945 | { |
778e9a9c | 946 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 | 947 | |
e60cbc5c TG |
948 | /* |
949 | * Userspace might have messed up non-PI and PI futexes [3] | |
950 | */ | |
951 | if (unlikely(!pi_state)) | |
952 | return -EINVAL; | |
06a9ec29 | 953 | |
e60cbc5c | 954 | WARN_ON(!atomic_read(&pi_state->refcount)); |
59647b6a | 955 | |
e60cbc5c TG |
956 | /* |
957 | * Handle the owner died case: | |
958 | */ | |
959 | if (uval & FUTEX_OWNER_DIED) { | |
bd1dbcc6 | 960 | /* |
e60cbc5c TG |
961 | * exit_pi_state_list sets owner to NULL and wakes the |
962 | * topmost waiter. The task which acquires the | |
963 | * pi_state->rt_mutex will fixup owner. | |
bd1dbcc6 | 964 | */ |
e60cbc5c | 965 | if (!pi_state->owner) { |
59647b6a | 966 | /* |
e60cbc5c TG |
967 | * No pi state owner, but the user space TID |
968 | * is not 0. Inconsistent state. [5] | |
59647b6a | 969 | */ |
e60cbc5c TG |
970 | if (pid) |
971 | return -EINVAL; | |
bd1dbcc6 | 972 | /* |
e60cbc5c | 973 | * Take a ref on the state and return success. [4] |
866293ee | 974 | */ |
e60cbc5c | 975 | goto out_state; |
c87e2837 | 976 | } |
bd1dbcc6 TG |
977 | |
978 | /* | |
e60cbc5c TG |
979 | * If TID is 0, then either the dying owner has not |
980 | * yet executed exit_pi_state_list() or some waiter | |
981 | * acquired the rtmutex in the pi state, but did not | |
982 | * yet fixup the TID in user space. | |
983 | * | |
984 | * Take a ref on the state and return success. [6] | |
985 | */ | |
986 | if (!pid) | |
987 | goto out_state; | |
988 | } else { | |
989 | /* | |
990 | * If the owner died bit is not set, then the pi_state | |
991 | * must have an owner. [7] | |
bd1dbcc6 | 992 | */ |
e60cbc5c | 993 | if (!pi_state->owner) |
bd1dbcc6 | 994 | return -EINVAL; |
c87e2837 IM |
995 | } |
996 | ||
e60cbc5c TG |
997 | /* |
998 | * Bail out if user space manipulated the futex value. If pi | |
999 | * state exists then the owner TID must be the same as the | |
1000 | * user space TID. [9/10] | |
1001 | */ | |
1002 | if (pid != task_pid_vnr(pi_state->owner)) | |
1003 | return -EINVAL; | |
1004 | out_state: | |
1005 | atomic_inc(&pi_state->refcount); | |
1006 | *ps = pi_state; | |
1007 | return 0; | |
1008 | } | |
1009 | ||
04e1b2e5 TG |
1010 | /* |
1011 | * Lookup the task for the TID provided from user space and attach to | |
1012 | * it after doing proper sanity checks. | |
1013 | */ | |
1014 | static int attach_to_pi_owner(u32 uval, union futex_key *key, | |
1015 | struct futex_pi_state **ps) | |
e60cbc5c | 1016 | { |
e60cbc5c | 1017 | pid_t pid = uval & FUTEX_TID_MASK; |
04e1b2e5 TG |
1018 | struct futex_pi_state *pi_state; |
1019 | struct task_struct *p; | |
e60cbc5c | 1020 | |
c87e2837 | 1021 | /* |
e3f2ddea | 1022 | * We are the first waiter - try to look up the real owner and attach |
54a21788 | 1023 | * the new pi_state to it, but bail out when TID = 0 [1] |
c87e2837 | 1024 | */ |
778e9a9c | 1025 | if (!pid) |
e3f2ddea | 1026 | return -ESRCH; |
c87e2837 | 1027 | p = futex_find_get_task(pid); |
7a0ea09a MH |
1028 | if (!p) |
1029 | return -ESRCH; | |
778e9a9c | 1030 | |
a2129464 | 1031 | if (unlikely(p->flags & PF_KTHREAD)) { |
f0d71b3d TG |
1032 | put_task_struct(p); |
1033 | return -EPERM; | |
1034 | } | |
1035 | ||
778e9a9c AK |
1036 | /* |
1037 | * We need to look at the task state flags to figure out, | |
1038 | * whether the task is exiting. To protect against the do_exit | |
1039 | * change of the task flags, we do this protected by | |
1040 | * p->pi_lock: | |
1041 | */ | |
1d615482 | 1042 | raw_spin_lock_irq(&p->pi_lock); |
778e9a9c AK |
1043 | if (unlikely(p->flags & PF_EXITING)) { |
1044 | /* | |
1045 | * The task is on the way out. When PF_EXITPIDONE is | |
1046 | * set, we know that the task has finished the | |
1047 | * cleanup: | |
1048 | */ | |
1049 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
1050 | ||
1d615482 | 1051 | raw_spin_unlock_irq(&p->pi_lock); |
778e9a9c AK |
1052 | put_task_struct(p); |
1053 | return ret; | |
1054 | } | |
c87e2837 | 1055 | |
54a21788 TG |
1056 | /* |
1057 | * No existing pi state. First waiter. [2] | |
1058 | */ | |
c87e2837 IM |
1059 | pi_state = alloc_pi_state(); |
1060 | ||
1061 | /* | |
04e1b2e5 | 1062 | * Initialize the pi_mutex in locked state and make @p |
c87e2837 IM |
1063 | * the owner of it: |
1064 | */ | |
1065 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
1066 | ||
1067 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 1068 | pi_state->key = *key; |
c87e2837 | 1069 | |
627371d7 | 1070 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
1071 | list_add(&pi_state->list, &p->pi_state_list); |
1072 | pi_state->owner = p; | |
1d615482 | 1073 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
1074 | |
1075 | put_task_struct(p); | |
1076 | ||
d0aa7a70 | 1077 | *ps = pi_state; |
c87e2837 IM |
1078 | |
1079 | return 0; | |
1080 | } | |
1081 | ||
04e1b2e5 TG |
1082 | static int lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
1083 | union futex_key *key, struct futex_pi_state **ps) | |
1084 | { | |
1085 | struct futex_q *match = futex_top_waiter(hb, key); | |
1086 | ||
1087 | /* | |
1088 | * If there is a waiter on that futex, validate it and | |
1089 | * attach to the pi_state when the validation succeeds. | |
1090 | */ | |
1091 | if (match) | |
1092 | return attach_to_pi_state(uval, match->pi_state, ps); | |
1093 | ||
1094 | /* | |
1095 | * We are the first waiter - try to look up the owner based on | |
1096 | * @uval and attach to it. | |
1097 | */ | |
1098 | return attach_to_pi_owner(uval, key, ps); | |
1099 | } | |
1100 | ||
af54d6a1 TG |
1101 | static int lock_pi_update_atomic(u32 __user *uaddr, u32 uval, u32 newval) |
1102 | { | |
1103 | u32 uninitialized_var(curval); | |
1104 | ||
ab51fbab DB |
1105 | if (unlikely(should_fail_futex(true))) |
1106 | return -EFAULT; | |
1107 | ||
af54d6a1 TG |
1108 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))) |
1109 | return -EFAULT; | |
1110 | ||
1111 | /*If user space value changed, let the caller retry */ | |
1112 | return curval != uval ? -EAGAIN : 0; | |
1113 | } | |
1114 | ||
1a52084d | 1115 | /** |
d96ee56c | 1116 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
1117 | * @uaddr: the pi futex user address |
1118 | * @hb: the pi futex hash bucket | |
1119 | * @key: the futex key associated with uaddr and hb | |
1120 | * @ps: the pi_state pointer where we store the result of the | |
1121 | * lookup | |
1122 | * @task: the task to perform the atomic lock work for. This will | |
1123 | * be "current" except in the case of requeue pi. | |
1124 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d | 1125 | * |
6c23cbbd RD |
1126 | * Return: |
1127 | * 0 - ready to wait; | |
1128 | * 1 - acquired the lock; | |
1a52084d DH |
1129 | * <0 - error |
1130 | * | |
1131 | * The hb->lock and futex_key refs shall be held by the caller. | |
1132 | */ | |
1133 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
1134 | union futex_key *key, | |
1135 | struct futex_pi_state **ps, | |
bab5bc9e | 1136 | struct task_struct *task, int set_waiters) |
1a52084d | 1137 | { |
af54d6a1 TG |
1138 | u32 uval, newval, vpid = task_pid_vnr(task); |
1139 | struct futex_q *match; | |
1140 | int ret; | |
1a52084d DH |
1141 | |
1142 | /* | |
af54d6a1 TG |
1143 | * Read the user space value first so we can validate a few |
1144 | * things before proceeding further. | |
1a52084d | 1145 | */ |
af54d6a1 | 1146 | if (get_futex_value_locked(&uval, uaddr)) |
1a52084d DH |
1147 | return -EFAULT; |
1148 | ||
ab51fbab DB |
1149 | if (unlikely(should_fail_futex(true))) |
1150 | return -EFAULT; | |
1151 | ||
1a52084d DH |
1152 | /* |
1153 | * Detect deadlocks. | |
1154 | */ | |
af54d6a1 | 1155 | if ((unlikely((uval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
1156 | return -EDEADLK; |
1157 | ||
ab51fbab DB |
1158 | if ((unlikely(should_fail_futex(true)))) |
1159 | return -EDEADLK; | |
1160 | ||
1a52084d | 1161 | /* |
af54d6a1 TG |
1162 | * Lookup existing state first. If it exists, try to attach to |
1163 | * its pi_state. | |
1a52084d | 1164 | */ |
af54d6a1 TG |
1165 | match = futex_top_waiter(hb, key); |
1166 | if (match) | |
1167 | return attach_to_pi_state(uval, match->pi_state, ps); | |
1a52084d DH |
1168 | |
1169 | /* | |
af54d6a1 TG |
1170 | * No waiter and user TID is 0. We are here because the |
1171 | * waiters or the owner died bit is set or called from | |
1172 | * requeue_cmp_pi or for whatever reason something took the | |
1173 | * syscall. | |
1a52084d | 1174 | */ |
af54d6a1 | 1175 | if (!(uval & FUTEX_TID_MASK)) { |
59fa6245 | 1176 | /* |
af54d6a1 TG |
1177 | * We take over the futex. No other waiters and the user space |
1178 | * TID is 0. We preserve the owner died bit. | |
59fa6245 | 1179 | */ |
af54d6a1 TG |
1180 | newval = uval & FUTEX_OWNER_DIED; |
1181 | newval |= vpid; | |
1a52084d | 1182 | |
af54d6a1 TG |
1183 | /* The futex requeue_pi code can enforce the waiters bit */ |
1184 | if (set_waiters) | |
1185 | newval |= FUTEX_WAITERS; | |
1186 | ||
1187 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1188 | /* If the take over worked, return 1 */ | |
1189 | return ret < 0 ? ret : 1; | |
1190 | } | |
1a52084d DH |
1191 | |
1192 | /* | |
af54d6a1 TG |
1193 | * First waiter. Set the waiters bit before attaching ourself to |
1194 | * the owner. If owner tries to unlock, it will be forced into | |
1195 | * the kernel and blocked on hb->lock. | |
1a52084d | 1196 | */ |
af54d6a1 TG |
1197 | newval = uval | FUTEX_WAITERS; |
1198 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1199 | if (ret) | |
1200 | return ret; | |
1a52084d | 1201 | /* |
af54d6a1 TG |
1202 | * If the update of the user space value succeeded, we try to |
1203 | * attach to the owner. If that fails, no harm done, we only | |
1204 | * set the FUTEX_WAITERS bit in the user space variable. | |
1a52084d | 1205 | */ |
af54d6a1 | 1206 | return attach_to_pi_owner(uval, key, ps); |
1a52084d DH |
1207 | } |
1208 | ||
2e12978a LJ |
1209 | /** |
1210 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
1211 | * @q: The futex_q to unqueue | |
1212 | * | |
1213 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
1214 | */ | |
1215 | static void __unqueue_futex(struct futex_q *q) | |
1216 | { | |
1217 | struct futex_hash_bucket *hb; | |
1218 | ||
29096202 SR |
1219 | if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr)) |
1220 | || WARN_ON(plist_node_empty(&q->list))) | |
2e12978a LJ |
1221 | return; |
1222 | ||
1223 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
1224 | plist_del(&q->list, &hb->chain); | |
11d4616b | 1225 | hb_waiters_dec(hb); |
2e12978a LJ |
1226 | } |
1227 | ||
1da177e4 LT |
1228 | /* |
1229 | * The hash bucket lock must be held when this is called. | |
1d0dcb3a DB |
1230 | * Afterwards, the futex_q must not be accessed. Callers |
1231 | * must ensure to later call wake_up_q() for the actual | |
1232 | * wakeups to occur. | |
1da177e4 | 1233 | */ |
1d0dcb3a | 1234 | static void mark_wake_futex(struct wake_q_head *wake_q, struct futex_q *q) |
1da177e4 | 1235 | { |
f1a11e05 TG |
1236 | struct task_struct *p = q->task; |
1237 | ||
aa10990e DH |
1238 | if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n")) |
1239 | return; | |
1240 | ||
1da177e4 | 1241 | /* |
1d0dcb3a DB |
1242 | * Queue the task for later wakeup for after we've released |
1243 | * the hb->lock. wake_q_add() grabs reference to p. | |
1da177e4 | 1244 | */ |
1d0dcb3a | 1245 | wake_q_add(wake_q, p); |
2e12978a | 1246 | __unqueue_futex(q); |
1da177e4 | 1247 | /* |
f1a11e05 TG |
1248 | * The waiting task can free the futex_q as soon as |
1249 | * q->lock_ptr = NULL is written, without taking any locks. A | |
1250 | * memory barrier is required here to prevent the following | |
1251 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 1252 | */ |
ccdea2f8 | 1253 | smp_wmb(); |
1da177e4 LT |
1254 | q->lock_ptr = NULL; |
1255 | } | |
1256 | ||
802ab58d SAS |
1257 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this, |
1258 | struct futex_hash_bucket *hb) | |
c87e2837 IM |
1259 | { |
1260 | struct task_struct *new_owner; | |
1261 | struct futex_pi_state *pi_state = this->pi_state; | |
7cfdaf38 | 1262 | u32 uninitialized_var(curval), newval; |
802ab58d SAS |
1263 | WAKE_Q(wake_q); |
1264 | bool deboost; | |
13fbca4c | 1265 | int ret = 0; |
c87e2837 IM |
1266 | |
1267 | if (!pi_state) | |
1268 | return -EINVAL; | |
1269 | ||
51246bfd TG |
1270 | /* |
1271 | * If current does not own the pi_state then the futex is | |
1272 | * inconsistent and user space fiddled with the futex value. | |
1273 | */ | |
1274 | if (pi_state->owner != current) | |
1275 | return -EINVAL; | |
1276 | ||
b4abf910 | 1277 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
1278 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
1279 | ||
1280 | /* | |
f123c98e SR |
1281 | * It is possible that the next waiter (the one that brought |
1282 | * this owner to the kernel) timed out and is no longer | |
1283 | * waiting on the lock. | |
c87e2837 IM |
1284 | */ |
1285 | if (!new_owner) | |
1286 | new_owner = this->task; | |
1287 | ||
1288 | /* | |
13fbca4c TG |
1289 | * We pass it to the next owner. The WAITERS bit is always |
1290 | * kept enabled while there is PI state around. We cleanup the | |
1291 | * owner died bit, because we are the owner. | |
c87e2837 | 1292 | */ |
13fbca4c | 1293 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 1294 | |
ab51fbab DB |
1295 | if (unlikely(should_fail_futex(true))) |
1296 | ret = -EFAULT; | |
1297 | ||
13fbca4c TG |
1298 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1299 | ret = -EFAULT; | |
1300 | else if (curval != uval) | |
1301 | ret = -EINVAL; | |
1302 | if (ret) { | |
b4abf910 | 1303 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
13fbca4c | 1304 | return ret; |
e3f2ddea | 1305 | } |
c87e2837 | 1306 | |
b4abf910 | 1307 | raw_spin_lock(&pi_state->owner->pi_lock); |
627371d7 IM |
1308 | WARN_ON(list_empty(&pi_state->list)); |
1309 | list_del_init(&pi_state->list); | |
b4abf910 | 1310 | raw_spin_unlock(&pi_state->owner->pi_lock); |
627371d7 | 1311 | |
b4abf910 | 1312 | raw_spin_lock(&new_owner->pi_lock); |
627371d7 | 1313 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
1314 | list_add(&pi_state->list, &new_owner->pi_state_list); |
1315 | pi_state->owner = new_owner; | |
b4abf910 | 1316 | raw_spin_unlock(&new_owner->pi_lock); |
627371d7 | 1317 | |
b4abf910 | 1318 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
802ab58d SAS |
1319 | |
1320 | deboost = rt_mutex_futex_unlock(&pi_state->pi_mutex, &wake_q); | |
1321 | ||
1322 | /* | |
1323 | * First unlock HB so the waiter does not spin on it once he got woken | |
1324 | * up. Second wake up the waiter before the priority is adjusted. If we | |
1325 | * deboost first (and lose our higher priority), then the task might get | |
1326 | * scheduled away before the wake up can take place. | |
1327 | */ | |
1328 | spin_unlock(&hb->lock); | |
1329 | wake_up_q(&wake_q); | |
1330 | if (deboost) | |
1331 | rt_mutex_adjust_prio(current); | |
c87e2837 IM |
1332 | |
1333 | return 0; | |
1334 | } | |
1335 | ||
8b8f319f IM |
1336 | /* |
1337 | * Express the locking dependencies for lockdep: | |
1338 | */ | |
1339 | static inline void | |
1340 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1341 | { | |
1342 | if (hb1 <= hb2) { | |
1343 | spin_lock(&hb1->lock); | |
1344 | if (hb1 < hb2) | |
1345 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
1346 | } else { /* hb1 > hb2 */ | |
1347 | spin_lock(&hb2->lock); | |
1348 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
1349 | } | |
1350 | } | |
1351 | ||
5eb3dc62 DH |
1352 | static inline void |
1353 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1354 | { | |
f061d351 | 1355 | spin_unlock(&hb1->lock); |
88f502fe IM |
1356 | if (hb1 != hb2) |
1357 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
1358 | } |
1359 | ||
1da177e4 | 1360 | /* |
b2d0994b | 1361 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 1362 | */ |
b41277dc DH |
1363 | static int |
1364 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 1365 | { |
e2970f2f | 1366 | struct futex_hash_bucket *hb; |
1da177e4 | 1367 | struct futex_q *this, *next; |
38d47c1b | 1368 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 | 1369 | int ret; |
1d0dcb3a | 1370 | WAKE_Q(wake_q); |
1da177e4 | 1371 | |
cd689985 TG |
1372 | if (!bitset) |
1373 | return -EINVAL; | |
1374 | ||
9ea71503 | 1375 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ); |
1da177e4 LT |
1376 | if (unlikely(ret != 0)) |
1377 | goto out; | |
1378 | ||
e2970f2f | 1379 | hb = hash_futex(&key); |
b0c29f79 DB |
1380 | |
1381 | /* Make sure we really have tasks to wakeup */ | |
1382 | if (!hb_waiters_pending(hb)) | |
1383 | goto out_put_key; | |
1384 | ||
e2970f2f | 1385 | spin_lock(&hb->lock); |
1da177e4 | 1386 | |
0d00c7b2 | 1387 | plist_for_each_entry_safe(this, next, &hb->chain, list) { |
1da177e4 | 1388 | if (match_futex (&this->key, &key)) { |
52400ba9 | 1389 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
1390 | ret = -EINVAL; |
1391 | break; | |
1392 | } | |
cd689985 TG |
1393 | |
1394 | /* Check if one of the bits is set in both bitsets */ | |
1395 | if (!(this->bitset & bitset)) | |
1396 | continue; | |
1397 | ||
1d0dcb3a | 1398 | mark_wake_futex(&wake_q, this); |
1da177e4 LT |
1399 | if (++ret >= nr_wake) |
1400 | break; | |
1401 | } | |
1402 | } | |
1403 | ||
e2970f2f | 1404 | spin_unlock(&hb->lock); |
1d0dcb3a | 1405 | wake_up_q(&wake_q); |
b0c29f79 | 1406 | out_put_key: |
ae791a2d | 1407 | put_futex_key(&key); |
42d35d48 | 1408 | out: |
1da177e4 LT |
1409 | return ret; |
1410 | } | |
1411 | ||
4732efbe JJ |
1412 | /* |
1413 | * Wake up all waiters hashed on the physical page that is mapped | |
1414 | * to this virtual address: | |
1415 | */ | |
e2970f2f | 1416 | static int |
b41277dc | 1417 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 1418 | int nr_wake, int nr_wake2, int op) |
4732efbe | 1419 | { |
38d47c1b | 1420 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 1421 | struct futex_hash_bucket *hb1, *hb2; |
4732efbe | 1422 | struct futex_q *this, *next; |
e4dc5b7a | 1423 | int ret, op_ret; |
1d0dcb3a | 1424 | WAKE_Q(wake_q); |
4732efbe | 1425 | |
e4dc5b7a | 1426 | retry: |
9ea71503 | 1427 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
4732efbe JJ |
1428 | if (unlikely(ret != 0)) |
1429 | goto out; | |
9ea71503 | 1430 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
4732efbe | 1431 | if (unlikely(ret != 0)) |
42d35d48 | 1432 | goto out_put_key1; |
4732efbe | 1433 | |
e2970f2f IM |
1434 | hb1 = hash_futex(&key1); |
1435 | hb2 = hash_futex(&key2); | |
4732efbe | 1436 | |
e4dc5b7a | 1437 | retry_private: |
eaaea803 | 1438 | double_lock_hb(hb1, hb2); |
e2970f2f | 1439 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1440 | if (unlikely(op_ret < 0)) { |
4732efbe | 1441 | |
5eb3dc62 | 1442 | double_unlock_hb(hb1, hb2); |
4732efbe | 1443 | |
7ee1dd3f | 1444 | #ifndef CONFIG_MMU |
e2970f2f IM |
1445 | /* |
1446 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
1447 | * but we might get them from range checking | |
1448 | */ | |
7ee1dd3f | 1449 | ret = op_ret; |
42d35d48 | 1450 | goto out_put_keys; |
7ee1dd3f DH |
1451 | #endif |
1452 | ||
796f8d9b DG |
1453 | if (unlikely(op_ret != -EFAULT)) { |
1454 | ret = op_ret; | |
42d35d48 | 1455 | goto out_put_keys; |
796f8d9b DG |
1456 | } |
1457 | ||
d0725992 | 1458 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 1459 | if (ret) |
de87fcc1 | 1460 | goto out_put_keys; |
4732efbe | 1461 | |
b41277dc | 1462 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1463 | goto retry_private; |
1464 | ||
ae791a2d TG |
1465 | put_futex_key(&key2); |
1466 | put_futex_key(&key1); | |
e4dc5b7a | 1467 | goto retry; |
4732efbe JJ |
1468 | } |
1469 | ||
0d00c7b2 | 1470 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
4732efbe | 1471 | if (match_futex (&this->key, &key1)) { |
aa10990e DH |
1472 | if (this->pi_state || this->rt_waiter) { |
1473 | ret = -EINVAL; | |
1474 | goto out_unlock; | |
1475 | } | |
1d0dcb3a | 1476 | mark_wake_futex(&wake_q, this); |
4732efbe JJ |
1477 | if (++ret >= nr_wake) |
1478 | break; | |
1479 | } | |
1480 | } | |
1481 | ||
1482 | if (op_ret > 0) { | |
4732efbe | 1483 | op_ret = 0; |
0d00c7b2 | 1484 | plist_for_each_entry_safe(this, next, &hb2->chain, list) { |
4732efbe | 1485 | if (match_futex (&this->key, &key2)) { |
aa10990e DH |
1486 | if (this->pi_state || this->rt_waiter) { |
1487 | ret = -EINVAL; | |
1488 | goto out_unlock; | |
1489 | } | |
1d0dcb3a | 1490 | mark_wake_futex(&wake_q, this); |
4732efbe JJ |
1491 | if (++op_ret >= nr_wake2) |
1492 | break; | |
1493 | } | |
1494 | } | |
1495 | ret += op_ret; | |
1496 | } | |
1497 | ||
aa10990e | 1498 | out_unlock: |
5eb3dc62 | 1499 | double_unlock_hb(hb1, hb2); |
1d0dcb3a | 1500 | wake_up_q(&wake_q); |
42d35d48 | 1501 | out_put_keys: |
ae791a2d | 1502 | put_futex_key(&key2); |
42d35d48 | 1503 | out_put_key1: |
ae791a2d | 1504 | put_futex_key(&key1); |
42d35d48 | 1505 | out: |
4732efbe JJ |
1506 | return ret; |
1507 | } | |
1508 | ||
9121e478 DH |
1509 | /** |
1510 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1511 | * @q: the futex_q to requeue | |
1512 | * @hb1: the source hash_bucket | |
1513 | * @hb2: the target hash_bucket | |
1514 | * @key2: the new key for the requeued futex_q | |
1515 | */ | |
1516 | static inline | |
1517 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1518 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1519 | { | |
1520 | ||
1521 | /* | |
1522 | * If key1 and key2 hash to the same bucket, no need to | |
1523 | * requeue. | |
1524 | */ | |
1525 | if (likely(&hb1->chain != &hb2->chain)) { | |
1526 | plist_del(&q->list, &hb1->chain); | |
11d4616b | 1527 | hb_waiters_dec(hb1); |
9121e478 | 1528 | plist_add(&q->list, &hb2->chain); |
11d4616b | 1529 | hb_waiters_inc(hb2); |
9121e478 | 1530 | q->lock_ptr = &hb2->lock; |
9121e478 DH |
1531 | } |
1532 | get_futex_key_refs(key2); | |
1533 | q->key = *key2; | |
1534 | } | |
1535 | ||
52400ba9 DH |
1536 | /** |
1537 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1538 | * @q: the futex_q |
1539 | * @key: the key of the requeue target futex | |
1540 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1541 | * |
1542 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1543 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1544 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1545 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1546 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1547 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1548 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1549 | */ |
1550 | static inline | |
beda2c7e DH |
1551 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1552 | struct futex_hash_bucket *hb) | |
52400ba9 | 1553 | { |
52400ba9 DH |
1554 | get_futex_key_refs(key); |
1555 | q->key = *key; | |
1556 | ||
2e12978a | 1557 | __unqueue_futex(q); |
52400ba9 DH |
1558 | |
1559 | WARN_ON(!q->rt_waiter); | |
1560 | q->rt_waiter = NULL; | |
1561 | ||
beda2c7e | 1562 | q->lock_ptr = &hb->lock; |
beda2c7e | 1563 | |
f1a11e05 | 1564 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1565 | } |
1566 | ||
1567 | /** | |
1568 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1569 | * @pifutex: the user address of the to futex |
1570 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1571 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1572 | * @key1: the from futex key | |
1573 | * @key2: the to futex key | |
1574 | * @ps: address to store the pi_state pointer | |
1575 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1576 | * |
1577 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1578 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1579 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1580 | * hb1 and hb2 must be held by the caller. | |
52400ba9 | 1581 | * |
6c23cbbd RD |
1582 | * Return: |
1583 | * 0 - failed to acquire the lock atomically; | |
866293ee | 1584 | * >0 - acquired the lock, return value is vpid of the top_waiter |
52400ba9 DH |
1585 | * <0 - error |
1586 | */ | |
1587 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1588 | struct futex_hash_bucket *hb1, | |
1589 | struct futex_hash_bucket *hb2, | |
1590 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1591 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1592 | { |
bab5bc9e | 1593 | struct futex_q *top_waiter = NULL; |
52400ba9 | 1594 | u32 curval; |
866293ee | 1595 | int ret, vpid; |
52400ba9 DH |
1596 | |
1597 | if (get_futex_value_locked(&curval, pifutex)) | |
1598 | return -EFAULT; | |
1599 | ||
ab51fbab DB |
1600 | if (unlikely(should_fail_futex(true))) |
1601 | return -EFAULT; | |
1602 | ||
bab5bc9e DH |
1603 | /* |
1604 | * Find the top_waiter and determine if there are additional waiters. | |
1605 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1606 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1607 | * as we have means to handle the possible fault. If not, don't set | |
1608 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1609 | * the kernel. | |
1610 | */ | |
52400ba9 DH |
1611 | top_waiter = futex_top_waiter(hb1, key1); |
1612 | ||
1613 | /* There are no waiters, nothing for us to do. */ | |
1614 | if (!top_waiter) | |
1615 | return 0; | |
1616 | ||
84bc4af5 DH |
1617 | /* Ensure we requeue to the expected futex. */ |
1618 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1619 | return -EINVAL; | |
1620 | ||
52400ba9 | 1621 | /* |
bab5bc9e DH |
1622 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1623 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1624 | * in ps in contended cases. | |
52400ba9 | 1625 | */ |
866293ee | 1626 | vpid = task_pid_vnr(top_waiter->task); |
bab5bc9e DH |
1627 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1628 | set_waiters); | |
866293ee | 1629 | if (ret == 1) { |
beda2c7e | 1630 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
866293ee TG |
1631 | return vpid; |
1632 | } | |
52400ba9 DH |
1633 | return ret; |
1634 | } | |
1635 | ||
1636 | /** | |
1637 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1638 | * @uaddr1: source futex user address |
b41277dc | 1639 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1640 | * @uaddr2: target futex user address |
1641 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1642 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1643 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1644 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1645 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1646 | * |
1647 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1648 | * uaddr2 atomically on behalf of the top waiter. | |
1649 | * | |
6c23cbbd RD |
1650 | * Return: |
1651 | * >=0 - on success, the number of tasks requeued or woken; | |
52400ba9 | 1652 | * <0 - on error |
1da177e4 | 1653 | */ |
b41277dc DH |
1654 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
1655 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
1656 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 1657 | { |
38d47c1b | 1658 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1659 | int drop_count = 0, task_count = 0, ret; |
1660 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1661 | struct futex_hash_bucket *hb1, *hb2; |
1da177e4 | 1662 | struct futex_q *this, *next; |
1d0dcb3a | 1663 | WAKE_Q(wake_q); |
52400ba9 DH |
1664 | |
1665 | if (requeue_pi) { | |
e9c243a5 TG |
1666 | /* |
1667 | * Requeue PI only works on two distinct uaddrs. This | |
1668 | * check is only valid for private futexes. See below. | |
1669 | */ | |
1670 | if (uaddr1 == uaddr2) | |
1671 | return -EINVAL; | |
1672 | ||
52400ba9 DH |
1673 | /* |
1674 | * requeue_pi requires a pi_state, try to allocate it now | |
1675 | * without any locks in case it fails. | |
1676 | */ | |
1677 | if (refill_pi_state_cache()) | |
1678 | return -ENOMEM; | |
1679 | /* | |
1680 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1681 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1682 | * returning to userspace, so as to not leave the rt_mutex with | |
1683 | * waiters and no owner. However, second and third wake-ups | |
1684 | * cannot be predicted as they involve race conditions with the | |
1685 | * first wake and a fault while looking up the pi_state. Both | |
1686 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1687 | * use nr_wake=1. | |
1688 | */ | |
1689 | if (nr_wake != 1) | |
1690 | return -EINVAL; | |
1691 | } | |
1da177e4 | 1692 | |
42d35d48 | 1693 | retry: |
9ea71503 | 1694 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
1da177e4 LT |
1695 | if (unlikely(ret != 0)) |
1696 | goto out; | |
9ea71503 SB |
1697 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, |
1698 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1699 | if (unlikely(ret != 0)) |
42d35d48 | 1700 | goto out_put_key1; |
1da177e4 | 1701 | |
e9c243a5 TG |
1702 | /* |
1703 | * The check above which compares uaddrs is not sufficient for | |
1704 | * shared futexes. We need to compare the keys: | |
1705 | */ | |
1706 | if (requeue_pi && match_futex(&key1, &key2)) { | |
1707 | ret = -EINVAL; | |
1708 | goto out_put_keys; | |
1709 | } | |
1710 | ||
e2970f2f IM |
1711 | hb1 = hash_futex(&key1); |
1712 | hb2 = hash_futex(&key2); | |
1da177e4 | 1713 | |
e4dc5b7a | 1714 | retry_private: |
69cd9eba | 1715 | hb_waiters_inc(hb2); |
8b8f319f | 1716 | double_lock_hb(hb1, hb2); |
1da177e4 | 1717 | |
e2970f2f IM |
1718 | if (likely(cmpval != NULL)) { |
1719 | u32 curval; | |
1da177e4 | 1720 | |
e2970f2f | 1721 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1722 | |
1723 | if (unlikely(ret)) { | |
5eb3dc62 | 1724 | double_unlock_hb(hb1, hb2); |
69cd9eba | 1725 | hb_waiters_dec(hb2); |
1da177e4 | 1726 | |
e2970f2f | 1727 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1728 | if (ret) |
1729 | goto out_put_keys; | |
1da177e4 | 1730 | |
b41277dc | 1731 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 1732 | goto retry_private; |
1da177e4 | 1733 | |
ae791a2d TG |
1734 | put_futex_key(&key2); |
1735 | put_futex_key(&key1); | |
e4dc5b7a | 1736 | goto retry; |
1da177e4 | 1737 | } |
e2970f2f | 1738 | if (curval != *cmpval) { |
1da177e4 LT |
1739 | ret = -EAGAIN; |
1740 | goto out_unlock; | |
1741 | } | |
1742 | } | |
1743 | ||
52400ba9 | 1744 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1745 | /* |
1746 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1747 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1748 | * bit. We force this here where we are able to easily handle | |
1749 | * faults rather in the requeue loop below. | |
1750 | */ | |
52400ba9 | 1751 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1752 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1753 | |
1754 | /* | |
1755 | * At this point the top_waiter has either taken uaddr2 or is | |
1756 | * waiting on it. If the former, then the pi_state will not | |
1757 | * exist yet, look it up one more time to ensure we have a | |
866293ee TG |
1758 | * reference to it. If the lock was taken, ret contains the |
1759 | * vpid of the top waiter task. | |
ecb38b78 TG |
1760 | * If the lock was not taken, we have pi_state and an initial |
1761 | * refcount on it. In case of an error we have nothing. | |
52400ba9 | 1762 | */ |
866293ee | 1763 | if (ret > 0) { |
52400ba9 | 1764 | WARN_ON(pi_state); |
89061d3d | 1765 | drop_count++; |
52400ba9 | 1766 | task_count++; |
866293ee | 1767 | /* |
ecb38b78 TG |
1768 | * If we acquired the lock, then the user space value |
1769 | * of uaddr2 should be vpid. It cannot be changed by | |
1770 | * the top waiter as it is blocked on hb2 lock if it | |
1771 | * tries to do so. If something fiddled with it behind | |
1772 | * our back the pi state lookup might unearth it. So | |
1773 | * we rather use the known value than rereading and | |
1774 | * handing potential crap to lookup_pi_state. | |
1775 | * | |
1776 | * If that call succeeds then we have pi_state and an | |
1777 | * initial refcount on it. | |
866293ee | 1778 | */ |
54a21788 | 1779 | ret = lookup_pi_state(ret, hb2, &key2, &pi_state); |
52400ba9 DH |
1780 | } |
1781 | ||
1782 | switch (ret) { | |
1783 | case 0: | |
ecb38b78 | 1784 | /* We hold a reference on the pi state. */ |
52400ba9 | 1785 | break; |
4959f2de TG |
1786 | |
1787 | /* If the above failed, then pi_state is NULL */ | |
52400ba9 DH |
1788 | case -EFAULT: |
1789 | double_unlock_hb(hb1, hb2); | |
69cd9eba | 1790 | hb_waiters_dec(hb2); |
ae791a2d TG |
1791 | put_futex_key(&key2); |
1792 | put_futex_key(&key1); | |
d0725992 | 1793 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
1794 | if (!ret) |
1795 | goto retry; | |
1796 | goto out; | |
1797 | case -EAGAIN: | |
af54d6a1 TG |
1798 | /* |
1799 | * Two reasons for this: | |
1800 | * - Owner is exiting and we just wait for the | |
1801 | * exit to complete. | |
1802 | * - The user space value changed. | |
1803 | */ | |
52400ba9 | 1804 | double_unlock_hb(hb1, hb2); |
69cd9eba | 1805 | hb_waiters_dec(hb2); |
ae791a2d TG |
1806 | put_futex_key(&key2); |
1807 | put_futex_key(&key1); | |
52400ba9 DH |
1808 | cond_resched(); |
1809 | goto retry; | |
1810 | default: | |
1811 | goto out_unlock; | |
1812 | } | |
1813 | } | |
1814 | ||
0d00c7b2 | 1815 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
52400ba9 DH |
1816 | if (task_count - nr_wake >= nr_requeue) |
1817 | break; | |
1818 | ||
1819 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1820 | continue; |
52400ba9 | 1821 | |
392741e0 DH |
1822 | /* |
1823 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
1824 | * be paired with each other and no other futex ops. | |
aa10990e DH |
1825 | * |
1826 | * We should never be requeueing a futex_q with a pi_state, | |
1827 | * which is awaiting a futex_unlock_pi(). | |
392741e0 DH |
1828 | */ |
1829 | if ((requeue_pi && !this->rt_waiter) || | |
aa10990e DH |
1830 | (!requeue_pi && this->rt_waiter) || |
1831 | this->pi_state) { | |
392741e0 DH |
1832 | ret = -EINVAL; |
1833 | break; | |
1834 | } | |
52400ba9 DH |
1835 | |
1836 | /* | |
1837 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1838 | * lock, we already woke the top_waiter. If not, it will be | |
1839 | * woken by futex_unlock_pi(). | |
1840 | */ | |
1841 | if (++task_count <= nr_wake && !requeue_pi) { | |
1d0dcb3a | 1842 | mark_wake_futex(&wake_q, this); |
52400ba9 DH |
1843 | continue; |
1844 | } | |
1da177e4 | 1845 | |
84bc4af5 DH |
1846 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
1847 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
1848 | ret = -EINVAL; | |
1849 | break; | |
1850 | } | |
1851 | ||
52400ba9 DH |
1852 | /* |
1853 | * Requeue nr_requeue waiters and possibly one more in the case | |
1854 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1855 | */ | |
1856 | if (requeue_pi) { | |
ecb38b78 TG |
1857 | /* |
1858 | * Prepare the waiter to take the rt_mutex. Take a | |
1859 | * refcount on the pi_state and store the pointer in | |
1860 | * the futex_q object of the waiter. | |
1861 | */ | |
52400ba9 DH |
1862 | atomic_inc(&pi_state->refcount); |
1863 | this->pi_state = pi_state; | |
1864 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1865 | this->rt_waiter, | |
c051b21f | 1866 | this->task); |
52400ba9 | 1867 | if (ret == 1) { |
ecb38b78 TG |
1868 | /* |
1869 | * We got the lock. We do neither drop the | |
1870 | * refcount on pi_state nor clear | |
1871 | * this->pi_state because the waiter needs the | |
1872 | * pi_state for cleaning up the user space | |
1873 | * value. It will drop the refcount after | |
1874 | * doing so. | |
1875 | */ | |
beda2c7e | 1876 | requeue_pi_wake_futex(this, &key2, hb2); |
89061d3d | 1877 | drop_count++; |
52400ba9 DH |
1878 | continue; |
1879 | } else if (ret) { | |
ecb38b78 TG |
1880 | /* |
1881 | * rt_mutex_start_proxy_lock() detected a | |
1882 | * potential deadlock when we tried to queue | |
1883 | * that waiter. Drop the pi_state reference | |
1884 | * which we took above and remove the pointer | |
1885 | * to the state from the waiters futex_q | |
1886 | * object. | |
1887 | */ | |
52400ba9 | 1888 | this->pi_state = NULL; |
29e9ee5d | 1889 | put_pi_state(pi_state); |
885c2cb7 TG |
1890 | /* |
1891 | * We stop queueing more waiters and let user | |
1892 | * space deal with the mess. | |
1893 | */ | |
1894 | break; | |
52400ba9 | 1895 | } |
1da177e4 | 1896 | } |
52400ba9 DH |
1897 | requeue_futex(this, hb1, hb2, &key2); |
1898 | drop_count++; | |
1da177e4 LT |
1899 | } |
1900 | ||
ecb38b78 TG |
1901 | /* |
1902 | * We took an extra initial reference to the pi_state either | |
1903 | * in futex_proxy_trylock_atomic() or in lookup_pi_state(). We | |
1904 | * need to drop it here again. | |
1905 | */ | |
29e9ee5d | 1906 | put_pi_state(pi_state); |
885c2cb7 TG |
1907 | |
1908 | out_unlock: | |
5eb3dc62 | 1909 | double_unlock_hb(hb1, hb2); |
1d0dcb3a | 1910 | wake_up_q(&wake_q); |
69cd9eba | 1911 | hb_waiters_dec(hb2); |
1da177e4 | 1912 | |
cd84a42f DH |
1913 | /* |
1914 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1915 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1916 | * one at key2 and updated their key pointer. We no longer need to | |
1917 | * hold the references to key1. | |
1918 | */ | |
1da177e4 | 1919 | while (--drop_count >= 0) |
9adef58b | 1920 | drop_futex_key_refs(&key1); |
1da177e4 | 1921 | |
42d35d48 | 1922 | out_put_keys: |
ae791a2d | 1923 | put_futex_key(&key2); |
42d35d48 | 1924 | out_put_key1: |
ae791a2d | 1925 | put_futex_key(&key1); |
42d35d48 | 1926 | out: |
52400ba9 | 1927 | return ret ? ret : task_count; |
1da177e4 LT |
1928 | } |
1929 | ||
1930 | /* The key must be already stored in q->key. */ | |
82af7aca | 1931 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 1932 | __acquires(&hb->lock) |
1da177e4 | 1933 | { |
e2970f2f | 1934 | struct futex_hash_bucket *hb; |
1da177e4 | 1935 | |
e2970f2f | 1936 | hb = hash_futex(&q->key); |
11d4616b LT |
1937 | |
1938 | /* | |
1939 | * Increment the counter before taking the lock so that | |
1940 | * a potential waker won't miss a to-be-slept task that is | |
1941 | * waiting for the spinlock. This is safe as all queue_lock() | |
1942 | * users end up calling queue_me(). Similarly, for housekeeping, | |
1943 | * decrement the counter at queue_unlock() when some error has | |
1944 | * occurred and we don't end up adding the task to the list. | |
1945 | */ | |
1946 | hb_waiters_inc(hb); | |
1947 | ||
e2970f2f | 1948 | q->lock_ptr = &hb->lock; |
1da177e4 | 1949 | |
8ad7b378 | 1950 | spin_lock(&hb->lock); /* implies smp_mb(); (A) */ |
e2970f2f | 1951 | return hb; |
1da177e4 LT |
1952 | } |
1953 | ||
d40d65c8 | 1954 | static inline void |
0d00c7b2 | 1955 | queue_unlock(struct futex_hash_bucket *hb) |
15e408cd | 1956 | __releases(&hb->lock) |
d40d65c8 DH |
1957 | { |
1958 | spin_unlock(&hb->lock); | |
11d4616b | 1959 | hb_waiters_dec(hb); |
d40d65c8 DH |
1960 | } |
1961 | ||
1962 | /** | |
1963 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
1964 | * @q: The futex_q to enqueue | |
1965 | * @hb: The destination hash bucket | |
1966 | * | |
1967 | * The hb->lock must be held by the caller, and is released here. A call to | |
1968 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
1969 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
1970 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
1971 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
1972 | * an example). | |
1973 | */ | |
82af7aca | 1974 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
15e408cd | 1975 | __releases(&hb->lock) |
1da177e4 | 1976 | { |
ec92d082 PP |
1977 | int prio; |
1978 | ||
1979 | /* | |
1980 | * The priority used to register this element is | |
1981 | * - either the real thread-priority for the real-time threads | |
1982 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
1983 | * - or MAX_RT_PRIO for non-RT threads. | |
1984 | * Thus, all RT-threads are woken first in priority order, and | |
1985 | * the others are woken last, in FIFO order. | |
1986 | */ | |
1987 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1988 | ||
1989 | plist_node_init(&q->list, prio); | |
ec92d082 | 1990 | plist_add(&q->list, &hb->chain); |
c87e2837 | 1991 | q->task = current; |
e2970f2f | 1992 | spin_unlock(&hb->lock); |
1da177e4 LT |
1993 | } |
1994 | ||
d40d65c8 DH |
1995 | /** |
1996 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
1997 | * @q: The futex_q to unqueue | |
1998 | * | |
1999 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
2000 | * be paired with exactly one earlier call to queue_me(). | |
2001 | * | |
6c23cbbd RD |
2002 | * Return: |
2003 | * 1 - if the futex_q was still queued (and we removed unqueued it); | |
d40d65c8 | 2004 | * 0 - if the futex_q was already removed by the waking thread |
1da177e4 | 2005 | */ |
1da177e4 LT |
2006 | static int unqueue_me(struct futex_q *q) |
2007 | { | |
1da177e4 | 2008 | spinlock_t *lock_ptr; |
e2970f2f | 2009 | int ret = 0; |
1da177e4 LT |
2010 | |
2011 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 2012 | retry: |
29b75eb2 JZ |
2013 | /* |
2014 | * q->lock_ptr can change between this read and the following spin_lock. | |
2015 | * Use READ_ONCE to forbid the compiler from reloading q->lock_ptr and | |
2016 | * optimizing lock_ptr out of the logic below. | |
2017 | */ | |
2018 | lock_ptr = READ_ONCE(q->lock_ptr); | |
c80544dc | 2019 | if (lock_ptr != NULL) { |
1da177e4 LT |
2020 | spin_lock(lock_ptr); |
2021 | /* | |
2022 | * q->lock_ptr can change between reading it and | |
2023 | * spin_lock(), causing us to take the wrong lock. This | |
2024 | * corrects the race condition. | |
2025 | * | |
2026 | * Reasoning goes like this: if we have the wrong lock, | |
2027 | * q->lock_ptr must have changed (maybe several times) | |
2028 | * between reading it and the spin_lock(). It can | |
2029 | * change again after the spin_lock() but only if it was | |
2030 | * already changed before the spin_lock(). It cannot, | |
2031 | * however, change back to the original value. Therefore | |
2032 | * we can detect whether we acquired the correct lock. | |
2033 | */ | |
2034 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
2035 | spin_unlock(lock_ptr); | |
2036 | goto retry; | |
2037 | } | |
2e12978a | 2038 | __unqueue_futex(q); |
c87e2837 IM |
2039 | |
2040 | BUG_ON(q->pi_state); | |
2041 | ||
1da177e4 LT |
2042 | spin_unlock(lock_ptr); |
2043 | ret = 1; | |
2044 | } | |
2045 | ||
9adef58b | 2046 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
2047 | return ret; |
2048 | } | |
2049 | ||
c87e2837 IM |
2050 | /* |
2051 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
2052 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
2053 | * and dropped here. | |
c87e2837 | 2054 | */ |
d0aa7a70 | 2055 | static void unqueue_me_pi(struct futex_q *q) |
15e408cd | 2056 | __releases(q->lock_ptr) |
c87e2837 | 2057 | { |
2e12978a | 2058 | __unqueue_futex(q); |
c87e2837 IM |
2059 | |
2060 | BUG_ON(!q->pi_state); | |
29e9ee5d | 2061 | put_pi_state(q->pi_state); |
c87e2837 IM |
2062 | q->pi_state = NULL; |
2063 | ||
d0aa7a70 | 2064 | spin_unlock(q->lock_ptr); |
c87e2837 IM |
2065 | } |
2066 | ||
d0aa7a70 | 2067 | /* |
cdf71a10 | 2068 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 2069 | * |
778e9a9c AK |
2070 | * Must be called with hash bucket lock held and mm->sem held for non |
2071 | * private futexes. | |
d0aa7a70 | 2072 | */ |
778e9a9c | 2073 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
ae791a2d | 2074 | struct task_struct *newowner) |
d0aa7a70 | 2075 | { |
cdf71a10 | 2076 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 2077 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 2078 | struct task_struct *oldowner = pi_state->owner; |
7cfdaf38 | 2079 | u32 uval, uninitialized_var(curval), newval; |
e4dc5b7a | 2080 | int ret; |
d0aa7a70 PP |
2081 | |
2082 | /* Owner died? */ | |
1b7558e4 TG |
2083 | if (!pi_state->owner) |
2084 | newtid |= FUTEX_OWNER_DIED; | |
2085 | ||
2086 | /* | |
2087 | * We are here either because we stole the rtmutex from the | |
8161239a LJ |
2088 | * previous highest priority waiter or we are the highest priority |
2089 | * waiter but failed to get the rtmutex the first time. | |
2090 | * We have to replace the newowner TID in the user space variable. | |
2091 | * This must be atomic as we have to preserve the owner died bit here. | |
1b7558e4 | 2092 | * |
b2d0994b DH |
2093 | * Note: We write the user space value _before_ changing the pi_state |
2094 | * because we can fault here. Imagine swapped out pages or a fork | |
2095 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
2096 | * |
2097 | * Modifying pi_state _before_ the user space value would | |
2098 | * leave the pi_state in an inconsistent state when we fault | |
2099 | * here, because we need to drop the hash bucket lock to | |
2100 | * handle the fault. This might be observed in the PID check | |
2101 | * in lookup_pi_state. | |
2102 | */ | |
2103 | retry: | |
2104 | if (get_futex_value_locked(&uval, uaddr)) | |
2105 | goto handle_fault; | |
2106 | ||
2107 | while (1) { | |
2108 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
2109 | ||
37a9d912 | 2110 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1b7558e4 TG |
2111 | goto handle_fault; |
2112 | if (curval == uval) | |
2113 | break; | |
2114 | uval = curval; | |
2115 | } | |
2116 | ||
2117 | /* | |
2118 | * We fixed up user space. Now we need to fix the pi_state | |
2119 | * itself. | |
2120 | */ | |
d0aa7a70 | 2121 | if (pi_state->owner != NULL) { |
1d615482 | 2122 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
d0aa7a70 PP |
2123 | WARN_ON(list_empty(&pi_state->list)); |
2124 | list_del_init(&pi_state->list); | |
1d615482 | 2125 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
1b7558e4 | 2126 | } |
d0aa7a70 | 2127 | |
cdf71a10 | 2128 | pi_state->owner = newowner; |
d0aa7a70 | 2129 | |
1d615482 | 2130 | raw_spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 2131 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 | 2132 | list_add(&pi_state->list, &newowner->pi_state_list); |
1d615482 | 2133 | raw_spin_unlock_irq(&newowner->pi_lock); |
1b7558e4 | 2134 | return 0; |
d0aa7a70 | 2135 | |
d0aa7a70 | 2136 | /* |
1b7558e4 | 2137 | * To handle the page fault we need to drop the hash bucket |
8161239a LJ |
2138 | * lock here. That gives the other task (either the highest priority |
2139 | * waiter itself or the task which stole the rtmutex) the | |
1b7558e4 TG |
2140 | * chance to try the fixup of the pi_state. So once we are |
2141 | * back from handling the fault we need to check the pi_state | |
2142 | * after reacquiring the hash bucket lock and before trying to | |
2143 | * do another fixup. When the fixup has been done already we | |
2144 | * simply return. | |
d0aa7a70 | 2145 | */ |
1b7558e4 TG |
2146 | handle_fault: |
2147 | spin_unlock(q->lock_ptr); | |
778e9a9c | 2148 | |
d0725992 | 2149 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 2150 | |
1b7558e4 | 2151 | spin_lock(q->lock_ptr); |
778e9a9c | 2152 | |
1b7558e4 TG |
2153 | /* |
2154 | * Check if someone else fixed it for us: | |
2155 | */ | |
2156 | if (pi_state->owner != oldowner) | |
2157 | return 0; | |
2158 | ||
2159 | if (ret) | |
2160 | return ret; | |
2161 | ||
2162 | goto retry; | |
d0aa7a70 PP |
2163 | } |
2164 | ||
72c1bbf3 | 2165 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 2166 | |
dd973998 DH |
2167 | /** |
2168 | * fixup_owner() - Post lock pi_state and corner case management | |
2169 | * @uaddr: user address of the futex | |
dd973998 DH |
2170 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
2171 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
2172 | * | |
2173 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
2174 | * the pi_state owner as well as handle race conditions that may allow us to | |
2175 | * acquire the lock. Must be called with the hb lock held. | |
2176 | * | |
6c23cbbd RD |
2177 | * Return: |
2178 | * 1 - success, lock taken; | |
2179 | * 0 - success, lock not taken; | |
dd973998 DH |
2180 | * <0 - on error (-EFAULT) |
2181 | */ | |
ae791a2d | 2182 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 DH |
2183 | { |
2184 | struct task_struct *owner; | |
2185 | int ret = 0; | |
2186 | ||
2187 | if (locked) { | |
2188 | /* | |
2189 | * Got the lock. We might not be the anticipated owner if we | |
2190 | * did a lock-steal - fix up the PI-state in that case: | |
2191 | */ | |
2192 | if (q->pi_state->owner != current) | |
ae791a2d | 2193 | ret = fixup_pi_state_owner(uaddr, q, current); |
dd973998 DH |
2194 | goto out; |
2195 | } | |
2196 | ||
2197 | /* | |
2198 | * Catch the rare case, where the lock was released when we were on the | |
2199 | * way back before we locked the hash bucket. | |
2200 | */ | |
2201 | if (q->pi_state->owner == current) { | |
2202 | /* | |
2203 | * Try to get the rt_mutex now. This might fail as some other | |
2204 | * task acquired the rt_mutex after we removed ourself from the | |
2205 | * rt_mutex waiters list. | |
2206 | */ | |
2207 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
2208 | locked = 1; | |
2209 | goto out; | |
2210 | } | |
2211 | ||
2212 | /* | |
2213 | * pi_state is incorrect, some other task did a lock steal and | |
2214 | * we returned due to timeout or signal without taking the | |
8161239a | 2215 | * rt_mutex. Too late. |
dd973998 | 2216 | */ |
b4abf910 | 2217 | raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); |
dd973998 | 2218 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); |
8161239a LJ |
2219 | if (!owner) |
2220 | owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); | |
b4abf910 | 2221 | raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); |
ae791a2d | 2222 | ret = fixup_pi_state_owner(uaddr, q, owner); |
dd973998 DH |
2223 | goto out; |
2224 | } | |
2225 | ||
2226 | /* | |
2227 | * Paranoia check. If we did not take the lock, then we should not be | |
8161239a | 2228 | * the owner of the rt_mutex. |
dd973998 DH |
2229 | */ |
2230 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
2231 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
2232 | "pi-state %p\n", ret, | |
2233 | q->pi_state->pi_mutex.owner, | |
2234 | q->pi_state->owner); | |
2235 | ||
2236 | out: | |
2237 | return ret ? ret : locked; | |
2238 | } | |
2239 | ||
ca5f9524 DH |
2240 | /** |
2241 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
2242 | * @hb: the futex hash bucket, must be locked by the caller | |
2243 | * @q: the futex_q to queue up on | |
2244 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
2245 | */ |
2246 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 2247 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 2248 | { |
9beba3c5 DH |
2249 | /* |
2250 | * The task state is guaranteed to be set before another task can | |
b92b8b35 | 2251 | * wake it. set_current_state() is implemented using smp_store_mb() and |
9beba3c5 DH |
2252 | * queue_me() calls spin_unlock() upon completion, both serializing |
2253 | * access to the hash list and forcing another memory barrier. | |
2254 | */ | |
f1a11e05 | 2255 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 2256 | queue_me(q, hb); |
ca5f9524 DH |
2257 | |
2258 | /* Arm the timer */ | |
2e4b0d3f | 2259 | if (timeout) |
ca5f9524 | 2260 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); |
ca5f9524 DH |
2261 | |
2262 | /* | |
0729e196 DH |
2263 | * If we have been removed from the hash list, then another task |
2264 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
2265 | */ |
2266 | if (likely(!plist_node_empty(&q->list))) { | |
2267 | /* | |
2268 | * If the timer has already expired, current will already be | |
2269 | * flagged for rescheduling. Only call schedule if there | |
2270 | * is no timeout, or if it has yet to expire. | |
2271 | */ | |
2272 | if (!timeout || timeout->task) | |
88c8004f | 2273 | freezable_schedule(); |
ca5f9524 DH |
2274 | } |
2275 | __set_current_state(TASK_RUNNING); | |
2276 | } | |
2277 | ||
f801073f DH |
2278 | /** |
2279 | * futex_wait_setup() - Prepare to wait on a futex | |
2280 | * @uaddr: the futex userspace address | |
2281 | * @val: the expected value | |
b41277dc | 2282 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
2283 | * @q: the associated futex_q |
2284 | * @hb: storage for hash_bucket pointer to be returned to caller | |
2285 | * | |
2286 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
2287 | * compare it with the expected value. Handle atomic faults internally. | |
2288 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
2289 | * with no q.key reference on failure. | |
2290 | * | |
6c23cbbd RD |
2291 | * Return: |
2292 | * 0 - uaddr contains val and hb has been locked; | |
ca4a04cf | 2293 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked |
f801073f | 2294 | */ |
b41277dc | 2295 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 2296 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 2297 | { |
e2970f2f IM |
2298 | u32 uval; |
2299 | int ret; | |
1da177e4 | 2300 | |
1da177e4 | 2301 | /* |
b2d0994b | 2302 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
2303 | * Order is important: |
2304 | * | |
2305 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
2306 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
2307 | * | |
2308 | * The basic logical guarantee of a futex is that it blocks ONLY | |
2309 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
2310 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
2311 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
2312 | * cond(var) false, which would violate the guarantee. |
2313 | * | |
8fe8f545 ML |
2314 | * On the other hand, we insert q and release the hash-bucket only |
2315 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
2316 | * absorb a wakeup if *uaddr does not match the desired values | |
2317 | * while the syscall executes. | |
1da177e4 | 2318 | */ |
f801073f | 2319 | retry: |
9ea71503 | 2320 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ); |
f801073f | 2321 | if (unlikely(ret != 0)) |
a5a2a0c7 | 2322 | return ret; |
f801073f DH |
2323 | |
2324 | retry_private: | |
2325 | *hb = queue_lock(q); | |
2326 | ||
e2970f2f | 2327 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 2328 | |
f801073f | 2329 | if (ret) { |
0d00c7b2 | 2330 | queue_unlock(*hb); |
1da177e4 | 2331 | |
e2970f2f | 2332 | ret = get_user(uval, uaddr); |
e4dc5b7a | 2333 | if (ret) |
f801073f | 2334 | goto out; |
1da177e4 | 2335 | |
b41277dc | 2336 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2337 | goto retry_private; |
2338 | ||
ae791a2d | 2339 | put_futex_key(&q->key); |
e4dc5b7a | 2340 | goto retry; |
1da177e4 | 2341 | } |
ca5f9524 | 2342 | |
f801073f | 2343 | if (uval != val) { |
0d00c7b2 | 2344 | queue_unlock(*hb); |
f801073f | 2345 | ret = -EWOULDBLOCK; |
2fff78c7 | 2346 | } |
1da177e4 | 2347 | |
f801073f DH |
2348 | out: |
2349 | if (ret) | |
ae791a2d | 2350 | put_futex_key(&q->key); |
f801073f DH |
2351 | return ret; |
2352 | } | |
2353 | ||
b41277dc DH |
2354 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
2355 | ktime_t *abs_time, u32 bitset) | |
f801073f DH |
2356 | { |
2357 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
2358 | struct restart_block *restart; |
2359 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 2360 | struct futex_q q = futex_q_init; |
f801073f DH |
2361 | int ret; |
2362 | ||
2363 | if (!bitset) | |
2364 | return -EINVAL; | |
f801073f DH |
2365 | q.bitset = bitset; |
2366 | ||
2367 | if (abs_time) { | |
2368 | to = &timeout; | |
2369 | ||
b41277dc DH |
2370 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2371 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2372 | HRTIMER_MODE_ABS); | |
f801073f DH |
2373 | hrtimer_init_sleeper(to, current); |
2374 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2375 | current->timer_slack_ns); | |
2376 | } | |
2377 | ||
d58e6576 | 2378 | retry: |
7ada876a DH |
2379 | /* |
2380 | * Prepare to wait on uaddr. On success, holds hb lock and increments | |
2381 | * q.key refs. | |
2382 | */ | |
b41277dc | 2383 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
2384 | if (ret) |
2385 | goto out; | |
2386 | ||
ca5f9524 | 2387 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 2388 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
2389 | |
2390 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 2391 | ret = 0; |
7ada876a | 2392 | /* unqueue_me() drops q.key ref */ |
1da177e4 | 2393 | if (!unqueue_me(&q)) |
7ada876a | 2394 | goto out; |
2fff78c7 | 2395 | ret = -ETIMEDOUT; |
ca5f9524 | 2396 | if (to && !to->task) |
7ada876a | 2397 | goto out; |
72c1bbf3 | 2398 | |
e2970f2f | 2399 | /* |
d58e6576 TG |
2400 | * We expect signal_pending(current), but we might be the |
2401 | * victim of a spurious wakeup as well. | |
e2970f2f | 2402 | */ |
7ada876a | 2403 | if (!signal_pending(current)) |
d58e6576 | 2404 | goto retry; |
d58e6576 | 2405 | |
2fff78c7 | 2406 | ret = -ERESTARTSYS; |
c19384b5 | 2407 | if (!abs_time) |
7ada876a | 2408 | goto out; |
1da177e4 | 2409 | |
f56141e3 | 2410 | restart = ¤t->restart_block; |
2fff78c7 | 2411 | restart->fn = futex_wait_restart; |
a3c74c52 | 2412 | restart->futex.uaddr = uaddr; |
2fff78c7 PZ |
2413 | restart->futex.val = val; |
2414 | restart->futex.time = abs_time->tv64; | |
2415 | restart->futex.bitset = bitset; | |
0cd9c649 | 2416 | restart->futex.flags = flags | FLAGS_HAS_TIMEOUT; |
42d35d48 | 2417 | |
2fff78c7 PZ |
2418 | ret = -ERESTART_RESTARTBLOCK; |
2419 | ||
42d35d48 | 2420 | out: |
ca5f9524 DH |
2421 | if (to) { |
2422 | hrtimer_cancel(&to->timer); | |
2423 | destroy_hrtimer_on_stack(&to->timer); | |
2424 | } | |
c87e2837 IM |
2425 | return ret; |
2426 | } | |
2427 | ||
72c1bbf3 NP |
2428 | |
2429 | static long futex_wait_restart(struct restart_block *restart) | |
2430 | { | |
a3c74c52 | 2431 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 2432 | ktime_t t, *tp = NULL; |
72c1bbf3 | 2433 | |
a72188d8 DH |
2434 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
2435 | t.tv64 = restart->futex.time; | |
2436 | tp = &t; | |
2437 | } | |
72c1bbf3 | 2438 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
2439 | |
2440 | return (long)futex_wait(uaddr, restart->futex.flags, | |
2441 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
2442 | } |
2443 | ||
2444 | ||
c87e2837 IM |
2445 | /* |
2446 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
2447 | * and failed. The kernel side here does the whole locking operation: | |
767f509c DB |
2448 | * if there are waiters then it will block as a consequence of relying |
2449 | * on rt-mutexes, it does PI, etc. (Due to races the kernel might see | |
2450 | * a 0 value of the futex too.). | |
2451 | * | |
2452 | * Also serves as futex trylock_pi()'ing, and due semantics. | |
c87e2837 | 2453 | */ |
996636dd | 2454 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, |
b41277dc | 2455 | ktime_t *time, int trylock) |
c87e2837 | 2456 | { |
c5780e97 | 2457 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 2458 | struct futex_hash_bucket *hb; |
5bdb05f9 | 2459 | struct futex_q q = futex_q_init; |
dd973998 | 2460 | int res, ret; |
c87e2837 IM |
2461 | |
2462 | if (refill_pi_state_cache()) | |
2463 | return -ENOMEM; | |
2464 | ||
c19384b5 | 2465 | if (time) { |
c5780e97 | 2466 | to = &timeout; |
237fc6e7 TG |
2467 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
2468 | HRTIMER_MODE_ABS); | |
c5780e97 | 2469 | hrtimer_init_sleeper(to, current); |
cc584b21 | 2470 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
2471 | } |
2472 | ||
42d35d48 | 2473 | retry: |
9ea71503 | 2474 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE); |
c87e2837 | 2475 | if (unlikely(ret != 0)) |
42d35d48 | 2476 | goto out; |
c87e2837 | 2477 | |
e4dc5b7a | 2478 | retry_private: |
82af7aca | 2479 | hb = queue_lock(&q); |
c87e2837 | 2480 | |
bab5bc9e | 2481 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 2482 | if (unlikely(ret)) { |
767f509c DB |
2483 | /* |
2484 | * Atomic work succeeded and we got the lock, | |
2485 | * or failed. Either way, we do _not_ block. | |
2486 | */ | |
778e9a9c | 2487 | switch (ret) { |
1a52084d DH |
2488 | case 1: |
2489 | /* We got the lock. */ | |
2490 | ret = 0; | |
2491 | goto out_unlock_put_key; | |
2492 | case -EFAULT: | |
2493 | goto uaddr_faulted; | |
778e9a9c AK |
2494 | case -EAGAIN: |
2495 | /* | |
af54d6a1 TG |
2496 | * Two reasons for this: |
2497 | * - Task is exiting and we just wait for the | |
2498 | * exit to complete. | |
2499 | * - The user space value changed. | |
778e9a9c | 2500 | */ |
0d00c7b2 | 2501 | queue_unlock(hb); |
ae791a2d | 2502 | put_futex_key(&q.key); |
778e9a9c AK |
2503 | cond_resched(); |
2504 | goto retry; | |
778e9a9c | 2505 | default: |
42d35d48 | 2506 | goto out_unlock_put_key; |
c87e2837 | 2507 | } |
c87e2837 IM |
2508 | } |
2509 | ||
2510 | /* | |
2511 | * Only actually queue now that the atomic ops are done: | |
2512 | */ | |
82af7aca | 2513 | queue_me(&q, hb); |
c87e2837 | 2514 | |
c87e2837 IM |
2515 | WARN_ON(!q.pi_state); |
2516 | /* | |
2517 | * Block on the PI mutex: | |
2518 | */ | |
c051b21f TG |
2519 | if (!trylock) { |
2520 | ret = rt_mutex_timed_futex_lock(&q.pi_state->pi_mutex, to); | |
2521 | } else { | |
c87e2837 IM |
2522 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); |
2523 | /* Fixup the trylock return value: */ | |
2524 | ret = ret ? 0 : -EWOULDBLOCK; | |
2525 | } | |
2526 | ||
a99e4e41 | 2527 | spin_lock(q.lock_ptr); |
dd973998 DH |
2528 | /* |
2529 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2530 | * haven't already. | |
2531 | */ | |
ae791a2d | 2532 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 DH |
2533 | /* |
2534 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
2535 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
2536 | */ | |
2537 | if (res) | |
2538 | ret = (res < 0) ? res : 0; | |
c87e2837 | 2539 | |
e8f6386c | 2540 | /* |
dd973998 DH |
2541 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
2542 | * it and return the fault to userspace. | |
e8f6386c DH |
2543 | */ |
2544 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
2545 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
2546 | ||
778e9a9c AK |
2547 | /* Unqueue and drop the lock */ |
2548 | unqueue_me_pi(&q); | |
c87e2837 | 2549 | |
5ecb01cf | 2550 | goto out_put_key; |
c87e2837 | 2551 | |
42d35d48 | 2552 | out_unlock_put_key: |
0d00c7b2 | 2553 | queue_unlock(hb); |
c87e2837 | 2554 | |
42d35d48 | 2555 | out_put_key: |
ae791a2d | 2556 | put_futex_key(&q.key); |
42d35d48 | 2557 | out: |
237fc6e7 TG |
2558 | if (to) |
2559 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 2560 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 2561 | |
42d35d48 | 2562 | uaddr_faulted: |
0d00c7b2 | 2563 | queue_unlock(hb); |
778e9a9c | 2564 | |
d0725992 | 2565 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
2566 | if (ret) |
2567 | goto out_put_key; | |
c87e2837 | 2568 | |
b41277dc | 2569 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2570 | goto retry_private; |
2571 | ||
ae791a2d | 2572 | put_futex_key(&q.key); |
e4dc5b7a | 2573 | goto retry; |
c87e2837 IM |
2574 | } |
2575 | ||
c87e2837 IM |
2576 | /* |
2577 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
2578 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
2579 | * and do the rt-mutex unlock. | |
2580 | */ | |
b41277dc | 2581 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 | 2582 | { |
ccf9e6a8 | 2583 | u32 uninitialized_var(curval), uval, vpid = task_pid_vnr(current); |
38d47c1b | 2584 | union futex_key key = FUTEX_KEY_INIT; |
ccf9e6a8 TG |
2585 | struct futex_hash_bucket *hb; |
2586 | struct futex_q *match; | |
e4dc5b7a | 2587 | int ret; |
c87e2837 IM |
2588 | |
2589 | retry: | |
2590 | if (get_user(uval, uaddr)) | |
2591 | return -EFAULT; | |
2592 | /* | |
2593 | * We release only a lock we actually own: | |
2594 | */ | |
c0c9ed15 | 2595 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 2596 | return -EPERM; |
c87e2837 | 2597 | |
9ea71503 | 2598 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE); |
ccf9e6a8 TG |
2599 | if (ret) |
2600 | return ret; | |
c87e2837 IM |
2601 | |
2602 | hb = hash_futex(&key); | |
2603 | spin_lock(&hb->lock); | |
2604 | ||
c87e2837 | 2605 | /* |
ccf9e6a8 TG |
2606 | * Check waiters first. We do not trust user space values at |
2607 | * all and we at least want to know if user space fiddled | |
2608 | * with the futex value instead of blindly unlocking. | |
c87e2837 | 2609 | */ |
ccf9e6a8 TG |
2610 | match = futex_top_waiter(hb, &key); |
2611 | if (match) { | |
802ab58d SAS |
2612 | ret = wake_futex_pi(uaddr, uval, match, hb); |
2613 | /* | |
2614 | * In case of success wake_futex_pi dropped the hash | |
2615 | * bucket lock. | |
2616 | */ | |
2617 | if (!ret) | |
2618 | goto out_putkey; | |
c87e2837 | 2619 | /* |
ccf9e6a8 TG |
2620 | * The atomic access to the futex value generated a |
2621 | * pagefault, so retry the user-access and the wakeup: | |
c87e2837 IM |
2622 | */ |
2623 | if (ret == -EFAULT) | |
2624 | goto pi_faulted; | |
802ab58d SAS |
2625 | /* |
2626 | * wake_futex_pi has detected invalid state. Tell user | |
2627 | * space. | |
2628 | */ | |
c87e2837 IM |
2629 | goto out_unlock; |
2630 | } | |
ccf9e6a8 | 2631 | |
c87e2837 | 2632 | /* |
ccf9e6a8 TG |
2633 | * We have no kernel internal state, i.e. no waiters in the |
2634 | * kernel. Waiters which are about to queue themselves are stuck | |
2635 | * on hb->lock. So we can safely ignore them. We do neither | |
2636 | * preserve the WAITERS bit not the OWNER_DIED one. We are the | |
2637 | * owner. | |
c87e2837 | 2638 | */ |
ccf9e6a8 | 2639 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, 0)) |
13fbca4c | 2640 | goto pi_faulted; |
c87e2837 | 2641 | |
ccf9e6a8 TG |
2642 | /* |
2643 | * If uval has changed, let user space handle it. | |
2644 | */ | |
2645 | ret = (curval == uval) ? 0 : -EAGAIN; | |
2646 | ||
c87e2837 IM |
2647 | out_unlock: |
2648 | spin_unlock(&hb->lock); | |
802ab58d | 2649 | out_putkey: |
ae791a2d | 2650 | put_futex_key(&key); |
c87e2837 IM |
2651 | return ret; |
2652 | ||
2653 | pi_faulted: | |
778e9a9c | 2654 | spin_unlock(&hb->lock); |
ae791a2d | 2655 | put_futex_key(&key); |
c87e2837 | 2656 | |
d0725992 | 2657 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2658 | if (!ret) |
c87e2837 IM |
2659 | goto retry; |
2660 | ||
1da177e4 LT |
2661 | return ret; |
2662 | } | |
2663 | ||
52400ba9 DH |
2664 | /** |
2665 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2666 | * @hb: the hash_bucket futex_q was original enqueued on | |
2667 | * @q: the futex_q woken while waiting to be requeued | |
2668 | * @key2: the futex_key of the requeue target futex | |
2669 | * @timeout: the timeout associated with the wait (NULL if none) | |
2670 | * | |
2671 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2672 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2673 | * the wakeup and return the appropriate error code to the caller. Must be | |
2674 | * called with the hb lock held. | |
2675 | * | |
6c23cbbd RD |
2676 | * Return: |
2677 | * 0 = no early wakeup detected; | |
2678 | * <0 = -ETIMEDOUT or -ERESTARTNOINTR | |
52400ba9 DH |
2679 | */ |
2680 | static inline | |
2681 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2682 | struct futex_q *q, union futex_key *key2, | |
2683 | struct hrtimer_sleeper *timeout) | |
2684 | { | |
2685 | int ret = 0; | |
2686 | ||
2687 | /* | |
2688 | * With the hb lock held, we avoid races while we process the wakeup. | |
2689 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2690 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2691 | * It can't be requeued from uaddr2 to something else since we don't | |
2692 | * support a PI aware source futex for requeue. | |
2693 | */ | |
2694 | if (!match_futex(&q->key, key2)) { | |
2695 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2696 | /* | |
2697 | * We were woken prior to requeue by a timeout or a signal. | |
2698 | * Unqueue the futex_q and determine which it was. | |
2699 | */ | |
2e12978a | 2700 | plist_del(&q->list, &hb->chain); |
11d4616b | 2701 | hb_waiters_dec(hb); |
52400ba9 | 2702 | |
d58e6576 | 2703 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 2704 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2705 | if (timeout && !timeout->task) |
2706 | ret = -ETIMEDOUT; | |
d58e6576 | 2707 | else if (signal_pending(current)) |
1c840c14 | 2708 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
2709 | } |
2710 | return ret; | |
2711 | } | |
2712 | ||
2713 | /** | |
2714 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 2715 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 2716 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
ab51fbab | 2717 | * the same type, no requeueing from private to shared, etc. |
52400ba9 DH |
2718 | * @val: the expected value of uaddr |
2719 | * @abs_time: absolute timeout | |
56ec1607 | 2720 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
2721 | * @uaddr2: the pi futex we will take prior to returning to user-space |
2722 | * | |
2723 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
6f7b0a2a DH |
2724 | * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake |
2725 | * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to | |
2726 | * userspace. This ensures the rt_mutex maintains an owner when it has waiters; | |
2727 | * without one, the pi logic would not know which task to boost/deboost, if | |
2728 | * there was a need to. | |
52400ba9 DH |
2729 | * |
2730 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
6c23cbbd | 2731 | * via the following-- |
52400ba9 | 2732 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() |
cc6db4e6 DH |
2733 | * 2) wakeup on uaddr2 after a requeue |
2734 | * 3) signal | |
2735 | * 4) timeout | |
52400ba9 | 2736 | * |
cc6db4e6 | 2737 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
2738 | * |
2739 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2740 | * 5) successful lock | |
2741 | * 6) signal | |
2742 | * 7) timeout | |
2743 | * 8) other lock acquisition failure | |
2744 | * | |
cc6db4e6 | 2745 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
2746 | * |
2747 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2748 | * | |
6c23cbbd RD |
2749 | * Return: |
2750 | * 0 - On success; | |
52400ba9 DH |
2751 | * <0 - On error |
2752 | */ | |
b41277dc | 2753 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 2754 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 2755 | u32 __user *uaddr2) |
52400ba9 DH |
2756 | { |
2757 | struct hrtimer_sleeper timeout, *to = NULL; | |
2758 | struct rt_mutex_waiter rt_waiter; | |
2759 | struct rt_mutex *pi_mutex = NULL; | |
52400ba9 | 2760 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
2761 | union futex_key key2 = FUTEX_KEY_INIT; |
2762 | struct futex_q q = futex_q_init; | |
52400ba9 | 2763 | int res, ret; |
52400ba9 | 2764 | |
6f7b0a2a DH |
2765 | if (uaddr == uaddr2) |
2766 | return -EINVAL; | |
2767 | ||
52400ba9 DH |
2768 | if (!bitset) |
2769 | return -EINVAL; | |
2770 | ||
2771 | if (abs_time) { | |
2772 | to = &timeout; | |
b41277dc DH |
2773 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2774 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2775 | HRTIMER_MODE_ABS); | |
52400ba9 DH |
2776 | hrtimer_init_sleeper(to, current); |
2777 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2778 | current->timer_slack_ns); | |
2779 | } | |
2780 | ||
2781 | /* | |
2782 | * The waiter is allocated on our stack, manipulated by the requeue | |
2783 | * code while we sleep on uaddr. | |
2784 | */ | |
2785 | debug_rt_mutex_init_waiter(&rt_waiter); | |
fb00aca4 PZ |
2786 | RB_CLEAR_NODE(&rt_waiter.pi_tree_entry); |
2787 | RB_CLEAR_NODE(&rt_waiter.tree_entry); | |
52400ba9 DH |
2788 | rt_waiter.task = NULL; |
2789 | ||
9ea71503 | 2790 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
52400ba9 DH |
2791 | if (unlikely(ret != 0)) |
2792 | goto out; | |
2793 | ||
84bc4af5 DH |
2794 | q.bitset = bitset; |
2795 | q.rt_waiter = &rt_waiter; | |
2796 | q.requeue_pi_key = &key2; | |
2797 | ||
7ada876a DH |
2798 | /* |
2799 | * Prepare to wait on uaddr. On success, increments q.key (key1) ref | |
2800 | * count. | |
2801 | */ | |
b41277dc | 2802 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 TG |
2803 | if (ret) |
2804 | goto out_key2; | |
52400ba9 | 2805 | |
e9c243a5 TG |
2806 | /* |
2807 | * The check above which compares uaddrs is not sufficient for | |
2808 | * shared futexes. We need to compare the keys: | |
2809 | */ | |
2810 | if (match_futex(&q.key, &key2)) { | |
13c42c2f | 2811 | queue_unlock(hb); |
e9c243a5 TG |
2812 | ret = -EINVAL; |
2813 | goto out_put_keys; | |
2814 | } | |
2815 | ||
52400ba9 | 2816 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ |
f1a11e05 | 2817 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2818 | |
2819 | spin_lock(&hb->lock); | |
2820 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2821 | spin_unlock(&hb->lock); | |
2822 | if (ret) | |
2823 | goto out_put_keys; | |
2824 | ||
2825 | /* | |
2826 | * In order for us to be here, we know our q.key == key2, and since | |
2827 | * we took the hb->lock above, we also know that futex_requeue() has | |
2828 | * completed and we no longer have to concern ourselves with a wakeup | |
7ada876a DH |
2829 | * race with the atomic proxy lock acquisition by the requeue code. The |
2830 | * futex_requeue dropped our key1 reference and incremented our key2 | |
2831 | * reference count. | |
52400ba9 DH |
2832 | */ |
2833 | ||
2834 | /* Check if the requeue code acquired the second futex for us. */ | |
2835 | if (!q.rt_waiter) { | |
2836 | /* | |
2837 | * Got the lock. We might not be the anticipated owner if we | |
2838 | * did a lock-steal - fix up the PI-state in that case. | |
2839 | */ | |
2840 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2841 | spin_lock(q.lock_ptr); | |
ae791a2d | 2842 | ret = fixup_pi_state_owner(uaddr2, &q, current); |
fb75a428 TG |
2843 | /* |
2844 | * Drop the reference to the pi state which | |
2845 | * the requeue_pi() code acquired for us. | |
2846 | */ | |
29e9ee5d | 2847 | put_pi_state(q.pi_state); |
52400ba9 DH |
2848 | spin_unlock(q.lock_ptr); |
2849 | } | |
2850 | } else { | |
2851 | /* | |
2852 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2853 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2854 | * the pi_state. | |
2855 | */ | |
f27071cb | 2856 | WARN_ON(!q.pi_state); |
52400ba9 | 2857 | pi_mutex = &q.pi_state->pi_mutex; |
c051b21f | 2858 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter); |
52400ba9 DH |
2859 | debug_rt_mutex_free_waiter(&rt_waiter); |
2860 | ||
2861 | spin_lock(q.lock_ptr); | |
2862 | /* | |
2863 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2864 | * haven't already. | |
2865 | */ | |
ae791a2d | 2866 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 DH |
2867 | /* |
2868 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 2869 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
2870 | */ |
2871 | if (res) | |
2872 | ret = (res < 0) ? res : 0; | |
2873 | ||
2874 | /* Unqueue and drop the lock. */ | |
2875 | unqueue_me_pi(&q); | |
2876 | } | |
2877 | ||
2878 | /* | |
2879 | * If fixup_pi_state_owner() faulted and was unable to handle the | |
2880 | * fault, unlock the rt_mutex and return the fault to userspace. | |
2881 | */ | |
2882 | if (ret == -EFAULT) { | |
b6070a8d | 2883 | if (pi_mutex && rt_mutex_owner(pi_mutex) == current) |
52400ba9 DH |
2884 | rt_mutex_unlock(pi_mutex); |
2885 | } else if (ret == -EINTR) { | |
52400ba9 | 2886 | /* |
cc6db4e6 DH |
2887 | * We've already been requeued, but cannot restart by calling |
2888 | * futex_lock_pi() directly. We could restart this syscall, but | |
2889 | * it would detect that the user space "val" changed and return | |
2890 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
2891 | * -EWOULDBLOCK directly. | |
52400ba9 | 2892 | */ |
2070887f | 2893 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2894 | } |
2895 | ||
2896 | out_put_keys: | |
ae791a2d | 2897 | put_futex_key(&q.key); |
c8b15a70 | 2898 | out_key2: |
ae791a2d | 2899 | put_futex_key(&key2); |
52400ba9 DH |
2900 | |
2901 | out: | |
2902 | if (to) { | |
2903 | hrtimer_cancel(&to->timer); | |
2904 | destroy_hrtimer_on_stack(&to->timer); | |
2905 | } | |
2906 | return ret; | |
2907 | } | |
2908 | ||
0771dfef IM |
2909 | /* |
2910 | * Support for robust futexes: the kernel cleans up held futexes at | |
2911 | * thread exit time. | |
2912 | * | |
2913 | * Implementation: user-space maintains a per-thread list of locks it | |
2914 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2915 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2916 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2917 | * always manipulated with the lock held, so the list is private and |
2918 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2919 | * field, to allow the kernel to clean up if the thread dies after | |
2920 | * acquiring the lock, but just before it could have added itself to | |
2921 | * the list. There can only be one such pending lock. | |
2922 | */ | |
2923 | ||
2924 | /** | |
d96ee56c DH |
2925 | * sys_set_robust_list() - Set the robust-futex list head of a task |
2926 | * @head: pointer to the list-head | |
2927 | * @len: length of the list-head, as userspace expects | |
0771dfef | 2928 | */ |
836f92ad HC |
2929 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2930 | size_t, len) | |
0771dfef | 2931 | { |
a0c1e907 TG |
2932 | if (!futex_cmpxchg_enabled) |
2933 | return -ENOSYS; | |
0771dfef IM |
2934 | /* |
2935 | * The kernel knows only one size for now: | |
2936 | */ | |
2937 | if (unlikely(len != sizeof(*head))) | |
2938 | return -EINVAL; | |
2939 | ||
2940 | current->robust_list = head; | |
2941 | ||
2942 | return 0; | |
2943 | } | |
2944 | ||
2945 | /** | |
d96ee56c DH |
2946 | * sys_get_robust_list() - Get the robust-futex list head of a task |
2947 | * @pid: pid of the process [zero for current task] | |
2948 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
2949 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 2950 | */ |
836f92ad HC |
2951 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
2952 | struct robust_list_head __user * __user *, head_ptr, | |
2953 | size_t __user *, len_ptr) | |
0771dfef | 2954 | { |
ba46df98 | 2955 | struct robust_list_head __user *head; |
0771dfef | 2956 | unsigned long ret; |
bdbb776f | 2957 | struct task_struct *p; |
0771dfef | 2958 | |
a0c1e907 TG |
2959 | if (!futex_cmpxchg_enabled) |
2960 | return -ENOSYS; | |
2961 | ||
bdbb776f KC |
2962 | rcu_read_lock(); |
2963 | ||
2964 | ret = -ESRCH; | |
0771dfef | 2965 | if (!pid) |
bdbb776f | 2966 | p = current; |
0771dfef | 2967 | else { |
228ebcbe | 2968 | p = find_task_by_vpid(pid); |
0771dfef IM |
2969 | if (!p) |
2970 | goto err_unlock; | |
0771dfef IM |
2971 | } |
2972 | ||
bdbb776f | 2973 | ret = -EPERM; |
caaee623 | 2974 | if (!ptrace_may_access(p, PTRACE_MODE_READ_REALCREDS)) |
bdbb776f KC |
2975 | goto err_unlock; |
2976 | ||
2977 | head = p->robust_list; | |
2978 | rcu_read_unlock(); | |
2979 | ||
0771dfef IM |
2980 | if (put_user(sizeof(*head), len_ptr)) |
2981 | return -EFAULT; | |
2982 | return put_user(head, head_ptr); | |
2983 | ||
2984 | err_unlock: | |
aaa2a97e | 2985 | rcu_read_unlock(); |
0771dfef IM |
2986 | |
2987 | return ret; | |
2988 | } | |
2989 | ||
2990 | /* | |
2991 | * Process a futex-list entry, check whether it's owned by the | |
2992 | * dying task, and do notification if so: | |
2993 | */ | |
e3f2ddea | 2994 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 2995 | { |
7cfdaf38 | 2996 | u32 uval, uninitialized_var(nval), mval; |
0771dfef | 2997 | |
8f17d3a5 IM |
2998 | retry: |
2999 | if (get_user(uval, uaddr)) | |
0771dfef IM |
3000 | return -1; |
3001 | ||
b488893a | 3002 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
3003 | /* |
3004 | * Ok, this dying thread is truly holding a futex | |
3005 | * of interest. Set the OWNER_DIED bit atomically | |
3006 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
3007 | * set, wake up a waiter (if any). (We have to do a | |
3008 | * futex_wake() even if OWNER_DIED is already set - | |
3009 | * to handle the rare but possible case of recursive | |
3010 | * thread-death.) The rest of the cleanup is done in | |
3011 | * userspace. | |
3012 | */ | |
e3f2ddea | 3013 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
6e0aa9f8 TG |
3014 | /* |
3015 | * We are not holding a lock here, but we want to have | |
3016 | * the pagefault_disable/enable() protection because | |
3017 | * we want to handle the fault gracefully. If the | |
3018 | * access fails we try to fault in the futex with R/W | |
3019 | * verification via get_user_pages. get_user() above | |
3020 | * does not guarantee R/W access. If that fails we | |
3021 | * give up and leave the futex locked. | |
3022 | */ | |
3023 | if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) { | |
3024 | if (fault_in_user_writeable(uaddr)) | |
3025 | return -1; | |
3026 | goto retry; | |
3027 | } | |
c87e2837 | 3028 | if (nval != uval) |
8f17d3a5 | 3029 | goto retry; |
0771dfef | 3030 | |
e3f2ddea IM |
3031 | /* |
3032 | * Wake robust non-PI futexes here. The wakeup of | |
3033 | * PI futexes happens in exit_pi_state(): | |
3034 | */ | |
36cf3b5c | 3035 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 3036 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
3037 | } |
3038 | return 0; | |
3039 | } | |
3040 | ||
e3f2ddea IM |
3041 | /* |
3042 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
3043 | */ | |
3044 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 3045 | struct robust_list __user * __user *head, |
1dcc41bb | 3046 | unsigned int *pi) |
e3f2ddea IM |
3047 | { |
3048 | unsigned long uentry; | |
3049 | ||
ba46df98 | 3050 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
3051 | return -EFAULT; |
3052 | ||
ba46df98 | 3053 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
3054 | *pi = uentry & 1; |
3055 | ||
3056 | return 0; | |
3057 | } | |
3058 | ||
0771dfef IM |
3059 | /* |
3060 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
3061 | * and mark any locks found there dead, and notify any waiters. | |
3062 | * | |
3063 | * We silently return on any sign of list-walking problem. | |
3064 | */ | |
3065 | void exit_robust_list(struct task_struct *curr) | |
3066 | { | |
3067 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 3068 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 DH |
3069 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
3070 | unsigned int uninitialized_var(next_pi); | |
0771dfef | 3071 | unsigned long futex_offset; |
9f96cb1e | 3072 | int rc; |
0771dfef | 3073 | |
a0c1e907 TG |
3074 | if (!futex_cmpxchg_enabled) |
3075 | return; | |
3076 | ||
0771dfef IM |
3077 | /* |
3078 | * Fetch the list head (which was registered earlier, via | |
3079 | * sys_set_robust_list()): | |
3080 | */ | |
e3f2ddea | 3081 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
3082 | return; |
3083 | /* | |
3084 | * Fetch the relative futex offset: | |
3085 | */ | |
3086 | if (get_user(futex_offset, &head->futex_offset)) | |
3087 | return; | |
3088 | /* | |
3089 | * Fetch any possibly pending lock-add first, and handle it | |
3090 | * if it exists: | |
3091 | */ | |
e3f2ddea | 3092 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 3093 | return; |
e3f2ddea | 3094 | |
9f96cb1e | 3095 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 3096 | while (entry != &head->list) { |
9f96cb1e MS |
3097 | /* |
3098 | * Fetch the next entry in the list before calling | |
3099 | * handle_futex_death: | |
3100 | */ | |
3101 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
3102 | /* |
3103 | * A pending lock might already be on the list, so | |
c87e2837 | 3104 | * don't process it twice: |
0771dfef IM |
3105 | */ |
3106 | if (entry != pending) | |
ba46df98 | 3107 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 3108 | curr, pi)) |
0771dfef | 3109 | return; |
9f96cb1e | 3110 | if (rc) |
0771dfef | 3111 | return; |
9f96cb1e MS |
3112 | entry = next_entry; |
3113 | pi = next_pi; | |
0771dfef IM |
3114 | /* |
3115 | * Avoid excessively long or circular lists: | |
3116 | */ | |
3117 | if (!--limit) | |
3118 | break; | |
3119 | ||
3120 | cond_resched(); | |
3121 | } | |
9f96cb1e MS |
3122 | |
3123 | if (pending) | |
3124 | handle_futex_death((void __user *)pending + futex_offset, | |
3125 | curr, pip); | |
0771dfef IM |
3126 | } |
3127 | ||
c19384b5 | 3128 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 3129 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 3130 | { |
81b40539 | 3131 | int cmd = op & FUTEX_CMD_MASK; |
b41277dc | 3132 | unsigned int flags = 0; |
34f01cc1 ED |
3133 | |
3134 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 3135 | flags |= FLAGS_SHARED; |
1da177e4 | 3136 | |
b41277dc DH |
3137 | if (op & FUTEX_CLOCK_REALTIME) { |
3138 | flags |= FLAGS_CLOCKRT; | |
337f1304 DH |
3139 | if (cmd != FUTEX_WAIT && cmd != FUTEX_WAIT_BITSET && \ |
3140 | cmd != FUTEX_WAIT_REQUEUE_PI) | |
b41277dc DH |
3141 | return -ENOSYS; |
3142 | } | |
1da177e4 | 3143 | |
59263b51 TG |
3144 | switch (cmd) { |
3145 | case FUTEX_LOCK_PI: | |
3146 | case FUTEX_UNLOCK_PI: | |
3147 | case FUTEX_TRYLOCK_PI: | |
3148 | case FUTEX_WAIT_REQUEUE_PI: | |
3149 | case FUTEX_CMP_REQUEUE_PI: | |
3150 | if (!futex_cmpxchg_enabled) | |
3151 | return -ENOSYS; | |
3152 | } | |
3153 | ||
34f01cc1 | 3154 | switch (cmd) { |
1da177e4 | 3155 | case FUTEX_WAIT: |
cd689985 TG |
3156 | val3 = FUTEX_BITSET_MATCH_ANY; |
3157 | case FUTEX_WAIT_BITSET: | |
81b40539 | 3158 | return futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 | 3159 | case FUTEX_WAKE: |
cd689985 TG |
3160 | val3 = FUTEX_BITSET_MATCH_ANY; |
3161 | case FUTEX_WAKE_BITSET: | |
81b40539 | 3162 | return futex_wake(uaddr, flags, val, val3); |
1da177e4 | 3163 | case FUTEX_REQUEUE: |
81b40539 | 3164 | return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 | 3165 | case FUTEX_CMP_REQUEUE: |
81b40539 | 3166 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
4732efbe | 3167 | case FUTEX_WAKE_OP: |
81b40539 | 3168 | return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
c87e2837 | 3169 | case FUTEX_LOCK_PI: |
996636dd | 3170 | return futex_lock_pi(uaddr, flags, timeout, 0); |
c87e2837 | 3171 | case FUTEX_UNLOCK_PI: |
81b40539 | 3172 | return futex_unlock_pi(uaddr, flags); |
c87e2837 | 3173 | case FUTEX_TRYLOCK_PI: |
996636dd | 3174 | return futex_lock_pi(uaddr, flags, NULL, 1); |
52400ba9 DH |
3175 | case FUTEX_WAIT_REQUEUE_PI: |
3176 | val3 = FUTEX_BITSET_MATCH_ANY; | |
81b40539 TG |
3177 | return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
3178 | uaddr2); | |
52400ba9 | 3179 | case FUTEX_CMP_REQUEUE_PI: |
81b40539 | 3180 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
1da177e4 | 3181 | } |
81b40539 | 3182 | return -ENOSYS; |
1da177e4 LT |
3183 | } |
3184 | ||
3185 | ||
17da2bd9 HC |
3186 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
3187 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
3188 | u32, val3) | |
1da177e4 | 3189 | { |
c19384b5 PP |
3190 | struct timespec ts; |
3191 | ktime_t t, *tp = NULL; | |
e2970f2f | 3192 | u32 val2 = 0; |
34f01cc1 | 3193 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 3194 | |
cd689985 | 3195 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
3196 | cmd == FUTEX_WAIT_BITSET || |
3197 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
ab51fbab DB |
3198 | if (unlikely(should_fail_futex(!(op & FUTEX_PRIVATE_FLAG)))) |
3199 | return -EFAULT; | |
c19384b5 | 3200 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 3201 | return -EFAULT; |
c19384b5 | 3202 | if (!timespec_valid(&ts)) |
9741ef96 | 3203 | return -EINVAL; |
c19384b5 PP |
3204 | |
3205 | t = timespec_to_ktime(ts); | |
34f01cc1 | 3206 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 3207 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 3208 | tp = &t; |
1da177e4 LT |
3209 | } |
3210 | /* | |
52400ba9 | 3211 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 3212 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 3213 | */ |
f54f0986 | 3214 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 3215 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 3216 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 3217 | |
c19384b5 | 3218 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
3219 | } |
3220 | ||
03b8c7b6 | 3221 | static void __init futex_detect_cmpxchg(void) |
1da177e4 | 3222 | { |
03b8c7b6 | 3223 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 3224 | u32 curval; |
03b8c7b6 HC |
3225 | |
3226 | /* | |
3227 | * This will fail and we want it. Some arch implementations do | |
3228 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
3229 | * functionality. We want to know that before we call in any | |
3230 | * of the complex code paths. Also we want to prevent | |
3231 | * registration of robust lists in that case. NULL is | |
3232 | * guaranteed to fault and we get -EFAULT on functional | |
3233 | * implementation, the non-functional ones will return | |
3234 | * -ENOSYS. | |
3235 | */ | |
3236 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) | |
3237 | futex_cmpxchg_enabled = 1; | |
3238 | #endif | |
3239 | } | |
3240 | ||
3241 | static int __init futex_init(void) | |
3242 | { | |
63b1a816 | 3243 | unsigned int futex_shift; |
a52b89eb DB |
3244 | unsigned long i; |
3245 | ||
3246 | #if CONFIG_BASE_SMALL | |
3247 | futex_hashsize = 16; | |
3248 | #else | |
3249 | futex_hashsize = roundup_pow_of_two(256 * num_possible_cpus()); | |
3250 | #endif | |
3251 | ||
3252 | futex_queues = alloc_large_system_hash("futex", sizeof(*futex_queues), | |
3253 | futex_hashsize, 0, | |
3254 | futex_hashsize < 256 ? HASH_SMALL : 0, | |
63b1a816 HC |
3255 | &futex_shift, NULL, |
3256 | futex_hashsize, futex_hashsize); | |
3257 | futex_hashsize = 1UL << futex_shift; | |
03b8c7b6 HC |
3258 | |
3259 | futex_detect_cmpxchg(); | |
a0c1e907 | 3260 | |
a52b89eb | 3261 | for (i = 0; i < futex_hashsize; i++) { |
11d4616b | 3262 | atomic_set(&futex_queues[i].waiters, 0); |
732375c6 | 3263 | plist_head_init(&futex_queues[i].chain); |
3e4ab747 TG |
3264 | spin_lock_init(&futex_queues[i].lock); |
3265 | } | |
3266 | ||
1da177e4 LT |
3267 | return 0; |
3268 | } | |
f6d107fb | 3269 | __initcall(futex_init); |