]>
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 | */ | |
f6d7b233 | 47 | #include <linux/compat.h> |
1da177e4 LT |
48 | #include <linux/slab.h> |
49 | #include <linux/poll.h> | |
50 | #include <linux/fs.h> | |
51 | #include <linux/file.h> | |
52 | #include <linux/jhash.h> | |
53 | #include <linux/init.h> | |
54 | #include <linux/futex.h> | |
55 | #include <linux/mount.h> | |
56 | #include <linux/pagemap.h> | |
57 | #include <linux/syscalls.h> | |
7ed20e1a | 58 | #include <linux/signal.h> |
9984de1a | 59 | #include <linux/export.h> |
fd5eea42 | 60 | #include <linux/magic.h> |
b488893a PE |
61 | #include <linux/pid.h> |
62 | #include <linux/nsproxy.h> | |
bdbb776f | 63 | #include <linux/ptrace.h> |
8bd75c77 | 64 | #include <linux/sched/rt.h> |
84f001e1 | 65 | #include <linux/sched/wake_q.h> |
6e84f315 | 66 | #include <linux/sched/mm.h> |
13d60f4b | 67 | #include <linux/hugetlb.h> |
88c8004f | 68 | #include <linux/freezer.h> |
a52b89eb | 69 | #include <linux/bootmem.h> |
ab51fbab | 70 | #include <linux/fault-inject.h> |
b488893a | 71 | |
4732efbe | 72 | #include <asm/futex.h> |
1da177e4 | 73 | |
1696a8be | 74 | #include "locking/rtmutex_common.h" |
c87e2837 | 75 | |
99b60ce6 | 76 | /* |
d7e8af1a DB |
77 | * READ this before attempting to hack on futexes! |
78 | * | |
79 | * Basic futex operation and ordering guarantees | |
80 | * ============================================= | |
99b60ce6 TG |
81 | * |
82 | * The waiter reads the futex value in user space and calls | |
83 | * futex_wait(). This function computes the hash bucket and acquires | |
84 | * the hash bucket lock. After that it reads the futex user space value | |
b0c29f79 DB |
85 | * again and verifies that the data has not changed. If it has not changed |
86 | * it enqueues itself into the hash bucket, releases the hash bucket lock | |
87 | * and schedules. | |
99b60ce6 TG |
88 | * |
89 | * The waker side modifies the user space value of the futex and calls | |
b0c29f79 DB |
90 | * futex_wake(). This function computes the hash bucket and acquires the |
91 | * hash bucket lock. Then it looks for waiters on that futex in the hash | |
92 | * bucket and wakes them. | |
99b60ce6 | 93 | * |
b0c29f79 DB |
94 | * In futex wake up scenarios where no tasks are blocked on a futex, taking |
95 | * the hb spinlock can be avoided and simply return. In order for this | |
96 | * optimization to work, ordering guarantees must exist so that the waiter | |
97 | * being added to the list is acknowledged when the list is concurrently being | |
98 | * checked by the waker, avoiding scenarios like the following: | |
99b60ce6 TG |
99 | * |
100 | * CPU 0 CPU 1 | |
101 | * val = *futex; | |
102 | * sys_futex(WAIT, futex, val); | |
103 | * futex_wait(futex, val); | |
104 | * uval = *futex; | |
105 | * *futex = newval; | |
106 | * sys_futex(WAKE, futex); | |
107 | * futex_wake(futex); | |
108 | * if (queue_empty()) | |
109 | * return; | |
110 | * if (uval == val) | |
111 | * lock(hash_bucket(futex)); | |
112 | * queue(); | |
113 | * unlock(hash_bucket(futex)); | |
114 | * schedule(); | |
115 | * | |
116 | * This would cause the waiter on CPU 0 to wait forever because it | |
117 | * missed the transition of the user space value from val to newval | |
118 | * and the waker did not find the waiter in the hash bucket queue. | |
99b60ce6 | 119 | * |
b0c29f79 DB |
120 | * The correct serialization ensures that a waiter either observes |
121 | * the changed user space value before blocking or is woken by a | |
122 | * concurrent waker: | |
123 | * | |
124 | * CPU 0 CPU 1 | |
99b60ce6 TG |
125 | * val = *futex; |
126 | * sys_futex(WAIT, futex, val); | |
127 | * futex_wait(futex, val); | |
b0c29f79 | 128 | * |
d7e8af1a | 129 | * waiters++; (a) |
8ad7b378 DB |
130 | * smp_mb(); (A) <-- paired with -. |
131 | * | | |
132 | * lock(hash_bucket(futex)); | | |
133 | * | | |
134 | * uval = *futex; | | |
135 | * | *futex = newval; | |
136 | * | sys_futex(WAKE, futex); | |
137 | * | futex_wake(futex); | |
138 | * | | |
139 | * `--------> smp_mb(); (B) | |
99b60ce6 | 140 | * if (uval == val) |
b0c29f79 | 141 | * queue(); |
99b60ce6 | 142 | * unlock(hash_bucket(futex)); |
b0c29f79 DB |
143 | * schedule(); if (waiters) |
144 | * lock(hash_bucket(futex)); | |
d7e8af1a DB |
145 | * else wake_waiters(futex); |
146 | * waiters--; (b) unlock(hash_bucket(futex)); | |
b0c29f79 | 147 | * |
d7e8af1a DB |
148 | * Where (A) orders the waiters increment and the futex value read through |
149 | * atomic operations (see hb_waiters_inc) and where (B) orders the write | |
993b2ff2 DB |
150 | * to futex and the waiters read -- this is done by the barriers for both |
151 | * shared and private futexes in get_futex_key_refs(). | |
b0c29f79 DB |
152 | * |
153 | * This yields the following case (where X:=waiters, Y:=futex): | |
154 | * | |
155 | * X = Y = 0 | |
156 | * | |
157 | * w[X]=1 w[Y]=1 | |
158 | * MB MB | |
159 | * r[Y]=y r[X]=x | |
160 | * | |
161 | * Which guarantees that x==0 && y==0 is impossible; which translates back into | |
162 | * the guarantee that we cannot both miss the futex variable change and the | |
163 | * enqueue. | |
d7e8af1a DB |
164 | * |
165 | * Note that a new waiter is accounted for in (a) even when it is possible that | |
166 | * the wait call can return error, in which case we backtrack from it in (b). | |
167 | * Refer to the comment in queue_lock(). | |
168 | * | |
169 | * Similarly, in order to account for waiters being requeued on another | |
170 | * address we always increment the waiters for the destination bucket before | |
171 | * acquiring the lock. It then decrements them again after releasing it - | |
172 | * the code that actually moves the futex(es) between hash buckets (requeue_futex) | |
173 | * will do the additional required waiter count housekeeping. This is done for | |
174 | * double_lock_hb() and double_unlock_hb(), respectively. | |
99b60ce6 TG |
175 | */ |
176 | ||
f6d7b233 AB |
177 | #ifdef CONFIG_HAVE_FUTEX_CMPXCHG |
178 | #define futex_cmpxchg_enabled 1 | |
179 | #else | |
180 | static int __read_mostly futex_cmpxchg_enabled; | |
03b8c7b6 | 181 | #endif |
a0c1e907 | 182 | |
b41277dc DH |
183 | /* |
184 | * Futex flags used to encode options to functions and preserve them across | |
185 | * restarts. | |
186 | */ | |
784bdf3b TG |
187 | #ifdef CONFIG_MMU |
188 | # define FLAGS_SHARED 0x01 | |
189 | #else | |
190 | /* | |
191 | * NOMMU does not have per process address space. Let the compiler optimize | |
192 | * code away. | |
193 | */ | |
194 | # define FLAGS_SHARED 0x00 | |
195 | #endif | |
b41277dc DH |
196 | #define FLAGS_CLOCKRT 0x02 |
197 | #define FLAGS_HAS_TIMEOUT 0x04 | |
198 | ||
c87e2837 IM |
199 | /* |
200 | * Priority Inheritance state: | |
201 | */ | |
202 | struct futex_pi_state { | |
203 | /* | |
204 | * list of 'owned' pi_state instances - these have to be | |
205 | * cleaned up in do_exit() if the task exits prematurely: | |
206 | */ | |
207 | struct list_head list; | |
208 | ||
209 | /* | |
210 | * The PI object: | |
211 | */ | |
212 | struct rt_mutex pi_mutex; | |
213 | ||
214 | struct task_struct *owner; | |
215 | atomic_t refcount; | |
216 | ||
217 | union futex_key key; | |
3859a271 | 218 | } __randomize_layout; |
c87e2837 | 219 | |
d8d88fbb DH |
220 | /** |
221 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 222 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
223 | * @task: the task waiting on the futex |
224 | * @lock_ptr: the hash bucket lock | |
225 | * @key: the key the futex is hashed on | |
226 | * @pi_state: optional priority inheritance state | |
227 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
228 | * @requeue_pi_key: the requeue_pi target futex key | |
229 | * @bitset: bitset for the optional bitmasked wakeup | |
230 | * | |
ac6424b9 | 231 | * We use this hashed waitqueue, instead of a normal wait_queue_entry_t, so |
1da177e4 LT |
232 | * we can wake only the relevant ones (hashed queues may be shared). |
233 | * | |
234 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 235 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 236 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
237 | * the second. |
238 | * | |
239 | * PI futexes are typically woken before they are removed from the hash list via | |
240 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
241 | */ |
242 | struct futex_q { | |
ec92d082 | 243 | struct plist_node list; |
1da177e4 | 244 | |
d8d88fbb | 245 | struct task_struct *task; |
1da177e4 | 246 | spinlock_t *lock_ptr; |
1da177e4 | 247 | union futex_key key; |
c87e2837 | 248 | struct futex_pi_state *pi_state; |
52400ba9 | 249 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 250 | union futex_key *requeue_pi_key; |
cd689985 | 251 | u32 bitset; |
3859a271 | 252 | } __randomize_layout; |
1da177e4 | 253 | |
5bdb05f9 DH |
254 | static const struct futex_q futex_q_init = { |
255 | /* list gets initialized in queue_me()*/ | |
256 | .key = FUTEX_KEY_INIT, | |
257 | .bitset = FUTEX_BITSET_MATCH_ANY | |
258 | }; | |
259 | ||
1da177e4 | 260 | /* |
b2d0994b DH |
261 | * Hash buckets are shared by all the futex_keys that hash to the same |
262 | * location. Each key may have multiple futex_q structures, one for each task | |
263 | * waiting on a futex. | |
1da177e4 LT |
264 | */ |
265 | struct futex_hash_bucket { | |
11d4616b | 266 | atomic_t waiters; |
ec92d082 PP |
267 | spinlock_t lock; |
268 | struct plist_head chain; | |
a52b89eb | 269 | } ____cacheline_aligned_in_smp; |
1da177e4 | 270 | |
ac742d37 RV |
271 | /* |
272 | * The base of the bucket array and its size are always used together | |
273 | * (after initialization only in hash_futex()), so ensure that they | |
274 | * reside in the same cacheline. | |
275 | */ | |
276 | static struct { | |
277 | struct futex_hash_bucket *queues; | |
278 | unsigned long hashsize; | |
279 | } __futex_data __read_mostly __aligned(2*sizeof(long)); | |
280 | #define futex_queues (__futex_data.queues) | |
281 | #define futex_hashsize (__futex_data.hashsize) | |
a52b89eb | 282 | |
1da177e4 | 283 | |
ab51fbab DB |
284 | /* |
285 | * Fault injections for futexes. | |
286 | */ | |
287 | #ifdef CONFIG_FAIL_FUTEX | |
288 | ||
289 | static struct { | |
290 | struct fault_attr attr; | |
291 | ||
621a5f7a | 292 | bool ignore_private; |
ab51fbab DB |
293 | } fail_futex = { |
294 | .attr = FAULT_ATTR_INITIALIZER, | |
621a5f7a | 295 | .ignore_private = false, |
ab51fbab DB |
296 | }; |
297 | ||
298 | static int __init setup_fail_futex(char *str) | |
299 | { | |
300 | return setup_fault_attr(&fail_futex.attr, str); | |
301 | } | |
302 | __setup("fail_futex=", setup_fail_futex); | |
303 | ||
5d285a7f | 304 | static bool should_fail_futex(bool fshared) |
ab51fbab DB |
305 | { |
306 | if (fail_futex.ignore_private && !fshared) | |
307 | return false; | |
308 | ||
309 | return should_fail(&fail_futex.attr, 1); | |
310 | } | |
311 | ||
312 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
313 | ||
314 | static int __init fail_futex_debugfs(void) | |
315 | { | |
316 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
317 | struct dentry *dir; | |
318 | ||
319 | dir = fault_create_debugfs_attr("fail_futex", NULL, | |
320 | &fail_futex.attr); | |
321 | if (IS_ERR(dir)) | |
322 | return PTR_ERR(dir); | |
323 | ||
324 | if (!debugfs_create_bool("ignore-private", mode, dir, | |
325 | &fail_futex.ignore_private)) { | |
326 | debugfs_remove_recursive(dir); | |
327 | return -ENOMEM; | |
328 | } | |
329 | ||
330 | return 0; | |
331 | } | |
332 | ||
333 | late_initcall(fail_futex_debugfs); | |
334 | ||
335 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
336 | ||
337 | #else | |
338 | static inline bool should_fail_futex(bool fshared) | |
339 | { | |
340 | return false; | |
341 | } | |
342 | #endif /* CONFIG_FAIL_FUTEX */ | |
343 | ||
e891c8f7 TG |
344 | #ifdef CONFIG_COMPAT |
345 | static void compat_exit_robust_list(struct task_struct *curr); | |
346 | #else | |
347 | static inline void compat_exit_robust_list(struct task_struct *curr) { } | |
348 | #endif | |
349 | ||
b0c29f79 DB |
350 | static inline void futex_get_mm(union futex_key *key) |
351 | { | |
f1f10076 | 352 | mmgrab(key->private.mm); |
b0c29f79 DB |
353 | /* |
354 | * Ensure futex_get_mm() implies a full barrier such that | |
355 | * get_futex_key() implies a full barrier. This is relied upon | |
8ad7b378 | 356 | * as smp_mb(); (B), see the ordering comment above. |
b0c29f79 | 357 | */ |
4e857c58 | 358 | smp_mb__after_atomic(); |
b0c29f79 DB |
359 | } |
360 | ||
11d4616b LT |
361 | /* |
362 | * Reflects a new waiter being added to the waitqueue. | |
363 | */ | |
364 | static inline void hb_waiters_inc(struct futex_hash_bucket *hb) | |
b0c29f79 DB |
365 | { |
366 | #ifdef CONFIG_SMP | |
11d4616b | 367 | atomic_inc(&hb->waiters); |
b0c29f79 | 368 | /* |
11d4616b | 369 | * Full barrier (A), see the ordering comment above. |
b0c29f79 | 370 | */ |
4e857c58 | 371 | smp_mb__after_atomic(); |
11d4616b LT |
372 | #endif |
373 | } | |
374 | ||
375 | /* | |
376 | * Reflects a waiter being removed from the waitqueue by wakeup | |
377 | * paths. | |
378 | */ | |
379 | static inline void hb_waiters_dec(struct futex_hash_bucket *hb) | |
380 | { | |
381 | #ifdef CONFIG_SMP | |
382 | atomic_dec(&hb->waiters); | |
383 | #endif | |
384 | } | |
b0c29f79 | 385 | |
11d4616b LT |
386 | static inline int hb_waiters_pending(struct futex_hash_bucket *hb) |
387 | { | |
388 | #ifdef CONFIG_SMP | |
389 | return atomic_read(&hb->waiters); | |
b0c29f79 | 390 | #else |
11d4616b | 391 | return 1; |
b0c29f79 DB |
392 | #endif |
393 | } | |
394 | ||
e8b61b3f TG |
395 | /** |
396 | * hash_futex - Return the hash bucket in the global hash | |
397 | * @key: Pointer to the futex key for which the hash is calculated | |
398 | * | |
399 | * We hash on the keys returned from get_futex_key (see below) and return the | |
400 | * corresponding hash bucket in the global hash. | |
1da177e4 LT |
401 | */ |
402 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
403 | { | |
404 | u32 hash = jhash2((u32*)&key->both.word, | |
405 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
406 | key->both.offset); | |
a52b89eb | 407 | return &futex_queues[hash & (futex_hashsize - 1)]; |
1da177e4 LT |
408 | } |
409 | ||
e8b61b3f TG |
410 | |
411 | /** | |
412 | * match_futex - Check whether two futex keys are equal | |
413 | * @key1: Pointer to key1 | |
414 | * @key2: Pointer to key2 | |
415 | * | |
1da177e4 LT |
416 | * Return 1 if two futex_keys are equal, 0 otherwise. |
417 | */ | |
418 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
419 | { | |
2bc87203 DH |
420 | return (key1 && key2 |
421 | && key1->both.word == key2->both.word | |
1da177e4 LT |
422 | && key1->both.ptr == key2->both.ptr |
423 | && key1->both.offset == key2->both.offset); | |
424 | } | |
425 | ||
38d47c1b PZ |
426 | /* |
427 | * Take a reference to the resource addressed by a key. | |
428 | * Can be called while holding spinlocks. | |
429 | * | |
430 | */ | |
431 | static void get_futex_key_refs(union futex_key *key) | |
432 | { | |
433 | if (!key->both.ptr) | |
434 | return; | |
435 | ||
784bdf3b TG |
436 | /* |
437 | * On MMU less systems futexes are always "private" as there is no per | |
438 | * process address space. We need the smp wmb nevertheless - yes, | |
439 | * arch/blackfin has MMU less SMP ... | |
440 | */ | |
441 | if (!IS_ENABLED(CONFIG_MMU)) { | |
442 | smp_mb(); /* explicit smp_mb(); (B) */ | |
443 | return; | |
444 | } | |
445 | ||
38d47c1b PZ |
446 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { |
447 | case FUT_OFF_INODE: | |
8ad7b378 | 448 | ihold(key->shared.inode); /* implies smp_mb(); (B) */ |
38d47c1b PZ |
449 | break; |
450 | case FUT_OFF_MMSHARED: | |
8ad7b378 | 451 | futex_get_mm(key); /* implies smp_mb(); (B) */ |
38d47c1b | 452 | break; |
76835b0e | 453 | default: |
993b2ff2 DB |
454 | /* |
455 | * Private futexes do not hold reference on an inode or | |
456 | * mm, therefore the only purpose of calling get_futex_key_refs | |
457 | * is because we need the barrier for the lockless waiter check. | |
458 | */ | |
8ad7b378 | 459 | smp_mb(); /* explicit smp_mb(); (B) */ |
38d47c1b PZ |
460 | } |
461 | } | |
462 | ||
463 | /* | |
464 | * Drop a reference to the resource addressed by a key. | |
993b2ff2 DB |
465 | * The hash bucket spinlock must not be held. This is |
466 | * a no-op for private futexes, see comment in the get | |
467 | * counterpart. | |
38d47c1b PZ |
468 | */ |
469 | static void drop_futex_key_refs(union futex_key *key) | |
470 | { | |
90621c40 DH |
471 | if (!key->both.ptr) { |
472 | /* If we're here then we tried to put a key we failed to get */ | |
473 | WARN_ON_ONCE(1); | |
38d47c1b | 474 | return; |
90621c40 | 475 | } |
38d47c1b | 476 | |
784bdf3b TG |
477 | if (!IS_ENABLED(CONFIG_MMU)) |
478 | return; | |
479 | ||
38d47c1b PZ |
480 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { |
481 | case FUT_OFF_INODE: | |
482 | iput(key->shared.inode); | |
483 | break; | |
484 | case FUT_OFF_MMSHARED: | |
485 | mmdrop(key->private.mm); | |
486 | break; | |
487 | } | |
488 | } | |
489 | ||
34f01cc1 | 490 | /** |
d96ee56c DH |
491 | * get_futex_key() - Get parameters which are the keys for a futex |
492 | * @uaddr: virtual address of the futex | |
493 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED | |
494 | * @key: address where result is stored. | |
9ea71503 SB |
495 | * @rw: mapping needs to be read/write (values: VERIFY_READ, |
496 | * VERIFY_WRITE) | |
34f01cc1 | 497 | * |
6c23cbbd RD |
498 | * Return: a negative error code or 0 |
499 | * | |
7b4ff1ad | 500 | * The key words are stored in @key on success. |
1da177e4 | 501 | * |
6131ffaa | 502 | * For shared mappings, it's (page->index, file_inode(vma->vm_file), |
1da177e4 LT |
503 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
504 | * We can usually work out the index without swapping in the page. | |
505 | * | |
b2d0994b | 506 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 507 | */ |
64d1304a | 508 | static int |
9ea71503 | 509 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) |
1da177e4 | 510 | { |
e2970f2f | 511 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 512 | struct mm_struct *mm = current->mm; |
077fa7ae | 513 | struct page *page, *tail; |
14d27abd | 514 | struct address_space *mapping; |
9ea71503 | 515 | int err, ro = 0; |
1da177e4 LT |
516 | |
517 | /* | |
518 | * The futex address must be "naturally" aligned. | |
519 | */ | |
e2970f2f | 520 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 521 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 522 | return -EINVAL; |
e2970f2f | 523 | address -= key->both.offset; |
1da177e4 | 524 | |
5cdec2d8 LT |
525 | if (unlikely(!access_ok(rw, uaddr, sizeof(u32)))) |
526 | return -EFAULT; | |
527 | ||
ab51fbab DB |
528 | if (unlikely(should_fail_futex(fshared))) |
529 | return -EFAULT; | |
530 | ||
34f01cc1 ED |
531 | /* |
532 | * PROCESS_PRIVATE futexes are fast. | |
533 | * As the mm cannot disappear under us and the 'key' only needs | |
534 | * virtual address, we dont even have to find the underlying vma. | |
535 | * Note : We do have to check 'uaddr' is a valid user address, | |
536 | * but access_ok() should be faster than find_vma() | |
537 | */ | |
538 | if (!fshared) { | |
34f01cc1 ED |
539 | key->private.mm = mm; |
540 | key->private.address = address; | |
8ad7b378 | 541 | get_futex_key_refs(key); /* implies smp_mb(); (B) */ |
34f01cc1 ED |
542 | return 0; |
543 | } | |
1da177e4 | 544 | |
38d47c1b | 545 | again: |
ab51fbab DB |
546 | /* Ignore any VERIFY_READ mapping (futex common case) */ |
547 | if (unlikely(should_fail_futex(fshared))) | |
548 | return -EFAULT; | |
549 | ||
7485d0d3 | 550 | err = get_user_pages_fast(address, 1, 1, &page); |
9ea71503 SB |
551 | /* |
552 | * If write access is not required (eg. FUTEX_WAIT), try | |
553 | * and get read-only access. | |
554 | */ | |
555 | if (err == -EFAULT && rw == VERIFY_READ) { | |
556 | err = get_user_pages_fast(address, 1, 0, &page); | |
557 | ro = 1; | |
558 | } | |
38d47c1b PZ |
559 | if (err < 0) |
560 | return err; | |
9ea71503 SB |
561 | else |
562 | err = 0; | |
38d47c1b | 563 | |
65d8fc77 MG |
564 | /* |
565 | * The treatment of mapping from this point on is critical. The page | |
566 | * lock protects many things but in this context the page lock | |
567 | * stabilizes mapping, prevents inode freeing in the shared | |
568 | * file-backed region case and guards against movement to swap cache. | |
569 | * | |
570 | * Strictly speaking the page lock is not needed in all cases being | |
571 | * considered here and page lock forces unnecessarily serialization | |
572 | * From this point on, mapping will be re-verified if necessary and | |
573 | * page lock will be acquired only if it is unavoidable | |
077fa7ae MG |
574 | * |
575 | * Mapping checks require the head page for any compound page so the | |
576 | * head page and mapping is looked up now. For anonymous pages, it | |
577 | * does not matter if the page splits in the future as the key is | |
578 | * based on the address. For filesystem-backed pages, the tail is | |
579 | * required as the index of the page determines the key. For | |
580 | * base pages, there is no tail page and tail == page. | |
65d8fc77 | 581 | */ |
077fa7ae | 582 | tail = page; |
65d8fc77 MG |
583 | page = compound_head(page); |
584 | mapping = READ_ONCE(page->mapping); | |
585 | ||
e6780f72 | 586 | /* |
14d27abd | 587 | * If page->mapping is NULL, then it cannot be a PageAnon |
e6780f72 HD |
588 | * page; but it might be the ZERO_PAGE or in the gate area or |
589 | * in a special mapping (all cases which we are happy to fail); | |
590 | * or it may have been a good file page when get_user_pages_fast | |
591 | * found it, but truncated or holepunched or subjected to | |
592 | * invalidate_complete_page2 before we got the page lock (also | |
593 | * cases which we are happy to fail). And we hold a reference, | |
594 | * so refcount care in invalidate_complete_page's remove_mapping | |
595 | * prevents drop_caches from setting mapping to NULL beneath us. | |
596 | * | |
597 | * The case we do have to guard against is when memory pressure made | |
598 | * shmem_writepage move it from filecache to swapcache beneath us: | |
14d27abd | 599 | * an unlikely race, but we do need to retry for page->mapping. |
e6780f72 | 600 | */ |
65d8fc77 MG |
601 | if (unlikely(!mapping)) { |
602 | int shmem_swizzled; | |
603 | ||
604 | /* | |
605 | * Page lock is required to identify which special case above | |
606 | * applies. If this is really a shmem page then the page lock | |
607 | * will prevent unexpected transitions. | |
608 | */ | |
609 | lock_page(page); | |
610 | shmem_swizzled = PageSwapCache(page) || page->mapping; | |
14d27abd KS |
611 | unlock_page(page); |
612 | put_page(page); | |
65d8fc77 | 613 | |
e6780f72 HD |
614 | if (shmem_swizzled) |
615 | goto again; | |
65d8fc77 | 616 | |
e6780f72 | 617 | return -EFAULT; |
38d47c1b | 618 | } |
1da177e4 LT |
619 | |
620 | /* | |
621 | * Private mappings are handled in a simple way. | |
622 | * | |
65d8fc77 MG |
623 | * If the futex key is stored on an anonymous page, then the associated |
624 | * object is the mm which is implicitly pinned by the calling process. | |
625 | * | |
1da177e4 LT |
626 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if |
627 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 628 | * the object not the particular process. |
1da177e4 | 629 | */ |
14d27abd | 630 | if (PageAnon(page)) { |
9ea71503 SB |
631 | /* |
632 | * A RO anonymous page will never change and thus doesn't make | |
633 | * sense for futex operations. | |
634 | */ | |
ab51fbab | 635 | if (unlikely(should_fail_futex(fshared)) || ro) { |
9ea71503 SB |
636 | err = -EFAULT; |
637 | goto out; | |
638 | } | |
639 | ||
38d47c1b | 640 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 641 | key->private.mm = mm; |
e2970f2f | 642 | key->private.address = address; |
65d8fc77 MG |
643 | |
644 | get_futex_key_refs(key); /* implies smp_mb(); (B) */ | |
645 | ||
38d47c1b | 646 | } else { |
65d8fc77 MG |
647 | struct inode *inode; |
648 | ||
649 | /* | |
650 | * The associated futex object in this case is the inode and | |
651 | * the page->mapping must be traversed. Ordinarily this should | |
652 | * be stabilised under page lock but it's not strictly | |
653 | * necessary in this case as we just want to pin the inode, not | |
654 | * update the radix tree or anything like that. | |
655 | * | |
656 | * The RCU read lock is taken as the inode is finally freed | |
657 | * under RCU. If the mapping still matches expectations then the | |
658 | * mapping->host can be safely accessed as being a valid inode. | |
659 | */ | |
660 | rcu_read_lock(); | |
661 | ||
662 | if (READ_ONCE(page->mapping) != mapping) { | |
663 | rcu_read_unlock(); | |
664 | put_page(page); | |
665 | ||
666 | goto again; | |
667 | } | |
668 | ||
669 | inode = READ_ONCE(mapping->host); | |
670 | if (!inode) { | |
671 | rcu_read_unlock(); | |
672 | put_page(page); | |
673 | ||
674 | goto again; | |
675 | } | |
676 | ||
677 | /* | |
678 | * Take a reference unless it is about to be freed. Previously | |
679 | * this reference was taken by ihold under the page lock | |
680 | * pinning the inode in place so i_lock was unnecessary. The | |
681 | * only way for this check to fail is if the inode was | |
48fb6f4d MG |
682 | * truncated in parallel which is almost certainly an |
683 | * application bug. In such a case, just retry. | |
65d8fc77 MG |
684 | * |
685 | * We are not calling into get_futex_key_refs() in file-backed | |
686 | * cases, therefore a successful atomic_inc return below will | |
687 | * guarantee that get_futex_key() will still imply smp_mb(); (B). | |
688 | */ | |
48fb6f4d | 689 | if (!atomic_inc_not_zero(&inode->i_count)) { |
65d8fc77 MG |
690 | rcu_read_unlock(); |
691 | put_page(page); | |
692 | ||
693 | goto again; | |
694 | } | |
695 | ||
696 | /* Should be impossible but lets be paranoid for now */ | |
697 | if (WARN_ON_ONCE(inode->i_mapping != mapping)) { | |
698 | err = -EFAULT; | |
699 | rcu_read_unlock(); | |
700 | iput(inode); | |
701 | ||
702 | goto out; | |
703 | } | |
704 | ||
38d47c1b | 705 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ |
65d8fc77 | 706 | key->shared.inode = inode; |
077fa7ae | 707 | key->shared.pgoff = basepage_index(tail); |
65d8fc77 | 708 | rcu_read_unlock(); |
1da177e4 LT |
709 | } |
710 | ||
9ea71503 | 711 | out: |
14d27abd | 712 | put_page(page); |
9ea71503 | 713 | return err; |
1da177e4 LT |
714 | } |
715 | ||
ae791a2d | 716 | static inline void put_futex_key(union futex_key *key) |
1da177e4 | 717 | { |
38d47c1b | 718 | drop_futex_key_refs(key); |
1da177e4 LT |
719 | } |
720 | ||
d96ee56c DH |
721 | /** |
722 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
723 | * @uaddr: pointer to faulting user space address |
724 | * | |
725 | * Slow path to fixup the fault we just took in the atomic write | |
726 | * access to @uaddr. | |
727 | * | |
fb62db2b | 728 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
729 | * user address. We know that we faulted in the atomic pagefault |
730 | * disabled section so we can as well avoid the #PF overhead by | |
731 | * calling get_user_pages() right away. | |
732 | */ | |
733 | static int fault_in_user_writeable(u32 __user *uaddr) | |
734 | { | |
722d0172 AK |
735 | struct mm_struct *mm = current->mm; |
736 | int ret; | |
737 | ||
738 | down_read(&mm->mmap_sem); | |
2efaca92 | 739 | ret = fixup_user_fault(current, mm, (unsigned long)uaddr, |
4a9e1cda | 740 | FAULT_FLAG_WRITE, NULL); |
722d0172 AK |
741 | up_read(&mm->mmap_sem); |
742 | ||
d0725992 TG |
743 | return ret < 0 ? ret : 0; |
744 | } | |
745 | ||
4b1c486b DH |
746 | /** |
747 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
748 | * @hb: the hash bucket the futex_q's reside in |
749 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
750 | * |
751 | * Must be called with the hb lock held. | |
752 | */ | |
753 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
754 | union futex_key *key) | |
755 | { | |
756 | struct futex_q *this; | |
757 | ||
758 | plist_for_each_entry(this, &hb->chain, list) { | |
759 | if (match_futex(&this->key, key)) | |
760 | return this; | |
761 | } | |
762 | return NULL; | |
763 | } | |
764 | ||
37a9d912 ML |
765 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
766 | u32 uval, u32 newval) | |
36cf3b5c | 767 | { |
37a9d912 | 768 | int ret; |
36cf3b5c TG |
769 | |
770 | pagefault_disable(); | |
37a9d912 | 771 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
772 | pagefault_enable(); |
773 | ||
37a9d912 | 774 | return ret; |
36cf3b5c TG |
775 | } |
776 | ||
777 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
778 | { |
779 | int ret; | |
780 | ||
a866374a | 781 | pagefault_disable(); |
bd28b145 | 782 | ret = __get_user(*dest, from); |
a866374a | 783 | pagefault_enable(); |
1da177e4 LT |
784 | |
785 | return ret ? -EFAULT : 0; | |
786 | } | |
787 | ||
c87e2837 IM |
788 | |
789 | /* | |
790 | * PI code: | |
791 | */ | |
792 | static int refill_pi_state_cache(void) | |
793 | { | |
794 | struct futex_pi_state *pi_state; | |
795 | ||
796 | if (likely(current->pi_state_cache)) | |
797 | return 0; | |
798 | ||
4668edc3 | 799 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
800 | |
801 | if (!pi_state) | |
802 | return -ENOMEM; | |
803 | ||
c87e2837 IM |
804 | INIT_LIST_HEAD(&pi_state->list); |
805 | /* pi_mutex gets initialized later */ | |
806 | pi_state->owner = NULL; | |
807 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 808 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
809 | |
810 | current->pi_state_cache = pi_state; | |
811 | ||
812 | return 0; | |
813 | } | |
814 | ||
bf92cf3a | 815 | static struct futex_pi_state *alloc_pi_state(void) |
c87e2837 IM |
816 | { |
817 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
818 | ||
819 | WARN_ON(!pi_state); | |
820 | current->pi_state_cache = NULL; | |
821 | ||
822 | return pi_state; | |
823 | } | |
824 | ||
bf92cf3a PZ |
825 | static void get_pi_state(struct futex_pi_state *pi_state) |
826 | { | |
827 | WARN_ON_ONCE(!atomic_inc_not_zero(&pi_state->refcount)); | |
828 | } | |
829 | ||
30a6b803 | 830 | /* |
29e9ee5d TG |
831 | * Drops a reference to the pi_state object and frees or caches it |
832 | * when the last reference is gone. | |
30a6b803 | 833 | */ |
29e9ee5d | 834 | static void put_pi_state(struct futex_pi_state *pi_state) |
c87e2837 | 835 | { |
30a6b803 BS |
836 | if (!pi_state) |
837 | return; | |
838 | ||
c87e2837 IM |
839 | if (!atomic_dec_and_test(&pi_state->refcount)) |
840 | return; | |
841 | ||
842 | /* | |
843 | * If pi_state->owner is NULL, the owner is most probably dying | |
844 | * and has cleaned up the pi_state already | |
845 | */ | |
846 | if (pi_state->owner) { | |
c74aef2d | 847 | struct task_struct *owner; |
c87e2837 | 848 | |
c74aef2d PZ |
849 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
850 | owner = pi_state->owner; | |
851 | if (owner) { | |
852 | raw_spin_lock(&owner->pi_lock); | |
853 | list_del_init(&pi_state->list); | |
854 | raw_spin_unlock(&owner->pi_lock); | |
855 | } | |
856 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, owner); | |
857 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); | |
c87e2837 IM |
858 | } |
859 | ||
c74aef2d | 860 | if (current->pi_state_cache) { |
c87e2837 | 861 | kfree(pi_state); |
c74aef2d | 862 | } else { |
c87e2837 IM |
863 | /* |
864 | * pi_state->list is already empty. | |
865 | * clear pi_state->owner. | |
866 | * refcount is at 0 - put it back to 1. | |
867 | */ | |
868 | pi_state->owner = NULL; | |
869 | atomic_set(&pi_state->refcount, 1); | |
870 | current->pi_state_cache = pi_state; | |
871 | } | |
872 | } | |
873 | ||
874 | /* | |
875 | * Look up the task based on what TID userspace gave us. | |
876 | * We dont trust it. | |
877 | */ | |
bf92cf3a | 878 | static struct task_struct *futex_find_get_task(pid_t pid) |
c87e2837 IM |
879 | { |
880 | struct task_struct *p; | |
881 | ||
d359b549 | 882 | rcu_read_lock(); |
228ebcbe | 883 | p = find_task_by_vpid(pid); |
7a0ea09a MH |
884 | if (p) |
885 | get_task_struct(p); | |
a06381fe | 886 | |
d359b549 | 887 | rcu_read_unlock(); |
c87e2837 IM |
888 | |
889 | return p; | |
890 | } | |
891 | ||
bc2eecd7 NP |
892 | #ifdef CONFIG_FUTEX_PI |
893 | ||
c87e2837 IM |
894 | /* |
895 | * This task is holding PI mutexes at exit time => bad. | |
896 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
897 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
898 | */ | |
e891c8f7 | 899 | static void exit_pi_state_list(struct task_struct *curr) |
c87e2837 | 900 | { |
c87e2837 IM |
901 | struct list_head *next, *head = &curr->pi_state_list; |
902 | struct futex_pi_state *pi_state; | |
627371d7 | 903 | struct futex_hash_bucket *hb; |
38d47c1b | 904 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 905 | |
a0c1e907 TG |
906 | if (!futex_cmpxchg_enabled) |
907 | return; | |
c87e2837 IM |
908 | /* |
909 | * We are a ZOMBIE and nobody can enqueue itself on | |
910 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 911 | * versus waiters unqueueing themselves: |
c87e2837 | 912 | */ |
1d615482 | 913 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 914 | while (!list_empty(head)) { |
c87e2837 IM |
915 | next = head->next; |
916 | pi_state = list_entry(next, struct futex_pi_state, list); | |
917 | key = pi_state->key; | |
627371d7 | 918 | hb = hash_futex(&key); |
153fbd12 PZ |
919 | |
920 | /* | |
921 | * We can race against put_pi_state() removing itself from the | |
922 | * list (a waiter going away). put_pi_state() will first | |
923 | * decrement the reference count and then modify the list, so | |
924 | * its possible to see the list entry but fail this reference | |
925 | * acquire. | |
926 | * | |
927 | * In that case; drop the locks to let put_pi_state() make | |
928 | * progress and retry the loop. | |
929 | */ | |
930 | if (!atomic_inc_not_zero(&pi_state->refcount)) { | |
931 | raw_spin_unlock_irq(&curr->pi_lock); | |
932 | cpu_relax(); | |
933 | raw_spin_lock_irq(&curr->pi_lock); | |
934 | continue; | |
935 | } | |
1d615482 | 936 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 937 | |
c87e2837 | 938 | spin_lock(&hb->lock); |
c74aef2d PZ |
939 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
940 | raw_spin_lock(&curr->pi_lock); | |
627371d7 IM |
941 | /* |
942 | * We dropped the pi-lock, so re-check whether this | |
943 | * task still owns the PI-state: | |
944 | */ | |
c87e2837 | 945 | if (head->next != next) { |
153fbd12 | 946 | /* retain curr->pi_lock for the loop invariant */ |
c74aef2d | 947 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 | 948 | spin_unlock(&hb->lock); |
153fbd12 | 949 | put_pi_state(pi_state); |
c87e2837 IM |
950 | continue; |
951 | } | |
952 | ||
c87e2837 | 953 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
954 | WARN_ON(list_empty(&pi_state->list)); |
955 | list_del_init(&pi_state->list); | |
c87e2837 | 956 | pi_state->owner = NULL; |
c87e2837 | 957 | |
153fbd12 | 958 | raw_spin_unlock(&curr->pi_lock); |
c74aef2d | 959 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
960 | spin_unlock(&hb->lock); |
961 | ||
16ffa12d PZ |
962 | rt_mutex_futex_unlock(&pi_state->pi_mutex); |
963 | put_pi_state(pi_state); | |
964 | ||
1d615482 | 965 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 966 | } |
1d615482 | 967 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 968 | } |
e891c8f7 TG |
969 | #else |
970 | static inline void exit_pi_state_list(struct task_struct *curr) { } | |
bc2eecd7 NP |
971 | #endif |
972 | ||
54a21788 TG |
973 | /* |
974 | * We need to check the following states: | |
975 | * | |
976 | * Waiter | pi_state | pi->owner | uTID | uODIED | ? | |
977 | * | |
978 | * [1] NULL | --- | --- | 0 | 0/1 | Valid | |
979 | * [2] NULL | --- | --- | >0 | 0/1 | Valid | |
980 | * | |
981 | * [3] Found | NULL | -- | Any | 0/1 | Invalid | |
982 | * | |
983 | * [4] Found | Found | NULL | 0 | 1 | Valid | |
984 | * [5] Found | Found | NULL | >0 | 1 | Invalid | |
985 | * | |
986 | * [6] Found | Found | task | 0 | 1 | Valid | |
987 | * | |
988 | * [7] Found | Found | NULL | Any | 0 | Invalid | |
989 | * | |
990 | * [8] Found | Found | task | ==taskTID | 0/1 | Valid | |
991 | * [9] Found | Found | task | 0 | 0 | Invalid | |
992 | * [10] Found | Found | task | !=taskTID | 0/1 | Invalid | |
993 | * | |
994 | * [1] Indicates that the kernel can acquire the futex atomically. We | |
995 | * came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. | |
996 | * | |
997 | * [2] Valid, if TID does not belong to a kernel thread. If no matching | |
998 | * thread is found then it indicates that the owner TID has died. | |
999 | * | |
1000 | * [3] Invalid. The waiter is queued on a non PI futex | |
1001 | * | |
1002 | * [4] Valid state after exit_robust_list(), which sets the user space | |
1003 | * value to FUTEX_WAITERS | FUTEX_OWNER_DIED. | |
1004 | * | |
1005 | * [5] The user space value got manipulated between exit_robust_list() | |
1006 | * and exit_pi_state_list() | |
1007 | * | |
1008 | * [6] Valid state after exit_pi_state_list() which sets the new owner in | |
1009 | * the pi_state but cannot access the user space value. | |
1010 | * | |
1011 | * [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. | |
1012 | * | |
1013 | * [8] Owner and user space value match | |
1014 | * | |
1015 | * [9] There is no transient state which sets the user space TID to 0 | |
1016 | * except exit_robust_list(), but this is indicated by the | |
1017 | * FUTEX_OWNER_DIED bit. See [4] | |
1018 | * | |
1019 | * [10] There is no transient state which leaves owner and user space | |
1020 | * TID out of sync. | |
734009e9 PZ |
1021 | * |
1022 | * | |
1023 | * Serialization and lifetime rules: | |
1024 | * | |
1025 | * hb->lock: | |
1026 | * | |
1027 | * hb -> futex_q, relation | |
1028 | * futex_q -> pi_state, relation | |
1029 | * | |
1030 | * (cannot be raw because hb can contain arbitrary amount | |
1031 | * of futex_q's) | |
1032 | * | |
1033 | * pi_mutex->wait_lock: | |
1034 | * | |
1035 | * {uval, pi_state} | |
1036 | * | |
1037 | * (and pi_mutex 'obviously') | |
1038 | * | |
1039 | * p->pi_lock: | |
1040 | * | |
1041 | * p->pi_state_list -> pi_state->list, relation | |
1042 | * | |
1043 | * pi_state->refcount: | |
1044 | * | |
1045 | * pi_state lifetime | |
1046 | * | |
1047 | * | |
1048 | * Lock order: | |
1049 | * | |
1050 | * hb->lock | |
1051 | * pi_mutex->wait_lock | |
1052 | * p->pi_lock | |
1053 | * | |
54a21788 | 1054 | */ |
e60cbc5c TG |
1055 | |
1056 | /* | |
1057 | * Validate that the existing waiter has a pi_state and sanity check | |
1058 | * the pi_state against the user space value. If correct, attach to | |
1059 | * it. | |
1060 | */ | |
734009e9 PZ |
1061 | static int attach_to_pi_state(u32 __user *uaddr, u32 uval, |
1062 | struct futex_pi_state *pi_state, | |
e60cbc5c | 1063 | struct futex_pi_state **ps) |
c87e2837 | 1064 | { |
778e9a9c | 1065 | pid_t pid = uval & FUTEX_TID_MASK; |
94ffac5d PZ |
1066 | u32 uval2; |
1067 | int ret; | |
c87e2837 | 1068 | |
e60cbc5c TG |
1069 | /* |
1070 | * Userspace might have messed up non-PI and PI futexes [3] | |
1071 | */ | |
1072 | if (unlikely(!pi_state)) | |
1073 | return -EINVAL; | |
06a9ec29 | 1074 | |
734009e9 PZ |
1075 | /* |
1076 | * We get here with hb->lock held, and having found a | |
1077 | * futex_top_waiter(). This means that futex_lock_pi() of said futex_q | |
1078 | * has dropped the hb->lock in between queue_me() and unqueue_me_pi(), | |
1079 | * which in turn means that futex_lock_pi() still has a reference on | |
1080 | * our pi_state. | |
16ffa12d PZ |
1081 | * |
1082 | * The waiter holding a reference on @pi_state also protects against | |
1083 | * the unlocked put_pi_state() in futex_unlock_pi(), futex_lock_pi() | |
1084 | * and futex_wait_requeue_pi() as it cannot go to 0 and consequently | |
1085 | * free pi_state before we can take a reference ourselves. | |
734009e9 | 1086 | */ |
e60cbc5c | 1087 | WARN_ON(!atomic_read(&pi_state->refcount)); |
59647b6a | 1088 | |
734009e9 PZ |
1089 | /* |
1090 | * Now that we have a pi_state, we can acquire wait_lock | |
1091 | * and do the state validation. | |
1092 | */ | |
1093 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); | |
1094 | ||
1095 | /* | |
1096 | * Since {uval, pi_state} is serialized by wait_lock, and our current | |
1097 | * uval was read without holding it, it can have changed. Verify it | |
1098 | * still is what we expect it to be, otherwise retry the entire | |
1099 | * operation. | |
1100 | */ | |
1101 | if (get_futex_value_locked(&uval2, uaddr)) | |
1102 | goto out_efault; | |
1103 | ||
1104 | if (uval != uval2) | |
1105 | goto out_eagain; | |
1106 | ||
e60cbc5c TG |
1107 | /* |
1108 | * Handle the owner died case: | |
1109 | */ | |
1110 | if (uval & FUTEX_OWNER_DIED) { | |
bd1dbcc6 | 1111 | /* |
e60cbc5c TG |
1112 | * exit_pi_state_list sets owner to NULL and wakes the |
1113 | * topmost waiter. The task which acquires the | |
1114 | * pi_state->rt_mutex will fixup owner. | |
bd1dbcc6 | 1115 | */ |
e60cbc5c | 1116 | if (!pi_state->owner) { |
59647b6a | 1117 | /* |
e60cbc5c TG |
1118 | * No pi state owner, but the user space TID |
1119 | * is not 0. Inconsistent state. [5] | |
59647b6a | 1120 | */ |
e60cbc5c | 1121 | if (pid) |
734009e9 | 1122 | goto out_einval; |
bd1dbcc6 | 1123 | /* |
e60cbc5c | 1124 | * Take a ref on the state and return success. [4] |
866293ee | 1125 | */ |
734009e9 | 1126 | goto out_attach; |
c87e2837 | 1127 | } |
bd1dbcc6 TG |
1128 | |
1129 | /* | |
e60cbc5c TG |
1130 | * If TID is 0, then either the dying owner has not |
1131 | * yet executed exit_pi_state_list() or some waiter | |
1132 | * acquired the rtmutex in the pi state, but did not | |
1133 | * yet fixup the TID in user space. | |
1134 | * | |
1135 | * Take a ref on the state and return success. [6] | |
1136 | */ | |
1137 | if (!pid) | |
734009e9 | 1138 | goto out_attach; |
e60cbc5c TG |
1139 | } else { |
1140 | /* | |
1141 | * If the owner died bit is not set, then the pi_state | |
1142 | * must have an owner. [7] | |
bd1dbcc6 | 1143 | */ |
e60cbc5c | 1144 | if (!pi_state->owner) |
734009e9 | 1145 | goto out_einval; |
c87e2837 IM |
1146 | } |
1147 | ||
e60cbc5c TG |
1148 | /* |
1149 | * Bail out if user space manipulated the futex value. If pi | |
1150 | * state exists then the owner TID must be the same as the | |
1151 | * user space TID. [9/10] | |
1152 | */ | |
1153 | if (pid != task_pid_vnr(pi_state->owner)) | |
734009e9 PZ |
1154 | goto out_einval; |
1155 | ||
1156 | out_attach: | |
bf92cf3a | 1157 | get_pi_state(pi_state); |
734009e9 | 1158 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
e60cbc5c TG |
1159 | *ps = pi_state; |
1160 | return 0; | |
734009e9 PZ |
1161 | |
1162 | out_einval: | |
1163 | ret = -EINVAL; | |
1164 | goto out_error; | |
1165 | ||
1166 | out_eagain: | |
1167 | ret = -EAGAIN; | |
1168 | goto out_error; | |
1169 | ||
1170 | out_efault: | |
1171 | ret = -EFAULT; | |
1172 | goto out_error; | |
1173 | ||
1174 | out_error: | |
1175 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); | |
1176 | return ret; | |
e60cbc5c TG |
1177 | } |
1178 | ||
ad906e7e TG |
1179 | /** |
1180 | * wait_for_owner_exiting - Block until the owner has exited | |
1181 | * @exiting: Pointer to the exiting task | |
1182 | * | |
1183 | * Caller must hold a refcount on @exiting. | |
1184 | */ | |
1185 | static void wait_for_owner_exiting(int ret, struct task_struct *exiting) | |
1186 | { | |
1187 | if (ret != -EBUSY) { | |
1188 | WARN_ON_ONCE(exiting); | |
1189 | return; | |
1190 | } | |
1191 | ||
1192 | if (WARN_ON_ONCE(ret == -EBUSY && !exiting)) | |
1193 | return; | |
1194 | ||
1195 | mutex_lock(&exiting->futex_exit_mutex); | |
1196 | /* | |
1197 | * No point in doing state checking here. If the waiter got here | |
1198 | * while the task was in exec()->exec_futex_release() then it can | |
1199 | * have any FUTEX_STATE_* value when the waiter has acquired the | |
1200 | * mutex. OK, if running, EXITING or DEAD if it reached exit() | |
1201 | * already. Highly unlikely and not a problem. Just one more round | |
1202 | * through the futex maze. | |
1203 | */ | |
1204 | mutex_unlock(&exiting->futex_exit_mutex); | |
1205 | ||
1206 | put_task_struct(exiting); | |
1207 | } | |
1208 | ||
d2dfa785 TG |
1209 | static int handle_exit_race(u32 __user *uaddr, u32 uval, |
1210 | struct task_struct *tsk) | |
1211 | { | |
1212 | u32 uval2; | |
1213 | ||
1214 | /* | |
8ef775af TG |
1215 | * If the futex exit state is not yet FUTEX_STATE_DEAD, tell the |
1216 | * caller that the alleged owner is busy. | |
d2dfa785 | 1217 | */ |
37ad9017 | 1218 | if (tsk && tsk->futex_state != FUTEX_STATE_DEAD) |
8ef775af | 1219 | return -EBUSY; |
d2dfa785 TG |
1220 | |
1221 | /* | |
1222 | * Reread the user space value to handle the following situation: | |
1223 | * | |
1224 | * CPU0 CPU1 | |
1225 | * | |
1226 | * sys_exit() sys_futex() | |
1227 | * do_exit() futex_lock_pi() | |
1228 | * futex_lock_pi_atomic() | |
1229 | * exit_signals(tsk) No waiters: | |
1230 | * tsk->flags |= PF_EXITING; *uaddr == 0x00000PID | |
1231 | * mm_release(tsk) Set waiter bit | |
1232 | * exit_robust_list(tsk) { *uaddr = 0x80000PID; | |
1233 | * Set owner died attach_to_pi_owner() { | |
1234 | * *uaddr = 0xC0000000; tsk = get_task(PID); | |
1235 | * } if (!tsk->flags & PF_EXITING) { | |
1236 | * ... attach(); | |
37ad9017 TG |
1237 | * tsk->futex_state = } else { |
1238 | * FUTEX_STATE_DEAD; if (tsk->futex_state != | |
1239 | * FUTEX_STATE_DEAD) | |
d2dfa785 TG |
1240 | * return -EAGAIN; |
1241 | * return -ESRCH; <--- FAIL | |
1242 | * } | |
1243 | * | |
1244 | * Returning ESRCH unconditionally is wrong here because the | |
1245 | * user space value has been changed by the exiting task. | |
1246 | * | |
1247 | * The same logic applies to the case where the exiting task is | |
1248 | * already gone. | |
1249 | */ | |
1250 | if (get_futex_value_locked(&uval2, uaddr)) | |
1251 | return -EFAULT; | |
1252 | ||
1253 | /* If the user space value has changed, try again. */ | |
1254 | if (uval2 != uval) | |
1255 | return -EAGAIN; | |
1256 | ||
1257 | /* | |
1258 | * The exiting task did not have a robust list, the robust list was | |
1259 | * corrupted or the user space value in *uaddr is simply bogus. | |
1260 | * Give up and tell user space. | |
1261 | */ | |
1262 | return -ESRCH; | |
1263 | } | |
1264 | ||
04e1b2e5 TG |
1265 | /* |
1266 | * Lookup the task for the TID provided from user space and attach to | |
1267 | * it after doing proper sanity checks. | |
1268 | */ | |
d2dfa785 | 1269 | static int attach_to_pi_owner(u32 __user *uaddr, u32 uval, union futex_key *key, |
ad906e7e TG |
1270 | struct futex_pi_state **ps, |
1271 | struct task_struct **exiting) | |
e60cbc5c | 1272 | { |
e60cbc5c | 1273 | pid_t pid = uval & FUTEX_TID_MASK; |
04e1b2e5 TG |
1274 | struct futex_pi_state *pi_state; |
1275 | struct task_struct *p; | |
e60cbc5c | 1276 | |
c87e2837 | 1277 | /* |
e3f2ddea | 1278 | * We are the first waiter - try to look up the real owner and attach |
54a21788 | 1279 | * the new pi_state to it, but bail out when TID = 0 [1] |
d2dfa785 TG |
1280 | * |
1281 | * The !pid check is paranoid. None of the call sites should end up | |
1282 | * with pid == 0, but better safe than sorry. Let the caller retry | |
c87e2837 | 1283 | */ |
778e9a9c | 1284 | if (!pid) |
d2dfa785 | 1285 | return -EAGAIN; |
c87e2837 | 1286 | p = futex_find_get_task(pid); |
7a0ea09a | 1287 | if (!p) |
d2dfa785 | 1288 | return handle_exit_race(uaddr, uval, NULL); |
778e9a9c | 1289 | |
a2129464 | 1290 | if (unlikely(p->flags & PF_KTHREAD)) { |
f0d71b3d TG |
1291 | put_task_struct(p); |
1292 | return -EPERM; | |
1293 | } | |
1294 | ||
778e9a9c | 1295 | /* |
37ad9017 TG |
1296 | * We need to look at the task state to figure out, whether the |
1297 | * task is exiting. To protect against the change of the task state | |
1298 | * in futex_exit_release(), we do this protected by p->pi_lock: | |
778e9a9c | 1299 | */ |
1d615482 | 1300 | raw_spin_lock_irq(&p->pi_lock); |
37ad9017 | 1301 | if (unlikely(p->futex_state != FUTEX_STATE_OK)) { |
778e9a9c | 1302 | /* |
37ad9017 TG |
1303 | * The task is on the way out. When the futex state is |
1304 | * FUTEX_STATE_DEAD, we know that the task has finished | |
1305 | * the cleanup: | |
778e9a9c | 1306 | */ |
d2dfa785 | 1307 | int ret = handle_exit_race(uaddr, uval, p); |
778e9a9c | 1308 | |
1d615482 | 1309 | raw_spin_unlock_irq(&p->pi_lock); |
ad906e7e TG |
1310 | /* |
1311 | * If the owner task is between FUTEX_STATE_EXITING and | |
1312 | * FUTEX_STATE_DEAD then store the task pointer and keep | |
1313 | * the reference on the task struct. The calling code will | |
1314 | * drop all locks, wait for the task to reach | |
1315 | * FUTEX_STATE_DEAD and then drop the refcount. This is | |
1316 | * required to prevent a live lock when the current task | |
1317 | * preempted the exiting task between the two states. | |
1318 | */ | |
1319 | if (ret == -EBUSY) | |
1320 | *exiting = p; | |
1321 | else | |
1322 | put_task_struct(p); | |
778e9a9c AK |
1323 | return ret; |
1324 | } | |
c87e2837 | 1325 | |
54a21788 TG |
1326 | /* |
1327 | * No existing pi state. First waiter. [2] | |
734009e9 PZ |
1328 | * |
1329 | * This creates pi_state, we have hb->lock held, this means nothing can | |
1330 | * observe this state, wait_lock is irrelevant. | |
54a21788 | 1331 | */ |
c87e2837 IM |
1332 | pi_state = alloc_pi_state(); |
1333 | ||
1334 | /* | |
04e1b2e5 | 1335 | * Initialize the pi_mutex in locked state and make @p |
c87e2837 IM |
1336 | * the owner of it: |
1337 | */ | |
1338 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
1339 | ||
1340 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 1341 | pi_state->key = *key; |
c87e2837 | 1342 | |
627371d7 | 1343 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 | 1344 | list_add(&pi_state->list, &p->pi_state_list); |
c74aef2d PZ |
1345 | /* |
1346 | * Assignment without holding pi_state->pi_mutex.wait_lock is safe | |
1347 | * because there is no concurrency as the object is not published yet. | |
1348 | */ | |
c87e2837 | 1349 | pi_state->owner = p; |
1d615482 | 1350 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
1351 | |
1352 | put_task_struct(p); | |
1353 | ||
d0aa7a70 | 1354 | *ps = pi_state; |
c87e2837 IM |
1355 | |
1356 | return 0; | |
1357 | } | |
1358 | ||
734009e9 PZ |
1359 | static int lookup_pi_state(u32 __user *uaddr, u32 uval, |
1360 | struct futex_hash_bucket *hb, | |
ad906e7e TG |
1361 | union futex_key *key, struct futex_pi_state **ps, |
1362 | struct task_struct **exiting) | |
04e1b2e5 | 1363 | { |
499f5aca | 1364 | struct futex_q *top_waiter = futex_top_waiter(hb, key); |
04e1b2e5 TG |
1365 | |
1366 | /* | |
1367 | * If there is a waiter on that futex, validate it and | |
1368 | * attach to the pi_state when the validation succeeds. | |
1369 | */ | |
499f5aca | 1370 | if (top_waiter) |
734009e9 | 1371 | return attach_to_pi_state(uaddr, uval, top_waiter->pi_state, ps); |
04e1b2e5 TG |
1372 | |
1373 | /* | |
1374 | * We are the first waiter - try to look up the owner based on | |
1375 | * @uval and attach to it. | |
1376 | */ | |
ad906e7e | 1377 | return attach_to_pi_owner(uaddr, uval, key, ps, exiting); |
04e1b2e5 TG |
1378 | } |
1379 | ||
af54d6a1 TG |
1380 | static int lock_pi_update_atomic(u32 __user *uaddr, u32 uval, u32 newval) |
1381 | { | |
e5733e53 | 1382 | int err; |
af54d6a1 TG |
1383 | u32 uninitialized_var(curval); |
1384 | ||
ab51fbab DB |
1385 | if (unlikely(should_fail_futex(true))) |
1386 | return -EFAULT; | |
1387 | ||
e5733e53 WD |
1388 | err = cmpxchg_futex_value_locked(&curval, uaddr, uval, newval); |
1389 | if (unlikely(err)) | |
1390 | return err; | |
af54d6a1 | 1391 | |
734009e9 | 1392 | /* If user space value changed, let the caller retry */ |
af54d6a1 TG |
1393 | return curval != uval ? -EAGAIN : 0; |
1394 | } | |
1395 | ||
1a52084d | 1396 | /** |
d96ee56c | 1397 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
1398 | * @uaddr: the pi futex user address |
1399 | * @hb: the pi futex hash bucket | |
1400 | * @key: the futex key associated with uaddr and hb | |
1401 | * @ps: the pi_state pointer where we store the result of the | |
1402 | * lookup | |
1403 | * @task: the task to perform the atomic lock work for. This will | |
1404 | * be "current" except in the case of requeue pi. | |
ad906e7e TG |
1405 | * @exiting: Pointer to store the task pointer of the owner task |
1406 | * which is in the middle of exiting | |
bab5bc9e | 1407 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) |
1a52084d | 1408 | * |
6c23cbbd | 1409 | * Return: |
7b4ff1ad MCC |
1410 | * - 0 - ready to wait; |
1411 | * - 1 - acquired the lock; | |
1412 | * - <0 - error | |
1a52084d DH |
1413 | * |
1414 | * The hb->lock and futex_key refs shall be held by the caller. | |
ad906e7e TG |
1415 | * |
1416 | * @exiting is only set when the return value is -EBUSY. If so, this holds | |
1417 | * a refcount on the exiting task on return and the caller needs to drop it | |
1418 | * after waiting for the exit to complete. | |
1a52084d DH |
1419 | */ |
1420 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
1421 | union futex_key *key, | |
1422 | struct futex_pi_state **ps, | |
ad906e7e TG |
1423 | struct task_struct *task, |
1424 | struct task_struct **exiting, | |
1425 | int set_waiters) | |
1a52084d | 1426 | { |
af54d6a1 | 1427 | u32 uval, newval, vpid = task_pid_vnr(task); |
499f5aca | 1428 | struct futex_q *top_waiter; |
af54d6a1 | 1429 | int ret; |
1a52084d DH |
1430 | |
1431 | /* | |
af54d6a1 TG |
1432 | * Read the user space value first so we can validate a few |
1433 | * things before proceeding further. | |
1a52084d | 1434 | */ |
af54d6a1 | 1435 | if (get_futex_value_locked(&uval, uaddr)) |
1a52084d DH |
1436 | return -EFAULT; |
1437 | ||
ab51fbab DB |
1438 | if (unlikely(should_fail_futex(true))) |
1439 | return -EFAULT; | |
1440 | ||
1a52084d DH |
1441 | /* |
1442 | * Detect deadlocks. | |
1443 | */ | |
af54d6a1 | 1444 | if ((unlikely((uval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
1445 | return -EDEADLK; |
1446 | ||
ab51fbab DB |
1447 | if ((unlikely(should_fail_futex(true)))) |
1448 | return -EDEADLK; | |
1449 | ||
1a52084d | 1450 | /* |
af54d6a1 TG |
1451 | * Lookup existing state first. If it exists, try to attach to |
1452 | * its pi_state. | |
1a52084d | 1453 | */ |
499f5aca PZ |
1454 | top_waiter = futex_top_waiter(hb, key); |
1455 | if (top_waiter) | |
734009e9 | 1456 | return attach_to_pi_state(uaddr, uval, top_waiter->pi_state, ps); |
1a52084d DH |
1457 | |
1458 | /* | |
af54d6a1 TG |
1459 | * No waiter and user TID is 0. We are here because the |
1460 | * waiters or the owner died bit is set or called from | |
1461 | * requeue_cmp_pi or for whatever reason something took the | |
1462 | * syscall. | |
1a52084d | 1463 | */ |
af54d6a1 | 1464 | if (!(uval & FUTEX_TID_MASK)) { |
59fa6245 | 1465 | /* |
af54d6a1 TG |
1466 | * We take over the futex. No other waiters and the user space |
1467 | * TID is 0. We preserve the owner died bit. | |
59fa6245 | 1468 | */ |
af54d6a1 TG |
1469 | newval = uval & FUTEX_OWNER_DIED; |
1470 | newval |= vpid; | |
1a52084d | 1471 | |
af54d6a1 TG |
1472 | /* The futex requeue_pi code can enforce the waiters bit */ |
1473 | if (set_waiters) | |
1474 | newval |= FUTEX_WAITERS; | |
1475 | ||
1476 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1477 | /* If the take over worked, return 1 */ | |
1478 | return ret < 0 ? ret : 1; | |
1479 | } | |
1a52084d DH |
1480 | |
1481 | /* | |
af54d6a1 TG |
1482 | * First waiter. Set the waiters bit before attaching ourself to |
1483 | * the owner. If owner tries to unlock, it will be forced into | |
1484 | * the kernel and blocked on hb->lock. | |
1a52084d | 1485 | */ |
af54d6a1 TG |
1486 | newval = uval | FUTEX_WAITERS; |
1487 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1488 | if (ret) | |
1489 | return ret; | |
1a52084d | 1490 | /* |
af54d6a1 TG |
1491 | * If the update of the user space value succeeded, we try to |
1492 | * attach to the owner. If that fails, no harm done, we only | |
1493 | * set the FUTEX_WAITERS bit in the user space variable. | |
1a52084d | 1494 | */ |
ad906e7e | 1495 | return attach_to_pi_owner(uaddr, newval, key, ps, exiting); |
1a52084d DH |
1496 | } |
1497 | ||
2e12978a LJ |
1498 | /** |
1499 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
1500 | * @q: The futex_q to unqueue | |
1501 | * | |
1502 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
1503 | */ | |
1504 | static void __unqueue_futex(struct futex_q *q) | |
1505 | { | |
1506 | struct futex_hash_bucket *hb; | |
1507 | ||
29096202 SR |
1508 | if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr)) |
1509 | || WARN_ON(plist_node_empty(&q->list))) | |
2e12978a LJ |
1510 | return; |
1511 | ||
1512 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
1513 | plist_del(&q->list, &hb->chain); | |
11d4616b | 1514 | hb_waiters_dec(hb); |
2e12978a LJ |
1515 | } |
1516 | ||
1da177e4 LT |
1517 | /* |
1518 | * The hash bucket lock must be held when this is called. | |
1d0dcb3a DB |
1519 | * Afterwards, the futex_q must not be accessed. Callers |
1520 | * must ensure to later call wake_up_q() for the actual | |
1521 | * wakeups to occur. | |
1da177e4 | 1522 | */ |
1d0dcb3a | 1523 | static void mark_wake_futex(struct wake_q_head *wake_q, struct futex_q *q) |
1da177e4 | 1524 | { |
f1a11e05 TG |
1525 | struct task_struct *p = q->task; |
1526 | ||
aa10990e DH |
1527 | if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n")) |
1528 | return; | |
1529 | ||
1408661c | 1530 | get_task_struct(p); |
2e12978a | 1531 | __unqueue_futex(q); |
1da177e4 | 1532 | /* |
38fcd06e DHV |
1533 | * The waiting task can free the futex_q as soon as q->lock_ptr = NULL |
1534 | * is written, without taking any locks. This is possible in the event | |
1535 | * of a spurious wakeup, for example. A memory barrier is required here | |
1536 | * to prevent the following store to lock_ptr from getting ahead of the | |
1537 | * plist_del in __unqueue_futex(). | |
1da177e4 | 1538 | */ |
1b367ece | 1539 | smp_store_release(&q->lock_ptr, NULL); |
1408661c PZ |
1540 | |
1541 | /* | |
1542 | * Queue the task for later wakeup for after we've released | |
1543 | * the hb->lock. wake_q_add() grabs reference to p. | |
1544 | */ | |
1545 | wake_q_add(wake_q, p); | |
1546 | put_task_struct(p); | |
1da177e4 LT |
1547 | } |
1548 | ||
16ffa12d PZ |
1549 | /* |
1550 | * Caller must hold a reference on @pi_state. | |
1551 | */ | |
1552 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_pi_state *pi_state) | |
c87e2837 | 1553 | { |
7cfdaf38 | 1554 | u32 uninitialized_var(curval), newval; |
16ffa12d | 1555 | struct task_struct *new_owner; |
aa2bfe55 | 1556 | bool postunlock = false; |
194a6b5b | 1557 | DEFINE_WAKE_Q(wake_q); |
13fbca4c | 1558 | int ret = 0; |
c87e2837 | 1559 | |
c87e2837 | 1560 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
bebe5b51 | 1561 | if (WARN_ON_ONCE(!new_owner)) { |
16ffa12d | 1562 | /* |
bebe5b51 | 1563 | * As per the comment in futex_unlock_pi() this should not happen. |
16ffa12d PZ |
1564 | * |
1565 | * When this happens, give up our locks and try again, giving | |
1566 | * the futex_lock_pi() instance time to complete, either by | |
1567 | * waiting on the rtmutex or removing itself from the futex | |
1568 | * queue. | |
1569 | */ | |
1570 | ret = -EAGAIN; | |
1571 | goto out_unlock; | |
73d786bd | 1572 | } |
c87e2837 IM |
1573 | |
1574 | /* | |
16ffa12d PZ |
1575 | * We pass it to the next owner. The WAITERS bit is always kept |
1576 | * enabled while there is PI state around. We cleanup the owner | |
1577 | * died bit, because we are the owner. | |
c87e2837 | 1578 | */ |
13fbca4c | 1579 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 1580 | |
ab51fbab DB |
1581 | if (unlikely(should_fail_futex(true))) |
1582 | ret = -EFAULT; | |
1583 | ||
e5733e53 WD |
1584 | ret = cmpxchg_futex_value_locked(&curval, uaddr, uval, newval); |
1585 | if (!ret && (curval != uval)) { | |
89e9e66b SAS |
1586 | /* |
1587 | * If a unconditional UNLOCK_PI operation (user space did not | |
1588 | * try the TID->0 transition) raced with a waiter setting the | |
1589 | * FUTEX_WAITERS flag between get_user() and locking the hash | |
1590 | * bucket lock, retry the operation. | |
1591 | */ | |
1592 | if ((FUTEX_TID_MASK & curval) == uval) | |
1593 | ret = -EAGAIN; | |
1594 | else | |
1595 | ret = -EINVAL; | |
1596 | } | |
734009e9 | 1597 | |
16ffa12d PZ |
1598 | if (ret) |
1599 | goto out_unlock; | |
c87e2837 | 1600 | |
94ffac5d PZ |
1601 | /* |
1602 | * This is a point of no return; once we modify the uval there is no | |
1603 | * going back and subsequent operations must not fail. | |
1604 | */ | |
1605 | ||
b4abf910 | 1606 | raw_spin_lock(&pi_state->owner->pi_lock); |
627371d7 IM |
1607 | WARN_ON(list_empty(&pi_state->list)); |
1608 | list_del_init(&pi_state->list); | |
b4abf910 | 1609 | raw_spin_unlock(&pi_state->owner->pi_lock); |
627371d7 | 1610 | |
b4abf910 | 1611 | raw_spin_lock(&new_owner->pi_lock); |
627371d7 | 1612 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
1613 | list_add(&pi_state->list, &new_owner->pi_state_list); |
1614 | pi_state->owner = new_owner; | |
b4abf910 | 1615 | raw_spin_unlock(&new_owner->pi_lock); |
627371d7 | 1616 | |
aa2bfe55 | 1617 | postunlock = __rt_mutex_futex_unlock(&pi_state->pi_mutex, &wake_q); |
5293c2ef | 1618 | |
16ffa12d | 1619 | out_unlock: |
5293c2ef | 1620 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
5293c2ef | 1621 | |
aa2bfe55 PZ |
1622 | if (postunlock) |
1623 | rt_mutex_postunlock(&wake_q); | |
c87e2837 | 1624 | |
16ffa12d | 1625 | return ret; |
c87e2837 IM |
1626 | } |
1627 | ||
8b8f319f IM |
1628 | /* |
1629 | * Express the locking dependencies for lockdep: | |
1630 | */ | |
1631 | static inline void | |
1632 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1633 | { | |
1634 | if (hb1 <= hb2) { | |
1635 | spin_lock(&hb1->lock); | |
1636 | if (hb1 < hb2) | |
1637 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
1638 | } else { /* hb1 > hb2 */ | |
1639 | spin_lock(&hb2->lock); | |
1640 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
1641 | } | |
1642 | } | |
1643 | ||
5eb3dc62 DH |
1644 | static inline void |
1645 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1646 | { | |
f061d351 | 1647 | spin_unlock(&hb1->lock); |
88f502fe IM |
1648 | if (hb1 != hb2) |
1649 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
1650 | } |
1651 | ||
1da177e4 | 1652 | /* |
b2d0994b | 1653 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 1654 | */ |
b41277dc DH |
1655 | static int |
1656 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 1657 | { |
e2970f2f | 1658 | struct futex_hash_bucket *hb; |
1da177e4 | 1659 | struct futex_q *this, *next; |
38d47c1b | 1660 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 | 1661 | int ret; |
194a6b5b | 1662 | DEFINE_WAKE_Q(wake_q); |
1da177e4 | 1663 | |
cd689985 TG |
1664 | if (!bitset) |
1665 | return -EINVAL; | |
1666 | ||
9ea71503 | 1667 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ); |
1da177e4 LT |
1668 | if (unlikely(ret != 0)) |
1669 | goto out; | |
1670 | ||
e2970f2f | 1671 | hb = hash_futex(&key); |
b0c29f79 DB |
1672 | |
1673 | /* Make sure we really have tasks to wakeup */ | |
1674 | if (!hb_waiters_pending(hb)) | |
1675 | goto out_put_key; | |
1676 | ||
e2970f2f | 1677 | spin_lock(&hb->lock); |
1da177e4 | 1678 | |
0d00c7b2 | 1679 | plist_for_each_entry_safe(this, next, &hb->chain, list) { |
1da177e4 | 1680 | if (match_futex (&this->key, &key)) { |
52400ba9 | 1681 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
1682 | ret = -EINVAL; |
1683 | break; | |
1684 | } | |
cd689985 TG |
1685 | |
1686 | /* Check if one of the bits is set in both bitsets */ | |
1687 | if (!(this->bitset & bitset)) | |
1688 | continue; | |
1689 | ||
1d0dcb3a | 1690 | mark_wake_futex(&wake_q, this); |
1da177e4 LT |
1691 | if (++ret >= nr_wake) |
1692 | break; | |
1693 | } | |
1694 | } | |
1695 | ||
e2970f2f | 1696 | spin_unlock(&hb->lock); |
1d0dcb3a | 1697 | wake_up_q(&wake_q); |
b0c29f79 | 1698 | out_put_key: |
ae791a2d | 1699 | put_futex_key(&key); |
42d35d48 | 1700 | out: |
1da177e4 LT |
1701 | return ret; |
1702 | } | |
1703 | ||
30d6e0a4 JS |
1704 | static int futex_atomic_op_inuser(unsigned int encoded_op, u32 __user *uaddr) |
1705 | { | |
1706 | unsigned int op = (encoded_op & 0x70000000) >> 28; | |
1707 | unsigned int cmp = (encoded_op & 0x0f000000) >> 24; | |
d70ef228 JS |
1708 | int oparg = sign_extend32((encoded_op & 0x00fff000) >> 12, 11); |
1709 | int cmparg = sign_extend32(encoded_op & 0x00000fff, 11); | |
30d6e0a4 JS |
1710 | int oldval, ret; |
1711 | ||
1712 | if (encoded_op & (FUTEX_OP_OPARG_SHIFT << 28)) { | |
e78c38f6 JS |
1713 | if (oparg < 0 || oparg > 31) { |
1714 | char comm[sizeof(current->comm)]; | |
1715 | /* | |
1716 | * kill this print and return -EINVAL when userspace | |
1717 | * is sane again | |
1718 | */ | |
1719 | pr_info_ratelimited("futex_wake_op: %s tries to shift op by %d; fix this program\n", | |
1720 | get_task_comm(comm, current), oparg); | |
1721 | oparg &= 31; | |
1722 | } | |
30d6e0a4 JS |
1723 | oparg = 1 << oparg; |
1724 | } | |
1725 | ||
1726 | if (!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))) | |
1727 | return -EFAULT; | |
1728 | ||
1729 | ret = arch_futex_atomic_op_inuser(op, oparg, &oldval, uaddr); | |
1730 | if (ret) | |
1731 | return ret; | |
1732 | ||
1733 | switch (cmp) { | |
1734 | case FUTEX_OP_CMP_EQ: | |
1735 | return oldval == cmparg; | |
1736 | case FUTEX_OP_CMP_NE: | |
1737 | return oldval != cmparg; | |
1738 | case FUTEX_OP_CMP_LT: | |
1739 | return oldval < cmparg; | |
1740 | case FUTEX_OP_CMP_GE: | |
1741 | return oldval >= cmparg; | |
1742 | case FUTEX_OP_CMP_LE: | |
1743 | return oldval <= cmparg; | |
1744 | case FUTEX_OP_CMP_GT: | |
1745 | return oldval > cmparg; | |
1746 | default: | |
1747 | return -ENOSYS; | |
1748 | } | |
1749 | } | |
1750 | ||
4732efbe JJ |
1751 | /* |
1752 | * Wake up all waiters hashed on the physical page that is mapped | |
1753 | * to this virtual address: | |
1754 | */ | |
e2970f2f | 1755 | static int |
b41277dc | 1756 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 1757 | int nr_wake, int nr_wake2, int op) |
4732efbe | 1758 | { |
38d47c1b | 1759 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 1760 | struct futex_hash_bucket *hb1, *hb2; |
4732efbe | 1761 | struct futex_q *this, *next; |
e4dc5b7a | 1762 | int ret, op_ret; |
194a6b5b | 1763 | DEFINE_WAKE_Q(wake_q); |
4732efbe | 1764 | |
e4dc5b7a | 1765 | retry: |
9ea71503 | 1766 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
4732efbe JJ |
1767 | if (unlikely(ret != 0)) |
1768 | goto out; | |
9ea71503 | 1769 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
4732efbe | 1770 | if (unlikely(ret != 0)) |
42d35d48 | 1771 | goto out_put_key1; |
4732efbe | 1772 | |
e2970f2f IM |
1773 | hb1 = hash_futex(&key1); |
1774 | hb2 = hash_futex(&key2); | |
4732efbe | 1775 | |
e4dc5b7a | 1776 | retry_private: |
eaaea803 | 1777 | double_lock_hb(hb1, hb2); |
e2970f2f | 1778 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1779 | if (unlikely(op_ret < 0)) { |
5eb3dc62 | 1780 | double_unlock_hb(hb1, hb2); |
4732efbe | 1781 | |
e5733e53 WD |
1782 | if (!IS_ENABLED(CONFIG_MMU) || |
1783 | unlikely(op_ret != -EFAULT && op_ret != -EAGAIN)) { | |
1784 | /* | |
1785 | * we don't get EFAULT from MMU faults if we don't have | |
1786 | * an MMU, but we might get them from range checking | |
1787 | */ | |
796f8d9b | 1788 | ret = op_ret; |
42d35d48 | 1789 | goto out_put_keys; |
796f8d9b DG |
1790 | } |
1791 | ||
e5733e53 WD |
1792 | if (op_ret == -EFAULT) { |
1793 | ret = fault_in_user_writeable(uaddr2); | |
1794 | if (ret) | |
1795 | goto out_put_keys; | |
1796 | } | |
4732efbe | 1797 | |
e5733e53 WD |
1798 | if (!(flags & FLAGS_SHARED)) { |
1799 | cond_resched(); | |
e4dc5b7a | 1800 | goto retry_private; |
e5733e53 | 1801 | } |
e4dc5b7a | 1802 | |
ae791a2d TG |
1803 | put_futex_key(&key2); |
1804 | put_futex_key(&key1); | |
e5733e53 | 1805 | cond_resched(); |
e4dc5b7a | 1806 | goto retry; |
4732efbe JJ |
1807 | } |
1808 | ||
0d00c7b2 | 1809 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
4732efbe | 1810 | if (match_futex (&this->key, &key1)) { |
aa10990e DH |
1811 | if (this->pi_state || this->rt_waiter) { |
1812 | ret = -EINVAL; | |
1813 | goto out_unlock; | |
1814 | } | |
1d0dcb3a | 1815 | mark_wake_futex(&wake_q, this); |
4732efbe JJ |
1816 | if (++ret >= nr_wake) |
1817 | break; | |
1818 | } | |
1819 | } | |
1820 | ||
1821 | if (op_ret > 0) { | |
4732efbe | 1822 | op_ret = 0; |
0d00c7b2 | 1823 | plist_for_each_entry_safe(this, next, &hb2->chain, list) { |
4732efbe | 1824 | if (match_futex (&this->key, &key2)) { |
aa10990e DH |
1825 | if (this->pi_state || this->rt_waiter) { |
1826 | ret = -EINVAL; | |
1827 | goto out_unlock; | |
1828 | } | |
1d0dcb3a | 1829 | mark_wake_futex(&wake_q, this); |
4732efbe JJ |
1830 | if (++op_ret >= nr_wake2) |
1831 | break; | |
1832 | } | |
1833 | } | |
1834 | ret += op_ret; | |
1835 | } | |
1836 | ||
aa10990e | 1837 | out_unlock: |
5eb3dc62 | 1838 | double_unlock_hb(hb1, hb2); |
1d0dcb3a | 1839 | wake_up_q(&wake_q); |
42d35d48 | 1840 | out_put_keys: |
ae791a2d | 1841 | put_futex_key(&key2); |
42d35d48 | 1842 | out_put_key1: |
ae791a2d | 1843 | put_futex_key(&key1); |
42d35d48 | 1844 | out: |
4732efbe JJ |
1845 | return ret; |
1846 | } | |
1847 | ||
9121e478 DH |
1848 | /** |
1849 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1850 | * @q: the futex_q to requeue | |
1851 | * @hb1: the source hash_bucket | |
1852 | * @hb2: the target hash_bucket | |
1853 | * @key2: the new key for the requeued futex_q | |
1854 | */ | |
1855 | static inline | |
1856 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1857 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1858 | { | |
1859 | ||
1860 | /* | |
1861 | * If key1 and key2 hash to the same bucket, no need to | |
1862 | * requeue. | |
1863 | */ | |
1864 | if (likely(&hb1->chain != &hb2->chain)) { | |
1865 | plist_del(&q->list, &hb1->chain); | |
11d4616b | 1866 | hb_waiters_dec(hb1); |
11d4616b | 1867 | hb_waiters_inc(hb2); |
fe1bce9e | 1868 | plist_add(&q->list, &hb2->chain); |
9121e478 | 1869 | q->lock_ptr = &hb2->lock; |
9121e478 DH |
1870 | } |
1871 | get_futex_key_refs(key2); | |
1872 | q->key = *key2; | |
1873 | } | |
1874 | ||
52400ba9 DH |
1875 | /** |
1876 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1877 | * @q: the futex_q |
1878 | * @key: the key of the requeue target futex | |
1879 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1880 | * |
1881 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1882 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1883 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1884 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1885 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1886 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1887 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1888 | */ |
1889 | static inline | |
beda2c7e DH |
1890 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1891 | struct futex_hash_bucket *hb) | |
52400ba9 | 1892 | { |
52400ba9 DH |
1893 | get_futex_key_refs(key); |
1894 | q->key = *key; | |
1895 | ||
2e12978a | 1896 | __unqueue_futex(q); |
52400ba9 DH |
1897 | |
1898 | WARN_ON(!q->rt_waiter); | |
1899 | q->rt_waiter = NULL; | |
1900 | ||
beda2c7e | 1901 | q->lock_ptr = &hb->lock; |
beda2c7e | 1902 | |
f1a11e05 | 1903 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1904 | } |
1905 | ||
1906 | /** | |
1907 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1908 | * @pifutex: the user address of the to futex |
1909 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1910 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1911 | * @key1: the from futex key | |
1912 | * @key2: the to futex key | |
1913 | * @ps: address to store the pi_state pointer | |
ad906e7e TG |
1914 | * @exiting: Pointer to store the task pointer of the owner task |
1915 | * which is in the middle of exiting | |
bab5bc9e | 1916 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) |
52400ba9 DH |
1917 | * |
1918 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1919 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1920 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1921 | * hb1 and hb2 must be held by the caller. | |
52400ba9 | 1922 | * |
ad906e7e TG |
1923 | * @exiting is only set when the return value is -EBUSY. If so, this holds |
1924 | * a refcount on the exiting task on return and the caller needs to drop it | |
1925 | * after waiting for the exit to complete. | |
1926 | * | |
6c23cbbd | 1927 | * Return: |
7b4ff1ad MCC |
1928 | * - 0 - failed to acquire the lock atomically; |
1929 | * - >0 - acquired the lock, return value is vpid of the top_waiter | |
1930 | * - <0 - error | |
52400ba9 | 1931 | */ |
ad906e7e TG |
1932 | static int |
1933 | futex_proxy_trylock_atomic(u32 __user *pifutex, struct futex_hash_bucket *hb1, | |
1934 | struct futex_hash_bucket *hb2, union futex_key *key1, | |
1935 | union futex_key *key2, struct futex_pi_state **ps, | |
1936 | struct task_struct **exiting, int set_waiters) | |
52400ba9 | 1937 | { |
bab5bc9e | 1938 | struct futex_q *top_waiter = NULL; |
52400ba9 | 1939 | u32 curval; |
866293ee | 1940 | int ret, vpid; |
52400ba9 DH |
1941 | |
1942 | if (get_futex_value_locked(&curval, pifutex)) | |
1943 | return -EFAULT; | |
1944 | ||
ab51fbab DB |
1945 | if (unlikely(should_fail_futex(true))) |
1946 | return -EFAULT; | |
1947 | ||
bab5bc9e DH |
1948 | /* |
1949 | * Find the top_waiter and determine if there are additional waiters. | |
1950 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1951 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1952 | * as we have means to handle the possible fault. If not, don't set | |
1953 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1954 | * the kernel. | |
1955 | */ | |
52400ba9 DH |
1956 | top_waiter = futex_top_waiter(hb1, key1); |
1957 | ||
1958 | /* There are no waiters, nothing for us to do. */ | |
1959 | if (!top_waiter) | |
1960 | return 0; | |
1961 | ||
84bc4af5 DH |
1962 | /* Ensure we requeue to the expected futex. */ |
1963 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1964 | return -EINVAL; | |
1965 | ||
52400ba9 | 1966 | /* |
bab5bc9e DH |
1967 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1968 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1969 | * in ps in contended cases. | |
52400ba9 | 1970 | */ |
866293ee | 1971 | vpid = task_pid_vnr(top_waiter->task); |
bab5bc9e | 1972 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
ad906e7e | 1973 | exiting, set_waiters); |
866293ee | 1974 | if (ret == 1) { |
beda2c7e | 1975 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
866293ee TG |
1976 | return vpid; |
1977 | } | |
52400ba9 DH |
1978 | return ret; |
1979 | } | |
1980 | ||
1981 | /** | |
1982 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1983 | * @uaddr1: source futex user address |
b41277dc | 1984 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1985 | * @uaddr2: target futex user address |
1986 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1987 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1988 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1989 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1990 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1991 | * |
1992 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1993 | * uaddr2 atomically on behalf of the top waiter. | |
1994 | * | |
6c23cbbd | 1995 | * Return: |
7b4ff1ad MCC |
1996 | * - >=0 - on success, the number of tasks requeued or woken; |
1997 | * - <0 - on error | |
1da177e4 | 1998 | */ |
b41277dc DH |
1999 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
2000 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
2001 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 2002 | { |
38d47c1b | 2003 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
2004 | int drop_count = 0, task_count = 0, ret; |
2005 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 2006 | struct futex_hash_bucket *hb1, *hb2; |
1da177e4 | 2007 | struct futex_q *this, *next; |
194a6b5b | 2008 | DEFINE_WAKE_Q(wake_q); |
52400ba9 | 2009 | |
fbe0e839 LJ |
2010 | if (nr_wake < 0 || nr_requeue < 0) |
2011 | return -EINVAL; | |
2012 | ||
bc2eecd7 NP |
2013 | /* |
2014 | * When PI not supported: return -ENOSYS if requeue_pi is true, | |
2015 | * consequently the compiler knows requeue_pi is always false past | |
2016 | * this point which will optimize away all the conditional code | |
2017 | * further down. | |
2018 | */ | |
2019 | if (!IS_ENABLED(CONFIG_FUTEX_PI) && requeue_pi) | |
2020 | return -ENOSYS; | |
2021 | ||
52400ba9 | 2022 | if (requeue_pi) { |
e9c243a5 TG |
2023 | /* |
2024 | * Requeue PI only works on two distinct uaddrs. This | |
2025 | * check is only valid for private futexes. See below. | |
2026 | */ | |
2027 | if (uaddr1 == uaddr2) | |
2028 | return -EINVAL; | |
2029 | ||
52400ba9 DH |
2030 | /* |
2031 | * requeue_pi requires a pi_state, try to allocate it now | |
2032 | * without any locks in case it fails. | |
2033 | */ | |
2034 | if (refill_pi_state_cache()) | |
2035 | return -ENOMEM; | |
2036 | /* | |
2037 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
2038 | * + nr_requeue, since it acquires the rt_mutex prior to | |
2039 | * returning to userspace, so as to not leave the rt_mutex with | |
2040 | * waiters and no owner. However, second and third wake-ups | |
2041 | * cannot be predicted as they involve race conditions with the | |
2042 | * first wake and a fault while looking up the pi_state. Both | |
2043 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
2044 | * use nr_wake=1. | |
2045 | */ | |
2046 | if (nr_wake != 1) | |
2047 | return -EINVAL; | |
2048 | } | |
1da177e4 | 2049 | |
42d35d48 | 2050 | retry: |
9ea71503 | 2051 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
1da177e4 LT |
2052 | if (unlikely(ret != 0)) |
2053 | goto out; | |
9ea71503 SB |
2054 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, |
2055 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 2056 | if (unlikely(ret != 0)) |
42d35d48 | 2057 | goto out_put_key1; |
1da177e4 | 2058 | |
e9c243a5 TG |
2059 | /* |
2060 | * The check above which compares uaddrs is not sufficient for | |
2061 | * shared futexes. We need to compare the keys: | |
2062 | */ | |
2063 | if (requeue_pi && match_futex(&key1, &key2)) { | |
2064 | ret = -EINVAL; | |
2065 | goto out_put_keys; | |
2066 | } | |
2067 | ||
e2970f2f IM |
2068 | hb1 = hash_futex(&key1); |
2069 | hb2 = hash_futex(&key2); | |
1da177e4 | 2070 | |
e4dc5b7a | 2071 | retry_private: |
69cd9eba | 2072 | hb_waiters_inc(hb2); |
8b8f319f | 2073 | double_lock_hb(hb1, hb2); |
1da177e4 | 2074 | |
e2970f2f IM |
2075 | if (likely(cmpval != NULL)) { |
2076 | u32 curval; | |
1da177e4 | 2077 | |
e2970f2f | 2078 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
2079 | |
2080 | if (unlikely(ret)) { | |
5eb3dc62 | 2081 | double_unlock_hb(hb1, hb2); |
69cd9eba | 2082 | hb_waiters_dec(hb2); |
1da177e4 | 2083 | |
e2970f2f | 2084 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
2085 | if (ret) |
2086 | goto out_put_keys; | |
1da177e4 | 2087 | |
b41277dc | 2088 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 2089 | goto retry_private; |
1da177e4 | 2090 | |
ae791a2d TG |
2091 | put_futex_key(&key2); |
2092 | put_futex_key(&key1); | |
e4dc5b7a | 2093 | goto retry; |
1da177e4 | 2094 | } |
e2970f2f | 2095 | if (curval != *cmpval) { |
1da177e4 LT |
2096 | ret = -EAGAIN; |
2097 | goto out_unlock; | |
2098 | } | |
2099 | } | |
2100 | ||
52400ba9 | 2101 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
ad906e7e TG |
2102 | struct task_struct *exiting = NULL; |
2103 | ||
bab5bc9e DH |
2104 | /* |
2105 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
2106 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
2107 | * bit. We force this here where we are able to easily handle | |
2108 | * faults rather in the requeue loop below. | |
2109 | */ | |
52400ba9 | 2110 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
ad906e7e TG |
2111 | &key2, &pi_state, |
2112 | &exiting, nr_requeue); | |
52400ba9 DH |
2113 | |
2114 | /* | |
2115 | * At this point the top_waiter has either taken uaddr2 or is | |
2116 | * waiting on it. If the former, then the pi_state will not | |
2117 | * exist yet, look it up one more time to ensure we have a | |
866293ee TG |
2118 | * reference to it. If the lock was taken, ret contains the |
2119 | * vpid of the top waiter task. | |
ecb38b78 TG |
2120 | * If the lock was not taken, we have pi_state and an initial |
2121 | * refcount on it. In case of an error we have nothing. | |
52400ba9 | 2122 | */ |
866293ee | 2123 | if (ret > 0) { |
52400ba9 | 2124 | WARN_ON(pi_state); |
89061d3d | 2125 | drop_count++; |
52400ba9 | 2126 | task_count++; |
866293ee | 2127 | /* |
ecb38b78 TG |
2128 | * If we acquired the lock, then the user space value |
2129 | * of uaddr2 should be vpid. It cannot be changed by | |
2130 | * the top waiter as it is blocked on hb2 lock if it | |
2131 | * tries to do so. If something fiddled with it behind | |
2132 | * our back the pi state lookup might unearth it. So | |
2133 | * we rather use the known value than rereading and | |
2134 | * handing potential crap to lookup_pi_state. | |
2135 | * | |
2136 | * If that call succeeds then we have pi_state and an | |
2137 | * initial refcount on it. | |
866293ee | 2138 | */ |
ad906e7e TG |
2139 | ret = lookup_pi_state(uaddr2, ret, hb2, &key2, |
2140 | &pi_state, &exiting); | |
52400ba9 DH |
2141 | } |
2142 | ||
2143 | switch (ret) { | |
2144 | case 0: | |
ecb38b78 | 2145 | /* We hold a reference on the pi state. */ |
52400ba9 | 2146 | break; |
4959f2de TG |
2147 | |
2148 | /* If the above failed, then pi_state is NULL */ | |
52400ba9 DH |
2149 | case -EFAULT: |
2150 | double_unlock_hb(hb1, hb2); | |
69cd9eba | 2151 | hb_waiters_dec(hb2); |
ae791a2d TG |
2152 | put_futex_key(&key2); |
2153 | put_futex_key(&key1); | |
d0725992 | 2154 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
2155 | if (!ret) |
2156 | goto retry; | |
2157 | goto out; | |
8ef775af | 2158 | case -EBUSY: |
52400ba9 | 2159 | case -EAGAIN: |
af54d6a1 TG |
2160 | /* |
2161 | * Two reasons for this: | |
8ef775af | 2162 | * - EBUSY: Owner is exiting and we just wait for the |
af54d6a1 | 2163 | * exit to complete. |
8ef775af | 2164 | * - EAGAIN: The user space value changed. |
af54d6a1 | 2165 | */ |
52400ba9 | 2166 | double_unlock_hb(hb1, hb2); |
69cd9eba | 2167 | hb_waiters_dec(hb2); |
ae791a2d TG |
2168 | put_futex_key(&key2); |
2169 | put_futex_key(&key1); | |
ad906e7e TG |
2170 | /* |
2171 | * Handle the case where the owner is in the middle of | |
2172 | * exiting. Wait for the exit to complete otherwise | |
2173 | * this task might loop forever, aka. live lock. | |
2174 | */ | |
2175 | wait_for_owner_exiting(ret, exiting); | |
52400ba9 DH |
2176 | cond_resched(); |
2177 | goto retry; | |
2178 | default: | |
2179 | goto out_unlock; | |
2180 | } | |
2181 | } | |
2182 | ||
0d00c7b2 | 2183 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
52400ba9 DH |
2184 | if (task_count - nr_wake >= nr_requeue) |
2185 | break; | |
2186 | ||
2187 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 2188 | continue; |
52400ba9 | 2189 | |
392741e0 DH |
2190 | /* |
2191 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
2192 | * be paired with each other and no other futex ops. | |
aa10990e DH |
2193 | * |
2194 | * We should never be requeueing a futex_q with a pi_state, | |
2195 | * which is awaiting a futex_unlock_pi(). | |
392741e0 DH |
2196 | */ |
2197 | if ((requeue_pi && !this->rt_waiter) || | |
aa10990e DH |
2198 | (!requeue_pi && this->rt_waiter) || |
2199 | this->pi_state) { | |
392741e0 DH |
2200 | ret = -EINVAL; |
2201 | break; | |
2202 | } | |
52400ba9 DH |
2203 | |
2204 | /* | |
2205 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
2206 | * lock, we already woke the top_waiter. If not, it will be | |
2207 | * woken by futex_unlock_pi(). | |
2208 | */ | |
2209 | if (++task_count <= nr_wake && !requeue_pi) { | |
1d0dcb3a | 2210 | mark_wake_futex(&wake_q, this); |
52400ba9 DH |
2211 | continue; |
2212 | } | |
1da177e4 | 2213 | |
84bc4af5 DH |
2214 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
2215 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
2216 | ret = -EINVAL; | |
2217 | break; | |
2218 | } | |
2219 | ||
52400ba9 DH |
2220 | /* |
2221 | * Requeue nr_requeue waiters and possibly one more in the case | |
2222 | * of requeue_pi if we couldn't acquire the lock atomically. | |
2223 | */ | |
2224 | if (requeue_pi) { | |
ecb38b78 TG |
2225 | /* |
2226 | * Prepare the waiter to take the rt_mutex. Take a | |
2227 | * refcount on the pi_state and store the pointer in | |
2228 | * the futex_q object of the waiter. | |
2229 | */ | |
bf92cf3a | 2230 | get_pi_state(pi_state); |
52400ba9 DH |
2231 | this->pi_state = pi_state; |
2232 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
2233 | this->rt_waiter, | |
c051b21f | 2234 | this->task); |
52400ba9 | 2235 | if (ret == 1) { |
ecb38b78 TG |
2236 | /* |
2237 | * We got the lock. We do neither drop the | |
2238 | * refcount on pi_state nor clear | |
2239 | * this->pi_state because the waiter needs the | |
2240 | * pi_state for cleaning up the user space | |
2241 | * value. It will drop the refcount after | |
2242 | * doing so. | |
2243 | */ | |
beda2c7e | 2244 | requeue_pi_wake_futex(this, &key2, hb2); |
89061d3d | 2245 | drop_count++; |
52400ba9 DH |
2246 | continue; |
2247 | } else if (ret) { | |
ecb38b78 TG |
2248 | /* |
2249 | * rt_mutex_start_proxy_lock() detected a | |
2250 | * potential deadlock when we tried to queue | |
2251 | * that waiter. Drop the pi_state reference | |
2252 | * which we took above and remove the pointer | |
2253 | * to the state from the waiters futex_q | |
2254 | * object. | |
2255 | */ | |
52400ba9 | 2256 | this->pi_state = NULL; |
29e9ee5d | 2257 | put_pi_state(pi_state); |
885c2cb7 TG |
2258 | /* |
2259 | * We stop queueing more waiters and let user | |
2260 | * space deal with the mess. | |
2261 | */ | |
2262 | break; | |
52400ba9 | 2263 | } |
1da177e4 | 2264 | } |
52400ba9 DH |
2265 | requeue_futex(this, hb1, hb2, &key2); |
2266 | drop_count++; | |
1da177e4 LT |
2267 | } |
2268 | ||
ecb38b78 TG |
2269 | /* |
2270 | * We took an extra initial reference to the pi_state either | |
2271 | * in futex_proxy_trylock_atomic() or in lookup_pi_state(). We | |
2272 | * need to drop it here again. | |
2273 | */ | |
29e9ee5d | 2274 | put_pi_state(pi_state); |
885c2cb7 TG |
2275 | |
2276 | out_unlock: | |
5eb3dc62 | 2277 | double_unlock_hb(hb1, hb2); |
1d0dcb3a | 2278 | wake_up_q(&wake_q); |
69cd9eba | 2279 | hb_waiters_dec(hb2); |
1da177e4 | 2280 | |
cd84a42f DH |
2281 | /* |
2282 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
2283 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
2284 | * one at key2 and updated their key pointer. We no longer need to | |
2285 | * hold the references to key1. | |
2286 | */ | |
1da177e4 | 2287 | while (--drop_count >= 0) |
9adef58b | 2288 | drop_futex_key_refs(&key1); |
1da177e4 | 2289 | |
42d35d48 | 2290 | out_put_keys: |
ae791a2d | 2291 | put_futex_key(&key2); |
42d35d48 | 2292 | out_put_key1: |
ae791a2d | 2293 | put_futex_key(&key1); |
42d35d48 | 2294 | out: |
52400ba9 | 2295 | return ret ? ret : task_count; |
1da177e4 LT |
2296 | } |
2297 | ||
2298 | /* The key must be already stored in q->key. */ | |
82af7aca | 2299 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 2300 | __acquires(&hb->lock) |
1da177e4 | 2301 | { |
e2970f2f | 2302 | struct futex_hash_bucket *hb; |
1da177e4 | 2303 | |
e2970f2f | 2304 | hb = hash_futex(&q->key); |
11d4616b LT |
2305 | |
2306 | /* | |
2307 | * Increment the counter before taking the lock so that | |
2308 | * a potential waker won't miss a to-be-slept task that is | |
2309 | * waiting for the spinlock. This is safe as all queue_lock() | |
2310 | * users end up calling queue_me(). Similarly, for housekeeping, | |
2311 | * decrement the counter at queue_unlock() when some error has | |
2312 | * occurred and we don't end up adding the task to the list. | |
2313 | */ | |
2314 | hb_waiters_inc(hb); | |
2315 | ||
e2970f2f | 2316 | q->lock_ptr = &hb->lock; |
1da177e4 | 2317 | |
8ad7b378 | 2318 | spin_lock(&hb->lock); /* implies smp_mb(); (A) */ |
e2970f2f | 2319 | return hb; |
1da177e4 LT |
2320 | } |
2321 | ||
d40d65c8 | 2322 | static inline void |
0d00c7b2 | 2323 | queue_unlock(struct futex_hash_bucket *hb) |
15e408cd | 2324 | __releases(&hb->lock) |
d40d65c8 DH |
2325 | { |
2326 | spin_unlock(&hb->lock); | |
11d4616b | 2327 | hb_waiters_dec(hb); |
d40d65c8 DH |
2328 | } |
2329 | ||
cfafcd11 | 2330 | static inline void __queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 2331 | { |
ec92d082 PP |
2332 | int prio; |
2333 | ||
2334 | /* | |
2335 | * The priority used to register this element is | |
2336 | * - either the real thread-priority for the real-time threads | |
2337 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
2338 | * - or MAX_RT_PRIO for non-RT threads. | |
2339 | * Thus, all RT-threads are woken first in priority order, and | |
2340 | * the others are woken last, in FIFO order. | |
2341 | */ | |
2342 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
2343 | ||
2344 | plist_node_init(&q->list, prio); | |
ec92d082 | 2345 | plist_add(&q->list, &hb->chain); |
c87e2837 | 2346 | q->task = current; |
cfafcd11 PZ |
2347 | } |
2348 | ||
2349 | /** | |
2350 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
2351 | * @q: The futex_q to enqueue | |
2352 | * @hb: The destination hash bucket | |
2353 | * | |
2354 | * The hb->lock must be held by the caller, and is released here. A call to | |
2355 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
2356 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
2357 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
2358 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
2359 | * an example). | |
2360 | */ | |
2361 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) | |
2362 | __releases(&hb->lock) | |
2363 | { | |
2364 | __queue_me(q, hb); | |
e2970f2f | 2365 | spin_unlock(&hb->lock); |
1da177e4 LT |
2366 | } |
2367 | ||
d40d65c8 DH |
2368 | /** |
2369 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
2370 | * @q: The futex_q to unqueue | |
2371 | * | |
2372 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
2373 | * be paired with exactly one earlier call to queue_me(). | |
2374 | * | |
6c23cbbd | 2375 | * Return: |
7b4ff1ad MCC |
2376 | * - 1 - if the futex_q was still queued (and we removed unqueued it); |
2377 | * - 0 - if the futex_q was already removed by the waking thread | |
1da177e4 | 2378 | */ |
1da177e4 LT |
2379 | static int unqueue_me(struct futex_q *q) |
2380 | { | |
1da177e4 | 2381 | spinlock_t *lock_ptr; |
e2970f2f | 2382 | int ret = 0; |
1da177e4 LT |
2383 | |
2384 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 2385 | retry: |
29b75eb2 JZ |
2386 | /* |
2387 | * q->lock_ptr can change between this read and the following spin_lock. | |
2388 | * Use READ_ONCE to forbid the compiler from reloading q->lock_ptr and | |
2389 | * optimizing lock_ptr out of the logic below. | |
2390 | */ | |
2391 | lock_ptr = READ_ONCE(q->lock_ptr); | |
c80544dc | 2392 | if (lock_ptr != NULL) { |
1da177e4 LT |
2393 | spin_lock(lock_ptr); |
2394 | /* | |
2395 | * q->lock_ptr can change between reading it and | |
2396 | * spin_lock(), causing us to take the wrong lock. This | |
2397 | * corrects the race condition. | |
2398 | * | |
2399 | * Reasoning goes like this: if we have the wrong lock, | |
2400 | * q->lock_ptr must have changed (maybe several times) | |
2401 | * between reading it and the spin_lock(). It can | |
2402 | * change again after the spin_lock() but only if it was | |
2403 | * already changed before the spin_lock(). It cannot, | |
2404 | * however, change back to the original value. Therefore | |
2405 | * we can detect whether we acquired the correct lock. | |
2406 | */ | |
2407 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
2408 | spin_unlock(lock_ptr); | |
2409 | goto retry; | |
2410 | } | |
2e12978a | 2411 | __unqueue_futex(q); |
c87e2837 IM |
2412 | |
2413 | BUG_ON(q->pi_state); | |
2414 | ||
1da177e4 LT |
2415 | spin_unlock(lock_ptr); |
2416 | ret = 1; | |
2417 | } | |
2418 | ||
9adef58b | 2419 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
2420 | return ret; |
2421 | } | |
2422 | ||
c87e2837 IM |
2423 | /* |
2424 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
2425 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
2426 | * and dropped here. | |
c87e2837 | 2427 | */ |
d0aa7a70 | 2428 | static void unqueue_me_pi(struct futex_q *q) |
15e408cd | 2429 | __releases(q->lock_ptr) |
c87e2837 | 2430 | { |
2e12978a | 2431 | __unqueue_futex(q); |
c87e2837 IM |
2432 | |
2433 | BUG_ON(!q->pi_state); | |
29e9ee5d | 2434 | put_pi_state(q->pi_state); |
c87e2837 IM |
2435 | q->pi_state = NULL; |
2436 | ||
d0aa7a70 | 2437 | spin_unlock(q->lock_ptr); |
c87e2837 IM |
2438 | } |
2439 | ||
778e9a9c | 2440 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
c1e2f0ea | 2441 | struct task_struct *argowner) |
d0aa7a70 | 2442 | { |
d0aa7a70 | 2443 | struct futex_pi_state *pi_state = q->pi_state; |
7cfdaf38 | 2444 | u32 uval, uninitialized_var(curval), newval; |
c1e2f0ea PZ |
2445 | struct task_struct *oldowner, *newowner; |
2446 | u32 newtid; | |
e5733e53 | 2447 | int ret, err = 0; |
d0aa7a70 | 2448 | |
c1e2f0ea PZ |
2449 | lockdep_assert_held(q->lock_ptr); |
2450 | ||
734009e9 PZ |
2451 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
2452 | ||
2453 | oldowner = pi_state->owner; | |
1b7558e4 TG |
2454 | |
2455 | /* | |
c1e2f0ea | 2456 | * We are here because either: |
16ffa12d | 2457 | * |
c1e2f0ea PZ |
2458 | * - we stole the lock and pi_state->owner needs updating to reflect |
2459 | * that (@argowner == current), | |
2460 | * | |
2461 | * or: | |
2462 | * | |
2463 | * - someone stole our lock and we need to fix things to point to the | |
2464 | * new owner (@argowner == NULL). | |
2465 | * | |
2466 | * Either way, we have to replace the TID in the user space variable. | |
8161239a | 2467 | * This must be atomic as we have to preserve the owner died bit here. |
1b7558e4 | 2468 | * |
b2d0994b DH |
2469 | * Note: We write the user space value _before_ changing the pi_state |
2470 | * because we can fault here. Imagine swapped out pages or a fork | |
2471 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 | 2472 | * |
734009e9 PZ |
2473 | * Modifying pi_state _before_ the user space value would leave the |
2474 | * pi_state in an inconsistent state when we fault here, because we | |
2475 | * need to drop the locks to handle the fault. This might be observed | |
2476 | * in the PID check in lookup_pi_state. | |
1b7558e4 TG |
2477 | */ |
2478 | retry: | |
c1e2f0ea PZ |
2479 | if (!argowner) { |
2480 | if (oldowner != current) { | |
2481 | /* | |
2482 | * We raced against a concurrent self; things are | |
2483 | * already fixed up. Nothing to do. | |
2484 | */ | |
2485 | ret = 0; | |
2486 | goto out_unlock; | |
2487 | } | |
2488 | ||
2489 | if (__rt_mutex_futex_trylock(&pi_state->pi_mutex)) { | |
2490 | /* We got the lock after all, nothing to fix. */ | |
2491 | ret = 0; | |
2492 | goto out_unlock; | |
2493 | } | |
2494 | ||
2495 | /* | |
2496 | * Since we just failed the trylock; there must be an owner. | |
2497 | */ | |
2498 | newowner = rt_mutex_owner(&pi_state->pi_mutex); | |
2499 | BUG_ON(!newowner); | |
2500 | } else { | |
2501 | WARN_ON_ONCE(argowner != current); | |
2502 | if (oldowner == current) { | |
2503 | /* | |
2504 | * We raced against a concurrent self; things are | |
2505 | * already fixed up. Nothing to do. | |
2506 | */ | |
2507 | ret = 0; | |
2508 | goto out_unlock; | |
2509 | } | |
2510 | newowner = argowner; | |
2511 | } | |
2512 | ||
2513 | newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; | |
a97cb0e7 PZ |
2514 | /* Owner died? */ |
2515 | if (!pi_state->owner) | |
2516 | newtid |= FUTEX_OWNER_DIED; | |
c1e2f0ea | 2517 | |
e5733e53 WD |
2518 | err = get_futex_value_locked(&uval, uaddr); |
2519 | if (err) | |
2520 | goto handle_err; | |
1b7558e4 | 2521 | |
16ffa12d | 2522 | for (;;) { |
1b7558e4 TG |
2523 | newval = (uval & FUTEX_OWNER_DIED) | newtid; |
2524 | ||
e5733e53 WD |
2525 | err = cmpxchg_futex_value_locked(&curval, uaddr, uval, newval); |
2526 | if (err) | |
2527 | goto handle_err; | |
2528 | ||
1b7558e4 TG |
2529 | if (curval == uval) |
2530 | break; | |
2531 | uval = curval; | |
2532 | } | |
2533 | ||
2534 | /* | |
2535 | * We fixed up user space. Now we need to fix the pi_state | |
2536 | * itself. | |
2537 | */ | |
d0aa7a70 | 2538 | if (pi_state->owner != NULL) { |
734009e9 | 2539 | raw_spin_lock(&pi_state->owner->pi_lock); |
d0aa7a70 PP |
2540 | WARN_ON(list_empty(&pi_state->list)); |
2541 | list_del_init(&pi_state->list); | |
734009e9 | 2542 | raw_spin_unlock(&pi_state->owner->pi_lock); |
1b7558e4 | 2543 | } |
d0aa7a70 | 2544 | |
cdf71a10 | 2545 | pi_state->owner = newowner; |
d0aa7a70 | 2546 | |
734009e9 | 2547 | raw_spin_lock(&newowner->pi_lock); |
d0aa7a70 | 2548 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 | 2549 | list_add(&pi_state->list, &newowner->pi_state_list); |
734009e9 PZ |
2550 | raw_spin_unlock(&newowner->pi_lock); |
2551 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); | |
2552 | ||
1b7558e4 | 2553 | return 0; |
d0aa7a70 | 2554 | |
d0aa7a70 | 2555 | /* |
e5733e53 WD |
2556 | * In order to reschedule or handle a page fault, we need to drop the |
2557 | * locks here. In the case of a fault, this gives the other task | |
2558 | * (either the highest priority waiter itself or the task which stole | |
2559 | * the rtmutex) the chance to try the fixup of the pi_state. So once we | |
2560 | * are back from handling the fault we need to check the pi_state after | |
2561 | * reacquiring the locks and before trying to do another fixup. When | |
2562 | * the fixup has been done already we simply return. | |
734009e9 PZ |
2563 | * |
2564 | * Note: we hold both hb->lock and pi_mutex->wait_lock. We can safely | |
2565 | * drop hb->lock since the caller owns the hb -> futex_q relation. | |
2566 | * Dropping the pi_mutex->wait_lock requires the state revalidate. | |
d0aa7a70 | 2567 | */ |
e5733e53 | 2568 | handle_err: |
734009e9 | 2569 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
1b7558e4 | 2570 | spin_unlock(q->lock_ptr); |
778e9a9c | 2571 | |
e5733e53 WD |
2572 | switch (err) { |
2573 | case -EFAULT: | |
2574 | ret = fault_in_user_writeable(uaddr); | |
2575 | break; | |
2576 | ||
2577 | case -EAGAIN: | |
2578 | cond_resched(); | |
2579 | ret = 0; | |
2580 | break; | |
2581 | ||
2582 | default: | |
2583 | WARN_ON_ONCE(1); | |
2584 | ret = err; | |
2585 | break; | |
2586 | } | |
778e9a9c | 2587 | |
1b7558e4 | 2588 | spin_lock(q->lock_ptr); |
734009e9 | 2589 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
778e9a9c | 2590 | |
1b7558e4 TG |
2591 | /* |
2592 | * Check if someone else fixed it for us: | |
2593 | */ | |
734009e9 PZ |
2594 | if (pi_state->owner != oldowner) { |
2595 | ret = 0; | |
2596 | goto out_unlock; | |
2597 | } | |
1b7558e4 TG |
2598 | |
2599 | if (ret) | |
734009e9 | 2600 | goto out_unlock; |
1b7558e4 TG |
2601 | |
2602 | goto retry; | |
734009e9 PZ |
2603 | |
2604 | out_unlock: | |
2605 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); | |
2606 | return ret; | |
d0aa7a70 PP |
2607 | } |
2608 | ||
72c1bbf3 | 2609 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 2610 | |
dd973998 DH |
2611 | /** |
2612 | * fixup_owner() - Post lock pi_state and corner case management | |
2613 | * @uaddr: user address of the futex | |
dd973998 DH |
2614 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
2615 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
2616 | * | |
2617 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
2618 | * the pi_state owner as well as handle race conditions that may allow us to | |
2619 | * acquire the lock. Must be called with the hb lock held. | |
2620 | * | |
6c23cbbd | 2621 | * Return: |
7b4ff1ad MCC |
2622 | * - 1 - success, lock taken; |
2623 | * - 0 - success, lock not taken; | |
2624 | * - <0 - on error (-EFAULT) | |
dd973998 | 2625 | */ |
ae791a2d | 2626 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 | 2627 | { |
dd973998 DH |
2628 | int ret = 0; |
2629 | ||
2630 | if (locked) { | |
2631 | /* | |
2632 | * Got the lock. We might not be the anticipated owner if we | |
2633 | * did a lock-steal - fix up the PI-state in that case: | |
16ffa12d | 2634 | * |
c1e2f0ea PZ |
2635 | * Speculative pi_state->owner read (we don't hold wait_lock); |
2636 | * since we own the lock pi_state->owner == current is the | |
2637 | * stable state, anything else needs more attention. | |
dd973998 DH |
2638 | */ |
2639 | if (q->pi_state->owner != current) | |
ae791a2d | 2640 | ret = fixup_pi_state_owner(uaddr, q, current); |
dd973998 DH |
2641 | goto out; |
2642 | } | |
2643 | ||
c1e2f0ea PZ |
2644 | /* |
2645 | * If we didn't get the lock; check if anybody stole it from us. In | |
2646 | * that case, we need to fix up the uval to point to them instead of | |
2647 | * us, otherwise bad things happen. [10] | |
2648 | * | |
2649 | * Another speculative read; pi_state->owner == current is unstable | |
2650 | * but needs our attention. | |
2651 | */ | |
2652 | if (q->pi_state->owner == current) { | |
2653 | ret = fixup_pi_state_owner(uaddr, q, NULL); | |
2654 | goto out; | |
2655 | } | |
2656 | ||
dd973998 DH |
2657 | /* |
2658 | * Paranoia check. If we did not take the lock, then we should not be | |
8161239a | 2659 | * the owner of the rt_mutex. |
dd973998 | 2660 | */ |
73d786bd | 2661 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) { |
dd973998 DH |
2662 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " |
2663 | "pi-state %p\n", ret, | |
2664 | q->pi_state->pi_mutex.owner, | |
2665 | q->pi_state->owner); | |
73d786bd | 2666 | } |
dd973998 DH |
2667 | |
2668 | out: | |
2669 | return ret ? ret : locked; | |
2670 | } | |
2671 | ||
ca5f9524 DH |
2672 | /** |
2673 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
2674 | * @hb: the futex hash bucket, must be locked by the caller | |
2675 | * @q: the futex_q to queue up on | |
2676 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
2677 | */ |
2678 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 2679 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 2680 | { |
9beba3c5 DH |
2681 | /* |
2682 | * The task state is guaranteed to be set before another task can | |
b92b8b35 | 2683 | * wake it. set_current_state() is implemented using smp_store_mb() and |
9beba3c5 DH |
2684 | * queue_me() calls spin_unlock() upon completion, both serializing |
2685 | * access to the hash list and forcing another memory barrier. | |
2686 | */ | |
f1a11e05 | 2687 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 2688 | queue_me(q, hb); |
ca5f9524 DH |
2689 | |
2690 | /* Arm the timer */ | |
2e4b0d3f | 2691 | if (timeout) |
ca5f9524 | 2692 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); |
ca5f9524 DH |
2693 | |
2694 | /* | |
0729e196 DH |
2695 | * If we have been removed from the hash list, then another task |
2696 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
2697 | */ |
2698 | if (likely(!plist_node_empty(&q->list))) { | |
2699 | /* | |
2700 | * If the timer has already expired, current will already be | |
2701 | * flagged for rescheduling. Only call schedule if there | |
2702 | * is no timeout, or if it has yet to expire. | |
2703 | */ | |
2704 | if (!timeout || timeout->task) | |
88c8004f | 2705 | freezable_schedule(); |
ca5f9524 DH |
2706 | } |
2707 | __set_current_state(TASK_RUNNING); | |
2708 | } | |
2709 | ||
f801073f DH |
2710 | /** |
2711 | * futex_wait_setup() - Prepare to wait on a futex | |
2712 | * @uaddr: the futex userspace address | |
2713 | * @val: the expected value | |
b41277dc | 2714 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
2715 | * @q: the associated futex_q |
2716 | * @hb: storage for hash_bucket pointer to be returned to caller | |
2717 | * | |
2718 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
2719 | * compare it with the expected value. Handle atomic faults internally. | |
2720 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
2721 | * with no q.key reference on failure. | |
2722 | * | |
6c23cbbd | 2723 | * Return: |
7b4ff1ad MCC |
2724 | * - 0 - uaddr contains val and hb has been locked; |
2725 | * - <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked | |
f801073f | 2726 | */ |
b41277dc | 2727 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 2728 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 2729 | { |
e2970f2f IM |
2730 | u32 uval; |
2731 | int ret; | |
1da177e4 | 2732 | |
1da177e4 | 2733 | /* |
b2d0994b | 2734 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
2735 | * Order is important: |
2736 | * | |
2737 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
2738 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
2739 | * | |
2740 | * The basic logical guarantee of a futex is that it blocks ONLY | |
2741 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
2742 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
2743 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
2744 | * cond(var) false, which would violate the guarantee. |
2745 | * | |
8fe8f545 ML |
2746 | * On the other hand, we insert q and release the hash-bucket only |
2747 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
2748 | * absorb a wakeup if *uaddr does not match the desired values | |
2749 | * while the syscall executes. | |
1da177e4 | 2750 | */ |
f801073f | 2751 | retry: |
9ea71503 | 2752 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ); |
f801073f | 2753 | if (unlikely(ret != 0)) |
a5a2a0c7 | 2754 | return ret; |
f801073f DH |
2755 | |
2756 | retry_private: | |
2757 | *hb = queue_lock(q); | |
2758 | ||
e2970f2f | 2759 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 2760 | |
f801073f | 2761 | if (ret) { |
0d00c7b2 | 2762 | queue_unlock(*hb); |
1da177e4 | 2763 | |
e2970f2f | 2764 | ret = get_user(uval, uaddr); |
e4dc5b7a | 2765 | if (ret) |
f801073f | 2766 | goto out; |
1da177e4 | 2767 | |
b41277dc | 2768 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2769 | goto retry_private; |
2770 | ||
ae791a2d | 2771 | put_futex_key(&q->key); |
e4dc5b7a | 2772 | goto retry; |
1da177e4 | 2773 | } |
ca5f9524 | 2774 | |
f801073f | 2775 | if (uval != val) { |
0d00c7b2 | 2776 | queue_unlock(*hb); |
f801073f | 2777 | ret = -EWOULDBLOCK; |
2fff78c7 | 2778 | } |
1da177e4 | 2779 | |
f801073f DH |
2780 | out: |
2781 | if (ret) | |
ae791a2d | 2782 | put_futex_key(&q->key); |
f801073f DH |
2783 | return ret; |
2784 | } | |
2785 | ||
b41277dc DH |
2786 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
2787 | ktime_t *abs_time, u32 bitset) | |
f801073f DH |
2788 | { |
2789 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
2790 | struct restart_block *restart; |
2791 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 2792 | struct futex_q q = futex_q_init; |
f801073f DH |
2793 | int ret; |
2794 | ||
2795 | if (!bitset) | |
2796 | return -EINVAL; | |
f801073f DH |
2797 | q.bitset = bitset; |
2798 | ||
2799 | if (abs_time) { | |
2800 | to = &timeout; | |
2801 | ||
b41277dc DH |
2802 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2803 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2804 | HRTIMER_MODE_ABS); | |
f801073f DH |
2805 | hrtimer_init_sleeper(to, current); |
2806 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2807 | current->timer_slack_ns); | |
2808 | } | |
2809 | ||
d58e6576 | 2810 | retry: |
7ada876a DH |
2811 | /* |
2812 | * Prepare to wait on uaddr. On success, holds hb lock and increments | |
2813 | * q.key refs. | |
2814 | */ | |
b41277dc | 2815 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
2816 | if (ret) |
2817 | goto out; | |
2818 | ||
ca5f9524 | 2819 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 2820 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
2821 | |
2822 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 2823 | ret = 0; |
7ada876a | 2824 | /* unqueue_me() drops q.key ref */ |
1da177e4 | 2825 | if (!unqueue_me(&q)) |
7ada876a | 2826 | goto out; |
2fff78c7 | 2827 | ret = -ETIMEDOUT; |
ca5f9524 | 2828 | if (to && !to->task) |
7ada876a | 2829 | goto out; |
72c1bbf3 | 2830 | |
e2970f2f | 2831 | /* |
d58e6576 TG |
2832 | * We expect signal_pending(current), but we might be the |
2833 | * victim of a spurious wakeup as well. | |
e2970f2f | 2834 | */ |
7ada876a | 2835 | if (!signal_pending(current)) |
d58e6576 | 2836 | goto retry; |
d58e6576 | 2837 | |
2fff78c7 | 2838 | ret = -ERESTARTSYS; |
c19384b5 | 2839 | if (!abs_time) |
7ada876a | 2840 | goto out; |
1da177e4 | 2841 | |
f56141e3 | 2842 | restart = ¤t->restart_block; |
2fff78c7 | 2843 | restart->fn = futex_wait_restart; |
a3c74c52 | 2844 | restart->futex.uaddr = uaddr; |
2fff78c7 | 2845 | restart->futex.val = val; |
2456e855 | 2846 | restart->futex.time = *abs_time; |
2fff78c7 | 2847 | restart->futex.bitset = bitset; |
0cd9c649 | 2848 | restart->futex.flags = flags | FLAGS_HAS_TIMEOUT; |
42d35d48 | 2849 | |
2fff78c7 PZ |
2850 | ret = -ERESTART_RESTARTBLOCK; |
2851 | ||
42d35d48 | 2852 | out: |
ca5f9524 DH |
2853 | if (to) { |
2854 | hrtimer_cancel(&to->timer); | |
2855 | destroy_hrtimer_on_stack(&to->timer); | |
2856 | } | |
c87e2837 IM |
2857 | return ret; |
2858 | } | |
2859 | ||
72c1bbf3 NP |
2860 | |
2861 | static long futex_wait_restart(struct restart_block *restart) | |
2862 | { | |
a3c74c52 | 2863 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 2864 | ktime_t t, *tp = NULL; |
72c1bbf3 | 2865 | |
a72188d8 | 2866 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
2456e855 | 2867 | t = restart->futex.time; |
a72188d8 DH |
2868 | tp = &t; |
2869 | } | |
72c1bbf3 | 2870 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
2871 | |
2872 | return (long)futex_wait(uaddr, restart->futex.flags, | |
2873 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
2874 | } |
2875 | ||
2876 | ||
c87e2837 IM |
2877 | /* |
2878 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
2879 | * and failed. The kernel side here does the whole locking operation: | |
767f509c DB |
2880 | * if there are waiters then it will block as a consequence of relying |
2881 | * on rt-mutexes, it does PI, etc. (Due to races the kernel might see | |
2882 | * a 0 value of the futex too.). | |
2883 | * | |
2884 | * Also serves as futex trylock_pi()'ing, and due semantics. | |
c87e2837 | 2885 | */ |
996636dd | 2886 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, |
b41277dc | 2887 | ktime_t *time, int trylock) |
c87e2837 | 2888 | { |
c5780e97 | 2889 | struct hrtimer_sleeper timeout, *to = NULL; |
16ffa12d | 2890 | struct futex_pi_state *pi_state = NULL; |
ad906e7e | 2891 | struct task_struct *exiting = NULL; |
cfafcd11 | 2892 | struct rt_mutex_waiter rt_waiter; |
c87e2837 | 2893 | struct futex_hash_bucket *hb; |
5bdb05f9 | 2894 | struct futex_q q = futex_q_init; |
dd973998 | 2895 | int res, ret; |
c87e2837 | 2896 | |
bc2eecd7 NP |
2897 | if (!IS_ENABLED(CONFIG_FUTEX_PI)) |
2898 | return -ENOSYS; | |
2899 | ||
c87e2837 IM |
2900 | if (refill_pi_state_cache()) |
2901 | return -ENOMEM; | |
2902 | ||
c19384b5 | 2903 | if (time) { |
c5780e97 | 2904 | to = &timeout; |
237fc6e7 TG |
2905 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
2906 | HRTIMER_MODE_ABS); | |
c5780e97 | 2907 | hrtimer_init_sleeper(to, current); |
cc584b21 | 2908 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
2909 | } |
2910 | ||
42d35d48 | 2911 | retry: |
9ea71503 | 2912 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE); |
c87e2837 | 2913 | if (unlikely(ret != 0)) |
42d35d48 | 2914 | goto out; |
c87e2837 | 2915 | |
e4dc5b7a | 2916 | retry_private: |
82af7aca | 2917 | hb = queue_lock(&q); |
c87e2837 | 2918 | |
ad906e7e TG |
2919 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, |
2920 | &exiting, 0); | |
c87e2837 | 2921 | if (unlikely(ret)) { |
767f509c DB |
2922 | /* |
2923 | * Atomic work succeeded and we got the lock, | |
2924 | * or failed. Either way, we do _not_ block. | |
2925 | */ | |
778e9a9c | 2926 | switch (ret) { |
1a52084d DH |
2927 | case 1: |
2928 | /* We got the lock. */ | |
2929 | ret = 0; | |
2930 | goto out_unlock_put_key; | |
2931 | case -EFAULT: | |
2932 | goto uaddr_faulted; | |
8ef775af | 2933 | case -EBUSY: |
778e9a9c AK |
2934 | case -EAGAIN: |
2935 | /* | |
af54d6a1 | 2936 | * Two reasons for this: |
8ef775af | 2937 | * - EBUSY: Task is exiting and we just wait for the |
af54d6a1 | 2938 | * exit to complete. |
8ef775af | 2939 | * - EAGAIN: The user space value changed. |
778e9a9c | 2940 | */ |
0d00c7b2 | 2941 | queue_unlock(hb); |
ae791a2d | 2942 | put_futex_key(&q.key); |
ad906e7e TG |
2943 | /* |
2944 | * Handle the case where the owner is in the middle of | |
2945 | * exiting. Wait for the exit to complete otherwise | |
2946 | * this task might loop forever, aka. live lock. | |
2947 | */ | |
2948 | wait_for_owner_exiting(ret, exiting); | |
778e9a9c AK |
2949 | cond_resched(); |
2950 | goto retry; | |
778e9a9c | 2951 | default: |
42d35d48 | 2952 | goto out_unlock_put_key; |
c87e2837 | 2953 | } |
c87e2837 IM |
2954 | } |
2955 | ||
cfafcd11 PZ |
2956 | WARN_ON(!q.pi_state); |
2957 | ||
c87e2837 IM |
2958 | /* |
2959 | * Only actually queue now that the atomic ops are done: | |
2960 | */ | |
cfafcd11 | 2961 | __queue_me(&q, hb); |
c87e2837 | 2962 | |
cfafcd11 | 2963 | if (trylock) { |
5293c2ef | 2964 | ret = rt_mutex_futex_trylock(&q.pi_state->pi_mutex); |
c87e2837 IM |
2965 | /* Fixup the trylock return value: */ |
2966 | ret = ret ? 0 : -EWOULDBLOCK; | |
cfafcd11 | 2967 | goto no_block; |
c87e2837 IM |
2968 | } |
2969 | ||
56222b21 PZ |
2970 | rt_mutex_init_waiter(&rt_waiter); |
2971 | ||
cfafcd11 | 2972 | /* |
56222b21 PZ |
2973 | * On PREEMPT_RT_FULL, when hb->lock becomes an rt_mutex, we must not |
2974 | * hold it while doing rt_mutex_start_proxy(), because then it will | |
2975 | * include hb->lock in the blocking chain, even through we'll not in | |
2976 | * fact hold it while blocking. This will lead it to report -EDEADLK | |
2977 | * and BUG when futex_unlock_pi() interleaves with this. | |
2978 | * | |
2979 | * Therefore acquire wait_lock while holding hb->lock, but drop the | |
6e37a39c TG |
2980 | * latter before calling __rt_mutex_start_proxy_lock(). This |
2981 | * interleaves with futex_unlock_pi() -- which does a similar lock | |
2982 | * handoff -- such that the latter can observe the futex_q::pi_state | |
2983 | * before __rt_mutex_start_proxy_lock() is done. | |
cfafcd11 | 2984 | */ |
56222b21 PZ |
2985 | raw_spin_lock_irq(&q.pi_state->pi_mutex.wait_lock); |
2986 | spin_unlock(q.lock_ptr); | |
6e37a39c TG |
2987 | /* |
2988 | * __rt_mutex_start_proxy_lock() unconditionally enqueues the @rt_waiter | |
2989 | * such that futex_unlock_pi() is guaranteed to observe the waiter when | |
2990 | * it sees the futex_q::pi_state. | |
2991 | */ | |
56222b21 PZ |
2992 | ret = __rt_mutex_start_proxy_lock(&q.pi_state->pi_mutex, &rt_waiter, current); |
2993 | raw_spin_unlock_irq(&q.pi_state->pi_mutex.wait_lock); | |
2994 | ||
cfafcd11 PZ |
2995 | if (ret) { |
2996 | if (ret == 1) | |
2997 | ret = 0; | |
6e37a39c | 2998 | goto cleanup; |
cfafcd11 PZ |
2999 | } |
3000 | ||
cfafcd11 PZ |
3001 | if (unlikely(to)) |
3002 | hrtimer_start_expires(&to->timer, HRTIMER_MODE_ABS); | |
3003 | ||
3004 | ret = rt_mutex_wait_proxy_lock(&q.pi_state->pi_mutex, to, &rt_waiter); | |
3005 | ||
6e37a39c | 3006 | cleanup: |
a99e4e41 | 3007 | spin_lock(q.lock_ptr); |
cfafcd11 | 3008 | /* |
6e37a39c | 3009 | * If we failed to acquire the lock (deadlock/signal/timeout), we must |
cfafcd11 | 3010 | * first acquire the hb->lock before removing the lock from the |
6e37a39c TG |
3011 | * rt_mutex waitqueue, such that we can keep the hb and rt_mutex wait |
3012 | * lists consistent. | |
56222b21 PZ |
3013 | * |
3014 | * In particular; it is important that futex_unlock_pi() can not | |
3015 | * observe this inconsistency. | |
cfafcd11 PZ |
3016 | */ |
3017 | if (ret && !rt_mutex_cleanup_proxy_lock(&q.pi_state->pi_mutex, &rt_waiter)) | |
3018 | ret = 0; | |
3019 | ||
3020 | no_block: | |
dd973998 DH |
3021 | /* |
3022 | * Fixup the pi_state owner and possibly acquire the lock if we | |
3023 | * haven't already. | |
3024 | */ | |
ae791a2d | 3025 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 DH |
3026 | /* |
3027 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
3028 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
3029 | */ | |
3030 | if (res) | |
3031 | ret = (res < 0) ? res : 0; | |
c87e2837 | 3032 | |
e8f6386c | 3033 | /* |
dd973998 DH |
3034 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
3035 | * it and return the fault to userspace. | |
e8f6386c | 3036 | */ |
16ffa12d PZ |
3037 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) { |
3038 | pi_state = q.pi_state; | |
3039 | get_pi_state(pi_state); | |
3040 | } | |
e8f6386c | 3041 | |
778e9a9c AK |
3042 | /* Unqueue and drop the lock */ |
3043 | unqueue_me_pi(&q); | |
c87e2837 | 3044 | |
16ffa12d PZ |
3045 | if (pi_state) { |
3046 | rt_mutex_futex_unlock(&pi_state->pi_mutex); | |
3047 | put_pi_state(pi_state); | |
3048 | } | |
3049 | ||
5ecb01cf | 3050 | goto out_put_key; |
c87e2837 | 3051 | |
42d35d48 | 3052 | out_unlock_put_key: |
0d00c7b2 | 3053 | queue_unlock(hb); |
c87e2837 | 3054 | |
42d35d48 | 3055 | out_put_key: |
ae791a2d | 3056 | put_futex_key(&q.key); |
42d35d48 | 3057 | out: |
97181f9b TG |
3058 | if (to) { |
3059 | hrtimer_cancel(&to->timer); | |
237fc6e7 | 3060 | destroy_hrtimer_on_stack(&to->timer); |
97181f9b | 3061 | } |
dd973998 | 3062 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 3063 | |
42d35d48 | 3064 | uaddr_faulted: |
0d00c7b2 | 3065 | queue_unlock(hb); |
778e9a9c | 3066 | |
d0725992 | 3067 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
3068 | if (ret) |
3069 | goto out_put_key; | |
c87e2837 | 3070 | |
b41277dc | 3071 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
3072 | goto retry_private; |
3073 | ||
ae791a2d | 3074 | put_futex_key(&q.key); |
e4dc5b7a | 3075 | goto retry; |
c87e2837 IM |
3076 | } |
3077 | ||
c87e2837 IM |
3078 | /* |
3079 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
3080 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
3081 | * and do the rt-mutex unlock. | |
3082 | */ | |
b41277dc | 3083 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 | 3084 | { |
ccf9e6a8 | 3085 | u32 uninitialized_var(curval), uval, vpid = task_pid_vnr(current); |
38d47c1b | 3086 | union futex_key key = FUTEX_KEY_INIT; |
ccf9e6a8 | 3087 | struct futex_hash_bucket *hb; |
499f5aca | 3088 | struct futex_q *top_waiter; |
e4dc5b7a | 3089 | int ret; |
c87e2837 | 3090 | |
bc2eecd7 NP |
3091 | if (!IS_ENABLED(CONFIG_FUTEX_PI)) |
3092 | return -ENOSYS; | |
3093 | ||
c87e2837 IM |
3094 | retry: |
3095 | if (get_user(uval, uaddr)) | |
3096 | return -EFAULT; | |
3097 | /* | |
3098 | * We release only a lock we actually own: | |
3099 | */ | |
c0c9ed15 | 3100 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 3101 | return -EPERM; |
c87e2837 | 3102 | |
9ea71503 | 3103 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE); |
ccf9e6a8 TG |
3104 | if (ret) |
3105 | return ret; | |
c87e2837 IM |
3106 | |
3107 | hb = hash_futex(&key); | |
3108 | spin_lock(&hb->lock); | |
3109 | ||
c87e2837 | 3110 | /* |
ccf9e6a8 TG |
3111 | * Check waiters first. We do not trust user space values at |
3112 | * all and we at least want to know if user space fiddled | |
3113 | * with the futex value instead of blindly unlocking. | |
c87e2837 | 3114 | */ |
499f5aca PZ |
3115 | top_waiter = futex_top_waiter(hb, &key); |
3116 | if (top_waiter) { | |
16ffa12d PZ |
3117 | struct futex_pi_state *pi_state = top_waiter->pi_state; |
3118 | ||
3119 | ret = -EINVAL; | |
3120 | if (!pi_state) | |
3121 | goto out_unlock; | |
3122 | ||
3123 | /* | |
3124 | * If current does not own the pi_state then the futex is | |
3125 | * inconsistent and user space fiddled with the futex value. | |
3126 | */ | |
3127 | if (pi_state->owner != current) | |
3128 | goto out_unlock; | |
3129 | ||
bebe5b51 | 3130 | get_pi_state(pi_state); |
802ab58d | 3131 | /* |
bebe5b51 PZ |
3132 | * By taking wait_lock while still holding hb->lock, we ensure |
3133 | * there is no point where we hold neither; and therefore | |
3134 | * wake_futex_pi() must observe a state consistent with what we | |
3135 | * observed. | |
6e37a39c TG |
3136 | * |
3137 | * In particular; this forces __rt_mutex_start_proxy() to | |
3138 | * complete such that we're guaranteed to observe the | |
3139 | * rt_waiter. Also see the WARN in wake_futex_pi(). | |
16ffa12d | 3140 | */ |
bebe5b51 | 3141 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
16ffa12d PZ |
3142 | spin_unlock(&hb->lock); |
3143 | ||
c74aef2d | 3144 | /* drops pi_state->pi_mutex.wait_lock */ |
16ffa12d PZ |
3145 | ret = wake_futex_pi(uaddr, uval, pi_state); |
3146 | ||
3147 | put_pi_state(pi_state); | |
3148 | ||
3149 | /* | |
3150 | * Success, we're done! No tricky corner cases. | |
802ab58d SAS |
3151 | */ |
3152 | if (!ret) | |
3153 | goto out_putkey; | |
c87e2837 | 3154 | /* |
ccf9e6a8 TG |
3155 | * The atomic access to the futex value generated a |
3156 | * pagefault, so retry the user-access and the wakeup: | |
c87e2837 IM |
3157 | */ |
3158 | if (ret == -EFAULT) | |
3159 | goto pi_faulted; | |
89e9e66b SAS |
3160 | /* |
3161 | * A unconditional UNLOCK_PI op raced against a waiter | |
3162 | * setting the FUTEX_WAITERS bit. Try again. | |
3163 | */ | |
e5733e53 WD |
3164 | if (ret == -EAGAIN) |
3165 | goto pi_retry; | |
802ab58d SAS |
3166 | /* |
3167 | * wake_futex_pi has detected invalid state. Tell user | |
3168 | * space. | |
3169 | */ | |
16ffa12d | 3170 | goto out_putkey; |
c87e2837 | 3171 | } |
ccf9e6a8 | 3172 | |
c87e2837 | 3173 | /* |
ccf9e6a8 TG |
3174 | * We have no kernel internal state, i.e. no waiters in the |
3175 | * kernel. Waiters which are about to queue themselves are stuck | |
3176 | * on hb->lock. So we can safely ignore them. We do neither | |
3177 | * preserve the WAITERS bit not the OWNER_DIED one. We are the | |
3178 | * owner. | |
c87e2837 | 3179 | */ |
e5733e53 | 3180 | if ((ret = cmpxchg_futex_value_locked(&curval, uaddr, uval, 0))) { |
16ffa12d | 3181 | spin_unlock(&hb->lock); |
e5733e53 WD |
3182 | switch (ret) { |
3183 | case -EFAULT: | |
3184 | goto pi_faulted; | |
3185 | ||
3186 | case -EAGAIN: | |
3187 | goto pi_retry; | |
3188 | ||
3189 | default: | |
3190 | WARN_ON_ONCE(1); | |
3191 | goto out_putkey; | |
3192 | } | |
16ffa12d | 3193 | } |
c87e2837 | 3194 | |
ccf9e6a8 TG |
3195 | /* |
3196 | * If uval has changed, let user space handle it. | |
3197 | */ | |
3198 | ret = (curval == uval) ? 0 : -EAGAIN; | |
3199 | ||
c87e2837 IM |
3200 | out_unlock: |
3201 | spin_unlock(&hb->lock); | |
802ab58d | 3202 | out_putkey: |
ae791a2d | 3203 | put_futex_key(&key); |
c87e2837 IM |
3204 | return ret; |
3205 | ||
e5733e53 WD |
3206 | pi_retry: |
3207 | put_futex_key(&key); | |
3208 | cond_resched(); | |
3209 | goto retry; | |
3210 | ||
c87e2837 | 3211 | pi_faulted: |
ae791a2d | 3212 | put_futex_key(&key); |
c87e2837 | 3213 | |
d0725992 | 3214 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 3215 | if (!ret) |
c87e2837 IM |
3216 | goto retry; |
3217 | ||
1da177e4 LT |
3218 | return ret; |
3219 | } | |
3220 | ||
52400ba9 DH |
3221 | /** |
3222 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
3223 | * @hb: the hash_bucket futex_q was original enqueued on | |
3224 | * @q: the futex_q woken while waiting to be requeued | |
3225 | * @key2: the futex_key of the requeue target futex | |
3226 | * @timeout: the timeout associated with the wait (NULL if none) | |
3227 | * | |
3228 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
3229 | * target futex. If so, determine if it was a timeout or a signal that caused | |
3230 | * the wakeup and return the appropriate error code to the caller. Must be | |
3231 | * called with the hb lock held. | |
3232 | * | |
6c23cbbd | 3233 | * Return: |
7b4ff1ad MCC |
3234 | * - 0 = no early wakeup detected; |
3235 | * - <0 = -ETIMEDOUT or -ERESTARTNOINTR | |
52400ba9 DH |
3236 | */ |
3237 | static inline | |
3238 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
3239 | struct futex_q *q, union futex_key *key2, | |
3240 | struct hrtimer_sleeper *timeout) | |
3241 | { | |
3242 | int ret = 0; | |
3243 | ||
3244 | /* | |
3245 | * With the hb lock held, we avoid races while we process the wakeup. | |
3246 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
3247 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
3248 | * It can't be requeued from uaddr2 to something else since we don't | |
3249 | * support a PI aware source futex for requeue. | |
3250 | */ | |
3251 | if (!match_futex(&q->key, key2)) { | |
3252 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
3253 | /* | |
3254 | * We were woken prior to requeue by a timeout or a signal. | |
3255 | * Unqueue the futex_q and determine which it was. | |
3256 | */ | |
2e12978a | 3257 | plist_del(&q->list, &hb->chain); |
11d4616b | 3258 | hb_waiters_dec(hb); |
52400ba9 | 3259 | |
d58e6576 | 3260 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 3261 | ret = -EWOULDBLOCK; |
52400ba9 DH |
3262 | if (timeout && !timeout->task) |
3263 | ret = -ETIMEDOUT; | |
d58e6576 | 3264 | else if (signal_pending(current)) |
1c840c14 | 3265 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
3266 | } |
3267 | return ret; | |
3268 | } | |
3269 | ||
3270 | /** | |
3271 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 3272 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 3273 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
ab51fbab | 3274 | * the same type, no requeueing from private to shared, etc. |
52400ba9 DH |
3275 | * @val: the expected value of uaddr |
3276 | * @abs_time: absolute timeout | |
56ec1607 | 3277 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
3278 | * @uaddr2: the pi futex we will take prior to returning to user-space |
3279 | * | |
3280 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
6f7b0a2a DH |
3281 | * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake |
3282 | * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to | |
3283 | * userspace. This ensures the rt_mutex maintains an owner when it has waiters; | |
3284 | * without one, the pi logic would not know which task to boost/deboost, if | |
3285 | * there was a need to. | |
52400ba9 DH |
3286 | * |
3287 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
6c23cbbd | 3288 | * via the following-- |
52400ba9 | 3289 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() |
cc6db4e6 DH |
3290 | * 2) wakeup on uaddr2 after a requeue |
3291 | * 3) signal | |
3292 | * 4) timeout | |
52400ba9 | 3293 | * |
cc6db4e6 | 3294 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
3295 | * |
3296 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
3297 | * 5) successful lock | |
3298 | * 6) signal | |
3299 | * 7) timeout | |
3300 | * 8) other lock acquisition failure | |
3301 | * | |
cc6db4e6 | 3302 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
3303 | * |
3304 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
3305 | * | |
6c23cbbd | 3306 | * Return: |
7b4ff1ad MCC |
3307 | * - 0 - On success; |
3308 | * - <0 - On error | |
52400ba9 | 3309 | */ |
b41277dc | 3310 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 3311 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 3312 | u32 __user *uaddr2) |
52400ba9 DH |
3313 | { |
3314 | struct hrtimer_sleeper timeout, *to = NULL; | |
16ffa12d | 3315 | struct futex_pi_state *pi_state = NULL; |
52400ba9 | 3316 | struct rt_mutex_waiter rt_waiter; |
52400ba9 | 3317 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
3318 | union futex_key key2 = FUTEX_KEY_INIT; |
3319 | struct futex_q q = futex_q_init; | |
52400ba9 | 3320 | int res, ret; |
52400ba9 | 3321 | |
bc2eecd7 NP |
3322 | if (!IS_ENABLED(CONFIG_FUTEX_PI)) |
3323 | return -ENOSYS; | |
3324 | ||
6f7b0a2a DH |
3325 | if (uaddr == uaddr2) |
3326 | return -EINVAL; | |
3327 | ||
52400ba9 DH |
3328 | if (!bitset) |
3329 | return -EINVAL; | |
3330 | ||
3331 | if (abs_time) { | |
3332 | to = &timeout; | |
b41277dc DH |
3333 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
3334 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
3335 | HRTIMER_MODE_ABS); | |
52400ba9 DH |
3336 | hrtimer_init_sleeper(to, current); |
3337 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
3338 | current->timer_slack_ns); | |
3339 | } | |
3340 | ||
3341 | /* | |
3342 | * The waiter is allocated on our stack, manipulated by the requeue | |
3343 | * code while we sleep on uaddr. | |
3344 | */ | |
50809358 | 3345 | rt_mutex_init_waiter(&rt_waiter); |
52400ba9 | 3346 | |
9ea71503 | 3347 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
52400ba9 DH |
3348 | if (unlikely(ret != 0)) |
3349 | goto out; | |
3350 | ||
84bc4af5 DH |
3351 | q.bitset = bitset; |
3352 | q.rt_waiter = &rt_waiter; | |
3353 | q.requeue_pi_key = &key2; | |
3354 | ||
7ada876a DH |
3355 | /* |
3356 | * Prepare to wait on uaddr. On success, increments q.key (key1) ref | |
3357 | * count. | |
3358 | */ | |
b41277dc | 3359 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 TG |
3360 | if (ret) |
3361 | goto out_key2; | |
52400ba9 | 3362 | |
e9c243a5 TG |
3363 | /* |
3364 | * The check above which compares uaddrs is not sufficient for | |
3365 | * shared futexes. We need to compare the keys: | |
3366 | */ | |
3367 | if (match_futex(&q.key, &key2)) { | |
13c42c2f | 3368 | queue_unlock(hb); |
e9c243a5 TG |
3369 | ret = -EINVAL; |
3370 | goto out_put_keys; | |
3371 | } | |
3372 | ||
52400ba9 | 3373 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ |
f1a11e05 | 3374 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
3375 | |
3376 | spin_lock(&hb->lock); | |
3377 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
3378 | spin_unlock(&hb->lock); | |
3379 | if (ret) | |
3380 | goto out_put_keys; | |
3381 | ||
3382 | /* | |
3383 | * In order for us to be here, we know our q.key == key2, and since | |
3384 | * we took the hb->lock above, we also know that futex_requeue() has | |
3385 | * completed and we no longer have to concern ourselves with a wakeup | |
7ada876a DH |
3386 | * race with the atomic proxy lock acquisition by the requeue code. The |
3387 | * futex_requeue dropped our key1 reference and incremented our key2 | |
3388 | * reference count. | |
52400ba9 DH |
3389 | */ |
3390 | ||
3391 | /* Check if the requeue code acquired the second futex for us. */ | |
3392 | if (!q.rt_waiter) { | |
3393 | /* | |
3394 | * Got the lock. We might not be the anticipated owner if we | |
3395 | * did a lock-steal - fix up the PI-state in that case. | |
3396 | */ | |
3397 | if (q.pi_state && (q.pi_state->owner != current)) { | |
3398 | spin_lock(q.lock_ptr); | |
ae791a2d | 3399 | ret = fixup_pi_state_owner(uaddr2, &q, current); |
16ffa12d PZ |
3400 | if (ret && rt_mutex_owner(&q.pi_state->pi_mutex) == current) { |
3401 | pi_state = q.pi_state; | |
3402 | get_pi_state(pi_state); | |
3403 | } | |
fb75a428 TG |
3404 | /* |
3405 | * Drop the reference to the pi state which | |
3406 | * the requeue_pi() code acquired for us. | |
3407 | */ | |
29e9ee5d | 3408 | put_pi_state(q.pi_state); |
52400ba9 DH |
3409 | spin_unlock(q.lock_ptr); |
3410 | } | |
3411 | } else { | |
c236c8e9 PZ |
3412 | struct rt_mutex *pi_mutex; |
3413 | ||
52400ba9 DH |
3414 | /* |
3415 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
3416 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
3417 | * the pi_state. | |
3418 | */ | |
f27071cb | 3419 | WARN_ON(!q.pi_state); |
52400ba9 | 3420 | pi_mutex = &q.pi_state->pi_mutex; |
38d589f2 | 3421 | ret = rt_mutex_wait_proxy_lock(pi_mutex, to, &rt_waiter); |
52400ba9 DH |
3422 | |
3423 | spin_lock(q.lock_ptr); | |
38d589f2 PZ |
3424 | if (ret && !rt_mutex_cleanup_proxy_lock(pi_mutex, &rt_waiter)) |
3425 | ret = 0; | |
3426 | ||
3427 | debug_rt_mutex_free_waiter(&rt_waiter); | |
52400ba9 DH |
3428 | /* |
3429 | * Fixup the pi_state owner and possibly acquire the lock if we | |
3430 | * haven't already. | |
3431 | */ | |
ae791a2d | 3432 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 DH |
3433 | /* |
3434 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 3435 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
3436 | */ |
3437 | if (res) | |
3438 | ret = (res < 0) ? res : 0; | |
3439 | ||
c236c8e9 PZ |
3440 | /* |
3441 | * If fixup_pi_state_owner() faulted and was unable to handle | |
3442 | * the fault, unlock the rt_mutex and return the fault to | |
3443 | * userspace. | |
3444 | */ | |
16ffa12d PZ |
3445 | if (ret && rt_mutex_owner(&q.pi_state->pi_mutex) == current) { |
3446 | pi_state = q.pi_state; | |
3447 | get_pi_state(pi_state); | |
3448 | } | |
c236c8e9 | 3449 | |
52400ba9 DH |
3450 | /* Unqueue and drop the lock. */ |
3451 | unqueue_me_pi(&q); | |
3452 | } | |
3453 | ||
16ffa12d PZ |
3454 | if (pi_state) { |
3455 | rt_mutex_futex_unlock(&pi_state->pi_mutex); | |
3456 | put_pi_state(pi_state); | |
3457 | } | |
3458 | ||
c236c8e9 | 3459 | if (ret == -EINTR) { |
52400ba9 | 3460 | /* |
cc6db4e6 DH |
3461 | * We've already been requeued, but cannot restart by calling |
3462 | * futex_lock_pi() directly. We could restart this syscall, but | |
3463 | * it would detect that the user space "val" changed and return | |
3464 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
3465 | * -EWOULDBLOCK directly. | |
52400ba9 | 3466 | */ |
2070887f | 3467 | ret = -EWOULDBLOCK; |
52400ba9 DH |
3468 | } |
3469 | ||
3470 | out_put_keys: | |
ae791a2d | 3471 | put_futex_key(&q.key); |
c8b15a70 | 3472 | out_key2: |
ae791a2d | 3473 | put_futex_key(&key2); |
52400ba9 DH |
3474 | |
3475 | out: | |
3476 | if (to) { | |
3477 | hrtimer_cancel(&to->timer); | |
3478 | destroy_hrtimer_on_stack(&to->timer); | |
3479 | } | |
3480 | return ret; | |
3481 | } | |
3482 | ||
0771dfef IM |
3483 | /* |
3484 | * Support for robust futexes: the kernel cleans up held futexes at | |
3485 | * thread exit time. | |
3486 | * | |
3487 | * Implementation: user-space maintains a per-thread list of locks it | |
3488 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
3489 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 3490 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
3491 | * always manipulated with the lock held, so the list is private and |
3492 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
3493 | * field, to allow the kernel to clean up if the thread dies after | |
3494 | * acquiring the lock, but just before it could have added itself to | |
3495 | * the list. There can only be one such pending lock. | |
3496 | */ | |
3497 | ||
3498 | /** | |
d96ee56c DH |
3499 | * sys_set_robust_list() - Set the robust-futex list head of a task |
3500 | * @head: pointer to the list-head | |
3501 | * @len: length of the list-head, as userspace expects | |
0771dfef | 3502 | */ |
836f92ad HC |
3503 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
3504 | size_t, len) | |
0771dfef | 3505 | { |
a0c1e907 TG |
3506 | if (!futex_cmpxchg_enabled) |
3507 | return -ENOSYS; | |
0771dfef IM |
3508 | /* |
3509 | * The kernel knows only one size for now: | |
3510 | */ | |
3511 | if (unlikely(len != sizeof(*head))) | |
3512 | return -EINVAL; | |
3513 | ||
3514 | current->robust_list = head; | |
3515 | ||
3516 | return 0; | |
3517 | } | |
3518 | ||
3519 | /** | |
d96ee56c DH |
3520 | * sys_get_robust_list() - Get the robust-futex list head of a task |
3521 | * @pid: pid of the process [zero for current task] | |
3522 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
3523 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 3524 | */ |
836f92ad HC |
3525 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
3526 | struct robust_list_head __user * __user *, head_ptr, | |
3527 | size_t __user *, len_ptr) | |
0771dfef | 3528 | { |
ba46df98 | 3529 | struct robust_list_head __user *head; |
0771dfef | 3530 | unsigned long ret; |
bdbb776f | 3531 | struct task_struct *p; |
0771dfef | 3532 | |
a0c1e907 TG |
3533 | if (!futex_cmpxchg_enabled) |
3534 | return -ENOSYS; | |
3535 | ||
bdbb776f KC |
3536 | rcu_read_lock(); |
3537 | ||
3538 | ret = -ESRCH; | |
0771dfef | 3539 | if (!pid) |
bdbb776f | 3540 | p = current; |
0771dfef | 3541 | else { |
228ebcbe | 3542 | p = find_task_by_vpid(pid); |
0771dfef IM |
3543 | if (!p) |
3544 | goto err_unlock; | |
0771dfef IM |
3545 | } |
3546 | ||
bdbb776f | 3547 | ret = -EPERM; |
caaee623 | 3548 | if (!ptrace_may_access(p, PTRACE_MODE_READ_REALCREDS)) |
bdbb776f KC |
3549 | goto err_unlock; |
3550 | ||
3551 | head = p->robust_list; | |
3552 | rcu_read_unlock(); | |
3553 | ||
0771dfef IM |
3554 | if (put_user(sizeof(*head), len_ptr)) |
3555 | return -EFAULT; | |
3556 | return put_user(head, head_ptr); | |
3557 | ||
3558 | err_unlock: | |
aaa2a97e | 3559 | rcu_read_unlock(); |
0771dfef IM |
3560 | |
3561 | return ret; | |
3562 | } | |
3563 | ||
6c4e1847 YT |
3564 | /* Constants for the pending_op argument of handle_futex_death */ |
3565 | #define HANDLE_DEATH_PENDING true | |
3566 | #define HANDLE_DEATH_LIST false | |
3567 | ||
0771dfef IM |
3568 | /* |
3569 | * Process a futex-list entry, check whether it's owned by the | |
3570 | * dying task, and do notification if so: | |
3571 | */ | |
6c4e1847 YT |
3572 | static int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, |
3573 | bool pi, bool pending_op) | |
0771dfef | 3574 | { |
7cfdaf38 | 3575 | u32 uval, uninitialized_var(nval), mval; |
e5733e53 | 3576 | int err; |
0771dfef | 3577 | |
087fad77 CJ |
3578 | /* Futex address must be 32bit aligned */ |
3579 | if ((((unsigned long)uaddr) % sizeof(*uaddr)) != 0) | |
3580 | return -1; | |
3581 | ||
8f17d3a5 IM |
3582 | retry: |
3583 | if (get_user(uval, uaddr)) | |
0771dfef IM |
3584 | return -1; |
3585 | ||
6c4e1847 YT |
3586 | /* |
3587 | * Special case for regular (non PI) futexes. The unlock path in | |
3588 | * user space has two race scenarios: | |
3589 | * | |
3590 | * 1. The unlock path releases the user space futex value and | |
3591 | * before it can execute the futex() syscall to wake up | |
3592 | * waiters it is killed. | |
3593 | * | |
3594 | * 2. A woken up waiter is killed before it can acquire the | |
3595 | * futex in user space. | |
3596 | * | |
3597 | * In both cases the TID validation below prevents a wakeup of | |
3598 | * potential waiters which can cause these waiters to block | |
3599 | * forever. | |
3600 | * | |
3601 | * In both cases the following conditions are met: | |
3602 | * | |
3603 | * 1) task->robust_list->list_op_pending != NULL | |
3604 | * @pending_op == true | |
3605 | * 2) User space futex value == 0 | |
3606 | * 3) Regular futex: @pi == false | |
3607 | * | |
3608 | * If these conditions are met, it is safe to attempt waking up a | |
3609 | * potential waiter without touching the user space futex value and | |
3610 | * trying to set the OWNER_DIED bit. The user space futex value is | |
3611 | * uncontended and the rest of the user space mutex state is | |
3612 | * consistent, so a woken waiter will just take over the | |
3613 | * uncontended futex. Setting the OWNER_DIED bit would create | |
3614 | * inconsistent state and malfunction of the user space owner died | |
3615 | * handling. | |
3616 | */ | |
3617 | if (pending_op && !pi && !uval) { | |
3618 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); | |
3619 | return 0; | |
3620 | } | |
3621 | ||
e5733e53 WD |
3622 | if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) |
3623 | return 0; | |
3624 | ||
3625 | /* | |
3626 | * Ok, this dying thread is truly holding a futex | |
3627 | * of interest. Set the OWNER_DIED bit atomically | |
3628 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
3629 | * set, wake up a waiter (if any). (We have to do a | |
3630 | * futex_wake() even if OWNER_DIED is already set - | |
3631 | * to handle the rare but possible case of recursive | |
3632 | * thread-death.) The rest of the cleanup is done in | |
3633 | * userspace. | |
3634 | */ | |
3635 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; | |
3636 | ||
3637 | /* | |
3638 | * We are not holding a lock here, but we want to have | |
3639 | * the pagefault_disable/enable() protection because | |
3640 | * we want to handle the fault gracefully. If the | |
3641 | * access fails we try to fault in the futex with R/W | |
3642 | * verification via get_user_pages. get_user() above | |
3643 | * does not guarantee R/W access. If that fails we | |
3644 | * give up and leave the futex locked. | |
3645 | */ | |
3646 | if ((err = cmpxchg_futex_value_locked(&nval, uaddr, uval, mval))) { | |
3647 | switch (err) { | |
3648 | case -EFAULT: | |
6e0aa9f8 TG |
3649 | if (fault_in_user_writeable(uaddr)) |
3650 | return -1; | |
3651 | goto retry; | |
e5733e53 WD |
3652 | |
3653 | case -EAGAIN: | |
3654 | cond_resched(); | |
8f17d3a5 | 3655 | goto retry; |
0771dfef | 3656 | |
e5733e53 WD |
3657 | default: |
3658 | WARN_ON_ONCE(1); | |
3659 | return err; | |
3660 | } | |
0771dfef | 3661 | } |
e5733e53 WD |
3662 | |
3663 | if (nval != uval) | |
3664 | goto retry; | |
3665 | ||
3666 | /* | |
3667 | * Wake robust non-PI futexes here. The wakeup of | |
3668 | * PI futexes happens in exit_pi_state(): | |
3669 | */ | |
3670 | if (!pi && (uval & FUTEX_WAITERS)) | |
3671 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); | |
3672 | ||
0771dfef IM |
3673 | return 0; |
3674 | } | |
3675 | ||
e3f2ddea IM |
3676 | /* |
3677 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
3678 | */ | |
3679 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 3680 | struct robust_list __user * __user *head, |
1dcc41bb | 3681 | unsigned int *pi) |
e3f2ddea IM |
3682 | { |
3683 | unsigned long uentry; | |
3684 | ||
ba46df98 | 3685 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
3686 | return -EFAULT; |
3687 | ||
ba46df98 | 3688 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
3689 | *pi = uentry & 1; |
3690 | ||
3691 | return 0; | |
3692 | } | |
3693 | ||
0771dfef IM |
3694 | /* |
3695 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
3696 | * and mark any locks found there dead, and notify any waiters. | |
3697 | * | |
3698 | * We silently return on any sign of list-walking problem. | |
3699 | */ | |
e891c8f7 | 3700 | static void exit_robust_list(struct task_struct *curr) |
0771dfef IM |
3701 | { |
3702 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 3703 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 DH |
3704 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
3705 | unsigned int uninitialized_var(next_pi); | |
0771dfef | 3706 | unsigned long futex_offset; |
9f96cb1e | 3707 | int rc; |
0771dfef | 3708 | |
a0c1e907 TG |
3709 | if (!futex_cmpxchg_enabled) |
3710 | return; | |
3711 | ||
0771dfef IM |
3712 | /* |
3713 | * Fetch the list head (which was registered earlier, via | |
3714 | * sys_set_robust_list()): | |
3715 | */ | |
e3f2ddea | 3716 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
3717 | return; |
3718 | /* | |
3719 | * Fetch the relative futex offset: | |
3720 | */ | |
3721 | if (get_user(futex_offset, &head->futex_offset)) | |
3722 | return; | |
3723 | /* | |
3724 | * Fetch any possibly pending lock-add first, and handle it | |
3725 | * if it exists: | |
3726 | */ | |
e3f2ddea | 3727 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 3728 | return; |
e3f2ddea | 3729 | |
9f96cb1e | 3730 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 3731 | while (entry != &head->list) { |
9f96cb1e MS |
3732 | /* |
3733 | * Fetch the next entry in the list before calling | |
3734 | * handle_futex_death: | |
3735 | */ | |
3736 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
3737 | /* |
3738 | * A pending lock might already be on the list, so | |
c87e2837 | 3739 | * don't process it twice: |
0771dfef | 3740 | */ |
6c4e1847 | 3741 | if (entry != pending) { |
ba46df98 | 3742 | if (handle_futex_death((void __user *)entry + futex_offset, |
6c4e1847 | 3743 | curr, pi, HANDLE_DEATH_LIST)) |
0771dfef | 3744 | return; |
6c4e1847 | 3745 | } |
9f96cb1e | 3746 | if (rc) |
0771dfef | 3747 | return; |
9f96cb1e MS |
3748 | entry = next_entry; |
3749 | pi = next_pi; | |
0771dfef IM |
3750 | /* |
3751 | * Avoid excessively long or circular lists: | |
3752 | */ | |
3753 | if (!--limit) | |
3754 | break; | |
3755 | ||
3756 | cond_resched(); | |
3757 | } | |
9f96cb1e | 3758 | |
6c4e1847 | 3759 | if (pending) { |
9f96cb1e | 3760 | handle_futex_death((void __user *)pending + futex_offset, |
6c4e1847 YT |
3761 | curr, pip, HANDLE_DEATH_PENDING); |
3762 | } | |
0771dfef IM |
3763 | } |
3764 | ||
7db3796f | 3765 | static void futex_cleanup(struct task_struct *tsk) |
e891c8f7 TG |
3766 | { |
3767 | if (unlikely(tsk->robust_list)) { | |
3768 | exit_robust_list(tsk); | |
3769 | tsk->robust_list = NULL; | |
3770 | } | |
3771 | ||
3772 | #ifdef CONFIG_COMPAT | |
3773 | if (unlikely(tsk->compat_robust_list)) { | |
3774 | compat_exit_robust_list(tsk); | |
3775 | tsk->compat_robust_list = NULL; | |
3776 | } | |
3777 | #endif | |
3778 | ||
3779 | if (unlikely(!list_empty(&tsk->pi_state_list))) | |
3780 | exit_pi_state_list(tsk); | |
3781 | } | |
3782 | ||
f34df84f TG |
3783 | /** |
3784 | * futex_exit_recursive - Set the tasks futex state to FUTEX_STATE_DEAD | |
3785 | * @tsk: task to set the state on | |
3786 | * | |
3787 | * Set the futex exit state of the task lockless. The futex waiter code | |
3788 | * observes that state when a task is exiting and loops until the task has | |
3789 | * actually finished the futex cleanup. The worst case for this is that the | |
3790 | * waiter runs through the wait loop until the state becomes visible. | |
3791 | * | |
3792 | * This is called from the recursive fault handling path in do_exit(). | |
3793 | * | |
3794 | * This is best effort. Either the futex exit code has run already or | |
3795 | * not. If the OWNER_DIED bit has been set on the futex then the waiter can | |
3796 | * take it over. If not, the problem is pushed back to user space. If the | |
3797 | * futex exit code did not run yet, then an already queued waiter might | |
3798 | * block forever, but there is nothing which can be done about that. | |
3799 | */ | |
3800 | void futex_exit_recursive(struct task_struct *tsk) | |
3801 | { | |
af823481 TG |
3802 | /* If the state is FUTEX_STATE_EXITING then futex_exit_mutex is held */ |
3803 | if (tsk->futex_state == FUTEX_STATE_EXITING) | |
3804 | mutex_unlock(&tsk->futex_exit_mutex); | |
f34df84f TG |
3805 | tsk->futex_state = FUTEX_STATE_DEAD; |
3806 | } | |
3807 | ||
7db3796f | 3808 | static void futex_cleanup_begin(struct task_struct *tsk) |
b19fd552 | 3809 | { |
af823481 TG |
3810 | /* |
3811 | * Prevent various race issues against a concurrent incoming waiter | |
3812 | * including live locks by forcing the waiter to block on | |
3813 | * tsk->futex_exit_mutex when it observes FUTEX_STATE_EXITING in | |
3814 | * attach_to_pi_owner(). | |
3815 | */ | |
3816 | mutex_lock(&tsk->futex_exit_mutex); | |
3817 | ||
f34df84f | 3818 | /* |
015b4b29 TG |
3819 | * Switch the state to FUTEX_STATE_EXITING under tsk->pi_lock. |
3820 | * | |
3821 | * This ensures that all subsequent checks of tsk->futex_state in | |
3822 | * attach_to_pi_owner() must observe FUTEX_STATE_EXITING with | |
3823 | * tsk->pi_lock held. | |
3824 | * | |
3825 | * It guarantees also that a pi_state which was queued right before | |
3826 | * the state change under tsk->pi_lock by a concurrent waiter must | |
3827 | * be observed in exit_pi_state_list(). | |
f34df84f TG |
3828 | */ |
3829 | raw_spin_lock_irq(&tsk->pi_lock); | |
015b4b29 | 3830 | tsk->futex_state = FUTEX_STATE_EXITING; |
f34df84f | 3831 | raw_spin_unlock_irq(&tsk->pi_lock); |
7db3796f | 3832 | } |
f34df84f | 3833 | |
7db3796f TG |
3834 | static void futex_cleanup_end(struct task_struct *tsk, int state) |
3835 | { | |
3836 | /* | |
3837 | * Lockless store. The only side effect is that an observer might | |
3838 | * take another loop until it becomes visible. | |
3839 | */ | |
3840 | tsk->futex_state = state; | |
af823481 TG |
3841 | /* |
3842 | * Drop the exit protection. This unblocks waiters which observed | |
3843 | * FUTEX_STATE_EXITING to reevaluate the state. | |
3844 | */ | |
3845 | mutex_unlock(&tsk->futex_exit_mutex); | |
7db3796f | 3846 | } |
f34df84f | 3847 | |
7db3796f TG |
3848 | void futex_exec_release(struct task_struct *tsk) |
3849 | { | |
3850 | /* | |
3851 | * The state handling is done for consistency, but in the case of | |
3852 | * exec() there is no way to prevent futher damage as the PID stays | |
3853 | * the same. But for the unlikely and arguably buggy case that a | |
3854 | * futex is held on exec(), this provides at least as much state | |
3855 | * consistency protection which is possible. | |
3856 | */ | |
3857 | futex_cleanup_begin(tsk); | |
3858 | futex_cleanup(tsk); | |
3859 | /* | |
3860 | * Reset the state to FUTEX_STATE_OK. The task is alive and about | |
3861 | * exec a new binary. | |
3862 | */ | |
3863 | futex_cleanup_end(tsk, FUTEX_STATE_OK); | |
3864 | } | |
3865 | ||
3866 | void futex_exit_release(struct task_struct *tsk) | |
3867 | { | |
3868 | futex_cleanup_begin(tsk); | |
3869 | futex_cleanup(tsk); | |
3870 | futex_cleanup_end(tsk, FUTEX_STATE_DEAD); | |
b19fd552 TG |
3871 | } |
3872 | ||
c19384b5 | 3873 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 3874 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 3875 | { |
81b40539 | 3876 | int cmd = op & FUTEX_CMD_MASK; |
b41277dc | 3877 | unsigned int flags = 0; |
34f01cc1 ED |
3878 | |
3879 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 3880 | flags |= FLAGS_SHARED; |
1da177e4 | 3881 | |
b41277dc DH |
3882 | if (op & FUTEX_CLOCK_REALTIME) { |
3883 | flags |= FLAGS_CLOCKRT; | |
337f1304 DH |
3884 | if (cmd != FUTEX_WAIT && cmd != FUTEX_WAIT_BITSET && \ |
3885 | cmd != FUTEX_WAIT_REQUEUE_PI) | |
b41277dc DH |
3886 | return -ENOSYS; |
3887 | } | |
1da177e4 | 3888 | |
59263b51 TG |
3889 | switch (cmd) { |
3890 | case FUTEX_LOCK_PI: | |
3891 | case FUTEX_UNLOCK_PI: | |
3892 | case FUTEX_TRYLOCK_PI: | |
3893 | case FUTEX_WAIT_REQUEUE_PI: | |
3894 | case FUTEX_CMP_REQUEUE_PI: | |
3895 | if (!futex_cmpxchg_enabled) | |
3896 | return -ENOSYS; | |
3897 | } | |
3898 | ||
34f01cc1 | 3899 | switch (cmd) { |
1da177e4 | 3900 | case FUTEX_WAIT: |
cd689985 TG |
3901 | val3 = FUTEX_BITSET_MATCH_ANY; |
3902 | case FUTEX_WAIT_BITSET: | |
81b40539 | 3903 | return futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 | 3904 | case FUTEX_WAKE: |
cd689985 TG |
3905 | val3 = FUTEX_BITSET_MATCH_ANY; |
3906 | case FUTEX_WAKE_BITSET: | |
81b40539 | 3907 | return futex_wake(uaddr, flags, val, val3); |
1da177e4 | 3908 | case FUTEX_REQUEUE: |
81b40539 | 3909 | return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 | 3910 | case FUTEX_CMP_REQUEUE: |
81b40539 | 3911 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
4732efbe | 3912 | case FUTEX_WAKE_OP: |
81b40539 | 3913 | return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
c87e2837 | 3914 | case FUTEX_LOCK_PI: |
996636dd | 3915 | return futex_lock_pi(uaddr, flags, timeout, 0); |
c87e2837 | 3916 | case FUTEX_UNLOCK_PI: |
81b40539 | 3917 | return futex_unlock_pi(uaddr, flags); |
c87e2837 | 3918 | case FUTEX_TRYLOCK_PI: |
996636dd | 3919 | return futex_lock_pi(uaddr, flags, NULL, 1); |
52400ba9 DH |
3920 | case FUTEX_WAIT_REQUEUE_PI: |
3921 | val3 = FUTEX_BITSET_MATCH_ANY; | |
81b40539 TG |
3922 | return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
3923 | uaddr2); | |
52400ba9 | 3924 | case FUTEX_CMP_REQUEUE_PI: |
81b40539 | 3925 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
1da177e4 | 3926 | } |
81b40539 | 3927 | return -ENOSYS; |
1da177e4 LT |
3928 | } |
3929 | ||
3930 | ||
17da2bd9 HC |
3931 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
3932 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
3933 | u32, val3) | |
1da177e4 | 3934 | { |
c19384b5 PP |
3935 | struct timespec ts; |
3936 | ktime_t t, *tp = NULL; | |
e2970f2f | 3937 | u32 val2 = 0; |
34f01cc1 | 3938 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 3939 | |
cd689985 | 3940 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
3941 | cmd == FUTEX_WAIT_BITSET || |
3942 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
ab51fbab DB |
3943 | if (unlikely(should_fail_futex(!(op & FUTEX_PRIVATE_FLAG)))) |
3944 | return -EFAULT; | |
c19384b5 | 3945 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 3946 | return -EFAULT; |
c19384b5 | 3947 | if (!timespec_valid(&ts)) |
9741ef96 | 3948 | return -EINVAL; |
c19384b5 PP |
3949 | |
3950 | t = timespec_to_ktime(ts); | |
34f01cc1 | 3951 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 3952 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 3953 | tp = &t; |
1da177e4 LT |
3954 | } |
3955 | /* | |
52400ba9 | 3956 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 3957 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 3958 | */ |
f54f0986 | 3959 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 3960 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 3961 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 3962 | |
c19384b5 | 3963 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
3964 | } |
3965 | ||
f6d7b233 AB |
3966 | #ifdef CONFIG_COMPAT |
3967 | /* | |
3968 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
3969 | */ | |
3970 | static inline int | |
3971 | compat_fetch_robust_entry(compat_uptr_t *uentry, struct robust_list __user **entry, | |
3972 | compat_uptr_t __user *head, unsigned int *pi) | |
3973 | { | |
3974 | if (get_user(*uentry, head)) | |
3975 | return -EFAULT; | |
3976 | ||
3977 | *entry = compat_ptr((*uentry) & ~1); | |
3978 | *pi = (unsigned int)(*uentry) & 1; | |
3979 | ||
3980 | return 0; | |
3981 | } | |
3982 | ||
3983 | static void __user *futex_uaddr(struct robust_list __user *entry, | |
3984 | compat_long_t futex_offset) | |
3985 | { | |
3986 | compat_uptr_t base = ptr_to_compat(entry); | |
3987 | void __user *uaddr = compat_ptr(base + futex_offset); | |
3988 | ||
3989 | return uaddr; | |
3990 | } | |
3991 | ||
3992 | /* | |
3993 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
3994 | * and mark any locks found there dead, and notify any waiters. | |
3995 | * | |
3996 | * We silently return on any sign of list-walking problem. | |
3997 | */ | |
e891c8f7 | 3998 | static void compat_exit_robust_list(struct task_struct *curr) |
f6d7b233 AB |
3999 | { |
4000 | struct compat_robust_list_head __user *head = curr->compat_robust_list; | |
4001 | struct robust_list __user *entry, *next_entry, *pending; | |
4002 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; | |
4003 | unsigned int uninitialized_var(next_pi); | |
4004 | compat_uptr_t uentry, next_uentry, upending; | |
4005 | compat_long_t futex_offset; | |
4006 | int rc; | |
4007 | ||
4008 | if (!futex_cmpxchg_enabled) | |
4009 | return; | |
4010 | ||
4011 | /* | |
4012 | * Fetch the list head (which was registered earlier, via | |
4013 | * sys_set_robust_list()): | |
4014 | */ | |
4015 | if (compat_fetch_robust_entry(&uentry, &entry, &head->list.next, &pi)) | |
4016 | return; | |
4017 | /* | |
4018 | * Fetch the relative futex offset: | |
4019 | */ | |
4020 | if (get_user(futex_offset, &head->futex_offset)) | |
4021 | return; | |
4022 | /* | |
4023 | * Fetch any possibly pending lock-add first, and handle it | |
4024 | * if it exists: | |
4025 | */ | |
4026 | if (compat_fetch_robust_entry(&upending, &pending, | |
4027 | &head->list_op_pending, &pip)) | |
4028 | return; | |
4029 | ||
4030 | next_entry = NULL; /* avoid warning with gcc */ | |
4031 | while (entry != (struct robust_list __user *) &head->list) { | |
4032 | /* | |
4033 | * Fetch the next entry in the list before calling | |
4034 | * handle_futex_death: | |
4035 | */ | |
4036 | rc = compat_fetch_robust_entry(&next_uentry, &next_entry, | |
4037 | (compat_uptr_t __user *)&entry->next, &next_pi); | |
4038 | /* | |
4039 | * A pending lock might already be on the list, so | |
4040 | * dont process it twice: | |
4041 | */ | |
4042 | if (entry != pending) { | |
4043 | void __user *uaddr = futex_uaddr(entry, futex_offset); | |
4044 | ||
6c4e1847 YT |
4045 | if (handle_futex_death(uaddr, curr, pi, |
4046 | HANDLE_DEATH_LIST)) | |
f6d7b233 AB |
4047 | return; |
4048 | } | |
4049 | if (rc) | |
4050 | return; | |
4051 | uentry = next_uentry; | |
4052 | entry = next_entry; | |
4053 | pi = next_pi; | |
4054 | /* | |
4055 | * Avoid excessively long or circular lists: | |
4056 | */ | |
4057 | if (!--limit) | |
4058 | break; | |
4059 | ||
4060 | cond_resched(); | |
4061 | } | |
4062 | if (pending) { | |
4063 | void __user *uaddr = futex_uaddr(pending, futex_offset); | |
4064 | ||
6c4e1847 | 4065 | handle_futex_death(uaddr, curr, pip, HANDLE_DEATH_PENDING); |
f6d7b233 AB |
4066 | } |
4067 | } | |
4068 | ||
4069 | COMPAT_SYSCALL_DEFINE2(set_robust_list, | |
4070 | struct compat_robust_list_head __user *, head, | |
4071 | compat_size_t, len) | |
4072 | { | |
4073 | if (!futex_cmpxchg_enabled) | |
4074 | return -ENOSYS; | |
4075 | ||
4076 | if (unlikely(len != sizeof(*head))) | |
4077 | return -EINVAL; | |
4078 | ||
4079 | current->compat_robust_list = head; | |
4080 | ||
4081 | return 0; | |
4082 | } | |
4083 | ||
4084 | COMPAT_SYSCALL_DEFINE3(get_robust_list, int, pid, | |
4085 | compat_uptr_t __user *, head_ptr, | |
4086 | compat_size_t __user *, len_ptr) | |
4087 | { | |
4088 | struct compat_robust_list_head __user *head; | |
4089 | unsigned long ret; | |
4090 | struct task_struct *p; | |
4091 | ||
4092 | if (!futex_cmpxchg_enabled) | |
4093 | return -ENOSYS; | |
4094 | ||
4095 | rcu_read_lock(); | |
4096 | ||
4097 | ret = -ESRCH; | |
4098 | if (!pid) | |
4099 | p = current; | |
4100 | else { | |
4101 | p = find_task_by_vpid(pid); | |
4102 | if (!p) | |
4103 | goto err_unlock; | |
4104 | } | |
4105 | ||
4106 | ret = -EPERM; | |
4107 | if (!ptrace_may_access(p, PTRACE_MODE_READ_REALCREDS)) | |
4108 | goto err_unlock; | |
4109 | ||
4110 | head = p->compat_robust_list; | |
4111 | rcu_read_unlock(); | |
4112 | ||
4113 | if (put_user(sizeof(*head), len_ptr)) | |
4114 | return -EFAULT; | |
4115 | return put_user(ptr_to_compat(head), head_ptr); | |
4116 | ||
4117 | err_unlock: | |
4118 | rcu_read_unlock(); | |
4119 | ||
4120 | return ret; | |
4121 | } | |
4122 | ||
4123 | COMPAT_SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, | |
4124 | struct compat_timespec __user *, utime, u32 __user *, uaddr2, | |
4125 | u32, val3) | |
4126 | { | |
4127 | struct timespec ts; | |
4128 | ktime_t t, *tp = NULL; | |
4129 | int val2 = 0; | |
4130 | int cmd = op & FUTEX_CMD_MASK; | |
4131 | ||
4132 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || | |
4133 | cmd == FUTEX_WAIT_BITSET || | |
4134 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
4135 | if (compat_get_timespec(&ts, utime)) | |
4136 | return -EFAULT; | |
4137 | if (!timespec_valid(&ts)) | |
4138 | return -EINVAL; | |
4139 | ||
4140 | t = timespec_to_ktime(ts); | |
4141 | if (cmd == FUTEX_WAIT) | |
4142 | t = ktime_add_safe(ktime_get(), t); | |
4143 | tp = &t; | |
4144 | } | |
4145 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || | |
4146 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) | |
4147 | val2 = (int) (unsigned long) utime; | |
4148 | ||
4149 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); | |
4150 | } | |
4151 | #endif /* CONFIG_COMPAT */ | |
4152 | ||
03b8c7b6 | 4153 | static void __init futex_detect_cmpxchg(void) |
1da177e4 | 4154 | { |
03b8c7b6 | 4155 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 4156 | u32 curval; |
03b8c7b6 HC |
4157 | |
4158 | /* | |
4159 | * This will fail and we want it. Some arch implementations do | |
4160 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
4161 | * functionality. We want to know that before we call in any | |
4162 | * of the complex code paths. Also we want to prevent | |
4163 | * registration of robust lists in that case. NULL is | |
4164 | * guaranteed to fault and we get -EFAULT on functional | |
4165 | * implementation, the non-functional ones will return | |
4166 | * -ENOSYS. | |
4167 | */ | |
4168 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) | |
4169 | futex_cmpxchg_enabled = 1; | |
4170 | #endif | |
4171 | } | |
4172 | ||
4173 | static int __init futex_init(void) | |
4174 | { | |
63b1a816 | 4175 | unsigned int futex_shift; |
a52b89eb DB |
4176 | unsigned long i; |
4177 | ||
4178 | #if CONFIG_BASE_SMALL | |
4179 | futex_hashsize = 16; | |
4180 | #else | |
4181 | futex_hashsize = roundup_pow_of_two(256 * num_possible_cpus()); | |
4182 | #endif | |
4183 | ||
4184 | futex_queues = alloc_large_system_hash("futex", sizeof(*futex_queues), | |
4185 | futex_hashsize, 0, | |
4186 | futex_hashsize < 256 ? HASH_SMALL : 0, | |
63b1a816 HC |
4187 | &futex_shift, NULL, |
4188 | futex_hashsize, futex_hashsize); | |
4189 | futex_hashsize = 1UL << futex_shift; | |
03b8c7b6 HC |
4190 | |
4191 | futex_detect_cmpxchg(); | |
a0c1e907 | 4192 | |
a52b89eb | 4193 | for (i = 0; i < futex_hashsize; i++) { |
11d4616b | 4194 | atomic_set(&futex_queues[i].waiters, 0); |
732375c6 | 4195 | plist_head_init(&futex_queues[i].chain); |
3e4ab747 TG |
4196 | spin_lock_init(&futex_queues[i].lock); |
4197 | } | |
4198 | ||
1da177e4 LT |
4199 | return 0; |
4200 | } | |
25f71d1c | 4201 | core_initcall(futex_init); |