]>
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> |
84f001e1 | 64 | #include <linux/sched/wake_q.h> |
6e84f315 | 65 | #include <linux/sched/mm.h> |
13d60f4b | 66 | #include <linux/hugetlb.h> |
88c8004f | 67 | #include <linux/freezer.h> |
a52b89eb | 68 | #include <linux/bootmem.h> |
ab51fbab | 69 | #include <linux/fault-inject.h> |
b488893a | 70 | |
4732efbe | 71 | #include <asm/futex.h> |
1da177e4 | 72 | |
1696a8be | 73 | #include "locking/rtmutex_common.h" |
c87e2837 | 74 | |
99b60ce6 | 75 | /* |
d7e8af1a DB |
76 | * READ this before attempting to hack on futexes! |
77 | * | |
78 | * Basic futex operation and ordering guarantees | |
79 | * ============================================= | |
99b60ce6 TG |
80 | * |
81 | * The waiter reads the futex value in user space and calls | |
82 | * futex_wait(). This function computes the hash bucket and acquires | |
83 | * the hash bucket lock. After that it reads the futex user space value | |
b0c29f79 DB |
84 | * again and verifies that the data has not changed. If it has not changed |
85 | * it enqueues itself into the hash bucket, releases the hash bucket lock | |
86 | * and schedules. | |
99b60ce6 TG |
87 | * |
88 | * The waker side modifies the user space value of the futex and calls | |
b0c29f79 DB |
89 | * futex_wake(). This function computes the hash bucket and acquires the |
90 | * hash bucket lock. Then it looks for waiters on that futex in the hash | |
91 | * bucket and wakes them. | |
99b60ce6 | 92 | * |
b0c29f79 DB |
93 | * In futex wake up scenarios where no tasks are blocked on a futex, taking |
94 | * the hb spinlock can be avoided and simply return. In order for this | |
95 | * optimization to work, ordering guarantees must exist so that the waiter | |
96 | * being added to the list is acknowledged when the list is concurrently being | |
97 | * checked by the waker, avoiding scenarios like the following: | |
99b60ce6 TG |
98 | * |
99 | * CPU 0 CPU 1 | |
100 | * val = *futex; | |
101 | * sys_futex(WAIT, futex, val); | |
102 | * futex_wait(futex, val); | |
103 | * uval = *futex; | |
104 | * *futex = newval; | |
105 | * sys_futex(WAKE, futex); | |
106 | * futex_wake(futex); | |
107 | * if (queue_empty()) | |
108 | * return; | |
109 | * if (uval == val) | |
110 | * lock(hash_bucket(futex)); | |
111 | * queue(); | |
112 | * unlock(hash_bucket(futex)); | |
113 | * schedule(); | |
114 | * | |
115 | * This would cause the waiter on CPU 0 to wait forever because it | |
116 | * missed the transition of the user space value from val to newval | |
117 | * and the waker did not find the waiter in the hash bucket queue. | |
99b60ce6 | 118 | * |
b0c29f79 DB |
119 | * The correct serialization ensures that a waiter either observes |
120 | * the changed user space value before blocking or is woken by a | |
121 | * concurrent waker: | |
122 | * | |
123 | * CPU 0 CPU 1 | |
99b60ce6 TG |
124 | * val = *futex; |
125 | * sys_futex(WAIT, futex, val); | |
126 | * futex_wait(futex, val); | |
b0c29f79 | 127 | * |
d7e8af1a | 128 | * waiters++; (a) |
8ad7b378 DB |
129 | * smp_mb(); (A) <-- paired with -. |
130 | * | | |
131 | * lock(hash_bucket(futex)); | | |
132 | * | | |
133 | * uval = *futex; | | |
134 | * | *futex = newval; | |
135 | * | sys_futex(WAKE, futex); | |
136 | * | futex_wake(futex); | |
137 | * | | |
138 | * `--------> smp_mb(); (B) | |
99b60ce6 | 139 | * if (uval == val) |
b0c29f79 | 140 | * queue(); |
99b60ce6 | 141 | * unlock(hash_bucket(futex)); |
b0c29f79 DB |
142 | * schedule(); if (waiters) |
143 | * lock(hash_bucket(futex)); | |
d7e8af1a DB |
144 | * else wake_waiters(futex); |
145 | * waiters--; (b) unlock(hash_bucket(futex)); | |
b0c29f79 | 146 | * |
d7e8af1a DB |
147 | * Where (A) orders the waiters increment and the futex value read through |
148 | * atomic operations (see hb_waiters_inc) and where (B) orders the write | |
993b2ff2 DB |
149 | * to futex and the waiters read -- this is done by the barriers for both |
150 | * shared and private futexes in get_futex_key_refs(). | |
b0c29f79 DB |
151 | * |
152 | * This yields the following case (where X:=waiters, Y:=futex): | |
153 | * | |
154 | * X = Y = 0 | |
155 | * | |
156 | * w[X]=1 w[Y]=1 | |
157 | * MB MB | |
158 | * r[Y]=y r[X]=x | |
159 | * | |
160 | * Which guarantees that x==0 && y==0 is impossible; which translates back into | |
161 | * the guarantee that we cannot both miss the futex variable change and the | |
162 | * enqueue. | |
d7e8af1a DB |
163 | * |
164 | * Note that a new waiter is accounted for in (a) even when it is possible that | |
165 | * the wait call can return error, in which case we backtrack from it in (b). | |
166 | * Refer to the comment in queue_lock(). | |
167 | * | |
168 | * Similarly, in order to account for waiters being requeued on another | |
169 | * address we always increment the waiters for the destination bucket before | |
170 | * acquiring the lock. It then decrements them again after releasing it - | |
171 | * the code that actually moves the futex(es) between hash buckets (requeue_futex) | |
172 | * will do the additional required waiter count housekeeping. This is done for | |
173 | * double_lock_hb() and double_unlock_hb(), respectively. | |
99b60ce6 TG |
174 | */ |
175 | ||
03b8c7b6 | 176 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 177 | int __read_mostly futex_cmpxchg_enabled; |
03b8c7b6 | 178 | #endif |
a0c1e907 | 179 | |
b41277dc DH |
180 | /* |
181 | * Futex flags used to encode options to functions and preserve them across | |
182 | * restarts. | |
183 | */ | |
784bdf3b TG |
184 | #ifdef CONFIG_MMU |
185 | # define FLAGS_SHARED 0x01 | |
186 | #else | |
187 | /* | |
188 | * NOMMU does not have per process address space. Let the compiler optimize | |
189 | * code away. | |
190 | */ | |
191 | # define FLAGS_SHARED 0x00 | |
192 | #endif | |
b41277dc DH |
193 | #define FLAGS_CLOCKRT 0x02 |
194 | #define FLAGS_HAS_TIMEOUT 0x04 | |
195 | ||
c87e2837 IM |
196 | /* |
197 | * Priority Inheritance state: | |
198 | */ | |
199 | struct futex_pi_state { | |
200 | /* | |
201 | * list of 'owned' pi_state instances - these have to be | |
202 | * cleaned up in do_exit() if the task exits prematurely: | |
203 | */ | |
204 | struct list_head list; | |
205 | ||
206 | /* | |
207 | * The PI object: | |
208 | */ | |
209 | struct rt_mutex pi_mutex; | |
210 | ||
211 | struct task_struct *owner; | |
212 | atomic_t refcount; | |
213 | ||
214 | union futex_key key; | |
3859a271 | 215 | } __randomize_layout; |
c87e2837 | 216 | |
d8d88fbb DH |
217 | /** |
218 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 219 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
220 | * @task: the task waiting on the futex |
221 | * @lock_ptr: the hash bucket lock | |
222 | * @key: the key the futex is hashed on | |
223 | * @pi_state: optional priority inheritance state | |
224 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
225 | * @requeue_pi_key: the requeue_pi target futex key | |
226 | * @bitset: bitset for the optional bitmasked wakeup | |
227 | * | |
ac6424b9 | 228 | * We use this hashed waitqueue, instead of a normal wait_queue_entry_t, so |
1da177e4 LT |
229 | * we can wake only the relevant ones (hashed queues may be shared). |
230 | * | |
231 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 232 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 233 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
234 | * the second. |
235 | * | |
236 | * PI futexes are typically woken before they are removed from the hash list via | |
237 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
238 | */ |
239 | struct futex_q { | |
ec92d082 | 240 | struct plist_node list; |
1da177e4 | 241 | |
d8d88fbb | 242 | struct task_struct *task; |
1da177e4 | 243 | spinlock_t *lock_ptr; |
1da177e4 | 244 | union futex_key key; |
c87e2837 | 245 | struct futex_pi_state *pi_state; |
52400ba9 | 246 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 247 | union futex_key *requeue_pi_key; |
cd689985 | 248 | u32 bitset; |
3859a271 | 249 | } __randomize_layout; |
1da177e4 | 250 | |
5bdb05f9 DH |
251 | static const struct futex_q futex_q_init = { |
252 | /* list gets initialized in queue_me()*/ | |
253 | .key = FUTEX_KEY_INIT, | |
254 | .bitset = FUTEX_BITSET_MATCH_ANY | |
255 | }; | |
256 | ||
1da177e4 | 257 | /* |
b2d0994b DH |
258 | * Hash buckets are shared by all the futex_keys that hash to the same |
259 | * location. Each key may have multiple futex_q structures, one for each task | |
260 | * waiting on a futex. | |
1da177e4 LT |
261 | */ |
262 | struct futex_hash_bucket { | |
11d4616b | 263 | atomic_t waiters; |
ec92d082 PP |
264 | spinlock_t lock; |
265 | struct plist_head chain; | |
a52b89eb | 266 | } ____cacheline_aligned_in_smp; |
1da177e4 | 267 | |
ac742d37 RV |
268 | /* |
269 | * The base of the bucket array and its size are always used together | |
270 | * (after initialization only in hash_futex()), so ensure that they | |
271 | * reside in the same cacheline. | |
272 | */ | |
273 | static struct { | |
274 | struct futex_hash_bucket *queues; | |
275 | unsigned long hashsize; | |
276 | } __futex_data __read_mostly __aligned(2*sizeof(long)); | |
277 | #define futex_queues (__futex_data.queues) | |
278 | #define futex_hashsize (__futex_data.hashsize) | |
a52b89eb | 279 | |
1da177e4 | 280 | |
ab51fbab DB |
281 | /* |
282 | * Fault injections for futexes. | |
283 | */ | |
284 | #ifdef CONFIG_FAIL_FUTEX | |
285 | ||
286 | static struct { | |
287 | struct fault_attr attr; | |
288 | ||
621a5f7a | 289 | bool ignore_private; |
ab51fbab DB |
290 | } fail_futex = { |
291 | .attr = FAULT_ATTR_INITIALIZER, | |
621a5f7a | 292 | .ignore_private = false, |
ab51fbab DB |
293 | }; |
294 | ||
295 | static int __init setup_fail_futex(char *str) | |
296 | { | |
297 | return setup_fault_attr(&fail_futex.attr, str); | |
298 | } | |
299 | __setup("fail_futex=", setup_fail_futex); | |
300 | ||
5d285a7f | 301 | static bool should_fail_futex(bool fshared) |
ab51fbab DB |
302 | { |
303 | if (fail_futex.ignore_private && !fshared) | |
304 | return false; | |
305 | ||
306 | return should_fail(&fail_futex.attr, 1); | |
307 | } | |
308 | ||
309 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
310 | ||
311 | static int __init fail_futex_debugfs(void) | |
312 | { | |
313 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
314 | struct dentry *dir; | |
315 | ||
316 | dir = fault_create_debugfs_attr("fail_futex", NULL, | |
317 | &fail_futex.attr); | |
318 | if (IS_ERR(dir)) | |
319 | return PTR_ERR(dir); | |
320 | ||
321 | if (!debugfs_create_bool("ignore-private", mode, dir, | |
322 | &fail_futex.ignore_private)) { | |
323 | debugfs_remove_recursive(dir); | |
324 | return -ENOMEM; | |
325 | } | |
326 | ||
327 | return 0; | |
328 | } | |
329 | ||
330 | late_initcall(fail_futex_debugfs); | |
331 | ||
332 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
333 | ||
334 | #else | |
335 | static inline bool should_fail_futex(bool fshared) | |
336 | { | |
337 | return false; | |
338 | } | |
339 | #endif /* CONFIG_FAIL_FUTEX */ | |
340 | ||
b0c29f79 DB |
341 | static inline void futex_get_mm(union futex_key *key) |
342 | { | |
f1f10076 | 343 | mmgrab(key->private.mm); |
b0c29f79 DB |
344 | /* |
345 | * Ensure futex_get_mm() implies a full barrier such that | |
346 | * get_futex_key() implies a full barrier. This is relied upon | |
8ad7b378 | 347 | * as smp_mb(); (B), see the ordering comment above. |
b0c29f79 | 348 | */ |
4e857c58 | 349 | smp_mb__after_atomic(); |
b0c29f79 DB |
350 | } |
351 | ||
11d4616b LT |
352 | /* |
353 | * Reflects a new waiter being added to the waitqueue. | |
354 | */ | |
355 | static inline void hb_waiters_inc(struct futex_hash_bucket *hb) | |
b0c29f79 DB |
356 | { |
357 | #ifdef CONFIG_SMP | |
11d4616b | 358 | atomic_inc(&hb->waiters); |
b0c29f79 | 359 | /* |
11d4616b | 360 | * Full barrier (A), see the ordering comment above. |
b0c29f79 | 361 | */ |
4e857c58 | 362 | smp_mb__after_atomic(); |
11d4616b LT |
363 | #endif |
364 | } | |
365 | ||
366 | /* | |
367 | * Reflects a waiter being removed from the waitqueue by wakeup | |
368 | * paths. | |
369 | */ | |
370 | static inline void hb_waiters_dec(struct futex_hash_bucket *hb) | |
371 | { | |
372 | #ifdef CONFIG_SMP | |
373 | atomic_dec(&hb->waiters); | |
374 | #endif | |
375 | } | |
b0c29f79 | 376 | |
11d4616b LT |
377 | static inline int hb_waiters_pending(struct futex_hash_bucket *hb) |
378 | { | |
379 | #ifdef CONFIG_SMP | |
380 | return atomic_read(&hb->waiters); | |
b0c29f79 | 381 | #else |
11d4616b | 382 | return 1; |
b0c29f79 DB |
383 | #endif |
384 | } | |
385 | ||
e8b61b3f TG |
386 | /** |
387 | * hash_futex - Return the hash bucket in the global hash | |
388 | * @key: Pointer to the futex key for which the hash is calculated | |
389 | * | |
390 | * We hash on the keys returned from get_futex_key (see below) and return the | |
391 | * corresponding hash bucket in the global hash. | |
1da177e4 LT |
392 | */ |
393 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
394 | { | |
395 | u32 hash = jhash2((u32*)&key->both.word, | |
396 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
397 | key->both.offset); | |
a52b89eb | 398 | return &futex_queues[hash & (futex_hashsize - 1)]; |
1da177e4 LT |
399 | } |
400 | ||
e8b61b3f TG |
401 | |
402 | /** | |
403 | * match_futex - Check whether two futex keys are equal | |
404 | * @key1: Pointer to key1 | |
405 | * @key2: Pointer to key2 | |
406 | * | |
1da177e4 LT |
407 | * Return 1 if two futex_keys are equal, 0 otherwise. |
408 | */ | |
409 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
410 | { | |
2bc87203 DH |
411 | return (key1 && key2 |
412 | && key1->both.word == key2->both.word | |
1da177e4 LT |
413 | && key1->both.ptr == key2->both.ptr |
414 | && key1->both.offset == key2->both.offset); | |
415 | } | |
416 | ||
38d47c1b PZ |
417 | /* |
418 | * Take a reference to the resource addressed by a key. | |
419 | * Can be called while holding spinlocks. | |
420 | * | |
421 | */ | |
422 | static void get_futex_key_refs(union futex_key *key) | |
423 | { | |
424 | if (!key->both.ptr) | |
425 | return; | |
426 | ||
784bdf3b TG |
427 | /* |
428 | * On MMU less systems futexes are always "private" as there is no per | |
429 | * process address space. We need the smp wmb nevertheless - yes, | |
430 | * arch/blackfin has MMU less SMP ... | |
431 | */ | |
432 | if (!IS_ENABLED(CONFIG_MMU)) { | |
433 | smp_mb(); /* explicit smp_mb(); (B) */ | |
434 | return; | |
435 | } | |
436 | ||
38d47c1b PZ |
437 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { |
438 | case FUT_OFF_INODE: | |
8ad7b378 | 439 | ihold(key->shared.inode); /* implies smp_mb(); (B) */ |
38d47c1b PZ |
440 | break; |
441 | case FUT_OFF_MMSHARED: | |
8ad7b378 | 442 | futex_get_mm(key); /* implies smp_mb(); (B) */ |
38d47c1b | 443 | break; |
76835b0e | 444 | default: |
993b2ff2 DB |
445 | /* |
446 | * Private futexes do not hold reference on an inode or | |
447 | * mm, therefore the only purpose of calling get_futex_key_refs | |
448 | * is because we need the barrier for the lockless waiter check. | |
449 | */ | |
8ad7b378 | 450 | smp_mb(); /* explicit smp_mb(); (B) */ |
38d47c1b PZ |
451 | } |
452 | } | |
453 | ||
454 | /* | |
455 | * Drop a reference to the resource addressed by a key. | |
993b2ff2 DB |
456 | * The hash bucket spinlock must not be held. This is |
457 | * a no-op for private futexes, see comment in the get | |
458 | * counterpart. | |
38d47c1b PZ |
459 | */ |
460 | static void drop_futex_key_refs(union futex_key *key) | |
461 | { | |
90621c40 DH |
462 | if (!key->both.ptr) { |
463 | /* If we're here then we tried to put a key we failed to get */ | |
464 | WARN_ON_ONCE(1); | |
38d47c1b | 465 | return; |
90621c40 | 466 | } |
38d47c1b | 467 | |
784bdf3b TG |
468 | if (!IS_ENABLED(CONFIG_MMU)) |
469 | return; | |
470 | ||
38d47c1b PZ |
471 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { |
472 | case FUT_OFF_INODE: | |
473 | iput(key->shared.inode); | |
474 | break; | |
475 | case FUT_OFF_MMSHARED: | |
476 | mmdrop(key->private.mm); | |
477 | break; | |
478 | } | |
479 | } | |
480 | ||
34f01cc1 | 481 | /** |
d96ee56c DH |
482 | * get_futex_key() - Get parameters which are the keys for a futex |
483 | * @uaddr: virtual address of the futex | |
484 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED | |
485 | * @key: address where result is stored. | |
9ea71503 SB |
486 | * @rw: mapping needs to be read/write (values: VERIFY_READ, |
487 | * VERIFY_WRITE) | |
34f01cc1 | 488 | * |
6c23cbbd RD |
489 | * Return: a negative error code or 0 |
490 | * | |
7b4ff1ad | 491 | * The key words are stored in @key on success. |
1da177e4 | 492 | * |
6131ffaa | 493 | * For shared mappings, it's (page->index, file_inode(vma->vm_file), |
1da177e4 LT |
494 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
495 | * We can usually work out the index without swapping in the page. | |
496 | * | |
b2d0994b | 497 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 498 | */ |
64d1304a | 499 | static int |
9ea71503 | 500 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) |
1da177e4 | 501 | { |
e2970f2f | 502 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 503 | struct mm_struct *mm = current->mm; |
077fa7ae | 504 | struct page *page, *tail; |
14d27abd | 505 | struct address_space *mapping; |
9ea71503 | 506 | int err, ro = 0; |
1da177e4 LT |
507 | |
508 | /* | |
509 | * The futex address must be "naturally" aligned. | |
510 | */ | |
e2970f2f | 511 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 512 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 513 | return -EINVAL; |
e2970f2f | 514 | address -= key->both.offset; |
1da177e4 | 515 | |
5cdec2d8 LT |
516 | if (unlikely(!access_ok(rw, uaddr, sizeof(u32)))) |
517 | return -EFAULT; | |
518 | ||
ab51fbab DB |
519 | if (unlikely(should_fail_futex(fshared))) |
520 | return -EFAULT; | |
521 | ||
34f01cc1 ED |
522 | /* |
523 | * PROCESS_PRIVATE futexes are fast. | |
524 | * As the mm cannot disappear under us and the 'key' only needs | |
525 | * virtual address, we dont even have to find the underlying vma. | |
526 | * Note : We do have to check 'uaddr' is a valid user address, | |
527 | * but access_ok() should be faster than find_vma() | |
528 | */ | |
529 | if (!fshared) { | |
34f01cc1 ED |
530 | key->private.mm = mm; |
531 | key->private.address = address; | |
8ad7b378 | 532 | get_futex_key_refs(key); /* implies smp_mb(); (B) */ |
34f01cc1 ED |
533 | return 0; |
534 | } | |
1da177e4 | 535 | |
38d47c1b | 536 | again: |
ab51fbab DB |
537 | /* Ignore any VERIFY_READ mapping (futex common case) */ |
538 | if (unlikely(should_fail_futex(fshared))) | |
539 | return -EFAULT; | |
540 | ||
7485d0d3 | 541 | err = get_user_pages_fast(address, 1, 1, &page); |
9ea71503 SB |
542 | /* |
543 | * If write access is not required (eg. FUTEX_WAIT), try | |
544 | * and get read-only access. | |
545 | */ | |
546 | if (err == -EFAULT && rw == VERIFY_READ) { | |
547 | err = get_user_pages_fast(address, 1, 0, &page); | |
548 | ro = 1; | |
549 | } | |
38d47c1b PZ |
550 | if (err < 0) |
551 | return err; | |
9ea71503 SB |
552 | else |
553 | err = 0; | |
38d47c1b | 554 | |
65d8fc77 MG |
555 | /* |
556 | * The treatment of mapping from this point on is critical. The page | |
557 | * lock protects many things but in this context the page lock | |
558 | * stabilizes mapping, prevents inode freeing in the shared | |
559 | * file-backed region case and guards against movement to swap cache. | |
560 | * | |
561 | * Strictly speaking the page lock is not needed in all cases being | |
562 | * considered here and page lock forces unnecessarily serialization | |
563 | * From this point on, mapping will be re-verified if necessary and | |
564 | * page lock will be acquired only if it is unavoidable | |
077fa7ae MG |
565 | * |
566 | * Mapping checks require the head page for any compound page so the | |
567 | * head page and mapping is looked up now. For anonymous pages, it | |
568 | * does not matter if the page splits in the future as the key is | |
569 | * based on the address. For filesystem-backed pages, the tail is | |
570 | * required as the index of the page determines the key. For | |
571 | * base pages, there is no tail page and tail == page. | |
65d8fc77 | 572 | */ |
077fa7ae | 573 | tail = page; |
65d8fc77 MG |
574 | page = compound_head(page); |
575 | mapping = READ_ONCE(page->mapping); | |
576 | ||
e6780f72 | 577 | /* |
14d27abd | 578 | * If page->mapping is NULL, then it cannot be a PageAnon |
e6780f72 HD |
579 | * page; but it might be the ZERO_PAGE or in the gate area or |
580 | * in a special mapping (all cases which we are happy to fail); | |
581 | * or it may have been a good file page when get_user_pages_fast | |
582 | * found it, but truncated or holepunched or subjected to | |
583 | * invalidate_complete_page2 before we got the page lock (also | |
584 | * cases which we are happy to fail). And we hold a reference, | |
585 | * so refcount care in invalidate_complete_page's remove_mapping | |
586 | * prevents drop_caches from setting mapping to NULL beneath us. | |
587 | * | |
588 | * The case we do have to guard against is when memory pressure made | |
589 | * shmem_writepage move it from filecache to swapcache beneath us: | |
14d27abd | 590 | * an unlikely race, but we do need to retry for page->mapping. |
e6780f72 | 591 | */ |
65d8fc77 MG |
592 | if (unlikely(!mapping)) { |
593 | int shmem_swizzled; | |
594 | ||
595 | /* | |
596 | * Page lock is required to identify which special case above | |
597 | * applies. If this is really a shmem page then the page lock | |
598 | * will prevent unexpected transitions. | |
599 | */ | |
600 | lock_page(page); | |
601 | shmem_swizzled = PageSwapCache(page) || page->mapping; | |
14d27abd KS |
602 | unlock_page(page); |
603 | put_page(page); | |
65d8fc77 | 604 | |
e6780f72 HD |
605 | if (shmem_swizzled) |
606 | goto again; | |
65d8fc77 | 607 | |
e6780f72 | 608 | return -EFAULT; |
38d47c1b | 609 | } |
1da177e4 LT |
610 | |
611 | /* | |
612 | * Private mappings are handled in a simple way. | |
613 | * | |
65d8fc77 MG |
614 | * If the futex key is stored on an anonymous page, then the associated |
615 | * object is the mm which is implicitly pinned by the calling process. | |
616 | * | |
1da177e4 LT |
617 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if |
618 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 619 | * the object not the particular process. |
1da177e4 | 620 | */ |
14d27abd | 621 | if (PageAnon(page)) { |
9ea71503 SB |
622 | /* |
623 | * A RO anonymous page will never change and thus doesn't make | |
624 | * sense for futex operations. | |
625 | */ | |
ab51fbab | 626 | if (unlikely(should_fail_futex(fshared)) || ro) { |
9ea71503 SB |
627 | err = -EFAULT; |
628 | goto out; | |
629 | } | |
630 | ||
38d47c1b | 631 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 632 | key->private.mm = mm; |
e2970f2f | 633 | key->private.address = address; |
65d8fc77 MG |
634 | |
635 | get_futex_key_refs(key); /* implies smp_mb(); (B) */ | |
636 | ||
38d47c1b | 637 | } else { |
65d8fc77 MG |
638 | struct inode *inode; |
639 | ||
640 | /* | |
641 | * The associated futex object in this case is the inode and | |
642 | * the page->mapping must be traversed. Ordinarily this should | |
643 | * be stabilised under page lock but it's not strictly | |
644 | * necessary in this case as we just want to pin the inode, not | |
645 | * update the radix tree or anything like that. | |
646 | * | |
647 | * The RCU read lock is taken as the inode is finally freed | |
648 | * under RCU. If the mapping still matches expectations then the | |
649 | * mapping->host can be safely accessed as being a valid inode. | |
650 | */ | |
651 | rcu_read_lock(); | |
652 | ||
653 | if (READ_ONCE(page->mapping) != mapping) { | |
654 | rcu_read_unlock(); | |
655 | put_page(page); | |
656 | ||
657 | goto again; | |
658 | } | |
659 | ||
660 | inode = READ_ONCE(mapping->host); | |
661 | if (!inode) { | |
662 | rcu_read_unlock(); | |
663 | put_page(page); | |
664 | ||
665 | goto again; | |
666 | } | |
667 | ||
668 | /* | |
669 | * Take a reference unless it is about to be freed. Previously | |
670 | * this reference was taken by ihold under the page lock | |
671 | * pinning the inode in place so i_lock was unnecessary. The | |
672 | * only way for this check to fail is if the inode was | |
48fb6f4d MG |
673 | * truncated in parallel which is almost certainly an |
674 | * application bug. In such a case, just retry. | |
65d8fc77 MG |
675 | * |
676 | * We are not calling into get_futex_key_refs() in file-backed | |
677 | * cases, therefore a successful atomic_inc return below will | |
678 | * guarantee that get_futex_key() will still imply smp_mb(); (B). | |
679 | */ | |
48fb6f4d | 680 | if (!atomic_inc_not_zero(&inode->i_count)) { |
65d8fc77 MG |
681 | rcu_read_unlock(); |
682 | put_page(page); | |
683 | ||
684 | goto again; | |
685 | } | |
686 | ||
687 | /* Should be impossible but lets be paranoid for now */ | |
688 | if (WARN_ON_ONCE(inode->i_mapping != mapping)) { | |
689 | err = -EFAULT; | |
690 | rcu_read_unlock(); | |
691 | iput(inode); | |
692 | ||
693 | goto out; | |
694 | } | |
695 | ||
38d47c1b | 696 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ |
65d8fc77 | 697 | key->shared.inode = inode; |
077fa7ae | 698 | key->shared.pgoff = basepage_index(tail); |
65d8fc77 | 699 | rcu_read_unlock(); |
1da177e4 LT |
700 | } |
701 | ||
9ea71503 | 702 | out: |
14d27abd | 703 | put_page(page); |
9ea71503 | 704 | return err; |
1da177e4 LT |
705 | } |
706 | ||
ae791a2d | 707 | static inline void put_futex_key(union futex_key *key) |
1da177e4 | 708 | { |
38d47c1b | 709 | drop_futex_key_refs(key); |
1da177e4 LT |
710 | } |
711 | ||
d96ee56c DH |
712 | /** |
713 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
714 | * @uaddr: pointer to faulting user space address |
715 | * | |
716 | * Slow path to fixup the fault we just took in the atomic write | |
717 | * access to @uaddr. | |
718 | * | |
fb62db2b | 719 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
720 | * user address. We know that we faulted in the atomic pagefault |
721 | * disabled section so we can as well avoid the #PF overhead by | |
722 | * calling get_user_pages() right away. | |
723 | */ | |
724 | static int fault_in_user_writeable(u32 __user *uaddr) | |
725 | { | |
722d0172 AK |
726 | struct mm_struct *mm = current->mm; |
727 | int ret; | |
728 | ||
729 | down_read(&mm->mmap_sem); | |
2efaca92 | 730 | ret = fixup_user_fault(current, mm, (unsigned long)uaddr, |
4a9e1cda | 731 | FAULT_FLAG_WRITE, NULL); |
722d0172 AK |
732 | up_read(&mm->mmap_sem); |
733 | ||
d0725992 TG |
734 | return ret < 0 ? ret : 0; |
735 | } | |
736 | ||
4b1c486b DH |
737 | /** |
738 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
739 | * @hb: the hash bucket the futex_q's reside in |
740 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
741 | * |
742 | * Must be called with the hb lock held. | |
743 | */ | |
744 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
745 | union futex_key *key) | |
746 | { | |
747 | struct futex_q *this; | |
748 | ||
749 | plist_for_each_entry(this, &hb->chain, list) { | |
750 | if (match_futex(&this->key, key)) | |
751 | return this; | |
752 | } | |
753 | return NULL; | |
754 | } | |
755 | ||
37a9d912 ML |
756 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
757 | u32 uval, u32 newval) | |
36cf3b5c | 758 | { |
37a9d912 | 759 | int ret; |
36cf3b5c TG |
760 | |
761 | pagefault_disable(); | |
37a9d912 | 762 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
763 | pagefault_enable(); |
764 | ||
37a9d912 | 765 | return ret; |
36cf3b5c TG |
766 | } |
767 | ||
768 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
769 | { |
770 | int ret; | |
771 | ||
a866374a | 772 | pagefault_disable(); |
bd28b145 | 773 | ret = __get_user(*dest, from); |
a866374a | 774 | pagefault_enable(); |
1da177e4 LT |
775 | |
776 | return ret ? -EFAULT : 0; | |
777 | } | |
778 | ||
c87e2837 IM |
779 | |
780 | /* | |
781 | * PI code: | |
782 | */ | |
783 | static int refill_pi_state_cache(void) | |
784 | { | |
785 | struct futex_pi_state *pi_state; | |
786 | ||
787 | if (likely(current->pi_state_cache)) | |
788 | return 0; | |
789 | ||
4668edc3 | 790 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
791 | |
792 | if (!pi_state) | |
793 | return -ENOMEM; | |
794 | ||
c87e2837 IM |
795 | INIT_LIST_HEAD(&pi_state->list); |
796 | /* pi_mutex gets initialized later */ | |
797 | pi_state->owner = NULL; | |
798 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 799 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
800 | |
801 | current->pi_state_cache = pi_state; | |
802 | ||
803 | return 0; | |
804 | } | |
805 | ||
bf92cf3a | 806 | static struct futex_pi_state *alloc_pi_state(void) |
c87e2837 IM |
807 | { |
808 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
809 | ||
810 | WARN_ON(!pi_state); | |
811 | current->pi_state_cache = NULL; | |
812 | ||
813 | return pi_state; | |
814 | } | |
815 | ||
bf92cf3a PZ |
816 | static void get_pi_state(struct futex_pi_state *pi_state) |
817 | { | |
818 | WARN_ON_ONCE(!atomic_inc_not_zero(&pi_state->refcount)); | |
819 | } | |
820 | ||
30a6b803 | 821 | /* |
29e9ee5d TG |
822 | * Drops a reference to the pi_state object and frees or caches it |
823 | * when the last reference is gone. | |
30a6b803 | 824 | */ |
29e9ee5d | 825 | static void put_pi_state(struct futex_pi_state *pi_state) |
c87e2837 | 826 | { |
30a6b803 BS |
827 | if (!pi_state) |
828 | return; | |
829 | ||
c87e2837 IM |
830 | if (!atomic_dec_and_test(&pi_state->refcount)) |
831 | return; | |
832 | ||
833 | /* | |
834 | * If pi_state->owner is NULL, the owner is most probably dying | |
835 | * and has cleaned up the pi_state already | |
836 | */ | |
837 | if (pi_state->owner) { | |
c74aef2d | 838 | struct task_struct *owner; |
c87e2837 | 839 | |
c74aef2d PZ |
840 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
841 | owner = pi_state->owner; | |
842 | if (owner) { | |
843 | raw_spin_lock(&owner->pi_lock); | |
844 | list_del_init(&pi_state->list); | |
845 | raw_spin_unlock(&owner->pi_lock); | |
846 | } | |
847 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, owner); | |
848 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); | |
c87e2837 IM |
849 | } |
850 | ||
c74aef2d | 851 | if (current->pi_state_cache) { |
c87e2837 | 852 | kfree(pi_state); |
c74aef2d | 853 | } else { |
c87e2837 IM |
854 | /* |
855 | * pi_state->list is already empty. | |
856 | * clear pi_state->owner. | |
857 | * refcount is at 0 - put it back to 1. | |
858 | */ | |
859 | pi_state->owner = NULL; | |
860 | atomic_set(&pi_state->refcount, 1); | |
861 | current->pi_state_cache = pi_state; | |
862 | } | |
863 | } | |
864 | ||
865 | /* | |
866 | * Look up the task based on what TID userspace gave us. | |
867 | * We dont trust it. | |
868 | */ | |
bf92cf3a | 869 | static struct task_struct *futex_find_get_task(pid_t pid) |
c87e2837 IM |
870 | { |
871 | struct task_struct *p; | |
872 | ||
d359b549 | 873 | rcu_read_lock(); |
228ebcbe | 874 | p = find_task_by_vpid(pid); |
7a0ea09a MH |
875 | if (p) |
876 | get_task_struct(p); | |
a06381fe | 877 | |
d359b549 | 878 | rcu_read_unlock(); |
c87e2837 IM |
879 | |
880 | return p; | |
881 | } | |
882 | ||
bc2eecd7 NP |
883 | #ifdef CONFIG_FUTEX_PI |
884 | ||
c87e2837 IM |
885 | /* |
886 | * This task is holding PI mutexes at exit time => bad. | |
887 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
888 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
889 | */ | |
890 | void exit_pi_state_list(struct task_struct *curr) | |
891 | { | |
c87e2837 IM |
892 | struct list_head *next, *head = &curr->pi_state_list; |
893 | struct futex_pi_state *pi_state; | |
627371d7 | 894 | struct futex_hash_bucket *hb; |
38d47c1b | 895 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 896 | |
a0c1e907 TG |
897 | if (!futex_cmpxchg_enabled) |
898 | return; | |
c87e2837 IM |
899 | /* |
900 | * We are a ZOMBIE and nobody can enqueue itself on | |
901 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 902 | * versus waiters unqueueing themselves: |
c87e2837 | 903 | */ |
1d615482 | 904 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 905 | while (!list_empty(head)) { |
c87e2837 IM |
906 | next = head->next; |
907 | pi_state = list_entry(next, struct futex_pi_state, list); | |
908 | key = pi_state->key; | |
627371d7 | 909 | hb = hash_futex(&key); |
153fbd12 PZ |
910 | |
911 | /* | |
912 | * We can race against put_pi_state() removing itself from the | |
913 | * list (a waiter going away). put_pi_state() will first | |
914 | * decrement the reference count and then modify the list, so | |
915 | * its possible to see the list entry but fail this reference | |
916 | * acquire. | |
917 | * | |
918 | * In that case; drop the locks to let put_pi_state() make | |
919 | * progress and retry the loop. | |
920 | */ | |
921 | if (!atomic_inc_not_zero(&pi_state->refcount)) { | |
922 | raw_spin_unlock_irq(&curr->pi_lock); | |
923 | cpu_relax(); | |
924 | raw_spin_lock_irq(&curr->pi_lock); | |
925 | continue; | |
926 | } | |
1d615482 | 927 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 928 | |
c87e2837 | 929 | spin_lock(&hb->lock); |
c74aef2d PZ |
930 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
931 | raw_spin_lock(&curr->pi_lock); | |
627371d7 IM |
932 | /* |
933 | * We dropped the pi-lock, so re-check whether this | |
934 | * task still owns the PI-state: | |
935 | */ | |
c87e2837 | 936 | if (head->next != next) { |
153fbd12 | 937 | /* retain curr->pi_lock for the loop invariant */ |
c74aef2d | 938 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 | 939 | spin_unlock(&hb->lock); |
153fbd12 | 940 | put_pi_state(pi_state); |
c87e2837 IM |
941 | continue; |
942 | } | |
943 | ||
c87e2837 | 944 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
945 | WARN_ON(list_empty(&pi_state->list)); |
946 | list_del_init(&pi_state->list); | |
c87e2837 | 947 | pi_state->owner = NULL; |
c87e2837 | 948 | |
153fbd12 | 949 | raw_spin_unlock(&curr->pi_lock); |
c74aef2d | 950 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
951 | spin_unlock(&hb->lock); |
952 | ||
16ffa12d PZ |
953 | rt_mutex_futex_unlock(&pi_state->pi_mutex); |
954 | put_pi_state(pi_state); | |
955 | ||
1d615482 | 956 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 957 | } |
1d615482 | 958 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
959 | } |
960 | ||
bc2eecd7 NP |
961 | #endif |
962 | ||
54a21788 TG |
963 | /* |
964 | * We need to check the following states: | |
965 | * | |
966 | * Waiter | pi_state | pi->owner | uTID | uODIED | ? | |
967 | * | |
968 | * [1] NULL | --- | --- | 0 | 0/1 | Valid | |
969 | * [2] NULL | --- | --- | >0 | 0/1 | Valid | |
970 | * | |
971 | * [3] Found | NULL | -- | Any | 0/1 | Invalid | |
972 | * | |
973 | * [4] Found | Found | NULL | 0 | 1 | Valid | |
974 | * [5] Found | Found | NULL | >0 | 1 | Invalid | |
975 | * | |
976 | * [6] Found | Found | task | 0 | 1 | Valid | |
977 | * | |
978 | * [7] Found | Found | NULL | Any | 0 | Invalid | |
979 | * | |
980 | * [8] Found | Found | task | ==taskTID | 0/1 | Valid | |
981 | * [9] Found | Found | task | 0 | 0 | Invalid | |
982 | * [10] Found | Found | task | !=taskTID | 0/1 | Invalid | |
983 | * | |
984 | * [1] Indicates that the kernel can acquire the futex atomically. We | |
985 | * came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. | |
986 | * | |
987 | * [2] Valid, if TID does not belong to a kernel thread. If no matching | |
988 | * thread is found then it indicates that the owner TID has died. | |
989 | * | |
990 | * [3] Invalid. The waiter is queued on a non PI futex | |
991 | * | |
992 | * [4] Valid state after exit_robust_list(), which sets the user space | |
993 | * value to FUTEX_WAITERS | FUTEX_OWNER_DIED. | |
994 | * | |
995 | * [5] The user space value got manipulated between exit_robust_list() | |
996 | * and exit_pi_state_list() | |
997 | * | |
998 | * [6] Valid state after exit_pi_state_list() which sets the new owner in | |
999 | * the pi_state but cannot access the user space value. | |
1000 | * | |
1001 | * [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. | |
1002 | * | |
1003 | * [8] Owner and user space value match | |
1004 | * | |
1005 | * [9] There is no transient state which sets the user space TID to 0 | |
1006 | * except exit_robust_list(), but this is indicated by the | |
1007 | * FUTEX_OWNER_DIED bit. See [4] | |
1008 | * | |
1009 | * [10] There is no transient state which leaves owner and user space | |
1010 | * TID out of sync. | |
734009e9 PZ |
1011 | * |
1012 | * | |
1013 | * Serialization and lifetime rules: | |
1014 | * | |
1015 | * hb->lock: | |
1016 | * | |
1017 | * hb -> futex_q, relation | |
1018 | * futex_q -> pi_state, relation | |
1019 | * | |
1020 | * (cannot be raw because hb can contain arbitrary amount | |
1021 | * of futex_q's) | |
1022 | * | |
1023 | * pi_mutex->wait_lock: | |
1024 | * | |
1025 | * {uval, pi_state} | |
1026 | * | |
1027 | * (and pi_mutex 'obviously') | |
1028 | * | |
1029 | * p->pi_lock: | |
1030 | * | |
1031 | * p->pi_state_list -> pi_state->list, relation | |
1032 | * | |
1033 | * pi_state->refcount: | |
1034 | * | |
1035 | * pi_state lifetime | |
1036 | * | |
1037 | * | |
1038 | * Lock order: | |
1039 | * | |
1040 | * hb->lock | |
1041 | * pi_mutex->wait_lock | |
1042 | * p->pi_lock | |
1043 | * | |
54a21788 | 1044 | */ |
e60cbc5c TG |
1045 | |
1046 | /* | |
1047 | * Validate that the existing waiter has a pi_state and sanity check | |
1048 | * the pi_state against the user space value. If correct, attach to | |
1049 | * it. | |
1050 | */ | |
734009e9 PZ |
1051 | static int attach_to_pi_state(u32 __user *uaddr, u32 uval, |
1052 | struct futex_pi_state *pi_state, | |
e60cbc5c | 1053 | struct futex_pi_state **ps) |
c87e2837 | 1054 | { |
778e9a9c | 1055 | pid_t pid = uval & FUTEX_TID_MASK; |
94ffac5d PZ |
1056 | u32 uval2; |
1057 | int ret; | |
c87e2837 | 1058 | |
e60cbc5c TG |
1059 | /* |
1060 | * Userspace might have messed up non-PI and PI futexes [3] | |
1061 | */ | |
1062 | if (unlikely(!pi_state)) | |
1063 | return -EINVAL; | |
06a9ec29 | 1064 | |
734009e9 PZ |
1065 | /* |
1066 | * We get here with hb->lock held, and having found a | |
1067 | * futex_top_waiter(). This means that futex_lock_pi() of said futex_q | |
1068 | * has dropped the hb->lock in between queue_me() and unqueue_me_pi(), | |
1069 | * which in turn means that futex_lock_pi() still has a reference on | |
1070 | * our pi_state. | |
16ffa12d PZ |
1071 | * |
1072 | * The waiter holding a reference on @pi_state also protects against | |
1073 | * the unlocked put_pi_state() in futex_unlock_pi(), futex_lock_pi() | |
1074 | * and futex_wait_requeue_pi() as it cannot go to 0 and consequently | |
1075 | * free pi_state before we can take a reference ourselves. | |
734009e9 | 1076 | */ |
e60cbc5c | 1077 | WARN_ON(!atomic_read(&pi_state->refcount)); |
59647b6a | 1078 | |
734009e9 PZ |
1079 | /* |
1080 | * Now that we have a pi_state, we can acquire wait_lock | |
1081 | * and do the state validation. | |
1082 | */ | |
1083 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); | |
1084 | ||
1085 | /* | |
1086 | * Since {uval, pi_state} is serialized by wait_lock, and our current | |
1087 | * uval was read without holding it, it can have changed. Verify it | |
1088 | * still is what we expect it to be, otherwise retry the entire | |
1089 | * operation. | |
1090 | */ | |
1091 | if (get_futex_value_locked(&uval2, uaddr)) | |
1092 | goto out_efault; | |
1093 | ||
1094 | if (uval != uval2) | |
1095 | goto out_eagain; | |
1096 | ||
e60cbc5c TG |
1097 | /* |
1098 | * Handle the owner died case: | |
1099 | */ | |
1100 | if (uval & FUTEX_OWNER_DIED) { | |
bd1dbcc6 | 1101 | /* |
e60cbc5c TG |
1102 | * exit_pi_state_list sets owner to NULL and wakes the |
1103 | * topmost waiter. The task which acquires the | |
1104 | * pi_state->rt_mutex will fixup owner. | |
bd1dbcc6 | 1105 | */ |
e60cbc5c | 1106 | if (!pi_state->owner) { |
59647b6a | 1107 | /* |
e60cbc5c TG |
1108 | * No pi state owner, but the user space TID |
1109 | * is not 0. Inconsistent state. [5] | |
59647b6a | 1110 | */ |
e60cbc5c | 1111 | if (pid) |
734009e9 | 1112 | goto out_einval; |
bd1dbcc6 | 1113 | /* |
e60cbc5c | 1114 | * Take a ref on the state and return success. [4] |
866293ee | 1115 | */ |
734009e9 | 1116 | goto out_attach; |
c87e2837 | 1117 | } |
bd1dbcc6 TG |
1118 | |
1119 | /* | |
e60cbc5c TG |
1120 | * If TID is 0, then either the dying owner has not |
1121 | * yet executed exit_pi_state_list() or some waiter | |
1122 | * acquired the rtmutex in the pi state, but did not | |
1123 | * yet fixup the TID in user space. | |
1124 | * | |
1125 | * Take a ref on the state and return success. [6] | |
1126 | */ | |
1127 | if (!pid) | |
734009e9 | 1128 | goto out_attach; |
e60cbc5c TG |
1129 | } else { |
1130 | /* | |
1131 | * If the owner died bit is not set, then the pi_state | |
1132 | * must have an owner. [7] | |
bd1dbcc6 | 1133 | */ |
e60cbc5c | 1134 | if (!pi_state->owner) |
734009e9 | 1135 | goto out_einval; |
c87e2837 IM |
1136 | } |
1137 | ||
e60cbc5c TG |
1138 | /* |
1139 | * Bail out if user space manipulated the futex value. If pi | |
1140 | * state exists then the owner TID must be the same as the | |
1141 | * user space TID. [9/10] | |
1142 | */ | |
1143 | if (pid != task_pid_vnr(pi_state->owner)) | |
734009e9 PZ |
1144 | goto out_einval; |
1145 | ||
1146 | out_attach: | |
bf92cf3a | 1147 | get_pi_state(pi_state); |
734009e9 | 1148 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
e60cbc5c TG |
1149 | *ps = pi_state; |
1150 | return 0; | |
734009e9 PZ |
1151 | |
1152 | out_einval: | |
1153 | ret = -EINVAL; | |
1154 | goto out_error; | |
1155 | ||
1156 | out_eagain: | |
1157 | ret = -EAGAIN; | |
1158 | goto out_error; | |
1159 | ||
1160 | out_efault: | |
1161 | ret = -EFAULT; | |
1162 | goto out_error; | |
1163 | ||
1164 | out_error: | |
1165 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); | |
1166 | return ret; | |
e60cbc5c TG |
1167 | } |
1168 | ||
04e1b2e5 TG |
1169 | /* |
1170 | * Lookup the task for the TID provided from user space and attach to | |
1171 | * it after doing proper sanity checks. | |
1172 | */ | |
1173 | static int attach_to_pi_owner(u32 uval, union futex_key *key, | |
1174 | struct futex_pi_state **ps) | |
e60cbc5c | 1175 | { |
e60cbc5c | 1176 | pid_t pid = uval & FUTEX_TID_MASK; |
04e1b2e5 TG |
1177 | struct futex_pi_state *pi_state; |
1178 | struct task_struct *p; | |
e60cbc5c | 1179 | |
c87e2837 | 1180 | /* |
e3f2ddea | 1181 | * We are the first waiter - try to look up the real owner and attach |
54a21788 | 1182 | * the new pi_state to it, but bail out when TID = 0 [1] |
c87e2837 | 1183 | */ |
778e9a9c | 1184 | if (!pid) |
e3f2ddea | 1185 | return -ESRCH; |
c87e2837 | 1186 | p = futex_find_get_task(pid); |
7a0ea09a MH |
1187 | if (!p) |
1188 | return -ESRCH; | |
778e9a9c | 1189 | |
a2129464 | 1190 | if (unlikely(p->flags & PF_KTHREAD)) { |
f0d71b3d TG |
1191 | put_task_struct(p); |
1192 | return -EPERM; | |
1193 | } | |
1194 | ||
778e9a9c AK |
1195 | /* |
1196 | * We need to look at the task state flags to figure out, | |
1197 | * whether the task is exiting. To protect against the do_exit | |
1198 | * change of the task flags, we do this protected by | |
1199 | * p->pi_lock: | |
1200 | */ | |
1d615482 | 1201 | raw_spin_lock_irq(&p->pi_lock); |
778e9a9c AK |
1202 | if (unlikely(p->flags & PF_EXITING)) { |
1203 | /* | |
1204 | * The task is on the way out. When PF_EXITPIDONE is | |
1205 | * set, we know that the task has finished the | |
1206 | * cleanup: | |
1207 | */ | |
1208 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
1209 | ||
1d615482 | 1210 | raw_spin_unlock_irq(&p->pi_lock); |
778e9a9c AK |
1211 | put_task_struct(p); |
1212 | return ret; | |
1213 | } | |
c87e2837 | 1214 | |
54a21788 TG |
1215 | /* |
1216 | * No existing pi state. First waiter. [2] | |
734009e9 PZ |
1217 | * |
1218 | * This creates pi_state, we have hb->lock held, this means nothing can | |
1219 | * observe this state, wait_lock is irrelevant. | |
54a21788 | 1220 | */ |
c87e2837 IM |
1221 | pi_state = alloc_pi_state(); |
1222 | ||
1223 | /* | |
04e1b2e5 | 1224 | * Initialize the pi_mutex in locked state and make @p |
c87e2837 IM |
1225 | * the owner of it: |
1226 | */ | |
1227 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
1228 | ||
1229 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 1230 | pi_state->key = *key; |
c87e2837 | 1231 | |
627371d7 | 1232 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 | 1233 | list_add(&pi_state->list, &p->pi_state_list); |
c74aef2d PZ |
1234 | /* |
1235 | * Assignment without holding pi_state->pi_mutex.wait_lock is safe | |
1236 | * because there is no concurrency as the object is not published yet. | |
1237 | */ | |
c87e2837 | 1238 | pi_state->owner = p; |
1d615482 | 1239 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
1240 | |
1241 | put_task_struct(p); | |
1242 | ||
d0aa7a70 | 1243 | *ps = pi_state; |
c87e2837 IM |
1244 | |
1245 | return 0; | |
1246 | } | |
1247 | ||
734009e9 PZ |
1248 | static int lookup_pi_state(u32 __user *uaddr, u32 uval, |
1249 | struct futex_hash_bucket *hb, | |
04e1b2e5 TG |
1250 | union futex_key *key, struct futex_pi_state **ps) |
1251 | { | |
499f5aca | 1252 | struct futex_q *top_waiter = futex_top_waiter(hb, key); |
04e1b2e5 TG |
1253 | |
1254 | /* | |
1255 | * If there is a waiter on that futex, validate it and | |
1256 | * attach to the pi_state when the validation succeeds. | |
1257 | */ | |
499f5aca | 1258 | if (top_waiter) |
734009e9 | 1259 | return attach_to_pi_state(uaddr, uval, top_waiter->pi_state, ps); |
04e1b2e5 TG |
1260 | |
1261 | /* | |
1262 | * We are the first waiter - try to look up the owner based on | |
1263 | * @uval and attach to it. | |
1264 | */ | |
1265 | return attach_to_pi_owner(uval, key, ps); | |
1266 | } | |
1267 | ||
af54d6a1 TG |
1268 | static int lock_pi_update_atomic(u32 __user *uaddr, u32 uval, u32 newval) |
1269 | { | |
1270 | u32 uninitialized_var(curval); | |
1271 | ||
ab51fbab DB |
1272 | if (unlikely(should_fail_futex(true))) |
1273 | return -EFAULT; | |
1274 | ||
af54d6a1 TG |
1275 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))) |
1276 | return -EFAULT; | |
1277 | ||
734009e9 | 1278 | /* If user space value changed, let the caller retry */ |
af54d6a1 TG |
1279 | return curval != uval ? -EAGAIN : 0; |
1280 | } | |
1281 | ||
1a52084d | 1282 | /** |
d96ee56c | 1283 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
1284 | * @uaddr: the pi futex user address |
1285 | * @hb: the pi futex hash bucket | |
1286 | * @key: the futex key associated with uaddr and hb | |
1287 | * @ps: the pi_state pointer where we store the result of the | |
1288 | * lookup | |
1289 | * @task: the task to perform the atomic lock work for. This will | |
1290 | * be "current" except in the case of requeue pi. | |
1291 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d | 1292 | * |
6c23cbbd | 1293 | * Return: |
7b4ff1ad MCC |
1294 | * - 0 - ready to wait; |
1295 | * - 1 - acquired the lock; | |
1296 | * - <0 - error | |
1a52084d DH |
1297 | * |
1298 | * The hb->lock and futex_key refs shall be held by the caller. | |
1299 | */ | |
1300 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
1301 | union futex_key *key, | |
1302 | struct futex_pi_state **ps, | |
bab5bc9e | 1303 | struct task_struct *task, int set_waiters) |
1a52084d | 1304 | { |
af54d6a1 | 1305 | u32 uval, newval, vpid = task_pid_vnr(task); |
499f5aca | 1306 | struct futex_q *top_waiter; |
af54d6a1 | 1307 | int ret; |
1a52084d DH |
1308 | |
1309 | /* | |
af54d6a1 TG |
1310 | * Read the user space value first so we can validate a few |
1311 | * things before proceeding further. | |
1a52084d | 1312 | */ |
af54d6a1 | 1313 | if (get_futex_value_locked(&uval, uaddr)) |
1a52084d DH |
1314 | return -EFAULT; |
1315 | ||
ab51fbab DB |
1316 | if (unlikely(should_fail_futex(true))) |
1317 | return -EFAULT; | |
1318 | ||
1a52084d DH |
1319 | /* |
1320 | * Detect deadlocks. | |
1321 | */ | |
af54d6a1 | 1322 | if ((unlikely((uval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
1323 | return -EDEADLK; |
1324 | ||
ab51fbab DB |
1325 | if ((unlikely(should_fail_futex(true)))) |
1326 | return -EDEADLK; | |
1327 | ||
1a52084d | 1328 | /* |
af54d6a1 TG |
1329 | * Lookup existing state first. If it exists, try to attach to |
1330 | * its pi_state. | |
1a52084d | 1331 | */ |
499f5aca PZ |
1332 | top_waiter = futex_top_waiter(hb, key); |
1333 | if (top_waiter) | |
734009e9 | 1334 | return attach_to_pi_state(uaddr, uval, top_waiter->pi_state, ps); |
1a52084d DH |
1335 | |
1336 | /* | |
af54d6a1 TG |
1337 | * No waiter and user TID is 0. We are here because the |
1338 | * waiters or the owner died bit is set or called from | |
1339 | * requeue_cmp_pi or for whatever reason something took the | |
1340 | * syscall. | |
1a52084d | 1341 | */ |
af54d6a1 | 1342 | if (!(uval & FUTEX_TID_MASK)) { |
59fa6245 | 1343 | /* |
af54d6a1 TG |
1344 | * We take over the futex. No other waiters and the user space |
1345 | * TID is 0. We preserve the owner died bit. | |
59fa6245 | 1346 | */ |
af54d6a1 TG |
1347 | newval = uval & FUTEX_OWNER_DIED; |
1348 | newval |= vpid; | |
1a52084d | 1349 | |
af54d6a1 TG |
1350 | /* The futex requeue_pi code can enforce the waiters bit */ |
1351 | if (set_waiters) | |
1352 | newval |= FUTEX_WAITERS; | |
1353 | ||
1354 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1355 | /* If the take over worked, return 1 */ | |
1356 | return ret < 0 ? ret : 1; | |
1357 | } | |
1a52084d DH |
1358 | |
1359 | /* | |
af54d6a1 TG |
1360 | * First waiter. Set the waiters bit before attaching ourself to |
1361 | * the owner. If owner tries to unlock, it will be forced into | |
1362 | * the kernel and blocked on hb->lock. | |
1a52084d | 1363 | */ |
af54d6a1 TG |
1364 | newval = uval | FUTEX_WAITERS; |
1365 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1366 | if (ret) | |
1367 | return ret; | |
1a52084d | 1368 | /* |
af54d6a1 TG |
1369 | * If the update of the user space value succeeded, we try to |
1370 | * attach to the owner. If that fails, no harm done, we only | |
1371 | * set the FUTEX_WAITERS bit in the user space variable. | |
1a52084d | 1372 | */ |
af54d6a1 | 1373 | return attach_to_pi_owner(uval, key, ps); |
1a52084d DH |
1374 | } |
1375 | ||
2e12978a LJ |
1376 | /** |
1377 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
1378 | * @q: The futex_q to unqueue | |
1379 | * | |
1380 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
1381 | */ | |
1382 | static void __unqueue_futex(struct futex_q *q) | |
1383 | { | |
1384 | struct futex_hash_bucket *hb; | |
1385 | ||
29096202 SR |
1386 | if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr)) |
1387 | || WARN_ON(plist_node_empty(&q->list))) | |
2e12978a LJ |
1388 | return; |
1389 | ||
1390 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
1391 | plist_del(&q->list, &hb->chain); | |
11d4616b | 1392 | hb_waiters_dec(hb); |
2e12978a LJ |
1393 | } |
1394 | ||
1da177e4 LT |
1395 | /* |
1396 | * The hash bucket lock must be held when this is called. | |
1d0dcb3a DB |
1397 | * Afterwards, the futex_q must not be accessed. Callers |
1398 | * must ensure to later call wake_up_q() for the actual | |
1399 | * wakeups to occur. | |
1da177e4 | 1400 | */ |
1d0dcb3a | 1401 | static void mark_wake_futex(struct wake_q_head *wake_q, struct futex_q *q) |
1da177e4 | 1402 | { |
f1a11e05 TG |
1403 | struct task_struct *p = q->task; |
1404 | ||
aa10990e DH |
1405 | if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n")) |
1406 | return; | |
1407 | ||
1da177e4 | 1408 | /* |
1d0dcb3a DB |
1409 | * Queue the task for later wakeup for after we've released |
1410 | * the hb->lock. wake_q_add() grabs reference to p. | |
1da177e4 | 1411 | */ |
1d0dcb3a | 1412 | wake_q_add(wake_q, p); |
2e12978a | 1413 | __unqueue_futex(q); |
1da177e4 | 1414 | /* |
38fcd06e DHV |
1415 | * The waiting task can free the futex_q as soon as q->lock_ptr = NULL |
1416 | * is written, without taking any locks. This is possible in the event | |
1417 | * of a spurious wakeup, for example. A memory barrier is required here | |
1418 | * to prevent the following store to lock_ptr from getting ahead of the | |
1419 | * plist_del in __unqueue_futex(). | |
1da177e4 | 1420 | */ |
1b367ece | 1421 | smp_store_release(&q->lock_ptr, NULL); |
1da177e4 LT |
1422 | } |
1423 | ||
16ffa12d PZ |
1424 | /* |
1425 | * Caller must hold a reference on @pi_state. | |
1426 | */ | |
1427 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_pi_state *pi_state) | |
c87e2837 | 1428 | { |
7cfdaf38 | 1429 | u32 uninitialized_var(curval), newval; |
16ffa12d | 1430 | struct task_struct *new_owner; |
aa2bfe55 | 1431 | bool postunlock = false; |
194a6b5b | 1432 | DEFINE_WAKE_Q(wake_q); |
13fbca4c | 1433 | int ret = 0; |
c87e2837 | 1434 | |
c87e2837 | 1435 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
bebe5b51 | 1436 | if (WARN_ON_ONCE(!new_owner)) { |
16ffa12d | 1437 | /* |
bebe5b51 | 1438 | * As per the comment in futex_unlock_pi() this should not happen. |
16ffa12d PZ |
1439 | * |
1440 | * When this happens, give up our locks and try again, giving | |
1441 | * the futex_lock_pi() instance time to complete, either by | |
1442 | * waiting on the rtmutex or removing itself from the futex | |
1443 | * queue. | |
1444 | */ | |
1445 | ret = -EAGAIN; | |
1446 | goto out_unlock; | |
73d786bd | 1447 | } |
c87e2837 IM |
1448 | |
1449 | /* | |
16ffa12d PZ |
1450 | * We pass it to the next owner. The WAITERS bit is always kept |
1451 | * enabled while there is PI state around. We cleanup the owner | |
1452 | * died bit, because we are the owner. | |
c87e2837 | 1453 | */ |
13fbca4c | 1454 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 1455 | |
ab51fbab DB |
1456 | if (unlikely(should_fail_futex(true))) |
1457 | ret = -EFAULT; | |
1458 | ||
89e9e66b | 1459 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) { |
13fbca4c | 1460 | ret = -EFAULT; |
734009e9 | 1461 | |
89e9e66b SAS |
1462 | } else if (curval != uval) { |
1463 | /* | |
1464 | * If a unconditional UNLOCK_PI operation (user space did not | |
1465 | * try the TID->0 transition) raced with a waiter setting the | |
1466 | * FUTEX_WAITERS flag between get_user() and locking the hash | |
1467 | * bucket lock, retry the operation. | |
1468 | */ | |
1469 | if ((FUTEX_TID_MASK & curval) == uval) | |
1470 | ret = -EAGAIN; | |
1471 | else | |
1472 | ret = -EINVAL; | |
1473 | } | |
734009e9 | 1474 | |
16ffa12d PZ |
1475 | if (ret) |
1476 | goto out_unlock; | |
c87e2837 | 1477 | |
94ffac5d PZ |
1478 | /* |
1479 | * This is a point of no return; once we modify the uval there is no | |
1480 | * going back and subsequent operations must not fail. | |
1481 | */ | |
1482 | ||
b4abf910 | 1483 | raw_spin_lock(&pi_state->owner->pi_lock); |
627371d7 IM |
1484 | WARN_ON(list_empty(&pi_state->list)); |
1485 | list_del_init(&pi_state->list); | |
b4abf910 | 1486 | raw_spin_unlock(&pi_state->owner->pi_lock); |
627371d7 | 1487 | |
b4abf910 | 1488 | raw_spin_lock(&new_owner->pi_lock); |
627371d7 | 1489 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
1490 | list_add(&pi_state->list, &new_owner->pi_state_list); |
1491 | pi_state->owner = new_owner; | |
b4abf910 | 1492 | raw_spin_unlock(&new_owner->pi_lock); |
627371d7 | 1493 | |
aa2bfe55 | 1494 | postunlock = __rt_mutex_futex_unlock(&pi_state->pi_mutex, &wake_q); |
5293c2ef | 1495 | |
16ffa12d | 1496 | out_unlock: |
5293c2ef | 1497 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
5293c2ef | 1498 | |
aa2bfe55 PZ |
1499 | if (postunlock) |
1500 | rt_mutex_postunlock(&wake_q); | |
c87e2837 | 1501 | |
16ffa12d | 1502 | return ret; |
c87e2837 IM |
1503 | } |
1504 | ||
8b8f319f IM |
1505 | /* |
1506 | * Express the locking dependencies for lockdep: | |
1507 | */ | |
1508 | static inline void | |
1509 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1510 | { | |
1511 | if (hb1 <= hb2) { | |
1512 | spin_lock(&hb1->lock); | |
1513 | if (hb1 < hb2) | |
1514 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
1515 | } else { /* hb1 > hb2 */ | |
1516 | spin_lock(&hb2->lock); | |
1517 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
1518 | } | |
1519 | } | |
1520 | ||
5eb3dc62 DH |
1521 | static inline void |
1522 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1523 | { | |
f061d351 | 1524 | spin_unlock(&hb1->lock); |
88f502fe IM |
1525 | if (hb1 != hb2) |
1526 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
1527 | } |
1528 | ||
1da177e4 | 1529 | /* |
b2d0994b | 1530 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 1531 | */ |
b41277dc DH |
1532 | static int |
1533 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 1534 | { |
e2970f2f | 1535 | struct futex_hash_bucket *hb; |
1da177e4 | 1536 | struct futex_q *this, *next; |
38d47c1b | 1537 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 | 1538 | int ret; |
194a6b5b | 1539 | DEFINE_WAKE_Q(wake_q); |
1da177e4 | 1540 | |
cd689985 TG |
1541 | if (!bitset) |
1542 | return -EINVAL; | |
1543 | ||
9ea71503 | 1544 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ); |
1da177e4 LT |
1545 | if (unlikely(ret != 0)) |
1546 | goto out; | |
1547 | ||
e2970f2f | 1548 | hb = hash_futex(&key); |
b0c29f79 DB |
1549 | |
1550 | /* Make sure we really have tasks to wakeup */ | |
1551 | if (!hb_waiters_pending(hb)) | |
1552 | goto out_put_key; | |
1553 | ||
e2970f2f | 1554 | spin_lock(&hb->lock); |
1da177e4 | 1555 | |
0d00c7b2 | 1556 | plist_for_each_entry_safe(this, next, &hb->chain, list) { |
1da177e4 | 1557 | if (match_futex (&this->key, &key)) { |
52400ba9 | 1558 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
1559 | ret = -EINVAL; |
1560 | break; | |
1561 | } | |
cd689985 TG |
1562 | |
1563 | /* Check if one of the bits is set in both bitsets */ | |
1564 | if (!(this->bitset & bitset)) | |
1565 | continue; | |
1566 | ||
1d0dcb3a | 1567 | mark_wake_futex(&wake_q, this); |
1da177e4 LT |
1568 | if (++ret >= nr_wake) |
1569 | break; | |
1570 | } | |
1571 | } | |
1572 | ||
e2970f2f | 1573 | spin_unlock(&hb->lock); |
1d0dcb3a | 1574 | wake_up_q(&wake_q); |
b0c29f79 | 1575 | out_put_key: |
ae791a2d | 1576 | put_futex_key(&key); |
42d35d48 | 1577 | out: |
1da177e4 LT |
1578 | return ret; |
1579 | } | |
1580 | ||
30d6e0a4 JS |
1581 | static int futex_atomic_op_inuser(unsigned int encoded_op, u32 __user *uaddr) |
1582 | { | |
1583 | unsigned int op = (encoded_op & 0x70000000) >> 28; | |
1584 | unsigned int cmp = (encoded_op & 0x0f000000) >> 24; | |
1585 | int oparg = sign_extend32((encoded_op & 0x00fff000) >> 12, 12); | |
1586 | int cmparg = sign_extend32(encoded_op & 0x00000fff, 12); | |
1587 | int oldval, ret; | |
1588 | ||
1589 | if (encoded_op & (FUTEX_OP_OPARG_SHIFT << 28)) { | |
e78c38f6 JS |
1590 | if (oparg < 0 || oparg > 31) { |
1591 | char comm[sizeof(current->comm)]; | |
1592 | /* | |
1593 | * kill this print and return -EINVAL when userspace | |
1594 | * is sane again | |
1595 | */ | |
1596 | pr_info_ratelimited("futex_wake_op: %s tries to shift op by %d; fix this program\n", | |
1597 | get_task_comm(comm, current), oparg); | |
1598 | oparg &= 31; | |
1599 | } | |
30d6e0a4 JS |
1600 | oparg = 1 << oparg; |
1601 | } | |
1602 | ||
1603 | if (!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))) | |
1604 | return -EFAULT; | |
1605 | ||
1606 | ret = arch_futex_atomic_op_inuser(op, oparg, &oldval, uaddr); | |
1607 | if (ret) | |
1608 | return ret; | |
1609 | ||
1610 | switch (cmp) { | |
1611 | case FUTEX_OP_CMP_EQ: | |
1612 | return oldval == cmparg; | |
1613 | case FUTEX_OP_CMP_NE: | |
1614 | return oldval != cmparg; | |
1615 | case FUTEX_OP_CMP_LT: | |
1616 | return oldval < cmparg; | |
1617 | case FUTEX_OP_CMP_GE: | |
1618 | return oldval >= cmparg; | |
1619 | case FUTEX_OP_CMP_LE: | |
1620 | return oldval <= cmparg; | |
1621 | case FUTEX_OP_CMP_GT: | |
1622 | return oldval > cmparg; | |
1623 | default: | |
1624 | return -ENOSYS; | |
1625 | } | |
1626 | } | |
1627 | ||
4732efbe JJ |
1628 | /* |
1629 | * Wake up all waiters hashed on the physical page that is mapped | |
1630 | * to this virtual address: | |
1631 | */ | |
e2970f2f | 1632 | static int |
b41277dc | 1633 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 1634 | int nr_wake, int nr_wake2, int op) |
4732efbe | 1635 | { |
38d47c1b | 1636 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 1637 | struct futex_hash_bucket *hb1, *hb2; |
4732efbe | 1638 | struct futex_q *this, *next; |
e4dc5b7a | 1639 | int ret, op_ret; |
194a6b5b | 1640 | DEFINE_WAKE_Q(wake_q); |
4732efbe | 1641 | |
e4dc5b7a | 1642 | retry: |
9ea71503 | 1643 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
4732efbe JJ |
1644 | if (unlikely(ret != 0)) |
1645 | goto out; | |
9ea71503 | 1646 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
4732efbe | 1647 | if (unlikely(ret != 0)) |
42d35d48 | 1648 | goto out_put_key1; |
4732efbe | 1649 | |
e2970f2f IM |
1650 | hb1 = hash_futex(&key1); |
1651 | hb2 = hash_futex(&key2); | |
4732efbe | 1652 | |
e4dc5b7a | 1653 | retry_private: |
eaaea803 | 1654 | double_lock_hb(hb1, hb2); |
e2970f2f | 1655 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1656 | if (unlikely(op_ret < 0)) { |
4732efbe | 1657 | |
5eb3dc62 | 1658 | double_unlock_hb(hb1, hb2); |
4732efbe | 1659 | |
7ee1dd3f | 1660 | #ifndef CONFIG_MMU |
e2970f2f IM |
1661 | /* |
1662 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
1663 | * but we might get them from range checking | |
1664 | */ | |
7ee1dd3f | 1665 | ret = op_ret; |
42d35d48 | 1666 | goto out_put_keys; |
7ee1dd3f DH |
1667 | #endif |
1668 | ||
796f8d9b DG |
1669 | if (unlikely(op_ret != -EFAULT)) { |
1670 | ret = op_ret; | |
42d35d48 | 1671 | goto out_put_keys; |
796f8d9b DG |
1672 | } |
1673 | ||
d0725992 | 1674 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 1675 | if (ret) |
de87fcc1 | 1676 | goto out_put_keys; |
4732efbe | 1677 | |
b41277dc | 1678 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1679 | goto retry_private; |
1680 | ||
ae791a2d TG |
1681 | put_futex_key(&key2); |
1682 | put_futex_key(&key1); | |
e4dc5b7a | 1683 | goto retry; |
4732efbe JJ |
1684 | } |
1685 | ||
0d00c7b2 | 1686 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
4732efbe | 1687 | if (match_futex (&this->key, &key1)) { |
aa10990e DH |
1688 | if (this->pi_state || this->rt_waiter) { |
1689 | ret = -EINVAL; | |
1690 | goto out_unlock; | |
1691 | } | |
1d0dcb3a | 1692 | mark_wake_futex(&wake_q, this); |
4732efbe JJ |
1693 | if (++ret >= nr_wake) |
1694 | break; | |
1695 | } | |
1696 | } | |
1697 | ||
1698 | if (op_ret > 0) { | |
4732efbe | 1699 | op_ret = 0; |
0d00c7b2 | 1700 | plist_for_each_entry_safe(this, next, &hb2->chain, list) { |
4732efbe | 1701 | if (match_futex (&this->key, &key2)) { |
aa10990e DH |
1702 | if (this->pi_state || this->rt_waiter) { |
1703 | ret = -EINVAL; | |
1704 | goto out_unlock; | |
1705 | } | |
1d0dcb3a | 1706 | mark_wake_futex(&wake_q, this); |
4732efbe JJ |
1707 | if (++op_ret >= nr_wake2) |
1708 | break; | |
1709 | } | |
1710 | } | |
1711 | ret += op_ret; | |
1712 | } | |
1713 | ||
aa10990e | 1714 | out_unlock: |
5eb3dc62 | 1715 | double_unlock_hb(hb1, hb2); |
1d0dcb3a | 1716 | wake_up_q(&wake_q); |
42d35d48 | 1717 | out_put_keys: |
ae791a2d | 1718 | put_futex_key(&key2); |
42d35d48 | 1719 | out_put_key1: |
ae791a2d | 1720 | put_futex_key(&key1); |
42d35d48 | 1721 | out: |
4732efbe JJ |
1722 | return ret; |
1723 | } | |
1724 | ||
9121e478 DH |
1725 | /** |
1726 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1727 | * @q: the futex_q to requeue | |
1728 | * @hb1: the source hash_bucket | |
1729 | * @hb2: the target hash_bucket | |
1730 | * @key2: the new key for the requeued futex_q | |
1731 | */ | |
1732 | static inline | |
1733 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1734 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1735 | { | |
1736 | ||
1737 | /* | |
1738 | * If key1 and key2 hash to the same bucket, no need to | |
1739 | * requeue. | |
1740 | */ | |
1741 | if (likely(&hb1->chain != &hb2->chain)) { | |
1742 | plist_del(&q->list, &hb1->chain); | |
11d4616b | 1743 | hb_waiters_dec(hb1); |
11d4616b | 1744 | hb_waiters_inc(hb2); |
fe1bce9e | 1745 | plist_add(&q->list, &hb2->chain); |
9121e478 | 1746 | q->lock_ptr = &hb2->lock; |
9121e478 DH |
1747 | } |
1748 | get_futex_key_refs(key2); | |
1749 | q->key = *key2; | |
1750 | } | |
1751 | ||
52400ba9 DH |
1752 | /** |
1753 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1754 | * @q: the futex_q |
1755 | * @key: the key of the requeue target futex | |
1756 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1757 | * |
1758 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1759 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1760 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1761 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1762 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1763 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1764 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1765 | */ |
1766 | static inline | |
beda2c7e DH |
1767 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1768 | struct futex_hash_bucket *hb) | |
52400ba9 | 1769 | { |
52400ba9 DH |
1770 | get_futex_key_refs(key); |
1771 | q->key = *key; | |
1772 | ||
2e12978a | 1773 | __unqueue_futex(q); |
52400ba9 DH |
1774 | |
1775 | WARN_ON(!q->rt_waiter); | |
1776 | q->rt_waiter = NULL; | |
1777 | ||
beda2c7e | 1778 | q->lock_ptr = &hb->lock; |
beda2c7e | 1779 | |
f1a11e05 | 1780 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1781 | } |
1782 | ||
1783 | /** | |
1784 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1785 | * @pifutex: the user address of the to futex |
1786 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1787 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1788 | * @key1: the from futex key | |
1789 | * @key2: the to futex key | |
1790 | * @ps: address to store the pi_state pointer | |
1791 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1792 | * |
1793 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1794 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1795 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1796 | * hb1 and hb2 must be held by the caller. | |
52400ba9 | 1797 | * |
6c23cbbd | 1798 | * Return: |
7b4ff1ad MCC |
1799 | * - 0 - failed to acquire the lock atomically; |
1800 | * - >0 - acquired the lock, return value is vpid of the top_waiter | |
1801 | * - <0 - error | |
52400ba9 DH |
1802 | */ |
1803 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1804 | struct futex_hash_bucket *hb1, | |
1805 | struct futex_hash_bucket *hb2, | |
1806 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1807 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1808 | { |
bab5bc9e | 1809 | struct futex_q *top_waiter = NULL; |
52400ba9 | 1810 | u32 curval; |
866293ee | 1811 | int ret, vpid; |
52400ba9 DH |
1812 | |
1813 | if (get_futex_value_locked(&curval, pifutex)) | |
1814 | return -EFAULT; | |
1815 | ||
ab51fbab DB |
1816 | if (unlikely(should_fail_futex(true))) |
1817 | return -EFAULT; | |
1818 | ||
bab5bc9e DH |
1819 | /* |
1820 | * Find the top_waiter and determine if there are additional waiters. | |
1821 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1822 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1823 | * as we have means to handle the possible fault. If not, don't set | |
1824 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1825 | * the kernel. | |
1826 | */ | |
52400ba9 DH |
1827 | top_waiter = futex_top_waiter(hb1, key1); |
1828 | ||
1829 | /* There are no waiters, nothing for us to do. */ | |
1830 | if (!top_waiter) | |
1831 | return 0; | |
1832 | ||
84bc4af5 DH |
1833 | /* Ensure we requeue to the expected futex. */ |
1834 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1835 | return -EINVAL; | |
1836 | ||
52400ba9 | 1837 | /* |
bab5bc9e DH |
1838 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1839 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1840 | * in ps in contended cases. | |
52400ba9 | 1841 | */ |
866293ee | 1842 | vpid = task_pid_vnr(top_waiter->task); |
bab5bc9e DH |
1843 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1844 | set_waiters); | |
866293ee | 1845 | if (ret == 1) { |
beda2c7e | 1846 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
866293ee TG |
1847 | return vpid; |
1848 | } | |
52400ba9 DH |
1849 | return ret; |
1850 | } | |
1851 | ||
1852 | /** | |
1853 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1854 | * @uaddr1: source futex user address |
b41277dc | 1855 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1856 | * @uaddr2: target futex user address |
1857 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1858 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1859 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1860 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1861 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1862 | * |
1863 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1864 | * uaddr2 atomically on behalf of the top waiter. | |
1865 | * | |
6c23cbbd | 1866 | * Return: |
7b4ff1ad MCC |
1867 | * - >=0 - on success, the number of tasks requeued or woken; |
1868 | * - <0 - on error | |
1da177e4 | 1869 | */ |
b41277dc DH |
1870 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
1871 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
1872 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 1873 | { |
38d47c1b | 1874 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1875 | int drop_count = 0, task_count = 0, ret; |
1876 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1877 | struct futex_hash_bucket *hb1, *hb2; |
1da177e4 | 1878 | struct futex_q *this, *next; |
194a6b5b | 1879 | DEFINE_WAKE_Q(wake_q); |
52400ba9 | 1880 | |
bc2eecd7 NP |
1881 | /* |
1882 | * When PI not supported: return -ENOSYS if requeue_pi is true, | |
1883 | * consequently the compiler knows requeue_pi is always false past | |
1884 | * this point which will optimize away all the conditional code | |
1885 | * further down. | |
1886 | */ | |
1887 | if (!IS_ENABLED(CONFIG_FUTEX_PI) && requeue_pi) | |
1888 | return -ENOSYS; | |
1889 | ||
52400ba9 | 1890 | if (requeue_pi) { |
e9c243a5 TG |
1891 | /* |
1892 | * Requeue PI only works on two distinct uaddrs. This | |
1893 | * check is only valid for private futexes. See below. | |
1894 | */ | |
1895 | if (uaddr1 == uaddr2) | |
1896 | return -EINVAL; | |
1897 | ||
52400ba9 DH |
1898 | /* |
1899 | * requeue_pi requires a pi_state, try to allocate it now | |
1900 | * without any locks in case it fails. | |
1901 | */ | |
1902 | if (refill_pi_state_cache()) | |
1903 | return -ENOMEM; | |
1904 | /* | |
1905 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1906 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1907 | * returning to userspace, so as to not leave the rt_mutex with | |
1908 | * waiters and no owner. However, second and third wake-ups | |
1909 | * cannot be predicted as they involve race conditions with the | |
1910 | * first wake and a fault while looking up the pi_state. Both | |
1911 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1912 | * use nr_wake=1. | |
1913 | */ | |
1914 | if (nr_wake != 1) | |
1915 | return -EINVAL; | |
1916 | } | |
1da177e4 | 1917 | |
42d35d48 | 1918 | retry: |
9ea71503 | 1919 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
1da177e4 LT |
1920 | if (unlikely(ret != 0)) |
1921 | goto out; | |
9ea71503 SB |
1922 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, |
1923 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1924 | if (unlikely(ret != 0)) |
42d35d48 | 1925 | goto out_put_key1; |
1da177e4 | 1926 | |
e9c243a5 TG |
1927 | /* |
1928 | * The check above which compares uaddrs is not sufficient for | |
1929 | * shared futexes. We need to compare the keys: | |
1930 | */ | |
1931 | if (requeue_pi && match_futex(&key1, &key2)) { | |
1932 | ret = -EINVAL; | |
1933 | goto out_put_keys; | |
1934 | } | |
1935 | ||
e2970f2f IM |
1936 | hb1 = hash_futex(&key1); |
1937 | hb2 = hash_futex(&key2); | |
1da177e4 | 1938 | |
e4dc5b7a | 1939 | retry_private: |
69cd9eba | 1940 | hb_waiters_inc(hb2); |
8b8f319f | 1941 | double_lock_hb(hb1, hb2); |
1da177e4 | 1942 | |
e2970f2f IM |
1943 | if (likely(cmpval != NULL)) { |
1944 | u32 curval; | |
1da177e4 | 1945 | |
e2970f2f | 1946 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1947 | |
1948 | if (unlikely(ret)) { | |
5eb3dc62 | 1949 | double_unlock_hb(hb1, hb2); |
69cd9eba | 1950 | hb_waiters_dec(hb2); |
1da177e4 | 1951 | |
e2970f2f | 1952 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1953 | if (ret) |
1954 | goto out_put_keys; | |
1da177e4 | 1955 | |
b41277dc | 1956 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 1957 | goto retry_private; |
1da177e4 | 1958 | |
ae791a2d TG |
1959 | put_futex_key(&key2); |
1960 | put_futex_key(&key1); | |
e4dc5b7a | 1961 | goto retry; |
1da177e4 | 1962 | } |
e2970f2f | 1963 | if (curval != *cmpval) { |
1da177e4 LT |
1964 | ret = -EAGAIN; |
1965 | goto out_unlock; | |
1966 | } | |
1967 | } | |
1968 | ||
52400ba9 | 1969 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1970 | /* |
1971 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1972 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1973 | * bit. We force this here where we are able to easily handle | |
1974 | * faults rather in the requeue loop below. | |
1975 | */ | |
52400ba9 | 1976 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1977 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1978 | |
1979 | /* | |
1980 | * At this point the top_waiter has either taken uaddr2 or is | |
1981 | * waiting on it. If the former, then the pi_state will not | |
1982 | * exist yet, look it up one more time to ensure we have a | |
866293ee TG |
1983 | * reference to it. If the lock was taken, ret contains the |
1984 | * vpid of the top waiter task. | |
ecb38b78 TG |
1985 | * If the lock was not taken, we have pi_state and an initial |
1986 | * refcount on it. In case of an error we have nothing. | |
52400ba9 | 1987 | */ |
866293ee | 1988 | if (ret > 0) { |
52400ba9 | 1989 | WARN_ON(pi_state); |
89061d3d | 1990 | drop_count++; |
52400ba9 | 1991 | task_count++; |
866293ee | 1992 | /* |
ecb38b78 TG |
1993 | * If we acquired the lock, then the user space value |
1994 | * of uaddr2 should be vpid. It cannot be changed by | |
1995 | * the top waiter as it is blocked on hb2 lock if it | |
1996 | * tries to do so. If something fiddled with it behind | |
1997 | * our back the pi state lookup might unearth it. So | |
1998 | * we rather use the known value than rereading and | |
1999 | * handing potential crap to lookup_pi_state. | |
2000 | * | |
2001 | * If that call succeeds then we have pi_state and an | |
2002 | * initial refcount on it. | |
866293ee | 2003 | */ |
734009e9 | 2004 | ret = lookup_pi_state(uaddr2, ret, hb2, &key2, &pi_state); |
52400ba9 DH |
2005 | } |
2006 | ||
2007 | switch (ret) { | |
2008 | case 0: | |
ecb38b78 | 2009 | /* We hold a reference on the pi state. */ |
52400ba9 | 2010 | break; |
4959f2de TG |
2011 | |
2012 | /* If the above failed, then pi_state is NULL */ | |
52400ba9 DH |
2013 | case -EFAULT: |
2014 | double_unlock_hb(hb1, hb2); | |
69cd9eba | 2015 | hb_waiters_dec(hb2); |
ae791a2d TG |
2016 | put_futex_key(&key2); |
2017 | put_futex_key(&key1); | |
d0725992 | 2018 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
2019 | if (!ret) |
2020 | goto retry; | |
2021 | goto out; | |
2022 | case -EAGAIN: | |
af54d6a1 TG |
2023 | /* |
2024 | * Two reasons for this: | |
2025 | * - Owner is exiting and we just wait for the | |
2026 | * exit to complete. | |
2027 | * - The user space value changed. | |
2028 | */ | |
52400ba9 | 2029 | double_unlock_hb(hb1, hb2); |
69cd9eba | 2030 | hb_waiters_dec(hb2); |
ae791a2d TG |
2031 | put_futex_key(&key2); |
2032 | put_futex_key(&key1); | |
52400ba9 DH |
2033 | cond_resched(); |
2034 | goto retry; | |
2035 | default: | |
2036 | goto out_unlock; | |
2037 | } | |
2038 | } | |
2039 | ||
0d00c7b2 | 2040 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
52400ba9 DH |
2041 | if (task_count - nr_wake >= nr_requeue) |
2042 | break; | |
2043 | ||
2044 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 2045 | continue; |
52400ba9 | 2046 | |
392741e0 DH |
2047 | /* |
2048 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
2049 | * be paired with each other and no other futex ops. | |
aa10990e DH |
2050 | * |
2051 | * We should never be requeueing a futex_q with a pi_state, | |
2052 | * which is awaiting a futex_unlock_pi(). | |
392741e0 DH |
2053 | */ |
2054 | if ((requeue_pi && !this->rt_waiter) || | |
aa10990e DH |
2055 | (!requeue_pi && this->rt_waiter) || |
2056 | this->pi_state) { | |
392741e0 DH |
2057 | ret = -EINVAL; |
2058 | break; | |
2059 | } | |
52400ba9 DH |
2060 | |
2061 | /* | |
2062 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
2063 | * lock, we already woke the top_waiter. If not, it will be | |
2064 | * woken by futex_unlock_pi(). | |
2065 | */ | |
2066 | if (++task_count <= nr_wake && !requeue_pi) { | |
1d0dcb3a | 2067 | mark_wake_futex(&wake_q, this); |
52400ba9 DH |
2068 | continue; |
2069 | } | |
1da177e4 | 2070 | |
84bc4af5 DH |
2071 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
2072 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
2073 | ret = -EINVAL; | |
2074 | break; | |
2075 | } | |
2076 | ||
52400ba9 DH |
2077 | /* |
2078 | * Requeue nr_requeue waiters and possibly one more in the case | |
2079 | * of requeue_pi if we couldn't acquire the lock atomically. | |
2080 | */ | |
2081 | if (requeue_pi) { | |
ecb38b78 TG |
2082 | /* |
2083 | * Prepare the waiter to take the rt_mutex. Take a | |
2084 | * refcount on the pi_state and store the pointer in | |
2085 | * the futex_q object of the waiter. | |
2086 | */ | |
bf92cf3a | 2087 | get_pi_state(pi_state); |
52400ba9 DH |
2088 | this->pi_state = pi_state; |
2089 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
2090 | this->rt_waiter, | |
c051b21f | 2091 | this->task); |
52400ba9 | 2092 | if (ret == 1) { |
ecb38b78 TG |
2093 | /* |
2094 | * We got the lock. We do neither drop the | |
2095 | * refcount on pi_state nor clear | |
2096 | * this->pi_state because the waiter needs the | |
2097 | * pi_state for cleaning up the user space | |
2098 | * value. It will drop the refcount after | |
2099 | * doing so. | |
2100 | */ | |
beda2c7e | 2101 | requeue_pi_wake_futex(this, &key2, hb2); |
89061d3d | 2102 | drop_count++; |
52400ba9 DH |
2103 | continue; |
2104 | } else if (ret) { | |
ecb38b78 TG |
2105 | /* |
2106 | * rt_mutex_start_proxy_lock() detected a | |
2107 | * potential deadlock when we tried to queue | |
2108 | * that waiter. Drop the pi_state reference | |
2109 | * which we took above and remove the pointer | |
2110 | * to the state from the waiters futex_q | |
2111 | * object. | |
2112 | */ | |
52400ba9 | 2113 | this->pi_state = NULL; |
29e9ee5d | 2114 | put_pi_state(pi_state); |
885c2cb7 TG |
2115 | /* |
2116 | * We stop queueing more waiters and let user | |
2117 | * space deal with the mess. | |
2118 | */ | |
2119 | break; | |
52400ba9 | 2120 | } |
1da177e4 | 2121 | } |
52400ba9 DH |
2122 | requeue_futex(this, hb1, hb2, &key2); |
2123 | drop_count++; | |
1da177e4 LT |
2124 | } |
2125 | ||
ecb38b78 TG |
2126 | /* |
2127 | * We took an extra initial reference to the pi_state either | |
2128 | * in futex_proxy_trylock_atomic() or in lookup_pi_state(). We | |
2129 | * need to drop it here again. | |
2130 | */ | |
29e9ee5d | 2131 | put_pi_state(pi_state); |
885c2cb7 TG |
2132 | |
2133 | out_unlock: | |
5eb3dc62 | 2134 | double_unlock_hb(hb1, hb2); |
1d0dcb3a | 2135 | wake_up_q(&wake_q); |
69cd9eba | 2136 | hb_waiters_dec(hb2); |
1da177e4 | 2137 | |
cd84a42f DH |
2138 | /* |
2139 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
2140 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
2141 | * one at key2 and updated their key pointer. We no longer need to | |
2142 | * hold the references to key1. | |
2143 | */ | |
1da177e4 | 2144 | while (--drop_count >= 0) |
9adef58b | 2145 | drop_futex_key_refs(&key1); |
1da177e4 | 2146 | |
42d35d48 | 2147 | out_put_keys: |
ae791a2d | 2148 | put_futex_key(&key2); |
42d35d48 | 2149 | out_put_key1: |
ae791a2d | 2150 | put_futex_key(&key1); |
42d35d48 | 2151 | out: |
52400ba9 | 2152 | return ret ? ret : task_count; |
1da177e4 LT |
2153 | } |
2154 | ||
2155 | /* The key must be already stored in q->key. */ | |
82af7aca | 2156 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 2157 | __acquires(&hb->lock) |
1da177e4 | 2158 | { |
e2970f2f | 2159 | struct futex_hash_bucket *hb; |
1da177e4 | 2160 | |
e2970f2f | 2161 | hb = hash_futex(&q->key); |
11d4616b LT |
2162 | |
2163 | /* | |
2164 | * Increment the counter before taking the lock so that | |
2165 | * a potential waker won't miss a to-be-slept task that is | |
2166 | * waiting for the spinlock. This is safe as all queue_lock() | |
2167 | * users end up calling queue_me(). Similarly, for housekeeping, | |
2168 | * decrement the counter at queue_unlock() when some error has | |
2169 | * occurred and we don't end up adding the task to the list. | |
2170 | */ | |
2171 | hb_waiters_inc(hb); | |
2172 | ||
e2970f2f | 2173 | q->lock_ptr = &hb->lock; |
1da177e4 | 2174 | |
8ad7b378 | 2175 | spin_lock(&hb->lock); /* implies smp_mb(); (A) */ |
e2970f2f | 2176 | return hb; |
1da177e4 LT |
2177 | } |
2178 | ||
d40d65c8 | 2179 | static inline void |
0d00c7b2 | 2180 | queue_unlock(struct futex_hash_bucket *hb) |
15e408cd | 2181 | __releases(&hb->lock) |
d40d65c8 DH |
2182 | { |
2183 | spin_unlock(&hb->lock); | |
11d4616b | 2184 | hb_waiters_dec(hb); |
d40d65c8 DH |
2185 | } |
2186 | ||
cfafcd11 | 2187 | static inline void __queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 2188 | { |
ec92d082 PP |
2189 | int prio; |
2190 | ||
2191 | /* | |
2192 | * The priority used to register this element is | |
2193 | * - either the real thread-priority for the real-time threads | |
2194 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
2195 | * - or MAX_RT_PRIO for non-RT threads. | |
2196 | * Thus, all RT-threads are woken first in priority order, and | |
2197 | * the others are woken last, in FIFO order. | |
2198 | */ | |
2199 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
2200 | ||
2201 | plist_node_init(&q->list, prio); | |
ec92d082 | 2202 | plist_add(&q->list, &hb->chain); |
c87e2837 | 2203 | q->task = current; |
cfafcd11 PZ |
2204 | } |
2205 | ||
2206 | /** | |
2207 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
2208 | * @q: The futex_q to enqueue | |
2209 | * @hb: The destination hash bucket | |
2210 | * | |
2211 | * The hb->lock must be held by the caller, and is released here. A call to | |
2212 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
2213 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
2214 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
2215 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
2216 | * an example). | |
2217 | */ | |
2218 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) | |
2219 | __releases(&hb->lock) | |
2220 | { | |
2221 | __queue_me(q, hb); | |
e2970f2f | 2222 | spin_unlock(&hb->lock); |
1da177e4 LT |
2223 | } |
2224 | ||
d40d65c8 DH |
2225 | /** |
2226 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
2227 | * @q: The futex_q to unqueue | |
2228 | * | |
2229 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
2230 | * be paired with exactly one earlier call to queue_me(). | |
2231 | * | |
6c23cbbd | 2232 | * Return: |
7b4ff1ad MCC |
2233 | * - 1 - if the futex_q was still queued (and we removed unqueued it); |
2234 | * - 0 - if the futex_q was already removed by the waking thread | |
1da177e4 | 2235 | */ |
1da177e4 LT |
2236 | static int unqueue_me(struct futex_q *q) |
2237 | { | |
1da177e4 | 2238 | spinlock_t *lock_ptr; |
e2970f2f | 2239 | int ret = 0; |
1da177e4 LT |
2240 | |
2241 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 2242 | retry: |
29b75eb2 JZ |
2243 | /* |
2244 | * q->lock_ptr can change between this read and the following spin_lock. | |
2245 | * Use READ_ONCE to forbid the compiler from reloading q->lock_ptr and | |
2246 | * optimizing lock_ptr out of the logic below. | |
2247 | */ | |
2248 | lock_ptr = READ_ONCE(q->lock_ptr); | |
c80544dc | 2249 | if (lock_ptr != NULL) { |
1da177e4 LT |
2250 | spin_lock(lock_ptr); |
2251 | /* | |
2252 | * q->lock_ptr can change between reading it and | |
2253 | * spin_lock(), causing us to take the wrong lock. This | |
2254 | * corrects the race condition. | |
2255 | * | |
2256 | * Reasoning goes like this: if we have the wrong lock, | |
2257 | * q->lock_ptr must have changed (maybe several times) | |
2258 | * between reading it and the spin_lock(). It can | |
2259 | * change again after the spin_lock() but only if it was | |
2260 | * already changed before the spin_lock(). It cannot, | |
2261 | * however, change back to the original value. Therefore | |
2262 | * we can detect whether we acquired the correct lock. | |
2263 | */ | |
2264 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
2265 | spin_unlock(lock_ptr); | |
2266 | goto retry; | |
2267 | } | |
2e12978a | 2268 | __unqueue_futex(q); |
c87e2837 IM |
2269 | |
2270 | BUG_ON(q->pi_state); | |
2271 | ||
1da177e4 LT |
2272 | spin_unlock(lock_ptr); |
2273 | ret = 1; | |
2274 | } | |
2275 | ||
9adef58b | 2276 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
2277 | return ret; |
2278 | } | |
2279 | ||
c87e2837 IM |
2280 | /* |
2281 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
2282 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
2283 | * and dropped here. | |
c87e2837 | 2284 | */ |
d0aa7a70 | 2285 | static void unqueue_me_pi(struct futex_q *q) |
15e408cd | 2286 | __releases(q->lock_ptr) |
c87e2837 | 2287 | { |
2e12978a | 2288 | __unqueue_futex(q); |
c87e2837 IM |
2289 | |
2290 | BUG_ON(!q->pi_state); | |
29e9ee5d | 2291 | put_pi_state(q->pi_state); |
c87e2837 IM |
2292 | q->pi_state = NULL; |
2293 | ||
d0aa7a70 | 2294 | spin_unlock(q->lock_ptr); |
c87e2837 IM |
2295 | } |
2296 | ||
d0aa7a70 | 2297 | /* |
cdf71a10 | 2298 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 2299 | * |
778e9a9c AK |
2300 | * Must be called with hash bucket lock held and mm->sem held for non |
2301 | * private futexes. | |
d0aa7a70 | 2302 | */ |
778e9a9c | 2303 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
ae791a2d | 2304 | struct task_struct *newowner) |
d0aa7a70 | 2305 | { |
cdf71a10 | 2306 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 2307 | struct futex_pi_state *pi_state = q->pi_state; |
7cfdaf38 | 2308 | u32 uval, uninitialized_var(curval), newval; |
734009e9 | 2309 | struct task_struct *oldowner; |
e4dc5b7a | 2310 | int ret; |
d0aa7a70 | 2311 | |
734009e9 PZ |
2312 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
2313 | ||
2314 | oldowner = pi_state->owner; | |
d0aa7a70 | 2315 | /* Owner died? */ |
1b7558e4 TG |
2316 | if (!pi_state->owner) |
2317 | newtid |= FUTEX_OWNER_DIED; | |
2318 | ||
2319 | /* | |
2320 | * We are here either because we stole the rtmutex from the | |
8161239a | 2321 | * previous highest priority waiter or we are the highest priority |
16ffa12d PZ |
2322 | * waiter but have failed to get the rtmutex the first time. |
2323 | * | |
8161239a LJ |
2324 | * We have to replace the newowner TID in the user space variable. |
2325 | * This must be atomic as we have to preserve the owner died bit here. | |
1b7558e4 | 2326 | * |
b2d0994b DH |
2327 | * Note: We write the user space value _before_ changing the pi_state |
2328 | * because we can fault here. Imagine swapped out pages or a fork | |
2329 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 | 2330 | * |
734009e9 PZ |
2331 | * Modifying pi_state _before_ the user space value would leave the |
2332 | * pi_state in an inconsistent state when we fault here, because we | |
2333 | * need to drop the locks to handle the fault. This might be observed | |
2334 | * in the PID check in lookup_pi_state. | |
1b7558e4 TG |
2335 | */ |
2336 | retry: | |
2337 | if (get_futex_value_locked(&uval, uaddr)) | |
2338 | goto handle_fault; | |
2339 | ||
16ffa12d | 2340 | for (;;) { |
1b7558e4 TG |
2341 | newval = (uval & FUTEX_OWNER_DIED) | newtid; |
2342 | ||
37a9d912 | 2343 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1b7558e4 TG |
2344 | goto handle_fault; |
2345 | if (curval == uval) | |
2346 | break; | |
2347 | uval = curval; | |
2348 | } | |
2349 | ||
2350 | /* | |
2351 | * We fixed up user space. Now we need to fix the pi_state | |
2352 | * itself. | |
2353 | */ | |
d0aa7a70 | 2354 | if (pi_state->owner != NULL) { |
734009e9 | 2355 | raw_spin_lock(&pi_state->owner->pi_lock); |
d0aa7a70 PP |
2356 | WARN_ON(list_empty(&pi_state->list)); |
2357 | list_del_init(&pi_state->list); | |
734009e9 | 2358 | raw_spin_unlock(&pi_state->owner->pi_lock); |
1b7558e4 | 2359 | } |
d0aa7a70 | 2360 | |
cdf71a10 | 2361 | pi_state->owner = newowner; |
d0aa7a70 | 2362 | |
734009e9 | 2363 | raw_spin_lock(&newowner->pi_lock); |
d0aa7a70 | 2364 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 | 2365 | list_add(&pi_state->list, &newowner->pi_state_list); |
734009e9 PZ |
2366 | raw_spin_unlock(&newowner->pi_lock); |
2367 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); | |
2368 | ||
1b7558e4 | 2369 | return 0; |
d0aa7a70 | 2370 | |
d0aa7a70 | 2371 | /* |
734009e9 PZ |
2372 | * To handle the page fault we need to drop the locks here. That gives |
2373 | * the other task (either the highest priority waiter itself or the | |
2374 | * task which stole the rtmutex) the chance to try the fixup of the | |
2375 | * pi_state. So once we are back from handling the fault we need to | |
2376 | * check the pi_state after reacquiring the locks and before trying to | |
2377 | * do another fixup. When the fixup has been done already we simply | |
2378 | * return. | |
2379 | * | |
2380 | * Note: we hold both hb->lock and pi_mutex->wait_lock. We can safely | |
2381 | * drop hb->lock since the caller owns the hb -> futex_q relation. | |
2382 | * Dropping the pi_mutex->wait_lock requires the state revalidate. | |
d0aa7a70 | 2383 | */ |
1b7558e4 | 2384 | handle_fault: |
734009e9 | 2385 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
1b7558e4 | 2386 | spin_unlock(q->lock_ptr); |
778e9a9c | 2387 | |
d0725992 | 2388 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 2389 | |
1b7558e4 | 2390 | spin_lock(q->lock_ptr); |
734009e9 | 2391 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
778e9a9c | 2392 | |
1b7558e4 TG |
2393 | /* |
2394 | * Check if someone else fixed it for us: | |
2395 | */ | |
734009e9 PZ |
2396 | if (pi_state->owner != oldowner) { |
2397 | ret = 0; | |
2398 | goto out_unlock; | |
2399 | } | |
1b7558e4 TG |
2400 | |
2401 | if (ret) | |
734009e9 | 2402 | goto out_unlock; |
1b7558e4 TG |
2403 | |
2404 | goto retry; | |
734009e9 PZ |
2405 | |
2406 | out_unlock: | |
2407 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); | |
2408 | return ret; | |
d0aa7a70 PP |
2409 | } |
2410 | ||
72c1bbf3 | 2411 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 2412 | |
dd973998 DH |
2413 | /** |
2414 | * fixup_owner() - Post lock pi_state and corner case management | |
2415 | * @uaddr: user address of the futex | |
dd973998 DH |
2416 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
2417 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
2418 | * | |
2419 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
2420 | * the pi_state owner as well as handle race conditions that may allow us to | |
2421 | * acquire the lock. Must be called with the hb lock held. | |
2422 | * | |
6c23cbbd | 2423 | * Return: |
7b4ff1ad MCC |
2424 | * - 1 - success, lock taken; |
2425 | * - 0 - success, lock not taken; | |
2426 | * - <0 - on error (-EFAULT) | |
dd973998 | 2427 | */ |
ae791a2d | 2428 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 | 2429 | { |
dd973998 DH |
2430 | int ret = 0; |
2431 | ||
2432 | if (locked) { | |
2433 | /* | |
2434 | * Got the lock. We might not be the anticipated owner if we | |
2435 | * did a lock-steal - fix up the PI-state in that case: | |
16ffa12d PZ |
2436 | * |
2437 | * We can safely read pi_state->owner without holding wait_lock | |
2438 | * because we now own the rt_mutex, only the owner will attempt | |
2439 | * to change it. | |
dd973998 DH |
2440 | */ |
2441 | if (q->pi_state->owner != current) | |
ae791a2d | 2442 | ret = fixup_pi_state_owner(uaddr, q, current); |
dd973998 DH |
2443 | goto out; |
2444 | } | |
2445 | ||
dd973998 DH |
2446 | /* |
2447 | * Paranoia check. If we did not take the lock, then we should not be | |
8161239a | 2448 | * the owner of the rt_mutex. |
dd973998 | 2449 | */ |
73d786bd | 2450 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) { |
dd973998 DH |
2451 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " |
2452 | "pi-state %p\n", ret, | |
2453 | q->pi_state->pi_mutex.owner, | |
2454 | q->pi_state->owner); | |
73d786bd | 2455 | } |
dd973998 DH |
2456 | |
2457 | out: | |
2458 | return ret ? ret : locked; | |
2459 | } | |
2460 | ||
ca5f9524 DH |
2461 | /** |
2462 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
2463 | * @hb: the futex hash bucket, must be locked by the caller | |
2464 | * @q: the futex_q to queue up on | |
2465 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
2466 | */ |
2467 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 2468 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 2469 | { |
9beba3c5 DH |
2470 | /* |
2471 | * The task state is guaranteed to be set before another task can | |
b92b8b35 | 2472 | * wake it. set_current_state() is implemented using smp_store_mb() and |
9beba3c5 DH |
2473 | * queue_me() calls spin_unlock() upon completion, both serializing |
2474 | * access to the hash list and forcing another memory barrier. | |
2475 | */ | |
f1a11e05 | 2476 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 2477 | queue_me(q, hb); |
ca5f9524 DH |
2478 | |
2479 | /* Arm the timer */ | |
2e4b0d3f | 2480 | if (timeout) |
ca5f9524 | 2481 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); |
ca5f9524 DH |
2482 | |
2483 | /* | |
0729e196 DH |
2484 | * If we have been removed from the hash list, then another task |
2485 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
2486 | */ |
2487 | if (likely(!plist_node_empty(&q->list))) { | |
2488 | /* | |
2489 | * If the timer has already expired, current will already be | |
2490 | * flagged for rescheduling. Only call schedule if there | |
2491 | * is no timeout, or if it has yet to expire. | |
2492 | */ | |
2493 | if (!timeout || timeout->task) | |
88c8004f | 2494 | freezable_schedule(); |
ca5f9524 DH |
2495 | } |
2496 | __set_current_state(TASK_RUNNING); | |
2497 | } | |
2498 | ||
f801073f DH |
2499 | /** |
2500 | * futex_wait_setup() - Prepare to wait on a futex | |
2501 | * @uaddr: the futex userspace address | |
2502 | * @val: the expected value | |
b41277dc | 2503 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
2504 | * @q: the associated futex_q |
2505 | * @hb: storage for hash_bucket pointer to be returned to caller | |
2506 | * | |
2507 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
2508 | * compare it with the expected value. Handle atomic faults internally. | |
2509 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
2510 | * with no q.key reference on failure. | |
2511 | * | |
6c23cbbd | 2512 | * Return: |
7b4ff1ad MCC |
2513 | * - 0 - uaddr contains val and hb has been locked; |
2514 | * - <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked | |
f801073f | 2515 | */ |
b41277dc | 2516 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 2517 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 2518 | { |
e2970f2f IM |
2519 | u32 uval; |
2520 | int ret; | |
1da177e4 | 2521 | |
1da177e4 | 2522 | /* |
b2d0994b | 2523 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
2524 | * Order is important: |
2525 | * | |
2526 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
2527 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
2528 | * | |
2529 | * The basic logical guarantee of a futex is that it blocks ONLY | |
2530 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
2531 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
2532 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
2533 | * cond(var) false, which would violate the guarantee. |
2534 | * | |
8fe8f545 ML |
2535 | * On the other hand, we insert q and release the hash-bucket only |
2536 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
2537 | * absorb a wakeup if *uaddr does not match the desired values | |
2538 | * while the syscall executes. | |
1da177e4 | 2539 | */ |
f801073f | 2540 | retry: |
9ea71503 | 2541 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ); |
f801073f | 2542 | if (unlikely(ret != 0)) |
a5a2a0c7 | 2543 | return ret; |
f801073f DH |
2544 | |
2545 | retry_private: | |
2546 | *hb = queue_lock(q); | |
2547 | ||
e2970f2f | 2548 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 2549 | |
f801073f | 2550 | if (ret) { |
0d00c7b2 | 2551 | queue_unlock(*hb); |
1da177e4 | 2552 | |
e2970f2f | 2553 | ret = get_user(uval, uaddr); |
e4dc5b7a | 2554 | if (ret) |
f801073f | 2555 | goto out; |
1da177e4 | 2556 | |
b41277dc | 2557 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2558 | goto retry_private; |
2559 | ||
ae791a2d | 2560 | put_futex_key(&q->key); |
e4dc5b7a | 2561 | goto retry; |
1da177e4 | 2562 | } |
ca5f9524 | 2563 | |
f801073f | 2564 | if (uval != val) { |
0d00c7b2 | 2565 | queue_unlock(*hb); |
f801073f | 2566 | ret = -EWOULDBLOCK; |
2fff78c7 | 2567 | } |
1da177e4 | 2568 | |
f801073f DH |
2569 | out: |
2570 | if (ret) | |
ae791a2d | 2571 | put_futex_key(&q->key); |
f801073f DH |
2572 | return ret; |
2573 | } | |
2574 | ||
b41277dc DH |
2575 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
2576 | ktime_t *abs_time, u32 bitset) | |
f801073f DH |
2577 | { |
2578 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
2579 | struct restart_block *restart; |
2580 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 2581 | struct futex_q q = futex_q_init; |
f801073f DH |
2582 | int ret; |
2583 | ||
2584 | if (!bitset) | |
2585 | return -EINVAL; | |
f801073f DH |
2586 | q.bitset = bitset; |
2587 | ||
2588 | if (abs_time) { | |
2589 | to = &timeout; | |
2590 | ||
b41277dc DH |
2591 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2592 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2593 | HRTIMER_MODE_ABS); | |
f801073f DH |
2594 | hrtimer_init_sleeper(to, current); |
2595 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2596 | current->timer_slack_ns); | |
2597 | } | |
2598 | ||
d58e6576 | 2599 | retry: |
7ada876a DH |
2600 | /* |
2601 | * Prepare to wait on uaddr. On success, holds hb lock and increments | |
2602 | * q.key refs. | |
2603 | */ | |
b41277dc | 2604 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
2605 | if (ret) |
2606 | goto out; | |
2607 | ||
ca5f9524 | 2608 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 2609 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
2610 | |
2611 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 2612 | ret = 0; |
7ada876a | 2613 | /* unqueue_me() drops q.key ref */ |
1da177e4 | 2614 | if (!unqueue_me(&q)) |
7ada876a | 2615 | goto out; |
2fff78c7 | 2616 | ret = -ETIMEDOUT; |
ca5f9524 | 2617 | if (to && !to->task) |
7ada876a | 2618 | goto out; |
72c1bbf3 | 2619 | |
e2970f2f | 2620 | /* |
d58e6576 TG |
2621 | * We expect signal_pending(current), but we might be the |
2622 | * victim of a spurious wakeup as well. | |
e2970f2f | 2623 | */ |
7ada876a | 2624 | if (!signal_pending(current)) |
d58e6576 | 2625 | goto retry; |
d58e6576 | 2626 | |
2fff78c7 | 2627 | ret = -ERESTARTSYS; |
c19384b5 | 2628 | if (!abs_time) |
7ada876a | 2629 | goto out; |
1da177e4 | 2630 | |
f56141e3 | 2631 | restart = ¤t->restart_block; |
2fff78c7 | 2632 | restart->fn = futex_wait_restart; |
a3c74c52 | 2633 | restart->futex.uaddr = uaddr; |
2fff78c7 | 2634 | restart->futex.val = val; |
2456e855 | 2635 | restart->futex.time = *abs_time; |
2fff78c7 | 2636 | restart->futex.bitset = bitset; |
0cd9c649 | 2637 | restart->futex.flags = flags | FLAGS_HAS_TIMEOUT; |
42d35d48 | 2638 | |
2fff78c7 PZ |
2639 | ret = -ERESTART_RESTARTBLOCK; |
2640 | ||
42d35d48 | 2641 | out: |
ca5f9524 DH |
2642 | if (to) { |
2643 | hrtimer_cancel(&to->timer); | |
2644 | destroy_hrtimer_on_stack(&to->timer); | |
2645 | } | |
c87e2837 IM |
2646 | return ret; |
2647 | } | |
2648 | ||
72c1bbf3 NP |
2649 | |
2650 | static long futex_wait_restart(struct restart_block *restart) | |
2651 | { | |
a3c74c52 | 2652 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 2653 | ktime_t t, *tp = NULL; |
72c1bbf3 | 2654 | |
a72188d8 | 2655 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
2456e855 | 2656 | t = restart->futex.time; |
a72188d8 DH |
2657 | tp = &t; |
2658 | } | |
72c1bbf3 | 2659 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
2660 | |
2661 | return (long)futex_wait(uaddr, restart->futex.flags, | |
2662 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
2663 | } |
2664 | ||
2665 | ||
c87e2837 IM |
2666 | /* |
2667 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
2668 | * and failed. The kernel side here does the whole locking operation: | |
767f509c DB |
2669 | * if there are waiters then it will block as a consequence of relying |
2670 | * on rt-mutexes, it does PI, etc. (Due to races the kernel might see | |
2671 | * a 0 value of the futex too.). | |
2672 | * | |
2673 | * Also serves as futex trylock_pi()'ing, and due semantics. | |
c87e2837 | 2674 | */ |
996636dd | 2675 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, |
b41277dc | 2676 | ktime_t *time, int trylock) |
c87e2837 | 2677 | { |
c5780e97 | 2678 | struct hrtimer_sleeper timeout, *to = NULL; |
16ffa12d | 2679 | struct futex_pi_state *pi_state = NULL; |
cfafcd11 | 2680 | struct rt_mutex_waiter rt_waiter; |
c87e2837 | 2681 | struct futex_hash_bucket *hb; |
5bdb05f9 | 2682 | struct futex_q q = futex_q_init; |
dd973998 | 2683 | int res, ret; |
c87e2837 | 2684 | |
bc2eecd7 NP |
2685 | if (!IS_ENABLED(CONFIG_FUTEX_PI)) |
2686 | return -ENOSYS; | |
2687 | ||
c87e2837 IM |
2688 | if (refill_pi_state_cache()) |
2689 | return -ENOMEM; | |
2690 | ||
c19384b5 | 2691 | if (time) { |
c5780e97 | 2692 | to = &timeout; |
237fc6e7 TG |
2693 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
2694 | HRTIMER_MODE_ABS); | |
c5780e97 | 2695 | hrtimer_init_sleeper(to, current); |
cc584b21 | 2696 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
2697 | } |
2698 | ||
42d35d48 | 2699 | retry: |
9ea71503 | 2700 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE); |
c87e2837 | 2701 | if (unlikely(ret != 0)) |
42d35d48 | 2702 | goto out; |
c87e2837 | 2703 | |
e4dc5b7a | 2704 | retry_private: |
82af7aca | 2705 | hb = queue_lock(&q); |
c87e2837 | 2706 | |
bab5bc9e | 2707 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 2708 | if (unlikely(ret)) { |
767f509c DB |
2709 | /* |
2710 | * Atomic work succeeded and we got the lock, | |
2711 | * or failed. Either way, we do _not_ block. | |
2712 | */ | |
778e9a9c | 2713 | switch (ret) { |
1a52084d DH |
2714 | case 1: |
2715 | /* We got the lock. */ | |
2716 | ret = 0; | |
2717 | goto out_unlock_put_key; | |
2718 | case -EFAULT: | |
2719 | goto uaddr_faulted; | |
778e9a9c AK |
2720 | case -EAGAIN: |
2721 | /* | |
af54d6a1 TG |
2722 | * Two reasons for this: |
2723 | * - Task is exiting and we just wait for the | |
2724 | * exit to complete. | |
2725 | * - The user space value changed. | |
778e9a9c | 2726 | */ |
0d00c7b2 | 2727 | queue_unlock(hb); |
ae791a2d | 2728 | put_futex_key(&q.key); |
778e9a9c AK |
2729 | cond_resched(); |
2730 | goto retry; | |
778e9a9c | 2731 | default: |
42d35d48 | 2732 | goto out_unlock_put_key; |
c87e2837 | 2733 | } |
c87e2837 IM |
2734 | } |
2735 | ||
cfafcd11 PZ |
2736 | WARN_ON(!q.pi_state); |
2737 | ||
c87e2837 IM |
2738 | /* |
2739 | * Only actually queue now that the atomic ops are done: | |
2740 | */ | |
cfafcd11 | 2741 | __queue_me(&q, hb); |
c87e2837 | 2742 | |
cfafcd11 | 2743 | if (trylock) { |
5293c2ef | 2744 | ret = rt_mutex_futex_trylock(&q.pi_state->pi_mutex); |
c87e2837 IM |
2745 | /* Fixup the trylock return value: */ |
2746 | ret = ret ? 0 : -EWOULDBLOCK; | |
cfafcd11 | 2747 | goto no_block; |
c87e2837 IM |
2748 | } |
2749 | ||
56222b21 PZ |
2750 | rt_mutex_init_waiter(&rt_waiter); |
2751 | ||
cfafcd11 | 2752 | /* |
56222b21 PZ |
2753 | * On PREEMPT_RT_FULL, when hb->lock becomes an rt_mutex, we must not |
2754 | * hold it while doing rt_mutex_start_proxy(), because then it will | |
2755 | * include hb->lock in the blocking chain, even through we'll not in | |
2756 | * fact hold it while blocking. This will lead it to report -EDEADLK | |
2757 | * and BUG when futex_unlock_pi() interleaves with this. | |
2758 | * | |
2759 | * Therefore acquire wait_lock while holding hb->lock, but drop the | |
2760 | * latter before calling rt_mutex_start_proxy_lock(). This still fully | |
2761 | * serializes against futex_unlock_pi() as that does the exact same | |
2762 | * lock handoff sequence. | |
cfafcd11 | 2763 | */ |
56222b21 PZ |
2764 | raw_spin_lock_irq(&q.pi_state->pi_mutex.wait_lock); |
2765 | spin_unlock(q.lock_ptr); | |
2766 | ret = __rt_mutex_start_proxy_lock(&q.pi_state->pi_mutex, &rt_waiter, current); | |
2767 | raw_spin_unlock_irq(&q.pi_state->pi_mutex.wait_lock); | |
2768 | ||
cfafcd11 PZ |
2769 | if (ret) { |
2770 | if (ret == 1) | |
2771 | ret = 0; | |
2772 | ||
56222b21 | 2773 | spin_lock(q.lock_ptr); |
cfafcd11 PZ |
2774 | goto no_block; |
2775 | } | |
2776 | ||
cfafcd11 PZ |
2777 | |
2778 | if (unlikely(to)) | |
2779 | hrtimer_start_expires(&to->timer, HRTIMER_MODE_ABS); | |
2780 | ||
2781 | ret = rt_mutex_wait_proxy_lock(&q.pi_state->pi_mutex, to, &rt_waiter); | |
2782 | ||
a99e4e41 | 2783 | spin_lock(q.lock_ptr); |
cfafcd11 PZ |
2784 | /* |
2785 | * If we failed to acquire the lock (signal/timeout), we must | |
2786 | * first acquire the hb->lock before removing the lock from the | |
2787 | * rt_mutex waitqueue, such that we can keep the hb and rt_mutex | |
2788 | * wait lists consistent. | |
56222b21 PZ |
2789 | * |
2790 | * In particular; it is important that futex_unlock_pi() can not | |
2791 | * observe this inconsistency. | |
cfafcd11 PZ |
2792 | */ |
2793 | if (ret && !rt_mutex_cleanup_proxy_lock(&q.pi_state->pi_mutex, &rt_waiter)) | |
2794 | ret = 0; | |
2795 | ||
2796 | no_block: | |
dd973998 DH |
2797 | /* |
2798 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2799 | * haven't already. | |
2800 | */ | |
ae791a2d | 2801 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 DH |
2802 | /* |
2803 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
2804 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
2805 | */ | |
2806 | if (res) | |
2807 | ret = (res < 0) ? res : 0; | |
c87e2837 | 2808 | |
e8f6386c | 2809 | /* |
dd973998 DH |
2810 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
2811 | * it and return the fault to userspace. | |
e8f6386c | 2812 | */ |
16ffa12d PZ |
2813 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) { |
2814 | pi_state = q.pi_state; | |
2815 | get_pi_state(pi_state); | |
2816 | } | |
e8f6386c | 2817 | |
778e9a9c AK |
2818 | /* Unqueue and drop the lock */ |
2819 | unqueue_me_pi(&q); | |
c87e2837 | 2820 | |
16ffa12d PZ |
2821 | if (pi_state) { |
2822 | rt_mutex_futex_unlock(&pi_state->pi_mutex); | |
2823 | put_pi_state(pi_state); | |
2824 | } | |
2825 | ||
5ecb01cf | 2826 | goto out_put_key; |
c87e2837 | 2827 | |
42d35d48 | 2828 | out_unlock_put_key: |
0d00c7b2 | 2829 | queue_unlock(hb); |
c87e2837 | 2830 | |
42d35d48 | 2831 | out_put_key: |
ae791a2d | 2832 | put_futex_key(&q.key); |
42d35d48 | 2833 | out: |
97181f9b TG |
2834 | if (to) { |
2835 | hrtimer_cancel(&to->timer); | |
237fc6e7 | 2836 | destroy_hrtimer_on_stack(&to->timer); |
97181f9b | 2837 | } |
dd973998 | 2838 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 2839 | |
42d35d48 | 2840 | uaddr_faulted: |
0d00c7b2 | 2841 | queue_unlock(hb); |
778e9a9c | 2842 | |
d0725992 | 2843 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
2844 | if (ret) |
2845 | goto out_put_key; | |
c87e2837 | 2846 | |
b41277dc | 2847 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2848 | goto retry_private; |
2849 | ||
ae791a2d | 2850 | put_futex_key(&q.key); |
e4dc5b7a | 2851 | goto retry; |
c87e2837 IM |
2852 | } |
2853 | ||
c87e2837 IM |
2854 | /* |
2855 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
2856 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
2857 | * and do the rt-mutex unlock. | |
2858 | */ | |
b41277dc | 2859 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 | 2860 | { |
ccf9e6a8 | 2861 | u32 uninitialized_var(curval), uval, vpid = task_pid_vnr(current); |
38d47c1b | 2862 | union futex_key key = FUTEX_KEY_INIT; |
ccf9e6a8 | 2863 | struct futex_hash_bucket *hb; |
499f5aca | 2864 | struct futex_q *top_waiter; |
e4dc5b7a | 2865 | int ret; |
c87e2837 | 2866 | |
bc2eecd7 NP |
2867 | if (!IS_ENABLED(CONFIG_FUTEX_PI)) |
2868 | return -ENOSYS; | |
2869 | ||
c87e2837 IM |
2870 | retry: |
2871 | if (get_user(uval, uaddr)) | |
2872 | return -EFAULT; | |
2873 | /* | |
2874 | * We release only a lock we actually own: | |
2875 | */ | |
c0c9ed15 | 2876 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 2877 | return -EPERM; |
c87e2837 | 2878 | |
9ea71503 | 2879 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE); |
ccf9e6a8 TG |
2880 | if (ret) |
2881 | return ret; | |
c87e2837 IM |
2882 | |
2883 | hb = hash_futex(&key); | |
2884 | spin_lock(&hb->lock); | |
2885 | ||
c87e2837 | 2886 | /* |
ccf9e6a8 TG |
2887 | * Check waiters first. We do not trust user space values at |
2888 | * all and we at least want to know if user space fiddled | |
2889 | * with the futex value instead of blindly unlocking. | |
c87e2837 | 2890 | */ |
499f5aca PZ |
2891 | top_waiter = futex_top_waiter(hb, &key); |
2892 | if (top_waiter) { | |
16ffa12d PZ |
2893 | struct futex_pi_state *pi_state = top_waiter->pi_state; |
2894 | ||
2895 | ret = -EINVAL; | |
2896 | if (!pi_state) | |
2897 | goto out_unlock; | |
2898 | ||
2899 | /* | |
2900 | * If current does not own the pi_state then the futex is | |
2901 | * inconsistent and user space fiddled with the futex value. | |
2902 | */ | |
2903 | if (pi_state->owner != current) | |
2904 | goto out_unlock; | |
2905 | ||
bebe5b51 | 2906 | get_pi_state(pi_state); |
802ab58d | 2907 | /* |
bebe5b51 PZ |
2908 | * By taking wait_lock while still holding hb->lock, we ensure |
2909 | * there is no point where we hold neither; and therefore | |
2910 | * wake_futex_pi() must observe a state consistent with what we | |
2911 | * observed. | |
16ffa12d | 2912 | */ |
bebe5b51 | 2913 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
16ffa12d PZ |
2914 | spin_unlock(&hb->lock); |
2915 | ||
c74aef2d | 2916 | /* drops pi_state->pi_mutex.wait_lock */ |
16ffa12d PZ |
2917 | ret = wake_futex_pi(uaddr, uval, pi_state); |
2918 | ||
2919 | put_pi_state(pi_state); | |
2920 | ||
2921 | /* | |
2922 | * Success, we're done! No tricky corner cases. | |
802ab58d SAS |
2923 | */ |
2924 | if (!ret) | |
2925 | goto out_putkey; | |
c87e2837 | 2926 | /* |
ccf9e6a8 TG |
2927 | * The atomic access to the futex value generated a |
2928 | * pagefault, so retry the user-access and the wakeup: | |
c87e2837 IM |
2929 | */ |
2930 | if (ret == -EFAULT) | |
2931 | goto pi_faulted; | |
89e9e66b SAS |
2932 | /* |
2933 | * A unconditional UNLOCK_PI op raced against a waiter | |
2934 | * setting the FUTEX_WAITERS bit. Try again. | |
2935 | */ | |
2936 | if (ret == -EAGAIN) { | |
89e9e66b SAS |
2937 | put_futex_key(&key); |
2938 | goto retry; | |
2939 | } | |
802ab58d SAS |
2940 | /* |
2941 | * wake_futex_pi has detected invalid state. Tell user | |
2942 | * space. | |
2943 | */ | |
16ffa12d | 2944 | goto out_putkey; |
c87e2837 | 2945 | } |
ccf9e6a8 | 2946 | |
c87e2837 | 2947 | /* |
ccf9e6a8 TG |
2948 | * We have no kernel internal state, i.e. no waiters in the |
2949 | * kernel. Waiters which are about to queue themselves are stuck | |
2950 | * on hb->lock. So we can safely ignore them. We do neither | |
2951 | * preserve the WAITERS bit not the OWNER_DIED one. We are the | |
2952 | * owner. | |
c87e2837 | 2953 | */ |
16ffa12d PZ |
2954 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, 0)) { |
2955 | spin_unlock(&hb->lock); | |
13fbca4c | 2956 | goto pi_faulted; |
16ffa12d | 2957 | } |
c87e2837 | 2958 | |
ccf9e6a8 TG |
2959 | /* |
2960 | * If uval has changed, let user space handle it. | |
2961 | */ | |
2962 | ret = (curval == uval) ? 0 : -EAGAIN; | |
2963 | ||
c87e2837 IM |
2964 | out_unlock: |
2965 | spin_unlock(&hb->lock); | |
802ab58d | 2966 | out_putkey: |
ae791a2d | 2967 | put_futex_key(&key); |
c87e2837 IM |
2968 | return ret; |
2969 | ||
2970 | pi_faulted: | |
ae791a2d | 2971 | put_futex_key(&key); |
c87e2837 | 2972 | |
d0725992 | 2973 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2974 | if (!ret) |
c87e2837 IM |
2975 | goto retry; |
2976 | ||
1da177e4 LT |
2977 | return ret; |
2978 | } | |
2979 | ||
52400ba9 DH |
2980 | /** |
2981 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2982 | * @hb: the hash_bucket futex_q was original enqueued on | |
2983 | * @q: the futex_q woken while waiting to be requeued | |
2984 | * @key2: the futex_key of the requeue target futex | |
2985 | * @timeout: the timeout associated with the wait (NULL if none) | |
2986 | * | |
2987 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2988 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2989 | * the wakeup and return the appropriate error code to the caller. Must be | |
2990 | * called with the hb lock held. | |
2991 | * | |
6c23cbbd | 2992 | * Return: |
7b4ff1ad MCC |
2993 | * - 0 = no early wakeup detected; |
2994 | * - <0 = -ETIMEDOUT or -ERESTARTNOINTR | |
52400ba9 DH |
2995 | */ |
2996 | static inline | |
2997 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2998 | struct futex_q *q, union futex_key *key2, | |
2999 | struct hrtimer_sleeper *timeout) | |
3000 | { | |
3001 | int ret = 0; | |
3002 | ||
3003 | /* | |
3004 | * With the hb lock held, we avoid races while we process the wakeup. | |
3005 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
3006 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
3007 | * It can't be requeued from uaddr2 to something else since we don't | |
3008 | * support a PI aware source futex for requeue. | |
3009 | */ | |
3010 | if (!match_futex(&q->key, key2)) { | |
3011 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
3012 | /* | |
3013 | * We were woken prior to requeue by a timeout or a signal. | |
3014 | * Unqueue the futex_q and determine which it was. | |
3015 | */ | |
2e12978a | 3016 | plist_del(&q->list, &hb->chain); |
11d4616b | 3017 | hb_waiters_dec(hb); |
52400ba9 | 3018 | |
d58e6576 | 3019 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 3020 | ret = -EWOULDBLOCK; |
52400ba9 DH |
3021 | if (timeout && !timeout->task) |
3022 | ret = -ETIMEDOUT; | |
d58e6576 | 3023 | else if (signal_pending(current)) |
1c840c14 | 3024 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
3025 | } |
3026 | return ret; | |
3027 | } | |
3028 | ||
3029 | /** | |
3030 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 3031 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 3032 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
ab51fbab | 3033 | * the same type, no requeueing from private to shared, etc. |
52400ba9 DH |
3034 | * @val: the expected value of uaddr |
3035 | * @abs_time: absolute timeout | |
56ec1607 | 3036 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
3037 | * @uaddr2: the pi futex we will take prior to returning to user-space |
3038 | * | |
3039 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
6f7b0a2a DH |
3040 | * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake |
3041 | * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to | |
3042 | * userspace. This ensures the rt_mutex maintains an owner when it has waiters; | |
3043 | * without one, the pi logic would not know which task to boost/deboost, if | |
3044 | * there was a need to. | |
52400ba9 DH |
3045 | * |
3046 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
6c23cbbd | 3047 | * via the following-- |
52400ba9 | 3048 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() |
cc6db4e6 DH |
3049 | * 2) wakeup on uaddr2 after a requeue |
3050 | * 3) signal | |
3051 | * 4) timeout | |
52400ba9 | 3052 | * |
cc6db4e6 | 3053 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
3054 | * |
3055 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
3056 | * 5) successful lock | |
3057 | * 6) signal | |
3058 | * 7) timeout | |
3059 | * 8) other lock acquisition failure | |
3060 | * | |
cc6db4e6 | 3061 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
3062 | * |
3063 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
3064 | * | |
6c23cbbd | 3065 | * Return: |
7b4ff1ad MCC |
3066 | * - 0 - On success; |
3067 | * - <0 - On error | |
52400ba9 | 3068 | */ |
b41277dc | 3069 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 3070 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 3071 | u32 __user *uaddr2) |
52400ba9 DH |
3072 | { |
3073 | struct hrtimer_sleeper timeout, *to = NULL; | |
16ffa12d | 3074 | struct futex_pi_state *pi_state = NULL; |
52400ba9 | 3075 | struct rt_mutex_waiter rt_waiter; |
52400ba9 | 3076 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
3077 | union futex_key key2 = FUTEX_KEY_INIT; |
3078 | struct futex_q q = futex_q_init; | |
52400ba9 | 3079 | int res, ret; |
52400ba9 | 3080 | |
bc2eecd7 NP |
3081 | if (!IS_ENABLED(CONFIG_FUTEX_PI)) |
3082 | return -ENOSYS; | |
3083 | ||
6f7b0a2a DH |
3084 | if (uaddr == uaddr2) |
3085 | return -EINVAL; | |
3086 | ||
52400ba9 DH |
3087 | if (!bitset) |
3088 | return -EINVAL; | |
3089 | ||
3090 | if (abs_time) { | |
3091 | to = &timeout; | |
b41277dc DH |
3092 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
3093 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
3094 | HRTIMER_MODE_ABS); | |
52400ba9 DH |
3095 | hrtimer_init_sleeper(to, current); |
3096 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
3097 | current->timer_slack_ns); | |
3098 | } | |
3099 | ||
3100 | /* | |
3101 | * The waiter is allocated on our stack, manipulated by the requeue | |
3102 | * code while we sleep on uaddr. | |
3103 | */ | |
50809358 | 3104 | rt_mutex_init_waiter(&rt_waiter); |
52400ba9 | 3105 | |
9ea71503 | 3106 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
52400ba9 DH |
3107 | if (unlikely(ret != 0)) |
3108 | goto out; | |
3109 | ||
84bc4af5 DH |
3110 | q.bitset = bitset; |
3111 | q.rt_waiter = &rt_waiter; | |
3112 | q.requeue_pi_key = &key2; | |
3113 | ||
7ada876a DH |
3114 | /* |
3115 | * Prepare to wait on uaddr. On success, increments q.key (key1) ref | |
3116 | * count. | |
3117 | */ | |
b41277dc | 3118 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 TG |
3119 | if (ret) |
3120 | goto out_key2; | |
52400ba9 | 3121 | |
e9c243a5 TG |
3122 | /* |
3123 | * The check above which compares uaddrs is not sufficient for | |
3124 | * shared futexes. We need to compare the keys: | |
3125 | */ | |
3126 | if (match_futex(&q.key, &key2)) { | |
13c42c2f | 3127 | queue_unlock(hb); |
e9c243a5 TG |
3128 | ret = -EINVAL; |
3129 | goto out_put_keys; | |
3130 | } | |
3131 | ||
52400ba9 | 3132 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ |
f1a11e05 | 3133 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
3134 | |
3135 | spin_lock(&hb->lock); | |
3136 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
3137 | spin_unlock(&hb->lock); | |
3138 | if (ret) | |
3139 | goto out_put_keys; | |
3140 | ||
3141 | /* | |
3142 | * In order for us to be here, we know our q.key == key2, and since | |
3143 | * we took the hb->lock above, we also know that futex_requeue() has | |
3144 | * completed and we no longer have to concern ourselves with a wakeup | |
7ada876a DH |
3145 | * race with the atomic proxy lock acquisition by the requeue code. The |
3146 | * futex_requeue dropped our key1 reference and incremented our key2 | |
3147 | * reference count. | |
52400ba9 DH |
3148 | */ |
3149 | ||
3150 | /* Check if the requeue code acquired the second futex for us. */ | |
3151 | if (!q.rt_waiter) { | |
3152 | /* | |
3153 | * Got the lock. We might not be the anticipated owner if we | |
3154 | * did a lock-steal - fix up the PI-state in that case. | |
3155 | */ | |
3156 | if (q.pi_state && (q.pi_state->owner != current)) { | |
3157 | spin_lock(q.lock_ptr); | |
ae791a2d | 3158 | ret = fixup_pi_state_owner(uaddr2, &q, current); |
16ffa12d PZ |
3159 | if (ret && rt_mutex_owner(&q.pi_state->pi_mutex) == current) { |
3160 | pi_state = q.pi_state; | |
3161 | get_pi_state(pi_state); | |
3162 | } | |
fb75a428 TG |
3163 | /* |
3164 | * Drop the reference to the pi state which | |
3165 | * the requeue_pi() code acquired for us. | |
3166 | */ | |
29e9ee5d | 3167 | put_pi_state(q.pi_state); |
52400ba9 DH |
3168 | spin_unlock(q.lock_ptr); |
3169 | } | |
3170 | } else { | |
c236c8e9 PZ |
3171 | struct rt_mutex *pi_mutex; |
3172 | ||
52400ba9 DH |
3173 | /* |
3174 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
3175 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
3176 | * the pi_state. | |
3177 | */ | |
f27071cb | 3178 | WARN_ON(!q.pi_state); |
52400ba9 | 3179 | pi_mutex = &q.pi_state->pi_mutex; |
38d589f2 | 3180 | ret = rt_mutex_wait_proxy_lock(pi_mutex, to, &rt_waiter); |
52400ba9 DH |
3181 | |
3182 | spin_lock(q.lock_ptr); | |
38d589f2 PZ |
3183 | if (ret && !rt_mutex_cleanup_proxy_lock(pi_mutex, &rt_waiter)) |
3184 | ret = 0; | |
3185 | ||
3186 | debug_rt_mutex_free_waiter(&rt_waiter); | |
52400ba9 DH |
3187 | /* |
3188 | * Fixup the pi_state owner and possibly acquire the lock if we | |
3189 | * haven't already. | |
3190 | */ | |
ae791a2d | 3191 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 DH |
3192 | /* |
3193 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 3194 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
3195 | */ |
3196 | if (res) | |
3197 | ret = (res < 0) ? res : 0; | |
3198 | ||
c236c8e9 PZ |
3199 | /* |
3200 | * If fixup_pi_state_owner() faulted and was unable to handle | |
3201 | * the fault, unlock the rt_mutex and return the fault to | |
3202 | * userspace. | |
3203 | */ | |
16ffa12d PZ |
3204 | if (ret && rt_mutex_owner(&q.pi_state->pi_mutex) == current) { |
3205 | pi_state = q.pi_state; | |
3206 | get_pi_state(pi_state); | |
3207 | } | |
c236c8e9 | 3208 | |
52400ba9 DH |
3209 | /* Unqueue and drop the lock. */ |
3210 | unqueue_me_pi(&q); | |
3211 | } | |
3212 | ||
16ffa12d PZ |
3213 | if (pi_state) { |
3214 | rt_mutex_futex_unlock(&pi_state->pi_mutex); | |
3215 | put_pi_state(pi_state); | |
3216 | } | |
3217 | ||
c236c8e9 | 3218 | if (ret == -EINTR) { |
52400ba9 | 3219 | /* |
cc6db4e6 DH |
3220 | * We've already been requeued, but cannot restart by calling |
3221 | * futex_lock_pi() directly. We could restart this syscall, but | |
3222 | * it would detect that the user space "val" changed and return | |
3223 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
3224 | * -EWOULDBLOCK directly. | |
52400ba9 | 3225 | */ |
2070887f | 3226 | ret = -EWOULDBLOCK; |
52400ba9 DH |
3227 | } |
3228 | ||
3229 | out_put_keys: | |
ae791a2d | 3230 | put_futex_key(&q.key); |
c8b15a70 | 3231 | out_key2: |
ae791a2d | 3232 | put_futex_key(&key2); |
52400ba9 DH |
3233 | |
3234 | out: | |
3235 | if (to) { | |
3236 | hrtimer_cancel(&to->timer); | |
3237 | destroy_hrtimer_on_stack(&to->timer); | |
3238 | } | |
3239 | return ret; | |
3240 | } | |
3241 | ||
0771dfef IM |
3242 | /* |
3243 | * Support for robust futexes: the kernel cleans up held futexes at | |
3244 | * thread exit time. | |
3245 | * | |
3246 | * Implementation: user-space maintains a per-thread list of locks it | |
3247 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
3248 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 3249 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
3250 | * always manipulated with the lock held, so the list is private and |
3251 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
3252 | * field, to allow the kernel to clean up if the thread dies after | |
3253 | * acquiring the lock, but just before it could have added itself to | |
3254 | * the list. There can only be one such pending lock. | |
3255 | */ | |
3256 | ||
3257 | /** | |
d96ee56c DH |
3258 | * sys_set_robust_list() - Set the robust-futex list head of a task |
3259 | * @head: pointer to the list-head | |
3260 | * @len: length of the list-head, as userspace expects | |
0771dfef | 3261 | */ |
836f92ad HC |
3262 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
3263 | size_t, len) | |
0771dfef | 3264 | { |
a0c1e907 TG |
3265 | if (!futex_cmpxchg_enabled) |
3266 | return -ENOSYS; | |
0771dfef IM |
3267 | /* |
3268 | * The kernel knows only one size for now: | |
3269 | */ | |
3270 | if (unlikely(len != sizeof(*head))) | |
3271 | return -EINVAL; | |
3272 | ||
3273 | current->robust_list = head; | |
3274 | ||
3275 | return 0; | |
3276 | } | |
3277 | ||
3278 | /** | |
d96ee56c DH |
3279 | * sys_get_robust_list() - Get the robust-futex list head of a task |
3280 | * @pid: pid of the process [zero for current task] | |
3281 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
3282 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 3283 | */ |
836f92ad HC |
3284 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
3285 | struct robust_list_head __user * __user *, head_ptr, | |
3286 | size_t __user *, len_ptr) | |
0771dfef | 3287 | { |
ba46df98 | 3288 | struct robust_list_head __user *head; |
0771dfef | 3289 | unsigned long ret; |
bdbb776f | 3290 | struct task_struct *p; |
0771dfef | 3291 | |
a0c1e907 TG |
3292 | if (!futex_cmpxchg_enabled) |
3293 | return -ENOSYS; | |
3294 | ||
bdbb776f KC |
3295 | rcu_read_lock(); |
3296 | ||
3297 | ret = -ESRCH; | |
0771dfef | 3298 | if (!pid) |
bdbb776f | 3299 | p = current; |
0771dfef | 3300 | else { |
228ebcbe | 3301 | p = find_task_by_vpid(pid); |
0771dfef IM |
3302 | if (!p) |
3303 | goto err_unlock; | |
0771dfef IM |
3304 | } |
3305 | ||
bdbb776f | 3306 | ret = -EPERM; |
caaee623 | 3307 | if (!ptrace_may_access(p, PTRACE_MODE_READ_REALCREDS)) |
bdbb776f KC |
3308 | goto err_unlock; |
3309 | ||
3310 | head = p->robust_list; | |
3311 | rcu_read_unlock(); | |
3312 | ||
0771dfef IM |
3313 | if (put_user(sizeof(*head), len_ptr)) |
3314 | return -EFAULT; | |
3315 | return put_user(head, head_ptr); | |
3316 | ||
3317 | err_unlock: | |
aaa2a97e | 3318 | rcu_read_unlock(); |
0771dfef IM |
3319 | |
3320 | return ret; | |
3321 | } | |
3322 | ||
3323 | /* | |
3324 | * Process a futex-list entry, check whether it's owned by the | |
3325 | * dying task, and do notification if so: | |
3326 | */ | |
e3f2ddea | 3327 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 3328 | { |
7cfdaf38 | 3329 | u32 uval, uninitialized_var(nval), mval; |
0771dfef | 3330 | |
8f17d3a5 IM |
3331 | retry: |
3332 | if (get_user(uval, uaddr)) | |
0771dfef IM |
3333 | return -1; |
3334 | ||
b488893a | 3335 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
3336 | /* |
3337 | * Ok, this dying thread is truly holding a futex | |
3338 | * of interest. Set the OWNER_DIED bit atomically | |
3339 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
3340 | * set, wake up a waiter (if any). (We have to do a | |
3341 | * futex_wake() even if OWNER_DIED is already set - | |
3342 | * to handle the rare but possible case of recursive | |
3343 | * thread-death.) The rest of the cleanup is done in | |
3344 | * userspace. | |
3345 | */ | |
e3f2ddea | 3346 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
6e0aa9f8 TG |
3347 | /* |
3348 | * We are not holding a lock here, but we want to have | |
3349 | * the pagefault_disable/enable() protection because | |
3350 | * we want to handle the fault gracefully. If the | |
3351 | * access fails we try to fault in the futex with R/W | |
3352 | * verification via get_user_pages. get_user() above | |
3353 | * does not guarantee R/W access. If that fails we | |
3354 | * give up and leave the futex locked. | |
3355 | */ | |
3356 | if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) { | |
3357 | if (fault_in_user_writeable(uaddr)) | |
3358 | return -1; | |
3359 | goto retry; | |
3360 | } | |
c87e2837 | 3361 | if (nval != uval) |
8f17d3a5 | 3362 | goto retry; |
0771dfef | 3363 | |
e3f2ddea IM |
3364 | /* |
3365 | * Wake robust non-PI futexes here. The wakeup of | |
3366 | * PI futexes happens in exit_pi_state(): | |
3367 | */ | |
36cf3b5c | 3368 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 3369 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
3370 | } |
3371 | return 0; | |
3372 | } | |
3373 | ||
e3f2ddea IM |
3374 | /* |
3375 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
3376 | */ | |
3377 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 3378 | struct robust_list __user * __user *head, |
1dcc41bb | 3379 | unsigned int *pi) |
e3f2ddea IM |
3380 | { |
3381 | unsigned long uentry; | |
3382 | ||
ba46df98 | 3383 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
3384 | return -EFAULT; |
3385 | ||
ba46df98 | 3386 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
3387 | *pi = uentry & 1; |
3388 | ||
3389 | return 0; | |
3390 | } | |
3391 | ||
0771dfef IM |
3392 | /* |
3393 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
3394 | * and mark any locks found there dead, and notify any waiters. | |
3395 | * | |
3396 | * We silently return on any sign of list-walking problem. | |
3397 | */ | |
3398 | void exit_robust_list(struct task_struct *curr) | |
3399 | { | |
3400 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 3401 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 DH |
3402 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
3403 | unsigned int uninitialized_var(next_pi); | |
0771dfef | 3404 | unsigned long futex_offset; |
9f96cb1e | 3405 | int rc; |
0771dfef | 3406 | |
a0c1e907 TG |
3407 | if (!futex_cmpxchg_enabled) |
3408 | return; | |
3409 | ||
0771dfef IM |
3410 | /* |
3411 | * Fetch the list head (which was registered earlier, via | |
3412 | * sys_set_robust_list()): | |
3413 | */ | |
e3f2ddea | 3414 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
3415 | return; |
3416 | /* | |
3417 | * Fetch the relative futex offset: | |
3418 | */ | |
3419 | if (get_user(futex_offset, &head->futex_offset)) | |
3420 | return; | |
3421 | /* | |
3422 | * Fetch any possibly pending lock-add first, and handle it | |
3423 | * if it exists: | |
3424 | */ | |
e3f2ddea | 3425 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 3426 | return; |
e3f2ddea | 3427 | |
9f96cb1e | 3428 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 3429 | while (entry != &head->list) { |
9f96cb1e MS |
3430 | /* |
3431 | * Fetch the next entry in the list before calling | |
3432 | * handle_futex_death: | |
3433 | */ | |
3434 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
3435 | /* |
3436 | * A pending lock might already be on the list, so | |
c87e2837 | 3437 | * don't process it twice: |
0771dfef IM |
3438 | */ |
3439 | if (entry != pending) | |
ba46df98 | 3440 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 3441 | curr, pi)) |
0771dfef | 3442 | return; |
9f96cb1e | 3443 | if (rc) |
0771dfef | 3444 | return; |
9f96cb1e MS |
3445 | entry = next_entry; |
3446 | pi = next_pi; | |
0771dfef IM |
3447 | /* |
3448 | * Avoid excessively long or circular lists: | |
3449 | */ | |
3450 | if (!--limit) | |
3451 | break; | |
3452 | ||
3453 | cond_resched(); | |
3454 | } | |
9f96cb1e MS |
3455 | |
3456 | if (pending) | |
3457 | handle_futex_death((void __user *)pending + futex_offset, | |
3458 | curr, pip); | |
0771dfef IM |
3459 | } |
3460 | ||
c19384b5 | 3461 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 3462 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 3463 | { |
81b40539 | 3464 | int cmd = op & FUTEX_CMD_MASK; |
b41277dc | 3465 | unsigned int flags = 0; |
34f01cc1 ED |
3466 | |
3467 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 3468 | flags |= FLAGS_SHARED; |
1da177e4 | 3469 | |
b41277dc DH |
3470 | if (op & FUTEX_CLOCK_REALTIME) { |
3471 | flags |= FLAGS_CLOCKRT; | |
337f1304 DH |
3472 | if (cmd != FUTEX_WAIT && cmd != FUTEX_WAIT_BITSET && \ |
3473 | cmd != FUTEX_WAIT_REQUEUE_PI) | |
b41277dc DH |
3474 | return -ENOSYS; |
3475 | } | |
1da177e4 | 3476 | |
59263b51 TG |
3477 | switch (cmd) { |
3478 | case FUTEX_LOCK_PI: | |
3479 | case FUTEX_UNLOCK_PI: | |
3480 | case FUTEX_TRYLOCK_PI: | |
3481 | case FUTEX_WAIT_REQUEUE_PI: | |
3482 | case FUTEX_CMP_REQUEUE_PI: | |
3483 | if (!futex_cmpxchg_enabled) | |
3484 | return -ENOSYS; | |
3485 | } | |
3486 | ||
34f01cc1 | 3487 | switch (cmd) { |
1da177e4 | 3488 | case FUTEX_WAIT: |
cd689985 TG |
3489 | val3 = FUTEX_BITSET_MATCH_ANY; |
3490 | case FUTEX_WAIT_BITSET: | |
81b40539 | 3491 | return futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 | 3492 | case FUTEX_WAKE: |
cd689985 TG |
3493 | val3 = FUTEX_BITSET_MATCH_ANY; |
3494 | case FUTEX_WAKE_BITSET: | |
81b40539 | 3495 | return futex_wake(uaddr, flags, val, val3); |
1da177e4 | 3496 | case FUTEX_REQUEUE: |
81b40539 | 3497 | return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 | 3498 | case FUTEX_CMP_REQUEUE: |
81b40539 | 3499 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
4732efbe | 3500 | case FUTEX_WAKE_OP: |
81b40539 | 3501 | return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
c87e2837 | 3502 | case FUTEX_LOCK_PI: |
996636dd | 3503 | return futex_lock_pi(uaddr, flags, timeout, 0); |
c87e2837 | 3504 | case FUTEX_UNLOCK_PI: |
81b40539 | 3505 | return futex_unlock_pi(uaddr, flags); |
c87e2837 | 3506 | case FUTEX_TRYLOCK_PI: |
996636dd | 3507 | return futex_lock_pi(uaddr, flags, NULL, 1); |
52400ba9 DH |
3508 | case FUTEX_WAIT_REQUEUE_PI: |
3509 | val3 = FUTEX_BITSET_MATCH_ANY; | |
81b40539 TG |
3510 | return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
3511 | uaddr2); | |
52400ba9 | 3512 | case FUTEX_CMP_REQUEUE_PI: |
81b40539 | 3513 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
1da177e4 | 3514 | } |
81b40539 | 3515 | return -ENOSYS; |
1da177e4 LT |
3516 | } |
3517 | ||
3518 | ||
17da2bd9 HC |
3519 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
3520 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
3521 | u32, val3) | |
1da177e4 | 3522 | { |
c19384b5 PP |
3523 | struct timespec ts; |
3524 | ktime_t t, *tp = NULL; | |
e2970f2f | 3525 | u32 val2 = 0; |
34f01cc1 | 3526 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 3527 | |
cd689985 | 3528 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
3529 | cmd == FUTEX_WAIT_BITSET || |
3530 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
ab51fbab DB |
3531 | if (unlikely(should_fail_futex(!(op & FUTEX_PRIVATE_FLAG)))) |
3532 | return -EFAULT; | |
c19384b5 | 3533 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 3534 | return -EFAULT; |
c19384b5 | 3535 | if (!timespec_valid(&ts)) |
9741ef96 | 3536 | return -EINVAL; |
c19384b5 PP |
3537 | |
3538 | t = timespec_to_ktime(ts); | |
34f01cc1 | 3539 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 3540 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 3541 | tp = &t; |
1da177e4 LT |
3542 | } |
3543 | /* | |
52400ba9 | 3544 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 3545 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 3546 | */ |
f54f0986 | 3547 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 3548 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 3549 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 3550 | |
c19384b5 | 3551 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
3552 | } |
3553 | ||
03b8c7b6 | 3554 | static void __init futex_detect_cmpxchg(void) |
1da177e4 | 3555 | { |
03b8c7b6 | 3556 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 3557 | u32 curval; |
03b8c7b6 HC |
3558 | |
3559 | /* | |
3560 | * This will fail and we want it. Some arch implementations do | |
3561 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
3562 | * functionality. We want to know that before we call in any | |
3563 | * of the complex code paths. Also we want to prevent | |
3564 | * registration of robust lists in that case. NULL is | |
3565 | * guaranteed to fault and we get -EFAULT on functional | |
3566 | * implementation, the non-functional ones will return | |
3567 | * -ENOSYS. | |
3568 | */ | |
3569 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) | |
3570 | futex_cmpxchg_enabled = 1; | |
3571 | #endif | |
3572 | } | |
3573 | ||
3574 | static int __init futex_init(void) | |
3575 | { | |
63b1a816 | 3576 | unsigned int futex_shift; |
a52b89eb DB |
3577 | unsigned long i; |
3578 | ||
3579 | #if CONFIG_BASE_SMALL | |
3580 | futex_hashsize = 16; | |
3581 | #else | |
3582 | futex_hashsize = roundup_pow_of_two(256 * num_possible_cpus()); | |
3583 | #endif | |
3584 | ||
3585 | futex_queues = alloc_large_system_hash("futex", sizeof(*futex_queues), | |
3586 | futex_hashsize, 0, | |
3587 | futex_hashsize < 256 ? HASH_SMALL : 0, | |
63b1a816 HC |
3588 | &futex_shift, NULL, |
3589 | futex_hashsize, futex_hashsize); | |
3590 | futex_hashsize = 1UL << futex_shift; | |
03b8c7b6 HC |
3591 | |
3592 | futex_detect_cmpxchg(); | |
a0c1e907 | 3593 | |
a52b89eb | 3594 | for (i = 0; i < futex_hashsize; i++) { |
11d4616b | 3595 | atomic_set(&futex_queues[i].waiters, 0); |
732375c6 | 3596 | plist_head_init(&futex_queues[i].chain); |
3e4ab747 TG |
3597 | spin_lock_init(&futex_queues[i].lock); |
3598 | } | |
3599 | ||
1da177e4 LT |
3600 | return 0; |
3601 | } | |
25f71d1c | 3602 | core_initcall(futex_init); |