Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Fast Userspace Mutexes (which I call "Futexes!"). | |
3 | * (C) Rusty Russell, IBM 2002 | |
4 | * | |
5 | * Generalized futexes, futex requeueing, misc fixes by Ingo Molnar | |
6 | * (C) Copyright 2003 Red Hat Inc, All Rights Reserved | |
7 | * | |
8 | * Removed page pinning, fix privately mapped COW pages and other cleanups | |
9 | * (C) Copyright 2003, 2004 Jamie Lokier | |
10 | * | |
0771dfef IM |
11 | * Robust futex support started by Ingo Molnar |
12 | * (C) Copyright 2006 Red Hat Inc, All Rights Reserved | |
13 | * Thanks to Thomas Gleixner for suggestions, analysis and fixes. | |
14 | * | |
c87e2837 IM |
15 | * PI-futex support started by Ingo Molnar and Thomas Gleixner |
16 | * Copyright (C) 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com> | |
17 | * Copyright (C) 2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com> | |
18 | * | |
34f01cc1 ED |
19 | * PRIVATE futexes by Eric Dumazet |
20 | * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com> | |
21 | * | |
52400ba9 DH |
22 | * Requeue-PI support by Darren Hart <dvhltc@us.ibm.com> |
23 | * Copyright (C) IBM Corporation, 2009 | |
24 | * Thanks to Thomas Gleixner for conceptual design and careful reviews. | |
25 | * | |
1da177e4 LT |
26 | * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly |
27 | * enough at me, Linus for the original (flawed) idea, Matthew | |
28 | * Kirkwood for proof-of-concept implementation. | |
29 | * | |
30 | * "The futexes are also cursed." | |
31 | * "But they come in a choice of three flavours!" | |
32 | * | |
33 | * This program is free software; you can redistribute it and/or modify | |
34 | * it under the terms of the GNU General Public License as published by | |
35 | * the Free Software Foundation; either version 2 of the License, or | |
36 | * (at your option) any later version. | |
37 | * | |
38 | * This program is distributed in the hope that it will be useful, | |
39 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
40 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
41 | * GNU General Public License for more details. | |
42 | * | |
43 | * You should have received a copy of the GNU General Public License | |
44 | * along with this program; if not, write to the Free Software | |
45 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
46 | */ | |
47 | #include <linux/slab.h> | |
48 | #include <linux/poll.h> | |
49 | #include <linux/fs.h> | |
50 | #include <linux/file.h> | |
51 | #include <linux/jhash.h> | |
52 | #include <linux/init.h> | |
53 | #include <linux/futex.h> | |
54 | #include <linux/mount.h> | |
55 | #include <linux/pagemap.h> | |
56 | #include <linux/syscalls.h> | |
7ed20e1a | 57 | #include <linux/signal.h> |
9984de1a | 58 | #include <linux/export.h> |
fd5eea42 | 59 | #include <linux/magic.h> |
b488893a PE |
60 | #include <linux/pid.h> |
61 | #include <linux/nsproxy.h> | |
bdbb776f | 62 | #include <linux/ptrace.h> |
8bd75c77 | 63 | #include <linux/sched/rt.h> |
13d60f4b | 64 | #include <linux/hugetlb.h> |
88c8004f | 65 | #include <linux/freezer.h> |
a52b89eb | 66 | #include <linux/bootmem.h> |
b488893a | 67 | |
4732efbe | 68 | #include <asm/futex.h> |
1da177e4 | 69 | |
1696a8be | 70 | #include "locking/rtmutex_common.h" |
c87e2837 | 71 | |
99b60ce6 | 72 | /* |
d7e8af1a DB |
73 | * READ this before attempting to hack on futexes! |
74 | * | |
75 | * Basic futex operation and ordering guarantees | |
76 | * ============================================= | |
99b60ce6 TG |
77 | * |
78 | * The waiter reads the futex value in user space and calls | |
79 | * futex_wait(). This function computes the hash bucket and acquires | |
80 | * the hash bucket lock. After that it reads the futex user space value | |
b0c29f79 DB |
81 | * again and verifies that the data has not changed. If it has not changed |
82 | * it enqueues itself into the hash bucket, releases the hash bucket lock | |
83 | * and schedules. | |
99b60ce6 TG |
84 | * |
85 | * The waker side modifies the user space value of the futex and calls | |
b0c29f79 DB |
86 | * futex_wake(). This function computes the hash bucket and acquires the |
87 | * hash bucket lock. Then it looks for waiters on that futex in the hash | |
88 | * bucket and wakes them. | |
99b60ce6 | 89 | * |
b0c29f79 DB |
90 | * In futex wake up scenarios where no tasks are blocked on a futex, taking |
91 | * the hb spinlock can be avoided and simply return. In order for this | |
92 | * optimization to work, ordering guarantees must exist so that the waiter | |
93 | * being added to the list is acknowledged when the list is concurrently being | |
94 | * checked by the waker, avoiding scenarios like the following: | |
99b60ce6 TG |
95 | * |
96 | * CPU 0 CPU 1 | |
97 | * val = *futex; | |
98 | * sys_futex(WAIT, futex, val); | |
99 | * futex_wait(futex, val); | |
100 | * uval = *futex; | |
101 | * *futex = newval; | |
102 | * sys_futex(WAKE, futex); | |
103 | * futex_wake(futex); | |
104 | * if (queue_empty()) | |
105 | * return; | |
106 | * if (uval == val) | |
107 | * lock(hash_bucket(futex)); | |
108 | * queue(); | |
109 | * unlock(hash_bucket(futex)); | |
110 | * schedule(); | |
111 | * | |
112 | * This would cause the waiter on CPU 0 to wait forever because it | |
113 | * missed the transition of the user space value from val to newval | |
114 | * and the waker did not find the waiter in the hash bucket queue. | |
99b60ce6 | 115 | * |
b0c29f79 DB |
116 | * The correct serialization ensures that a waiter either observes |
117 | * the changed user space value before blocking or is woken by a | |
118 | * concurrent waker: | |
119 | * | |
120 | * CPU 0 CPU 1 | |
99b60ce6 TG |
121 | * val = *futex; |
122 | * sys_futex(WAIT, futex, val); | |
123 | * futex_wait(futex, val); | |
b0c29f79 | 124 | * |
d7e8af1a | 125 | * waiters++; (a) |
b0c29f79 DB |
126 | * mb(); (A) <-- paired with -. |
127 | * | | |
128 | * lock(hash_bucket(futex)); | | |
129 | * | | |
130 | * uval = *futex; | | |
131 | * | *futex = newval; | |
132 | * | sys_futex(WAKE, futex); | |
133 | * | futex_wake(futex); | |
134 | * | | |
135 | * `-------> mb(); (B) | |
99b60ce6 | 136 | * if (uval == val) |
b0c29f79 | 137 | * queue(); |
99b60ce6 | 138 | * unlock(hash_bucket(futex)); |
b0c29f79 DB |
139 | * schedule(); if (waiters) |
140 | * lock(hash_bucket(futex)); | |
d7e8af1a DB |
141 | * else wake_waiters(futex); |
142 | * waiters--; (b) unlock(hash_bucket(futex)); | |
b0c29f79 | 143 | * |
d7e8af1a DB |
144 | * Where (A) orders the waiters increment and the futex value read through |
145 | * atomic operations (see hb_waiters_inc) and where (B) orders the write | |
146 | * to futex and the waiters read -- this is done by the barriers in | |
147 | * get_futex_key_refs(), through either ihold or atomic_inc, depending on the | |
148 | * futex type. | |
b0c29f79 DB |
149 | * |
150 | * This yields the following case (where X:=waiters, Y:=futex): | |
151 | * | |
152 | * X = Y = 0 | |
153 | * | |
154 | * w[X]=1 w[Y]=1 | |
155 | * MB MB | |
156 | * r[Y]=y r[X]=x | |
157 | * | |
158 | * Which guarantees that x==0 && y==0 is impossible; which translates back into | |
159 | * the guarantee that we cannot both miss the futex variable change and the | |
160 | * enqueue. | |
d7e8af1a DB |
161 | * |
162 | * Note that a new waiter is accounted for in (a) even when it is possible that | |
163 | * the wait call can return error, in which case we backtrack from it in (b). | |
164 | * Refer to the comment in queue_lock(). | |
165 | * | |
166 | * Similarly, in order to account for waiters being requeued on another | |
167 | * address we always increment the waiters for the destination bucket before | |
168 | * acquiring the lock. It then decrements them again after releasing it - | |
169 | * the code that actually moves the futex(es) between hash buckets (requeue_futex) | |
170 | * will do the additional required waiter count housekeeping. This is done for | |
171 | * double_lock_hb() and double_unlock_hb(), respectively. | |
99b60ce6 TG |
172 | */ |
173 | ||
03b8c7b6 | 174 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 175 | int __read_mostly futex_cmpxchg_enabled; |
03b8c7b6 | 176 | #endif |
a0c1e907 | 177 | |
b41277dc DH |
178 | /* |
179 | * Futex flags used to encode options to functions and preserve them across | |
180 | * restarts. | |
181 | */ | |
182 | #define FLAGS_SHARED 0x01 | |
183 | #define FLAGS_CLOCKRT 0x02 | |
184 | #define FLAGS_HAS_TIMEOUT 0x04 | |
185 | ||
c87e2837 IM |
186 | /* |
187 | * Priority Inheritance state: | |
188 | */ | |
189 | struct futex_pi_state { | |
190 | /* | |
191 | * list of 'owned' pi_state instances - these have to be | |
192 | * cleaned up in do_exit() if the task exits prematurely: | |
193 | */ | |
194 | struct list_head list; | |
195 | ||
196 | /* | |
197 | * The PI object: | |
198 | */ | |
199 | struct rt_mutex pi_mutex; | |
200 | ||
201 | struct task_struct *owner; | |
202 | atomic_t refcount; | |
203 | ||
204 | union futex_key key; | |
205 | }; | |
206 | ||
d8d88fbb DH |
207 | /** |
208 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 209 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
210 | * @task: the task waiting on the futex |
211 | * @lock_ptr: the hash bucket lock | |
212 | * @key: the key the futex is hashed on | |
213 | * @pi_state: optional priority inheritance state | |
214 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
215 | * @requeue_pi_key: the requeue_pi target futex key | |
216 | * @bitset: bitset for the optional bitmasked wakeup | |
217 | * | |
218 | * We use this hashed waitqueue, instead of a normal wait_queue_t, so | |
1da177e4 LT |
219 | * we can wake only the relevant ones (hashed queues may be shared). |
220 | * | |
221 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 222 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 223 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
224 | * the second. |
225 | * | |
226 | * PI futexes are typically woken before they are removed from the hash list via | |
227 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
228 | */ |
229 | struct futex_q { | |
ec92d082 | 230 | struct plist_node list; |
1da177e4 | 231 | |
d8d88fbb | 232 | struct task_struct *task; |
1da177e4 | 233 | spinlock_t *lock_ptr; |
1da177e4 | 234 | union futex_key key; |
c87e2837 | 235 | struct futex_pi_state *pi_state; |
52400ba9 | 236 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 237 | union futex_key *requeue_pi_key; |
cd689985 | 238 | u32 bitset; |
1da177e4 LT |
239 | }; |
240 | ||
5bdb05f9 DH |
241 | static const struct futex_q futex_q_init = { |
242 | /* list gets initialized in queue_me()*/ | |
243 | .key = FUTEX_KEY_INIT, | |
244 | .bitset = FUTEX_BITSET_MATCH_ANY | |
245 | }; | |
246 | ||
1da177e4 | 247 | /* |
b2d0994b DH |
248 | * Hash buckets are shared by all the futex_keys that hash to the same |
249 | * location. Each key may have multiple futex_q structures, one for each task | |
250 | * waiting on a futex. | |
1da177e4 LT |
251 | */ |
252 | struct futex_hash_bucket { | |
11d4616b | 253 | atomic_t waiters; |
ec92d082 PP |
254 | spinlock_t lock; |
255 | struct plist_head chain; | |
a52b89eb | 256 | } ____cacheline_aligned_in_smp; |
1da177e4 | 257 | |
a52b89eb DB |
258 | static unsigned long __read_mostly futex_hashsize; |
259 | ||
260 | static struct futex_hash_bucket *futex_queues; | |
1da177e4 | 261 | |
b0c29f79 DB |
262 | static inline void futex_get_mm(union futex_key *key) |
263 | { | |
264 | atomic_inc(&key->private.mm->mm_count); | |
265 | /* | |
266 | * Ensure futex_get_mm() implies a full barrier such that | |
267 | * get_futex_key() implies a full barrier. This is relied upon | |
268 | * as full barrier (B), see the ordering comment above. | |
269 | */ | |
4e857c58 | 270 | smp_mb__after_atomic(); |
b0c29f79 DB |
271 | } |
272 | ||
11d4616b LT |
273 | /* |
274 | * Reflects a new waiter being added to the waitqueue. | |
275 | */ | |
276 | static inline void hb_waiters_inc(struct futex_hash_bucket *hb) | |
b0c29f79 DB |
277 | { |
278 | #ifdef CONFIG_SMP | |
11d4616b | 279 | atomic_inc(&hb->waiters); |
b0c29f79 | 280 | /* |
11d4616b | 281 | * Full barrier (A), see the ordering comment above. |
b0c29f79 | 282 | */ |
4e857c58 | 283 | smp_mb__after_atomic(); |
11d4616b LT |
284 | #endif |
285 | } | |
286 | ||
287 | /* | |
288 | * Reflects a waiter being removed from the waitqueue by wakeup | |
289 | * paths. | |
290 | */ | |
291 | static inline void hb_waiters_dec(struct futex_hash_bucket *hb) | |
292 | { | |
293 | #ifdef CONFIG_SMP | |
294 | atomic_dec(&hb->waiters); | |
295 | #endif | |
296 | } | |
b0c29f79 | 297 | |
11d4616b LT |
298 | static inline int hb_waiters_pending(struct futex_hash_bucket *hb) |
299 | { | |
300 | #ifdef CONFIG_SMP | |
301 | return atomic_read(&hb->waiters); | |
b0c29f79 | 302 | #else |
11d4616b | 303 | return 1; |
b0c29f79 DB |
304 | #endif |
305 | } | |
306 | ||
1da177e4 LT |
307 | /* |
308 | * We hash on the keys returned from get_futex_key (see below). | |
309 | */ | |
310 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
311 | { | |
312 | u32 hash = jhash2((u32*)&key->both.word, | |
313 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
314 | key->both.offset); | |
a52b89eb | 315 | return &futex_queues[hash & (futex_hashsize - 1)]; |
1da177e4 LT |
316 | } |
317 | ||
318 | /* | |
319 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
320 | */ | |
321 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
322 | { | |
2bc87203 DH |
323 | return (key1 && key2 |
324 | && key1->both.word == key2->both.word | |
1da177e4 LT |
325 | && key1->both.ptr == key2->both.ptr |
326 | && key1->both.offset == key2->both.offset); | |
327 | } | |
328 | ||
38d47c1b PZ |
329 | /* |
330 | * Take a reference to the resource addressed by a key. | |
331 | * Can be called while holding spinlocks. | |
332 | * | |
333 | */ | |
334 | static void get_futex_key_refs(union futex_key *key) | |
335 | { | |
336 | if (!key->both.ptr) | |
337 | return; | |
338 | ||
339 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
340 | case FUT_OFF_INODE: | |
b0c29f79 | 341 | ihold(key->shared.inode); /* implies MB (B) */ |
38d47c1b PZ |
342 | break; |
343 | case FUT_OFF_MMSHARED: | |
b0c29f79 | 344 | futex_get_mm(key); /* implies MB (B) */ |
38d47c1b | 345 | break; |
76835b0e CM |
346 | default: |
347 | smp_mb(); /* explicit MB (B) */ | |
38d47c1b PZ |
348 | } |
349 | } | |
350 | ||
351 | /* | |
352 | * Drop a reference to the resource addressed by a key. | |
353 | * The hash bucket spinlock must not be held. | |
354 | */ | |
355 | static void drop_futex_key_refs(union futex_key *key) | |
356 | { | |
90621c40 DH |
357 | if (!key->both.ptr) { |
358 | /* If we're here then we tried to put a key we failed to get */ | |
359 | WARN_ON_ONCE(1); | |
38d47c1b | 360 | return; |
90621c40 | 361 | } |
38d47c1b PZ |
362 | |
363 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
364 | case FUT_OFF_INODE: | |
365 | iput(key->shared.inode); | |
366 | break; | |
367 | case FUT_OFF_MMSHARED: | |
368 | mmdrop(key->private.mm); | |
369 | break; | |
370 | } | |
371 | } | |
372 | ||
34f01cc1 | 373 | /** |
d96ee56c DH |
374 | * get_futex_key() - Get parameters which are the keys for a futex |
375 | * @uaddr: virtual address of the futex | |
376 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED | |
377 | * @key: address where result is stored. | |
9ea71503 SB |
378 | * @rw: mapping needs to be read/write (values: VERIFY_READ, |
379 | * VERIFY_WRITE) | |
34f01cc1 | 380 | * |
6c23cbbd RD |
381 | * Return: a negative error code or 0 |
382 | * | |
34f01cc1 | 383 | * The key words are stored in *key on success. |
1da177e4 | 384 | * |
6131ffaa | 385 | * For shared mappings, it's (page->index, file_inode(vma->vm_file), |
1da177e4 LT |
386 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
387 | * We can usually work out the index without swapping in the page. | |
388 | * | |
b2d0994b | 389 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 390 | */ |
64d1304a | 391 | static int |
9ea71503 | 392 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) |
1da177e4 | 393 | { |
e2970f2f | 394 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 395 | struct mm_struct *mm = current->mm; |
a5b338f2 | 396 | struct page *page, *page_head; |
9ea71503 | 397 | int err, ro = 0; |
1da177e4 LT |
398 | |
399 | /* | |
400 | * The futex address must be "naturally" aligned. | |
401 | */ | |
e2970f2f | 402 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 403 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 404 | return -EINVAL; |
e2970f2f | 405 | address -= key->both.offset; |
1da177e4 | 406 | |
5cdec2d8 LT |
407 | if (unlikely(!access_ok(rw, uaddr, sizeof(u32)))) |
408 | return -EFAULT; | |
409 | ||
34f01cc1 ED |
410 | /* |
411 | * PROCESS_PRIVATE futexes are fast. | |
412 | * As the mm cannot disappear under us and the 'key' only needs | |
413 | * virtual address, we dont even have to find the underlying vma. | |
414 | * Note : We do have to check 'uaddr' is a valid user address, | |
415 | * but access_ok() should be faster than find_vma() | |
416 | */ | |
417 | if (!fshared) { | |
34f01cc1 ED |
418 | key->private.mm = mm; |
419 | key->private.address = address; | |
b0c29f79 | 420 | get_futex_key_refs(key); /* implies MB (B) */ |
34f01cc1 ED |
421 | return 0; |
422 | } | |
1da177e4 | 423 | |
38d47c1b | 424 | again: |
7485d0d3 | 425 | err = get_user_pages_fast(address, 1, 1, &page); |
9ea71503 SB |
426 | /* |
427 | * If write access is not required (eg. FUTEX_WAIT), try | |
428 | * and get read-only access. | |
429 | */ | |
430 | if (err == -EFAULT && rw == VERIFY_READ) { | |
431 | err = get_user_pages_fast(address, 1, 0, &page); | |
432 | ro = 1; | |
433 | } | |
38d47c1b PZ |
434 | if (err < 0) |
435 | return err; | |
9ea71503 SB |
436 | else |
437 | err = 0; | |
38d47c1b | 438 | |
a5b338f2 AA |
439 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
440 | page_head = page; | |
441 | if (unlikely(PageTail(page))) { | |
38d47c1b | 442 | put_page(page); |
a5b338f2 AA |
443 | /* serialize against __split_huge_page_splitting() */ |
444 | local_irq_disable(); | |
f12d5bfc | 445 | if (likely(__get_user_pages_fast(address, 1, !ro, &page) == 1)) { |
a5b338f2 AA |
446 | page_head = compound_head(page); |
447 | /* | |
448 | * page_head is valid pointer but we must pin | |
449 | * it before taking the PG_lock and/or | |
450 | * PG_compound_lock. The moment we re-enable | |
451 | * irqs __split_huge_page_splitting() can | |
452 | * return and the head page can be freed from | |
453 | * under us. We can't take the PG_lock and/or | |
454 | * PG_compound_lock on a page that could be | |
455 | * freed from under us. | |
456 | */ | |
457 | if (page != page_head) { | |
458 | get_page(page_head); | |
459 | put_page(page); | |
460 | } | |
461 | local_irq_enable(); | |
462 | } else { | |
463 | local_irq_enable(); | |
464 | goto again; | |
465 | } | |
466 | } | |
467 | #else | |
468 | page_head = compound_head(page); | |
469 | if (page != page_head) { | |
470 | get_page(page_head); | |
471 | put_page(page); | |
472 | } | |
473 | #endif | |
474 | ||
475 | lock_page(page_head); | |
e6780f72 HD |
476 | |
477 | /* | |
478 | * If page_head->mapping is NULL, then it cannot be a PageAnon | |
479 | * page; but it might be the ZERO_PAGE or in the gate area or | |
480 | * in a special mapping (all cases which we are happy to fail); | |
481 | * or it may have been a good file page when get_user_pages_fast | |
482 | * found it, but truncated or holepunched or subjected to | |
483 | * invalidate_complete_page2 before we got the page lock (also | |
484 | * cases which we are happy to fail). And we hold a reference, | |
485 | * so refcount care in invalidate_complete_page's remove_mapping | |
486 | * prevents drop_caches from setting mapping to NULL beneath us. | |
487 | * | |
488 | * The case we do have to guard against is when memory pressure made | |
489 | * shmem_writepage move it from filecache to swapcache beneath us: | |
490 | * an unlikely race, but we do need to retry for page_head->mapping. | |
491 | */ | |
a5b338f2 | 492 | if (!page_head->mapping) { |
e6780f72 | 493 | int shmem_swizzled = PageSwapCache(page_head); |
a5b338f2 AA |
494 | unlock_page(page_head); |
495 | put_page(page_head); | |
e6780f72 HD |
496 | if (shmem_swizzled) |
497 | goto again; | |
498 | return -EFAULT; | |
38d47c1b | 499 | } |
1da177e4 LT |
500 | |
501 | /* | |
502 | * Private mappings are handled in a simple way. | |
503 | * | |
504 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
505 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 506 | * the object not the particular process. |
1da177e4 | 507 | */ |
a5b338f2 | 508 | if (PageAnon(page_head)) { |
9ea71503 SB |
509 | /* |
510 | * A RO anonymous page will never change and thus doesn't make | |
511 | * sense for futex operations. | |
512 | */ | |
513 | if (ro) { | |
514 | err = -EFAULT; | |
515 | goto out; | |
516 | } | |
517 | ||
38d47c1b | 518 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 519 | key->private.mm = mm; |
e2970f2f | 520 | key->private.address = address; |
38d47c1b PZ |
521 | } else { |
522 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ | |
a5b338f2 | 523 | key->shared.inode = page_head->mapping->host; |
13d60f4b | 524 | key->shared.pgoff = basepage_index(page); |
1da177e4 LT |
525 | } |
526 | ||
b0c29f79 | 527 | get_futex_key_refs(key); /* implies MB (B) */ |
1da177e4 | 528 | |
9ea71503 | 529 | out: |
a5b338f2 AA |
530 | unlock_page(page_head); |
531 | put_page(page_head); | |
9ea71503 | 532 | return err; |
1da177e4 LT |
533 | } |
534 | ||
ae791a2d | 535 | static inline void put_futex_key(union futex_key *key) |
1da177e4 | 536 | { |
38d47c1b | 537 | drop_futex_key_refs(key); |
1da177e4 LT |
538 | } |
539 | ||
d96ee56c DH |
540 | /** |
541 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
542 | * @uaddr: pointer to faulting user space address |
543 | * | |
544 | * Slow path to fixup the fault we just took in the atomic write | |
545 | * access to @uaddr. | |
546 | * | |
fb62db2b | 547 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
548 | * user address. We know that we faulted in the atomic pagefault |
549 | * disabled section so we can as well avoid the #PF overhead by | |
550 | * calling get_user_pages() right away. | |
551 | */ | |
552 | static int fault_in_user_writeable(u32 __user *uaddr) | |
553 | { | |
722d0172 AK |
554 | struct mm_struct *mm = current->mm; |
555 | int ret; | |
556 | ||
557 | down_read(&mm->mmap_sem); | |
2efaca92 BH |
558 | ret = fixup_user_fault(current, mm, (unsigned long)uaddr, |
559 | FAULT_FLAG_WRITE); | |
722d0172 AK |
560 | up_read(&mm->mmap_sem); |
561 | ||
d0725992 TG |
562 | return ret < 0 ? ret : 0; |
563 | } | |
564 | ||
4b1c486b DH |
565 | /** |
566 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
567 | * @hb: the hash bucket the futex_q's reside in |
568 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
569 | * |
570 | * Must be called with the hb lock held. | |
571 | */ | |
572 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
573 | union futex_key *key) | |
574 | { | |
575 | struct futex_q *this; | |
576 | ||
577 | plist_for_each_entry(this, &hb->chain, list) { | |
578 | if (match_futex(&this->key, key)) | |
579 | return this; | |
580 | } | |
581 | return NULL; | |
582 | } | |
583 | ||
37a9d912 ML |
584 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
585 | u32 uval, u32 newval) | |
36cf3b5c | 586 | { |
37a9d912 | 587 | int ret; |
36cf3b5c TG |
588 | |
589 | pagefault_disable(); | |
37a9d912 | 590 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
591 | pagefault_enable(); |
592 | ||
37a9d912 | 593 | return ret; |
36cf3b5c TG |
594 | } |
595 | ||
596 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
597 | { |
598 | int ret; | |
599 | ||
a866374a | 600 | pagefault_disable(); |
e2970f2f | 601 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 602 | pagefault_enable(); |
1da177e4 LT |
603 | |
604 | return ret ? -EFAULT : 0; | |
605 | } | |
606 | ||
c87e2837 IM |
607 | |
608 | /* | |
609 | * PI code: | |
610 | */ | |
611 | static int refill_pi_state_cache(void) | |
612 | { | |
613 | struct futex_pi_state *pi_state; | |
614 | ||
615 | if (likely(current->pi_state_cache)) | |
616 | return 0; | |
617 | ||
4668edc3 | 618 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
619 | |
620 | if (!pi_state) | |
621 | return -ENOMEM; | |
622 | ||
c87e2837 IM |
623 | INIT_LIST_HEAD(&pi_state->list); |
624 | /* pi_mutex gets initialized later */ | |
625 | pi_state->owner = NULL; | |
626 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 627 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
628 | |
629 | current->pi_state_cache = pi_state; | |
630 | ||
631 | return 0; | |
632 | } | |
633 | ||
634 | static struct futex_pi_state * alloc_pi_state(void) | |
635 | { | |
636 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
637 | ||
638 | WARN_ON(!pi_state); | |
639 | current->pi_state_cache = NULL; | |
640 | ||
641 | return pi_state; | |
642 | } | |
643 | ||
644 | static void free_pi_state(struct futex_pi_state *pi_state) | |
645 | { | |
646 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
647 | return; | |
648 | ||
649 | /* | |
650 | * If pi_state->owner is NULL, the owner is most probably dying | |
651 | * and has cleaned up the pi_state already | |
652 | */ | |
653 | if (pi_state->owner) { | |
1d615482 | 654 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
c87e2837 | 655 | list_del_init(&pi_state->list); |
1d615482 | 656 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
c87e2837 IM |
657 | |
658 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
659 | } | |
660 | ||
661 | if (current->pi_state_cache) | |
662 | kfree(pi_state); | |
663 | else { | |
664 | /* | |
665 | * pi_state->list is already empty. | |
666 | * clear pi_state->owner. | |
667 | * refcount is at 0 - put it back to 1. | |
668 | */ | |
669 | pi_state->owner = NULL; | |
670 | atomic_set(&pi_state->refcount, 1); | |
671 | current->pi_state_cache = pi_state; | |
672 | } | |
673 | } | |
674 | ||
675 | /* | |
676 | * Look up the task based on what TID userspace gave us. | |
677 | * We dont trust it. | |
678 | */ | |
679 | static struct task_struct * futex_find_get_task(pid_t pid) | |
680 | { | |
681 | struct task_struct *p; | |
682 | ||
d359b549 | 683 | rcu_read_lock(); |
228ebcbe | 684 | p = find_task_by_vpid(pid); |
7a0ea09a MH |
685 | if (p) |
686 | get_task_struct(p); | |
a06381fe | 687 | |
d359b549 | 688 | rcu_read_unlock(); |
c87e2837 IM |
689 | |
690 | return p; | |
691 | } | |
692 | ||
693 | /* | |
694 | * This task is holding PI mutexes at exit time => bad. | |
695 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
696 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
697 | */ | |
698 | void exit_pi_state_list(struct task_struct *curr) | |
699 | { | |
c87e2837 IM |
700 | struct list_head *next, *head = &curr->pi_state_list; |
701 | struct futex_pi_state *pi_state; | |
627371d7 | 702 | struct futex_hash_bucket *hb; |
38d47c1b | 703 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 704 | |
a0c1e907 TG |
705 | if (!futex_cmpxchg_enabled) |
706 | return; | |
c87e2837 IM |
707 | /* |
708 | * We are a ZOMBIE and nobody can enqueue itself on | |
709 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 710 | * versus waiters unqueueing themselves: |
c87e2837 | 711 | */ |
1d615482 | 712 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 IM |
713 | while (!list_empty(head)) { |
714 | ||
715 | next = head->next; | |
716 | pi_state = list_entry(next, struct futex_pi_state, list); | |
717 | key = pi_state->key; | |
627371d7 | 718 | hb = hash_futex(&key); |
1d615482 | 719 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 720 | |
c87e2837 IM |
721 | spin_lock(&hb->lock); |
722 | ||
1d615482 | 723 | raw_spin_lock_irq(&curr->pi_lock); |
627371d7 IM |
724 | /* |
725 | * We dropped the pi-lock, so re-check whether this | |
726 | * task still owns the PI-state: | |
727 | */ | |
c87e2837 IM |
728 | if (head->next != next) { |
729 | spin_unlock(&hb->lock); | |
730 | continue; | |
731 | } | |
732 | ||
c87e2837 | 733 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
734 | WARN_ON(list_empty(&pi_state->list)); |
735 | list_del_init(&pi_state->list); | |
c87e2837 | 736 | pi_state->owner = NULL; |
1d615482 | 737 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
738 | |
739 | rt_mutex_unlock(&pi_state->pi_mutex); | |
740 | ||
741 | spin_unlock(&hb->lock); | |
742 | ||
1d615482 | 743 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 744 | } |
1d615482 | 745 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
746 | } |
747 | ||
54a21788 TG |
748 | /* |
749 | * We need to check the following states: | |
750 | * | |
751 | * Waiter | pi_state | pi->owner | uTID | uODIED | ? | |
752 | * | |
753 | * [1] NULL | --- | --- | 0 | 0/1 | Valid | |
754 | * [2] NULL | --- | --- | >0 | 0/1 | Valid | |
755 | * | |
756 | * [3] Found | NULL | -- | Any | 0/1 | Invalid | |
757 | * | |
758 | * [4] Found | Found | NULL | 0 | 1 | Valid | |
759 | * [5] Found | Found | NULL | >0 | 1 | Invalid | |
760 | * | |
761 | * [6] Found | Found | task | 0 | 1 | Valid | |
762 | * | |
763 | * [7] Found | Found | NULL | Any | 0 | Invalid | |
764 | * | |
765 | * [8] Found | Found | task | ==taskTID | 0/1 | Valid | |
766 | * [9] Found | Found | task | 0 | 0 | Invalid | |
767 | * [10] Found | Found | task | !=taskTID | 0/1 | Invalid | |
768 | * | |
769 | * [1] Indicates that the kernel can acquire the futex atomically. We | |
770 | * came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. | |
771 | * | |
772 | * [2] Valid, if TID does not belong to a kernel thread. If no matching | |
773 | * thread is found then it indicates that the owner TID has died. | |
774 | * | |
775 | * [3] Invalid. The waiter is queued on a non PI futex | |
776 | * | |
777 | * [4] Valid state after exit_robust_list(), which sets the user space | |
778 | * value to FUTEX_WAITERS | FUTEX_OWNER_DIED. | |
779 | * | |
780 | * [5] The user space value got manipulated between exit_robust_list() | |
781 | * and exit_pi_state_list() | |
782 | * | |
783 | * [6] Valid state after exit_pi_state_list() which sets the new owner in | |
784 | * the pi_state but cannot access the user space value. | |
785 | * | |
786 | * [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. | |
787 | * | |
788 | * [8] Owner and user space value match | |
789 | * | |
790 | * [9] There is no transient state which sets the user space TID to 0 | |
791 | * except exit_robust_list(), but this is indicated by the | |
792 | * FUTEX_OWNER_DIED bit. See [4] | |
793 | * | |
794 | * [10] There is no transient state which leaves owner and user space | |
795 | * TID out of sync. | |
796 | */ | |
e60cbc5c TG |
797 | |
798 | /* | |
799 | * Validate that the existing waiter has a pi_state and sanity check | |
800 | * the pi_state against the user space value. If correct, attach to | |
801 | * it. | |
802 | */ | |
803 | static int attach_to_pi_state(u32 uval, struct futex_pi_state *pi_state, | |
804 | struct futex_pi_state **ps) | |
c87e2837 | 805 | { |
778e9a9c | 806 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 | 807 | |
e60cbc5c TG |
808 | /* |
809 | * Userspace might have messed up non-PI and PI futexes [3] | |
810 | */ | |
811 | if (unlikely(!pi_state)) | |
812 | return -EINVAL; | |
06a9ec29 | 813 | |
e60cbc5c | 814 | WARN_ON(!atomic_read(&pi_state->refcount)); |
59647b6a | 815 | |
e60cbc5c TG |
816 | /* |
817 | * Handle the owner died case: | |
818 | */ | |
819 | if (uval & FUTEX_OWNER_DIED) { | |
bd1dbcc6 | 820 | /* |
e60cbc5c TG |
821 | * exit_pi_state_list sets owner to NULL and wakes the |
822 | * topmost waiter. The task which acquires the | |
823 | * pi_state->rt_mutex will fixup owner. | |
bd1dbcc6 | 824 | */ |
e60cbc5c | 825 | if (!pi_state->owner) { |
59647b6a | 826 | /* |
e60cbc5c TG |
827 | * No pi state owner, but the user space TID |
828 | * is not 0. Inconsistent state. [5] | |
59647b6a | 829 | */ |
e60cbc5c TG |
830 | if (pid) |
831 | return -EINVAL; | |
bd1dbcc6 | 832 | /* |
e60cbc5c | 833 | * Take a ref on the state and return success. [4] |
866293ee | 834 | */ |
e60cbc5c | 835 | goto out_state; |
c87e2837 | 836 | } |
bd1dbcc6 TG |
837 | |
838 | /* | |
e60cbc5c TG |
839 | * If TID is 0, then either the dying owner has not |
840 | * yet executed exit_pi_state_list() or some waiter | |
841 | * acquired the rtmutex in the pi state, but did not | |
842 | * yet fixup the TID in user space. | |
843 | * | |
844 | * Take a ref on the state and return success. [6] | |
845 | */ | |
846 | if (!pid) | |
847 | goto out_state; | |
848 | } else { | |
849 | /* | |
850 | * If the owner died bit is not set, then the pi_state | |
851 | * must have an owner. [7] | |
bd1dbcc6 | 852 | */ |
e60cbc5c | 853 | if (!pi_state->owner) |
bd1dbcc6 | 854 | return -EINVAL; |
c87e2837 IM |
855 | } |
856 | ||
e60cbc5c TG |
857 | /* |
858 | * Bail out if user space manipulated the futex value. If pi | |
859 | * state exists then the owner TID must be the same as the | |
860 | * user space TID. [9/10] | |
861 | */ | |
862 | if (pid != task_pid_vnr(pi_state->owner)) | |
863 | return -EINVAL; | |
864 | out_state: | |
865 | atomic_inc(&pi_state->refcount); | |
866 | *ps = pi_state; | |
867 | return 0; | |
868 | } | |
869 | ||
04e1b2e5 TG |
870 | /* |
871 | * Lookup the task for the TID provided from user space and attach to | |
872 | * it after doing proper sanity checks. | |
873 | */ | |
874 | static int attach_to_pi_owner(u32 uval, union futex_key *key, | |
875 | struct futex_pi_state **ps) | |
e60cbc5c | 876 | { |
e60cbc5c | 877 | pid_t pid = uval & FUTEX_TID_MASK; |
04e1b2e5 TG |
878 | struct futex_pi_state *pi_state; |
879 | struct task_struct *p; | |
e60cbc5c | 880 | |
c87e2837 | 881 | /* |
e3f2ddea | 882 | * We are the first waiter - try to look up the real owner and attach |
54a21788 | 883 | * the new pi_state to it, but bail out when TID = 0 [1] |
c87e2837 | 884 | */ |
778e9a9c | 885 | if (!pid) |
e3f2ddea | 886 | return -ESRCH; |
c87e2837 | 887 | p = futex_find_get_task(pid); |
7a0ea09a MH |
888 | if (!p) |
889 | return -ESRCH; | |
778e9a9c | 890 | |
f0d71b3d TG |
891 | if (!p->mm) { |
892 | put_task_struct(p); | |
893 | return -EPERM; | |
894 | } | |
895 | ||
778e9a9c AK |
896 | /* |
897 | * We need to look at the task state flags to figure out, | |
898 | * whether the task is exiting. To protect against the do_exit | |
899 | * change of the task flags, we do this protected by | |
900 | * p->pi_lock: | |
901 | */ | |
1d615482 | 902 | raw_spin_lock_irq(&p->pi_lock); |
778e9a9c AK |
903 | if (unlikely(p->flags & PF_EXITING)) { |
904 | /* | |
905 | * The task is on the way out. When PF_EXITPIDONE is | |
906 | * set, we know that the task has finished the | |
907 | * cleanup: | |
908 | */ | |
909 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
910 | ||
1d615482 | 911 | raw_spin_unlock_irq(&p->pi_lock); |
778e9a9c AK |
912 | put_task_struct(p); |
913 | return ret; | |
914 | } | |
c87e2837 | 915 | |
54a21788 TG |
916 | /* |
917 | * No existing pi state. First waiter. [2] | |
918 | */ | |
c87e2837 IM |
919 | pi_state = alloc_pi_state(); |
920 | ||
921 | /* | |
04e1b2e5 | 922 | * Initialize the pi_mutex in locked state and make @p |
c87e2837 IM |
923 | * the owner of it: |
924 | */ | |
925 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
926 | ||
927 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 928 | pi_state->key = *key; |
c87e2837 | 929 | |
627371d7 | 930 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
931 | list_add(&pi_state->list, &p->pi_state_list); |
932 | pi_state->owner = p; | |
1d615482 | 933 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
934 | |
935 | put_task_struct(p); | |
936 | ||
d0aa7a70 | 937 | *ps = pi_state; |
c87e2837 IM |
938 | |
939 | return 0; | |
940 | } | |
941 | ||
04e1b2e5 TG |
942 | static int lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
943 | union futex_key *key, struct futex_pi_state **ps) | |
944 | { | |
945 | struct futex_q *match = futex_top_waiter(hb, key); | |
946 | ||
947 | /* | |
948 | * If there is a waiter on that futex, validate it and | |
949 | * attach to the pi_state when the validation succeeds. | |
950 | */ | |
951 | if (match) | |
952 | return attach_to_pi_state(uval, match->pi_state, ps); | |
953 | ||
954 | /* | |
955 | * We are the first waiter - try to look up the owner based on | |
956 | * @uval and attach to it. | |
957 | */ | |
958 | return attach_to_pi_owner(uval, key, ps); | |
959 | } | |
960 | ||
af54d6a1 TG |
961 | static int lock_pi_update_atomic(u32 __user *uaddr, u32 uval, u32 newval) |
962 | { | |
963 | u32 uninitialized_var(curval); | |
964 | ||
965 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))) | |
966 | return -EFAULT; | |
967 | ||
968 | /*If user space value changed, let the caller retry */ | |
969 | return curval != uval ? -EAGAIN : 0; | |
970 | } | |
971 | ||
1a52084d | 972 | /** |
d96ee56c | 973 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
974 | * @uaddr: the pi futex user address |
975 | * @hb: the pi futex hash bucket | |
976 | * @key: the futex key associated with uaddr and hb | |
977 | * @ps: the pi_state pointer where we store the result of the | |
978 | * lookup | |
979 | * @task: the task to perform the atomic lock work for. This will | |
980 | * be "current" except in the case of requeue pi. | |
981 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d | 982 | * |
6c23cbbd RD |
983 | * Return: |
984 | * 0 - ready to wait; | |
985 | * 1 - acquired the lock; | |
1a52084d DH |
986 | * <0 - error |
987 | * | |
988 | * The hb->lock and futex_key refs shall be held by the caller. | |
989 | */ | |
990 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
991 | union futex_key *key, | |
992 | struct futex_pi_state **ps, | |
bab5bc9e | 993 | struct task_struct *task, int set_waiters) |
1a52084d | 994 | { |
af54d6a1 TG |
995 | u32 uval, newval, vpid = task_pid_vnr(task); |
996 | struct futex_q *match; | |
997 | int ret; | |
1a52084d DH |
998 | |
999 | /* | |
af54d6a1 TG |
1000 | * Read the user space value first so we can validate a few |
1001 | * things before proceeding further. | |
1a52084d | 1002 | */ |
af54d6a1 | 1003 | if (get_futex_value_locked(&uval, uaddr)) |
1a52084d DH |
1004 | return -EFAULT; |
1005 | ||
1006 | /* | |
1007 | * Detect deadlocks. | |
1008 | */ | |
af54d6a1 | 1009 | if ((unlikely((uval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
1010 | return -EDEADLK; |
1011 | ||
1012 | /* | |
af54d6a1 TG |
1013 | * Lookup existing state first. If it exists, try to attach to |
1014 | * its pi_state. | |
1a52084d | 1015 | */ |
af54d6a1 TG |
1016 | match = futex_top_waiter(hb, key); |
1017 | if (match) | |
1018 | return attach_to_pi_state(uval, match->pi_state, ps); | |
1a52084d DH |
1019 | |
1020 | /* | |
af54d6a1 TG |
1021 | * No waiter and user TID is 0. We are here because the |
1022 | * waiters or the owner died bit is set or called from | |
1023 | * requeue_cmp_pi or for whatever reason something took the | |
1024 | * syscall. | |
1a52084d | 1025 | */ |
af54d6a1 | 1026 | if (!(uval & FUTEX_TID_MASK)) { |
59fa6245 | 1027 | /* |
af54d6a1 TG |
1028 | * We take over the futex. No other waiters and the user space |
1029 | * TID is 0. We preserve the owner died bit. | |
59fa6245 | 1030 | */ |
af54d6a1 TG |
1031 | newval = uval & FUTEX_OWNER_DIED; |
1032 | newval |= vpid; | |
1a52084d | 1033 | |
af54d6a1 TG |
1034 | /* The futex requeue_pi code can enforce the waiters bit */ |
1035 | if (set_waiters) | |
1036 | newval |= FUTEX_WAITERS; | |
1037 | ||
1038 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1039 | /* If the take over worked, return 1 */ | |
1040 | return ret < 0 ? ret : 1; | |
1041 | } | |
1a52084d DH |
1042 | |
1043 | /* | |
af54d6a1 TG |
1044 | * First waiter. Set the waiters bit before attaching ourself to |
1045 | * the owner. If owner tries to unlock, it will be forced into | |
1046 | * the kernel and blocked on hb->lock. | |
1a52084d | 1047 | */ |
af54d6a1 TG |
1048 | newval = uval | FUTEX_WAITERS; |
1049 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1050 | if (ret) | |
1051 | return ret; | |
1a52084d | 1052 | /* |
af54d6a1 TG |
1053 | * If the update of the user space value succeeded, we try to |
1054 | * attach to the owner. If that fails, no harm done, we only | |
1055 | * set the FUTEX_WAITERS bit in the user space variable. | |
1a52084d | 1056 | */ |
af54d6a1 | 1057 | return attach_to_pi_owner(uval, key, ps); |
1a52084d DH |
1058 | } |
1059 | ||
2e12978a LJ |
1060 | /** |
1061 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
1062 | * @q: The futex_q to unqueue | |
1063 | * | |
1064 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
1065 | */ | |
1066 | static void __unqueue_futex(struct futex_q *q) | |
1067 | { | |
1068 | struct futex_hash_bucket *hb; | |
1069 | ||
29096202 SR |
1070 | if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr)) |
1071 | || WARN_ON(plist_node_empty(&q->list))) | |
2e12978a LJ |
1072 | return; |
1073 | ||
1074 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
1075 | plist_del(&q->list, &hb->chain); | |
11d4616b | 1076 | hb_waiters_dec(hb); |
2e12978a LJ |
1077 | } |
1078 | ||
1da177e4 LT |
1079 | /* |
1080 | * The hash bucket lock must be held when this is called. | |
1081 | * Afterwards, the futex_q must not be accessed. | |
1082 | */ | |
1083 | static void wake_futex(struct futex_q *q) | |
1084 | { | |
f1a11e05 TG |
1085 | struct task_struct *p = q->task; |
1086 | ||
aa10990e DH |
1087 | if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n")) |
1088 | return; | |
1089 | ||
1da177e4 | 1090 | /* |
f1a11e05 | 1091 | * We set q->lock_ptr = NULL _before_ we wake up the task. If |
fb62db2b RD |
1092 | * a non-futex wake up happens on another CPU then the task |
1093 | * might exit and p would dereference a non-existing task | |
f1a11e05 TG |
1094 | * struct. Prevent this by holding a reference on p across the |
1095 | * wake up. | |
1da177e4 | 1096 | */ |
f1a11e05 TG |
1097 | get_task_struct(p); |
1098 | ||
2e12978a | 1099 | __unqueue_futex(q); |
1da177e4 | 1100 | /* |
f1a11e05 TG |
1101 | * The waiting task can free the futex_q as soon as |
1102 | * q->lock_ptr = NULL is written, without taking any locks. A | |
1103 | * memory barrier is required here to prevent the following | |
1104 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 1105 | */ |
ccdea2f8 | 1106 | smp_wmb(); |
1da177e4 | 1107 | q->lock_ptr = NULL; |
f1a11e05 TG |
1108 | |
1109 | wake_up_state(p, TASK_NORMAL); | |
1110 | put_task_struct(p); | |
1da177e4 LT |
1111 | } |
1112 | ||
c87e2837 IM |
1113 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
1114 | { | |
1115 | struct task_struct *new_owner; | |
1116 | struct futex_pi_state *pi_state = this->pi_state; | |
7cfdaf38 | 1117 | u32 uninitialized_var(curval), newval; |
13fbca4c | 1118 | int ret = 0; |
c87e2837 IM |
1119 | |
1120 | if (!pi_state) | |
1121 | return -EINVAL; | |
1122 | ||
51246bfd TG |
1123 | /* |
1124 | * If current does not own the pi_state then the futex is | |
1125 | * inconsistent and user space fiddled with the futex value. | |
1126 | */ | |
1127 | if (pi_state->owner != current) | |
1128 | return -EINVAL; | |
1129 | ||
d209d74d | 1130 | raw_spin_lock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
1131 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
1132 | ||
1133 | /* | |
f123c98e SR |
1134 | * It is possible that the next waiter (the one that brought |
1135 | * this owner to the kernel) timed out and is no longer | |
1136 | * waiting on the lock. | |
c87e2837 IM |
1137 | */ |
1138 | if (!new_owner) | |
1139 | new_owner = this->task; | |
1140 | ||
1141 | /* | |
13fbca4c TG |
1142 | * We pass it to the next owner. The WAITERS bit is always |
1143 | * kept enabled while there is PI state around. We cleanup the | |
1144 | * owner died bit, because we are the owner. | |
c87e2837 | 1145 | */ |
13fbca4c | 1146 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 1147 | |
13fbca4c TG |
1148 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1149 | ret = -EFAULT; | |
1150 | else if (curval != uval) | |
1151 | ret = -EINVAL; | |
1152 | if (ret) { | |
1153 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); | |
1154 | return ret; | |
e3f2ddea | 1155 | } |
c87e2837 | 1156 | |
1d615482 | 1157 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
627371d7 IM |
1158 | WARN_ON(list_empty(&pi_state->list)); |
1159 | list_del_init(&pi_state->list); | |
1d615482 | 1160 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
627371d7 | 1161 | |
1d615482 | 1162 | raw_spin_lock_irq(&new_owner->pi_lock); |
627371d7 | 1163 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
1164 | list_add(&pi_state->list, &new_owner->pi_state_list); |
1165 | pi_state->owner = new_owner; | |
1d615482 | 1166 | raw_spin_unlock_irq(&new_owner->pi_lock); |
627371d7 | 1167 | |
d209d74d | 1168 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
1169 | rt_mutex_unlock(&pi_state->pi_mutex); |
1170 | ||
1171 | return 0; | |
1172 | } | |
1173 | ||
8b8f319f IM |
1174 | /* |
1175 | * Express the locking dependencies for lockdep: | |
1176 | */ | |
1177 | static inline void | |
1178 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1179 | { | |
1180 | if (hb1 <= hb2) { | |
1181 | spin_lock(&hb1->lock); | |
1182 | if (hb1 < hb2) | |
1183 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
1184 | } else { /* hb1 > hb2 */ | |
1185 | spin_lock(&hb2->lock); | |
1186 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
1187 | } | |
1188 | } | |
1189 | ||
5eb3dc62 DH |
1190 | static inline void |
1191 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1192 | { | |
f061d351 | 1193 | spin_unlock(&hb1->lock); |
88f502fe IM |
1194 | if (hb1 != hb2) |
1195 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
1196 | } |
1197 | ||
1da177e4 | 1198 | /* |
b2d0994b | 1199 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 1200 | */ |
b41277dc DH |
1201 | static int |
1202 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 1203 | { |
e2970f2f | 1204 | struct futex_hash_bucket *hb; |
1da177e4 | 1205 | struct futex_q *this, *next; |
38d47c1b | 1206 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 LT |
1207 | int ret; |
1208 | ||
cd689985 TG |
1209 | if (!bitset) |
1210 | return -EINVAL; | |
1211 | ||
9ea71503 | 1212 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ); |
1da177e4 LT |
1213 | if (unlikely(ret != 0)) |
1214 | goto out; | |
1215 | ||
e2970f2f | 1216 | hb = hash_futex(&key); |
b0c29f79 DB |
1217 | |
1218 | /* Make sure we really have tasks to wakeup */ | |
1219 | if (!hb_waiters_pending(hb)) | |
1220 | goto out_put_key; | |
1221 | ||
e2970f2f | 1222 | spin_lock(&hb->lock); |
1da177e4 | 1223 | |
0d00c7b2 | 1224 | plist_for_each_entry_safe(this, next, &hb->chain, list) { |
1da177e4 | 1225 | if (match_futex (&this->key, &key)) { |
52400ba9 | 1226 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
1227 | ret = -EINVAL; |
1228 | break; | |
1229 | } | |
cd689985 TG |
1230 | |
1231 | /* Check if one of the bits is set in both bitsets */ | |
1232 | if (!(this->bitset & bitset)) | |
1233 | continue; | |
1234 | ||
1da177e4 LT |
1235 | wake_futex(this); |
1236 | if (++ret >= nr_wake) | |
1237 | break; | |
1238 | } | |
1239 | } | |
1240 | ||
e2970f2f | 1241 | spin_unlock(&hb->lock); |
b0c29f79 | 1242 | out_put_key: |
ae791a2d | 1243 | put_futex_key(&key); |
42d35d48 | 1244 | out: |
1da177e4 LT |
1245 | return ret; |
1246 | } | |
1247 | ||
4732efbe JJ |
1248 | /* |
1249 | * Wake up all waiters hashed on the physical page that is mapped | |
1250 | * to this virtual address: | |
1251 | */ | |
e2970f2f | 1252 | static int |
b41277dc | 1253 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 1254 | int nr_wake, int nr_wake2, int op) |
4732efbe | 1255 | { |
38d47c1b | 1256 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 1257 | struct futex_hash_bucket *hb1, *hb2; |
4732efbe | 1258 | struct futex_q *this, *next; |
e4dc5b7a | 1259 | int ret, op_ret; |
4732efbe | 1260 | |
e4dc5b7a | 1261 | retry: |
9ea71503 | 1262 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
4732efbe JJ |
1263 | if (unlikely(ret != 0)) |
1264 | goto out; | |
9ea71503 | 1265 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
4732efbe | 1266 | if (unlikely(ret != 0)) |
42d35d48 | 1267 | goto out_put_key1; |
4732efbe | 1268 | |
e2970f2f IM |
1269 | hb1 = hash_futex(&key1); |
1270 | hb2 = hash_futex(&key2); | |
4732efbe | 1271 | |
e4dc5b7a | 1272 | retry_private: |
eaaea803 | 1273 | double_lock_hb(hb1, hb2); |
e2970f2f | 1274 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1275 | if (unlikely(op_ret < 0)) { |
4732efbe | 1276 | |
5eb3dc62 | 1277 | double_unlock_hb(hb1, hb2); |
4732efbe | 1278 | |
7ee1dd3f | 1279 | #ifndef CONFIG_MMU |
e2970f2f IM |
1280 | /* |
1281 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
1282 | * but we might get them from range checking | |
1283 | */ | |
7ee1dd3f | 1284 | ret = op_ret; |
42d35d48 | 1285 | goto out_put_keys; |
7ee1dd3f DH |
1286 | #endif |
1287 | ||
796f8d9b DG |
1288 | if (unlikely(op_ret != -EFAULT)) { |
1289 | ret = op_ret; | |
42d35d48 | 1290 | goto out_put_keys; |
796f8d9b DG |
1291 | } |
1292 | ||
d0725992 | 1293 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 1294 | if (ret) |
de87fcc1 | 1295 | goto out_put_keys; |
4732efbe | 1296 | |
b41277dc | 1297 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1298 | goto retry_private; |
1299 | ||
ae791a2d TG |
1300 | put_futex_key(&key2); |
1301 | put_futex_key(&key1); | |
e4dc5b7a | 1302 | goto retry; |
4732efbe JJ |
1303 | } |
1304 | ||
0d00c7b2 | 1305 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
4732efbe | 1306 | if (match_futex (&this->key, &key1)) { |
aa10990e DH |
1307 | if (this->pi_state || this->rt_waiter) { |
1308 | ret = -EINVAL; | |
1309 | goto out_unlock; | |
1310 | } | |
4732efbe JJ |
1311 | wake_futex(this); |
1312 | if (++ret >= nr_wake) | |
1313 | break; | |
1314 | } | |
1315 | } | |
1316 | ||
1317 | if (op_ret > 0) { | |
4732efbe | 1318 | op_ret = 0; |
0d00c7b2 | 1319 | plist_for_each_entry_safe(this, next, &hb2->chain, list) { |
4732efbe | 1320 | if (match_futex (&this->key, &key2)) { |
aa10990e DH |
1321 | if (this->pi_state || this->rt_waiter) { |
1322 | ret = -EINVAL; | |
1323 | goto out_unlock; | |
1324 | } | |
4732efbe JJ |
1325 | wake_futex(this); |
1326 | if (++op_ret >= nr_wake2) | |
1327 | break; | |
1328 | } | |
1329 | } | |
1330 | ret += op_ret; | |
1331 | } | |
1332 | ||
aa10990e | 1333 | out_unlock: |
5eb3dc62 | 1334 | double_unlock_hb(hb1, hb2); |
42d35d48 | 1335 | out_put_keys: |
ae791a2d | 1336 | put_futex_key(&key2); |
42d35d48 | 1337 | out_put_key1: |
ae791a2d | 1338 | put_futex_key(&key1); |
42d35d48 | 1339 | out: |
4732efbe JJ |
1340 | return ret; |
1341 | } | |
1342 | ||
9121e478 DH |
1343 | /** |
1344 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1345 | * @q: the futex_q to requeue | |
1346 | * @hb1: the source hash_bucket | |
1347 | * @hb2: the target hash_bucket | |
1348 | * @key2: the new key for the requeued futex_q | |
1349 | */ | |
1350 | static inline | |
1351 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1352 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1353 | { | |
1354 | ||
1355 | /* | |
1356 | * If key1 and key2 hash to the same bucket, no need to | |
1357 | * requeue. | |
1358 | */ | |
1359 | if (likely(&hb1->chain != &hb2->chain)) { | |
1360 | plist_del(&q->list, &hb1->chain); | |
11d4616b | 1361 | hb_waiters_dec(hb1); |
9121e478 | 1362 | plist_add(&q->list, &hb2->chain); |
11d4616b | 1363 | hb_waiters_inc(hb2); |
9121e478 | 1364 | q->lock_ptr = &hb2->lock; |
9121e478 DH |
1365 | } |
1366 | get_futex_key_refs(key2); | |
1367 | q->key = *key2; | |
1368 | } | |
1369 | ||
52400ba9 DH |
1370 | /** |
1371 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1372 | * @q: the futex_q |
1373 | * @key: the key of the requeue target futex | |
1374 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1375 | * |
1376 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1377 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1378 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1379 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1380 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1381 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1382 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1383 | */ |
1384 | static inline | |
beda2c7e DH |
1385 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1386 | struct futex_hash_bucket *hb) | |
52400ba9 | 1387 | { |
52400ba9 DH |
1388 | get_futex_key_refs(key); |
1389 | q->key = *key; | |
1390 | ||
2e12978a | 1391 | __unqueue_futex(q); |
52400ba9 DH |
1392 | |
1393 | WARN_ON(!q->rt_waiter); | |
1394 | q->rt_waiter = NULL; | |
1395 | ||
beda2c7e | 1396 | q->lock_ptr = &hb->lock; |
beda2c7e | 1397 | |
f1a11e05 | 1398 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1399 | } |
1400 | ||
1401 | /** | |
1402 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1403 | * @pifutex: the user address of the to futex |
1404 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1405 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1406 | * @key1: the from futex key | |
1407 | * @key2: the to futex key | |
1408 | * @ps: address to store the pi_state pointer | |
1409 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1410 | * |
1411 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1412 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1413 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1414 | * hb1 and hb2 must be held by the caller. | |
52400ba9 | 1415 | * |
6c23cbbd RD |
1416 | * Return: |
1417 | * 0 - failed to acquire the lock atomically; | |
866293ee | 1418 | * >0 - acquired the lock, return value is vpid of the top_waiter |
52400ba9 DH |
1419 | * <0 - error |
1420 | */ | |
1421 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1422 | struct futex_hash_bucket *hb1, | |
1423 | struct futex_hash_bucket *hb2, | |
1424 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1425 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1426 | { |
bab5bc9e | 1427 | struct futex_q *top_waiter = NULL; |
52400ba9 | 1428 | u32 curval; |
866293ee | 1429 | int ret, vpid; |
52400ba9 DH |
1430 | |
1431 | if (get_futex_value_locked(&curval, pifutex)) | |
1432 | return -EFAULT; | |
1433 | ||
bab5bc9e DH |
1434 | /* |
1435 | * Find the top_waiter and determine if there are additional waiters. | |
1436 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1437 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1438 | * as we have means to handle the possible fault. If not, don't set | |
1439 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1440 | * the kernel. | |
1441 | */ | |
52400ba9 DH |
1442 | top_waiter = futex_top_waiter(hb1, key1); |
1443 | ||
1444 | /* There are no waiters, nothing for us to do. */ | |
1445 | if (!top_waiter) | |
1446 | return 0; | |
1447 | ||
84bc4af5 DH |
1448 | /* Ensure we requeue to the expected futex. */ |
1449 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1450 | return -EINVAL; | |
1451 | ||
52400ba9 | 1452 | /* |
bab5bc9e DH |
1453 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1454 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1455 | * in ps in contended cases. | |
52400ba9 | 1456 | */ |
866293ee | 1457 | vpid = task_pid_vnr(top_waiter->task); |
bab5bc9e DH |
1458 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1459 | set_waiters); | |
866293ee | 1460 | if (ret == 1) { |
beda2c7e | 1461 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
866293ee TG |
1462 | return vpid; |
1463 | } | |
52400ba9 DH |
1464 | return ret; |
1465 | } | |
1466 | ||
1467 | /** | |
1468 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1469 | * @uaddr1: source futex user address |
b41277dc | 1470 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1471 | * @uaddr2: target futex user address |
1472 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1473 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1474 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1475 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1476 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1477 | * |
1478 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1479 | * uaddr2 atomically on behalf of the top waiter. | |
1480 | * | |
6c23cbbd RD |
1481 | * Return: |
1482 | * >=0 - on success, the number of tasks requeued or woken; | |
52400ba9 | 1483 | * <0 - on error |
1da177e4 | 1484 | */ |
b41277dc DH |
1485 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
1486 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
1487 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 1488 | { |
38d47c1b | 1489 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1490 | int drop_count = 0, task_count = 0, ret; |
1491 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1492 | struct futex_hash_bucket *hb1, *hb2; |
1da177e4 | 1493 | struct futex_q *this, *next; |
52400ba9 DH |
1494 | |
1495 | if (requeue_pi) { | |
e9c243a5 TG |
1496 | /* |
1497 | * Requeue PI only works on two distinct uaddrs. This | |
1498 | * check is only valid for private futexes. See below. | |
1499 | */ | |
1500 | if (uaddr1 == uaddr2) | |
1501 | return -EINVAL; | |
1502 | ||
52400ba9 DH |
1503 | /* |
1504 | * requeue_pi requires a pi_state, try to allocate it now | |
1505 | * without any locks in case it fails. | |
1506 | */ | |
1507 | if (refill_pi_state_cache()) | |
1508 | return -ENOMEM; | |
1509 | /* | |
1510 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1511 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1512 | * returning to userspace, so as to not leave the rt_mutex with | |
1513 | * waiters and no owner. However, second and third wake-ups | |
1514 | * cannot be predicted as they involve race conditions with the | |
1515 | * first wake and a fault while looking up the pi_state. Both | |
1516 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1517 | * use nr_wake=1. | |
1518 | */ | |
1519 | if (nr_wake != 1) | |
1520 | return -EINVAL; | |
1521 | } | |
1da177e4 | 1522 | |
42d35d48 | 1523 | retry: |
52400ba9 DH |
1524 | if (pi_state != NULL) { |
1525 | /* | |
1526 | * We will have to lookup the pi_state again, so free this one | |
1527 | * to keep the accounting correct. | |
1528 | */ | |
1529 | free_pi_state(pi_state); | |
1530 | pi_state = NULL; | |
1531 | } | |
1532 | ||
9ea71503 | 1533 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
1da177e4 LT |
1534 | if (unlikely(ret != 0)) |
1535 | goto out; | |
9ea71503 SB |
1536 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, |
1537 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1538 | if (unlikely(ret != 0)) |
42d35d48 | 1539 | goto out_put_key1; |
1da177e4 | 1540 | |
e9c243a5 TG |
1541 | /* |
1542 | * The check above which compares uaddrs is not sufficient for | |
1543 | * shared futexes. We need to compare the keys: | |
1544 | */ | |
1545 | if (requeue_pi && match_futex(&key1, &key2)) { | |
1546 | ret = -EINVAL; | |
1547 | goto out_put_keys; | |
1548 | } | |
1549 | ||
e2970f2f IM |
1550 | hb1 = hash_futex(&key1); |
1551 | hb2 = hash_futex(&key2); | |
1da177e4 | 1552 | |
e4dc5b7a | 1553 | retry_private: |
69cd9eba | 1554 | hb_waiters_inc(hb2); |
8b8f319f | 1555 | double_lock_hb(hb1, hb2); |
1da177e4 | 1556 | |
e2970f2f IM |
1557 | if (likely(cmpval != NULL)) { |
1558 | u32 curval; | |
1da177e4 | 1559 | |
e2970f2f | 1560 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1561 | |
1562 | if (unlikely(ret)) { | |
5eb3dc62 | 1563 | double_unlock_hb(hb1, hb2); |
69cd9eba | 1564 | hb_waiters_dec(hb2); |
1da177e4 | 1565 | |
e2970f2f | 1566 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1567 | if (ret) |
1568 | goto out_put_keys; | |
1da177e4 | 1569 | |
b41277dc | 1570 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 1571 | goto retry_private; |
1da177e4 | 1572 | |
ae791a2d TG |
1573 | put_futex_key(&key2); |
1574 | put_futex_key(&key1); | |
e4dc5b7a | 1575 | goto retry; |
1da177e4 | 1576 | } |
e2970f2f | 1577 | if (curval != *cmpval) { |
1da177e4 LT |
1578 | ret = -EAGAIN; |
1579 | goto out_unlock; | |
1580 | } | |
1581 | } | |
1582 | ||
52400ba9 | 1583 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1584 | /* |
1585 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1586 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1587 | * bit. We force this here where we are able to easily handle | |
1588 | * faults rather in the requeue loop below. | |
1589 | */ | |
52400ba9 | 1590 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1591 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1592 | |
1593 | /* | |
1594 | * At this point the top_waiter has either taken uaddr2 or is | |
1595 | * waiting on it. If the former, then the pi_state will not | |
1596 | * exist yet, look it up one more time to ensure we have a | |
866293ee TG |
1597 | * reference to it. If the lock was taken, ret contains the |
1598 | * vpid of the top waiter task. | |
52400ba9 | 1599 | */ |
866293ee | 1600 | if (ret > 0) { |
52400ba9 | 1601 | WARN_ON(pi_state); |
89061d3d | 1602 | drop_count++; |
52400ba9 | 1603 | task_count++; |
866293ee TG |
1604 | /* |
1605 | * If we acquired the lock, then the user | |
1606 | * space value of uaddr2 should be vpid. It | |
1607 | * cannot be changed by the top waiter as it | |
1608 | * is blocked on hb2 lock if it tries to do | |
1609 | * so. If something fiddled with it behind our | |
1610 | * back the pi state lookup might unearth | |
1611 | * it. So we rather use the known value than | |
1612 | * rereading and handing potential crap to | |
1613 | * lookup_pi_state. | |
1614 | */ | |
54a21788 | 1615 | ret = lookup_pi_state(ret, hb2, &key2, &pi_state); |
52400ba9 DH |
1616 | } |
1617 | ||
1618 | switch (ret) { | |
1619 | case 0: | |
1620 | break; | |
1621 | case -EFAULT: | |
1622 | double_unlock_hb(hb1, hb2); | |
69cd9eba | 1623 | hb_waiters_dec(hb2); |
ae791a2d TG |
1624 | put_futex_key(&key2); |
1625 | put_futex_key(&key1); | |
d0725992 | 1626 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
1627 | if (!ret) |
1628 | goto retry; | |
1629 | goto out; | |
1630 | case -EAGAIN: | |
af54d6a1 TG |
1631 | /* |
1632 | * Two reasons for this: | |
1633 | * - Owner is exiting and we just wait for the | |
1634 | * exit to complete. | |
1635 | * - The user space value changed. | |
1636 | */ | |
52400ba9 | 1637 | double_unlock_hb(hb1, hb2); |
69cd9eba | 1638 | hb_waiters_dec(hb2); |
ae791a2d TG |
1639 | put_futex_key(&key2); |
1640 | put_futex_key(&key1); | |
52400ba9 DH |
1641 | cond_resched(); |
1642 | goto retry; | |
1643 | default: | |
1644 | goto out_unlock; | |
1645 | } | |
1646 | } | |
1647 | ||
0d00c7b2 | 1648 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
52400ba9 DH |
1649 | if (task_count - nr_wake >= nr_requeue) |
1650 | break; | |
1651 | ||
1652 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1653 | continue; |
52400ba9 | 1654 | |
392741e0 DH |
1655 | /* |
1656 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
1657 | * be paired with each other and no other futex ops. | |
aa10990e DH |
1658 | * |
1659 | * We should never be requeueing a futex_q with a pi_state, | |
1660 | * which is awaiting a futex_unlock_pi(). | |
392741e0 DH |
1661 | */ |
1662 | if ((requeue_pi && !this->rt_waiter) || | |
aa10990e DH |
1663 | (!requeue_pi && this->rt_waiter) || |
1664 | this->pi_state) { | |
392741e0 DH |
1665 | ret = -EINVAL; |
1666 | break; | |
1667 | } | |
52400ba9 DH |
1668 | |
1669 | /* | |
1670 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1671 | * lock, we already woke the top_waiter. If not, it will be | |
1672 | * woken by futex_unlock_pi(). | |
1673 | */ | |
1674 | if (++task_count <= nr_wake && !requeue_pi) { | |
1da177e4 | 1675 | wake_futex(this); |
52400ba9 DH |
1676 | continue; |
1677 | } | |
1da177e4 | 1678 | |
84bc4af5 DH |
1679 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
1680 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
1681 | ret = -EINVAL; | |
1682 | break; | |
1683 | } | |
1684 | ||
52400ba9 DH |
1685 | /* |
1686 | * Requeue nr_requeue waiters and possibly one more in the case | |
1687 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1688 | */ | |
1689 | if (requeue_pi) { | |
1690 | /* Prepare the waiter to take the rt_mutex. */ | |
1691 | atomic_inc(&pi_state->refcount); | |
1692 | this->pi_state = pi_state; | |
1693 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1694 | this->rt_waiter, | |
c051b21f | 1695 | this->task); |
52400ba9 DH |
1696 | if (ret == 1) { |
1697 | /* We got the lock. */ | |
beda2c7e | 1698 | requeue_pi_wake_futex(this, &key2, hb2); |
89061d3d | 1699 | drop_count++; |
52400ba9 DH |
1700 | continue; |
1701 | } else if (ret) { | |
1702 | /* -EDEADLK */ | |
1703 | this->pi_state = NULL; | |
1704 | free_pi_state(pi_state); | |
1705 | goto out_unlock; | |
1706 | } | |
1da177e4 | 1707 | } |
52400ba9 DH |
1708 | requeue_futex(this, hb1, hb2, &key2); |
1709 | drop_count++; | |
1da177e4 LT |
1710 | } |
1711 | ||
1712 | out_unlock: | |
5eb3dc62 | 1713 | double_unlock_hb(hb1, hb2); |
69cd9eba | 1714 | hb_waiters_dec(hb2); |
1da177e4 | 1715 | |
cd84a42f DH |
1716 | /* |
1717 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1718 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1719 | * one at key2 and updated their key pointer. We no longer need to | |
1720 | * hold the references to key1. | |
1721 | */ | |
1da177e4 | 1722 | while (--drop_count >= 0) |
9adef58b | 1723 | drop_futex_key_refs(&key1); |
1da177e4 | 1724 | |
42d35d48 | 1725 | out_put_keys: |
ae791a2d | 1726 | put_futex_key(&key2); |
42d35d48 | 1727 | out_put_key1: |
ae791a2d | 1728 | put_futex_key(&key1); |
42d35d48 | 1729 | out: |
52400ba9 DH |
1730 | if (pi_state != NULL) |
1731 | free_pi_state(pi_state); | |
1732 | return ret ? ret : task_count; | |
1da177e4 LT |
1733 | } |
1734 | ||
1735 | /* The key must be already stored in q->key. */ | |
82af7aca | 1736 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 1737 | __acquires(&hb->lock) |
1da177e4 | 1738 | { |
e2970f2f | 1739 | struct futex_hash_bucket *hb; |
1da177e4 | 1740 | |
e2970f2f | 1741 | hb = hash_futex(&q->key); |
11d4616b LT |
1742 | |
1743 | /* | |
1744 | * Increment the counter before taking the lock so that | |
1745 | * a potential waker won't miss a to-be-slept task that is | |
1746 | * waiting for the spinlock. This is safe as all queue_lock() | |
1747 | * users end up calling queue_me(). Similarly, for housekeeping, | |
1748 | * decrement the counter at queue_unlock() when some error has | |
1749 | * occurred and we don't end up adding the task to the list. | |
1750 | */ | |
1751 | hb_waiters_inc(hb); | |
1752 | ||
e2970f2f | 1753 | q->lock_ptr = &hb->lock; |
1da177e4 | 1754 | |
b0c29f79 | 1755 | spin_lock(&hb->lock); /* implies MB (A) */ |
e2970f2f | 1756 | return hb; |
1da177e4 LT |
1757 | } |
1758 | ||
d40d65c8 | 1759 | static inline void |
0d00c7b2 | 1760 | queue_unlock(struct futex_hash_bucket *hb) |
15e408cd | 1761 | __releases(&hb->lock) |
d40d65c8 DH |
1762 | { |
1763 | spin_unlock(&hb->lock); | |
11d4616b | 1764 | hb_waiters_dec(hb); |
d40d65c8 DH |
1765 | } |
1766 | ||
1767 | /** | |
1768 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
1769 | * @q: The futex_q to enqueue | |
1770 | * @hb: The destination hash bucket | |
1771 | * | |
1772 | * The hb->lock must be held by the caller, and is released here. A call to | |
1773 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
1774 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
1775 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
1776 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
1777 | * an example). | |
1778 | */ | |
82af7aca | 1779 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
15e408cd | 1780 | __releases(&hb->lock) |
1da177e4 | 1781 | { |
ec92d082 PP |
1782 | int prio; |
1783 | ||
1784 | /* | |
1785 | * The priority used to register this element is | |
1786 | * - either the real thread-priority for the real-time threads | |
1787 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
1788 | * - or MAX_RT_PRIO for non-RT threads. | |
1789 | * Thus, all RT-threads are woken first in priority order, and | |
1790 | * the others are woken last, in FIFO order. | |
1791 | */ | |
1792 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1793 | ||
1794 | plist_node_init(&q->list, prio); | |
ec92d082 | 1795 | plist_add(&q->list, &hb->chain); |
c87e2837 | 1796 | q->task = current; |
e2970f2f | 1797 | spin_unlock(&hb->lock); |
1da177e4 LT |
1798 | } |
1799 | ||
d40d65c8 DH |
1800 | /** |
1801 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
1802 | * @q: The futex_q to unqueue | |
1803 | * | |
1804 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
1805 | * be paired with exactly one earlier call to queue_me(). | |
1806 | * | |
6c23cbbd RD |
1807 | * Return: |
1808 | * 1 - if the futex_q was still queued (and we removed unqueued it); | |
d40d65c8 | 1809 | * 0 - if the futex_q was already removed by the waking thread |
1da177e4 | 1810 | */ |
1da177e4 LT |
1811 | static int unqueue_me(struct futex_q *q) |
1812 | { | |
1da177e4 | 1813 | spinlock_t *lock_ptr; |
e2970f2f | 1814 | int ret = 0; |
1da177e4 LT |
1815 | |
1816 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 1817 | retry: |
1da177e4 | 1818 | lock_ptr = q->lock_ptr; |
e91467ec | 1819 | barrier(); |
c80544dc | 1820 | if (lock_ptr != NULL) { |
1da177e4 LT |
1821 | spin_lock(lock_ptr); |
1822 | /* | |
1823 | * q->lock_ptr can change between reading it and | |
1824 | * spin_lock(), causing us to take the wrong lock. This | |
1825 | * corrects the race condition. | |
1826 | * | |
1827 | * Reasoning goes like this: if we have the wrong lock, | |
1828 | * q->lock_ptr must have changed (maybe several times) | |
1829 | * between reading it and the spin_lock(). It can | |
1830 | * change again after the spin_lock() but only if it was | |
1831 | * already changed before the spin_lock(). It cannot, | |
1832 | * however, change back to the original value. Therefore | |
1833 | * we can detect whether we acquired the correct lock. | |
1834 | */ | |
1835 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
1836 | spin_unlock(lock_ptr); | |
1837 | goto retry; | |
1838 | } | |
2e12978a | 1839 | __unqueue_futex(q); |
c87e2837 IM |
1840 | |
1841 | BUG_ON(q->pi_state); | |
1842 | ||
1da177e4 LT |
1843 | spin_unlock(lock_ptr); |
1844 | ret = 1; | |
1845 | } | |
1846 | ||
9adef58b | 1847 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1848 | return ret; |
1849 | } | |
1850 | ||
c87e2837 IM |
1851 | /* |
1852 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
1853 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
1854 | * and dropped here. | |
c87e2837 | 1855 | */ |
d0aa7a70 | 1856 | static void unqueue_me_pi(struct futex_q *q) |
15e408cd | 1857 | __releases(q->lock_ptr) |
c87e2837 | 1858 | { |
2e12978a | 1859 | __unqueue_futex(q); |
c87e2837 IM |
1860 | |
1861 | BUG_ON(!q->pi_state); | |
1862 | free_pi_state(q->pi_state); | |
1863 | q->pi_state = NULL; | |
1864 | ||
d0aa7a70 | 1865 | spin_unlock(q->lock_ptr); |
c87e2837 IM |
1866 | } |
1867 | ||
d0aa7a70 | 1868 | /* |
cdf71a10 | 1869 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 1870 | * |
778e9a9c AK |
1871 | * Must be called with hash bucket lock held and mm->sem held for non |
1872 | * private futexes. | |
d0aa7a70 | 1873 | */ |
778e9a9c | 1874 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
ae791a2d | 1875 | struct task_struct *newowner) |
d0aa7a70 | 1876 | { |
cdf71a10 | 1877 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 1878 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 1879 | struct task_struct *oldowner = pi_state->owner; |
7cfdaf38 | 1880 | u32 uval, uninitialized_var(curval), newval; |
e4dc5b7a | 1881 | int ret; |
d0aa7a70 PP |
1882 | |
1883 | /* Owner died? */ | |
1b7558e4 TG |
1884 | if (!pi_state->owner) |
1885 | newtid |= FUTEX_OWNER_DIED; | |
1886 | ||
1887 | /* | |
1888 | * We are here either because we stole the rtmutex from the | |
8161239a LJ |
1889 | * previous highest priority waiter or we are the highest priority |
1890 | * waiter but failed to get the rtmutex the first time. | |
1891 | * We have to replace the newowner TID in the user space variable. | |
1892 | * This must be atomic as we have to preserve the owner died bit here. | |
1b7558e4 | 1893 | * |
b2d0994b DH |
1894 | * Note: We write the user space value _before_ changing the pi_state |
1895 | * because we can fault here. Imagine swapped out pages or a fork | |
1896 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
1897 | * |
1898 | * Modifying pi_state _before_ the user space value would | |
1899 | * leave the pi_state in an inconsistent state when we fault | |
1900 | * here, because we need to drop the hash bucket lock to | |
1901 | * handle the fault. This might be observed in the PID check | |
1902 | * in lookup_pi_state. | |
1903 | */ | |
1904 | retry: | |
1905 | if (get_futex_value_locked(&uval, uaddr)) | |
1906 | goto handle_fault; | |
1907 | ||
1908 | while (1) { | |
1909 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
1910 | ||
37a9d912 | 1911 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1b7558e4 TG |
1912 | goto handle_fault; |
1913 | if (curval == uval) | |
1914 | break; | |
1915 | uval = curval; | |
1916 | } | |
1917 | ||
1918 | /* | |
1919 | * We fixed up user space. Now we need to fix the pi_state | |
1920 | * itself. | |
1921 | */ | |
d0aa7a70 | 1922 | if (pi_state->owner != NULL) { |
1d615482 | 1923 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
d0aa7a70 PP |
1924 | WARN_ON(list_empty(&pi_state->list)); |
1925 | list_del_init(&pi_state->list); | |
1d615482 | 1926 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
1b7558e4 | 1927 | } |
d0aa7a70 | 1928 | |
cdf71a10 | 1929 | pi_state->owner = newowner; |
d0aa7a70 | 1930 | |
1d615482 | 1931 | raw_spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 1932 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 | 1933 | list_add(&pi_state->list, &newowner->pi_state_list); |
1d615482 | 1934 | raw_spin_unlock_irq(&newowner->pi_lock); |
1b7558e4 | 1935 | return 0; |
d0aa7a70 | 1936 | |
d0aa7a70 | 1937 | /* |
1b7558e4 | 1938 | * To handle the page fault we need to drop the hash bucket |
8161239a LJ |
1939 | * lock here. That gives the other task (either the highest priority |
1940 | * waiter itself or the task which stole the rtmutex) the | |
1b7558e4 TG |
1941 | * chance to try the fixup of the pi_state. So once we are |
1942 | * back from handling the fault we need to check the pi_state | |
1943 | * after reacquiring the hash bucket lock and before trying to | |
1944 | * do another fixup. When the fixup has been done already we | |
1945 | * simply return. | |
d0aa7a70 | 1946 | */ |
1b7558e4 TG |
1947 | handle_fault: |
1948 | spin_unlock(q->lock_ptr); | |
778e9a9c | 1949 | |
d0725992 | 1950 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 1951 | |
1b7558e4 | 1952 | spin_lock(q->lock_ptr); |
778e9a9c | 1953 | |
1b7558e4 TG |
1954 | /* |
1955 | * Check if someone else fixed it for us: | |
1956 | */ | |
1957 | if (pi_state->owner != oldowner) | |
1958 | return 0; | |
1959 | ||
1960 | if (ret) | |
1961 | return ret; | |
1962 | ||
1963 | goto retry; | |
d0aa7a70 PP |
1964 | } |
1965 | ||
72c1bbf3 | 1966 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 1967 | |
dd973998 DH |
1968 | /** |
1969 | * fixup_owner() - Post lock pi_state and corner case management | |
1970 | * @uaddr: user address of the futex | |
dd973998 DH |
1971 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
1972 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
1973 | * | |
1974 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
1975 | * the pi_state owner as well as handle race conditions that may allow us to | |
1976 | * acquire the lock. Must be called with the hb lock held. | |
1977 | * | |
6c23cbbd RD |
1978 | * Return: |
1979 | * 1 - success, lock taken; | |
1980 | * 0 - success, lock not taken; | |
dd973998 DH |
1981 | * <0 - on error (-EFAULT) |
1982 | */ | |
ae791a2d | 1983 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 DH |
1984 | { |
1985 | struct task_struct *owner; | |
1986 | int ret = 0; | |
1987 | ||
1988 | if (locked) { | |
1989 | /* | |
1990 | * Got the lock. We might not be the anticipated owner if we | |
1991 | * did a lock-steal - fix up the PI-state in that case: | |
1992 | */ | |
1993 | if (q->pi_state->owner != current) | |
ae791a2d | 1994 | ret = fixup_pi_state_owner(uaddr, q, current); |
dd973998 DH |
1995 | goto out; |
1996 | } | |
1997 | ||
1998 | /* | |
1999 | * Catch the rare case, where the lock was released when we were on the | |
2000 | * way back before we locked the hash bucket. | |
2001 | */ | |
2002 | if (q->pi_state->owner == current) { | |
2003 | /* | |
2004 | * Try to get the rt_mutex now. This might fail as some other | |
2005 | * task acquired the rt_mutex after we removed ourself from the | |
2006 | * rt_mutex waiters list. | |
2007 | */ | |
2008 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
2009 | locked = 1; | |
2010 | goto out; | |
2011 | } | |
2012 | ||
2013 | /* | |
2014 | * pi_state is incorrect, some other task did a lock steal and | |
2015 | * we returned due to timeout or signal without taking the | |
8161239a | 2016 | * rt_mutex. Too late. |
dd973998 | 2017 | */ |
8161239a | 2018 | raw_spin_lock(&q->pi_state->pi_mutex.wait_lock); |
dd973998 | 2019 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); |
8161239a LJ |
2020 | if (!owner) |
2021 | owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); | |
2022 | raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock); | |
ae791a2d | 2023 | ret = fixup_pi_state_owner(uaddr, q, owner); |
dd973998 DH |
2024 | goto out; |
2025 | } | |
2026 | ||
2027 | /* | |
2028 | * Paranoia check. If we did not take the lock, then we should not be | |
8161239a | 2029 | * the owner of the rt_mutex. |
dd973998 DH |
2030 | */ |
2031 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
2032 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
2033 | "pi-state %p\n", ret, | |
2034 | q->pi_state->pi_mutex.owner, | |
2035 | q->pi_state->owner); | |
2036 | ||
2037 | out: | |
2038 | return ret ? ret : locked; | |
2039 | } | |
2040 | ||
ca5f9524 DH |
2041 | /** |
2042 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
2043 | * @hb: the futex hash bucket, must be locked by the caller | |
2044 | * @q: the futex_q to queue up on | |
2045 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
2046 | */ |
2047 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 2048 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 2049 | { |
9beba3c5 DH |
2050 | /* |
2051 | * The task state is guaranteed to be set before another task can | |
2052 | * wake it. set_current_state() is implemented using set_mb() and | |
2053 | * queue_me() calls spin_unlock() upon completion, both serializing | |
2054 | * access to the hash list and forcing another memory barrier. | |
2055 | */ | |
f1a11e05 | 2056 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 2057 | queue_me(q, hb); |
ca5f9524 DH |
2058 | |
2059 | /* Arm the timer */ | |
2060 | if (timeout) { | |
2061 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); | |
2062 | if (!hrtimer_active(&timeout->timer)) | |
2063 | timeout->task = NULL; | |
2064 | } | |
2065 | ||
2066 | /* | |
0729e196 DH |
2067 | * If we have been removed from the hash list, then another task |
2068 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
2069 | */ |
2070 | if (likely(!plist_node_empty(&q->list))) { | |
2071 | /* | |
2072 | * If the timer has already expired, current will already be | |
2073 | * flagged for rescheduling. Only call schedule if there | |
2074 | * is no timeout, or if it has yet to expire. | |
2075 | */ | |
2076 | if (!timeout || timeout->task) | |
88c8004f | 2077 | freezable_schedule(); |
ca5f9524 DH |
2078 | } |
2079 | __set_current_state(TASK_RUNNING); | |
2080 | } | |
2081 | ||
f801073f DH |
2082 | /** |
2083 | * futex_wait_setup() - Prepare to wait on a futex | |
2084 | * @uaddr: the futex userspace address | |
2085 | * @val: the expected value | |
b41277dc | 2086 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
2087 | * @q: the associated futex_q |
2088 | * @hb: storage for hash_bucket pointer to be returned to caller | |
2089 | * | |
2090 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
2091 | * compare it with the expected value. Handle atomic faults internally. | |
2092 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
2093 | * with no q.key reference on failure. | |
2094 | * | |
6c23cbbd RD |
2095 | * Return: |
2096 | * 0 - uaddr contains val and hb has been locked; | |
ca4a04cf | 2097 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked |
f801073f | 2098 | */ |
b41277dc | 2099 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 2100 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 2101 | { |
e2970f2f IM |
2102 | u32 uval; |
2103 | int ret; | |
1da177e4 | 2104 | |
1da177e4 | 2105 | /* |
b2d0994b | 2106 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
2107 | * Order is important: |
2108 | * | |
2109 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
2110 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
2111 | * | |
2112 | * The basic logical guarantee of a futex is that it blocks ONLY | |
2113 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
2114 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
2115 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
2116 | * cond(var) false, which would violate the guarantee. |
2117 | * | |
8fe8f545 ML |
2118 | * On the other hand, we insert q and release the hash-bucket only |
2119 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
2120 | * absorb a wakeup if *uaddr does not match the desired values | |
2121 | * while the syscall executes. | |
1da177e4 | 2122 | */ |
f801073f | 2123 | retry: |
9ea71503 | 2124 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ); |
f801073f | 2125 | if (unlikely(ret != 0)) |
a5a2a0c7 | 2126 | return ret; |
f801073f DH |
2127 | |
2128 | retry_private: | |
2129 | *hb = queue_lock(q); | |
2130 | ||
e2970f2f | 2131 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 2132 | |
f801073f | 2133 | if (ret) { |
0d00c7b2 | 2134 | queue_unlock(*hb); |
1da177e4 | 2135 | |
e2970f2f | 2136 | ret = get_user(uval, uaddr); |
e4dc5b7a | 2137 | if (ret) |
f801073f | 2138 | goto out; |
1da177e4 | 2139 | |
b41277dc | 2140 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2141 | goto retry_private; |
2142 | ||
ae791a2d | 2143 | put_futex_key(&q->key); |
e4dc5b7a | 2144 | goto retry; |
1da177e4 | 2145 | } |
ca5f9524 | 2146 | |
f801073f | 2147 | if (uval != val) { |
0d00c7b2 | 2148 | queue_unlock(*hb); |
f801073f | 2149 | ret = -EWOULDBLOCK; |
2fff78c7 | 2150 | } |
1da177e4 | 2151 | |
f801073f DH |
2152 | out: |
2153 | if (ret) | |
ae791a2d | 2154 | put_futex_key(&q->key); |
f801073f DH |
2155 | return ret; |
2156 | } | |
2157 | ||
b41277dc DH |
2158 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
2159 | ktime_t *abs_time, u32 bitset) | |
f801073f DH |
2160 | { |
2161 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
2162 | struct restart_block *restart; |
2163 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 2164 | struct futex_q q = futex_q_init; |
f801073f DH |
2165 | int ret; |
2166 | ||
2167 | if (!bitset) | |
2168 | return -EINVAL; | |
f801073f DH |
2169 | q.bitset = bitset; |
2170 | ||
2171 | if (abs_time) { | |
2172 | to = &timeout; | |
2173 | ||
b41277dc DH |
2174 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2175 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2176 | HRTIMER_MODE_ABS); | |
f801073f DH |
2177 | hrtimer_init_sleeper(to, current); |
2178 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2179 | current->timer_slack_ns); | |
2180 | } | |
2181 | ||
d58e6576 | 2182 | retry: |
7ada876a DH |
2183 | /* |
2184 | * Prepare to wait on uaddr. On success, holds hb lock and increments | |
2185 | * q.key refs. | |
2186 | */ | |
b41277dc | 2187 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
2188 | if (ret) |
2189 | goto out; | |
2190 | ||
ca5f9524 | 2191 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 2192 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
2193 | |
2194 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 2195 | ret = 0; |
7ada876a | 2196 | /* unqueue_me() drops q.key ref */ |
1da177e4 | 2197 | if (!unqueue_me(&q)) |
7ada876a | 2198 | goto out; |
2fff78c7 | 2199 | ret = -ETIMEDOUT; |
ca5f9524 | 2200 | if (to && !to->task) |
7ada876a | 2201 | goto out; |
72c1bbf3 | 2202 | |
e2970f2f | 2203 | /* |
d58e6576 TG |
2204 | * We expect signal_pending(current), but we might be the |
2205 | * victim of a spurious wakeup as well. | |
e2970f2f | 2206 | */ |
7ada876a | 2207 | if (!signal_pending(current)) |
d58e6576 | 2208 | goto retry; |
d58e6576 | 2209 | |
2fff78c7 | 2210 | ret = -ERESTARTSYS; |
c19384b5 | 2211 | if (!abs_time) |
7ada876a | 2212 | goto out; |
1da177e4 | 2213 | |
2fff78c7 PZ |
2214 | restart = ¤t_thread_info()->restart_block; |
2215 | restart->fn = futex_wait_restart; | |
a3c74c52 | 2216 | restart->futex.uaddr = uaddr; |
2fff78c7 PZ |
2217 | restart->futex.val = val; |
2218 | restart->futex.time = abs_time->tv64; | |
2219 | restart->futex.bitset = bitset; | |
0cd9c649 | 2220 | restart->futex.flags = flags | FLAGS_HAS_TIMEOUT; |
42d35d48 | 2221 | |
2fff78c7 PZ |
2222 | ret = -ERESTART_RESTARTBLOCK; |
2223 | ||
42d35d48 | 2224 | out: |
ca5f9524 DH |
2225 | if (to) { |
2226 | hrtimer_cancel(&to->timer); | |
2227 | destroy_hrtimer_on_stack(&to->timer); | |
2228 | } | |
c87e2837 IM |
2229 | return ret; |
2230 | } | |
2231 | ||
72c1bbf3 NP |
2232 | |
2233 | static long futex_wait_restart(struct restart_block *restart) | |
2234 | { | |
a3c74c52 | 2235 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 2236 | ktime_t t, *tp = NULL; |
72c1bbf3 | 2237 | |
a72188d8 DH |
2238 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
2239 | t.tv64 = restart->futex.time; | |
2240 | tp = &t; | |
2241 | } | |
72c1bbf3 | 2242 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
2243 | |
2244 | return (long)futex_wait(uaddr, restart->futex.flags, | |
2245 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
2246 | } |
2247 | ||
2248 | ||
c87e2837 IM |
2249 | /* |
2250 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
2251 | * and failed. The kernel side here does the whole locking operation: | |
2252 | * if there are waiters then it will block, it does PI, etc. (Due to | |
2253 | * races the kernel might see a 0 value of the futex too.) | |
2254 | */ | |
b41277dc DH |
2255 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect, |
2256 | ktime_t *time, int trylock) | |
c87e2837 | 2257 | { |
c5780e97 | 2258 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 2259 | struct futex_hash_bucket *hb; |
5bdb05f9 | 2260 | struct futex_q q = futex_q_init; |
dd973998 | 2261 | int res, ret; |
c87e2837 IM |
2262 | |
2263 | if (refill_pi_state_cache()) | |
2264 | return -ENOMEM; | |
2265 | ||
c19384b5 | 2266 | if (time) { |
c5780e97 | 2267 | to = &timeout; |
237fc6e7 TG |
2268 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
2269 | HRTIMER_MODE_ABS); | |
c5780e97 | 2270 | hrtimer_init_sleeper(to, current); |
cc584b21 | 2271 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
2272 | } |
2273 | ||
42d35d48 | 2274 | retry: |
9ea71503 | 2275 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE); |
c87e2837 | 2276 | if (unlikely(ret != 0)) |
42d35d48 | 2277 | goto out; |
c87e2837 | 2278 | |
e4dc5b7a | 2279 | retry_private: |
82af7aca | 2280 | hb = queue_lock(&q); |
c87e2837 | 2281 | |
bab5bc9e | 2282 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 2283 | if (unlikely(ret)) { |
778e9a9c | 2284 | switch (ret) { |
1a52084d DH |
2285 | case 1: |
2286 | /* We got the lock. */ | |
2287 | ret = 0; | |
2288 | goto out_unlock_put_key; | |
2289 | case -EFAULT: | |
2290 | goto uaddr_faulted; | |
778e9a9c AK |
2291 | case -EAGAIN: |
2292 | /* | |
af54d6a1 TG |
2293 | * Two reasons for this: |
2294 | * - Task is exiting and we just wait for the | |
2295 | * exit to complete. | |
2296 | * - The user space value changed. | |
778e9a9c | 2297 | */ |
0d00c7b2 | 2298 | queue_unlock(hb); |
ae791a2d | 2299 | put_futex_key(&q.key); |
778e9a9c AK |
2300 | cond_resched(); |
2301 | goto retry; | |
778e9a9c | 2302 | default: |
42d35d48 | 2303 | goto out_unlock_put_key; |
c87e2837 | 2304 | } |
c87e2837 IM |
2305 | } |
2306 | ||
2307 | /* | |
2308 | * Only actually queue now that the atomic ops are done: | |
2309 | */ | |
82af7aca | 2310 | queue_me(&q, hb); |
c87e2837 | 2311 | |
c87e2837 IM |
2312 | WARN_ON(!q.pi_state); |
2313 | /* | |
2314 | * Block on the PI mutex: | |
2315 | */ | |
c051b21f TG |
2316 | if (!trylock) { |
2317 | ret = rt_mutex_timed_futex_lock(&q.pi_state->pi_mutex, to); | |
2318 | } else { | |
c87e2837 IM |
2319 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); |
2320 | /* Fixup the trylock return value: */ | |
2321 | ret = ret ? 0 : -EWOULDBLOCK; | |
2322 | } | |
2323 | ||
a99e4e41 | 2324 | spin_lock(q.lock_ptr); |
dd973998 DH |
2325 | /* |
2326 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2327 | * haven't already. | |
2328 | */ | |
ae791a2d | 2329 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 DH |
2330 | /* |
2331 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
2332 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
2333 | */ | |
2334 | if (res) | |
2335 | ret = (res < 0) ? res : 0; | |
c87e2837 | 2336 | |
e8f6386c | 2337 | /* |
dd973998 DH |
2338 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
2339 | * it and return the fault to userspace. | |
e8f6386c DH |
2340 | */ |
2341 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
2342 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
2343 | ||
778e9a9c AK |
2344 | /* Unqueue and drop the lock */ |
2345 | unqueue_me_pi(&q); | |
c87e2837 | 2346 | |
5ecb01cf | 2347 | goto out_put_key; |
c87e2837 | 2348 | |
42d35d48 | 2349 | out_unlock_put_key: |
0d00c7b2 | 2350 | queue_unlock(hb); |
c87e2837 | 2351 | |
42d35d48 | 2352 | out_put_key: |
ae791a2d | 2353 | put_futex_key(&q.key); |
42d35d48 | 2354 | out: |
237fc6e7 TG |
2355 | if (to) |
2356 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 2357 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 2358 | |
42d35d48 | 2359 | uaddr_faulted: |
0d00c7b2 | 2360 | queue_unlock(hb); |
778e9a9c | 2361 | |
d0725992 | 2362 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
2363 | if (ret) |
2364 | goto out_put_key; | |
c87e2837 | 2365 | |
b41277dc | 2366 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2367 | goto retry_private; |
2368 | ||
ae791a2d | 2369 | put_futex_key(&q.key); |
e4dc5b7a | 2370 | goto retry; |
c87e2837 IM |
2371 | } |
2372 | ||
c87e2837 IM |
2373 | /* |
2374 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
2375 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
2376 | * and do the rt-mutex unlock. | |
2377 | */ | |
b41277dc | 2378 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 | 2379 | { |
ccf9e6a8 | 2380 | u32 uninitialized_var(curval), uval, vpid = task_pid_vnr(current); |
38d47c1b | 2381 | union futex_key key = FUTEX_KEY_INIT; |
ccf9e6a8 TG |
2382 | struct futex_hash_bucket *hb; |
2383 | struct futex_q *match; | |
e4dc5b7a | 2384 | int ret; |
c87e2837 IM |
2385 | |
2386 | retry: | |
2387 | if (get_user(uval, uaddr)) | |
2388 | return -EFAULT; | |
2389 | /* | |
2390 | * We release only a lock we actually own: | |
2391 | */ | |
c0c9ed15 | 2392 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 2393 | return -EPERM; |
c87e2837 | 2394 | |
9ea71503 | 2395 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE); |
ccf9e6a8 TG |
2396 | if (ret) |
2397 | return ret; | |
c87e2837 IM |
2398 | |
2399 | hb = hash_futex(&key); | |
2400 | spin_lock(&hb->lock); | |
2401 | ||
c87e2837 | 2402 | /* |
ccf9e6a8 TG |
2403 | * Check waiters first. We do not trust user space values at |
2404 | * all and we at least want to know if user space fiddled | |
2405 | * with the futex value instead of blindly unlocking. | |
c87e2837 | 2406 | */ |
ccf9e6a8 TG |
2407 | match = futex_top_waiter(hb, &key); |
2408 | if (match) { | |
2409 | ret = wake_futex_pi(uaddr, uval, match); | |
c87e2837 | 2410 | /* |
ccf9e6a8 TG |
2411 | * The atomic access to the futex value generated a |
2412 | * pagefault, so retry the user-access and the wakeup: | |
c87e2837 IM |
2413 | */ |
2414 | if (ret == -EFAULT) | |
2415 | goto pi_faulted; | |
2416 | goto out_unlock; | |
2417 | } | |
ccf9e6a8 | 2418 | |
c87e2837 | 2419 | /* |
ccf9e6a8 TG |
2420 | * We have no kernel internal state, i.e. no waiters in the |
2421 | * kernel. Waiters which are about to queue themselves are stuck | |
2422 | * on hb->lock. So we can safely ignore them. We do neither | |
2423 | * preserve the WAITERS bit not the OWNER_DIED one. We are the | |
2424 | * owner. | |
c87e2837 | 2425 | */ |
ccf9e6a8 | 2426 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, 0)) |
13fbca4c | 2427 | goto pi_faulted; |
c87e2837 | 2428 | |
ccf9e6a8 TG |
2429 | /* |
2430 | * If uval has changed, let user space handle it. | |
2431 | */ | |
2432 | ret = (curval == uval) ? 0 : -EAGAIN; | |
2433 | ||
c87e2837 IM |
2434 | out_unlock: |
2435 | spin_unlock(&hb->lock); | |
ae791a2d | 2436 | put_futex_key(&key); |
c87e2837 IM |
2437 | return ret; |
2438 | ||
2439 | pi_faulted: | |
778e9a9c | 2440 | spin_unlock(&hb->lock); |
ae791a2d | 2441 | put_futex_key(&key); |
c87e2837 | 2442 | |
d0725992 | 2443 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2444 | if (!ret) |
c87e2837 IM |
2445 | goto retry; |
2446 | ||
1da177e4 LT |
2447 | return ret; |
2448 | } | |
2449 | ||
52400ba9 DH |
2450 | /** |
2451 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2452 | * @hb: the hash_bucket futex_q was original enqueued on | |
2453 | * @q: the futex_q woken while waiting to be requeued | |
2454 | * @key2: the futex_key of the requeue target futex | |
2455 | * @timeout: the timeout associated with the wait (NULL if none) | |
2456 | * | |
2457 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2458 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2459 | * the wakeup and return the appropriate error code to the caller. Must be | |
2460 | * called with the hb lock held. | |
2461 | * | |
6c23cbbd RD |
2462 | * Return: |
2463 | * 0 = no early wakeup detected; | |
2464 | * <0 = -ETIMEDOUT or -ERESTARTNOINTR | |
52400ba9 DH |
2465 | */ |
2466 | static inline | |
2467 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2468 | struct futex_q *q, union futex_key *key2, | |
2469 | struct hrtimer_sleeper *timeout) | |
2470 | { | |
2471 | int ret = 0; | |
2472 | ||
2473 | /* | |
2474 | * With the hb lock held, we avoid races while we process the wakeup. | |
2475 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2476 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2477 | * It can't be requeued from uaddr2 to something else since we don't | |
2478 | * support a PI aware source futex for requeue. | |
2479 | */ | |
2480 | if (!match_futex(&q->key, key2)) { | |
2481 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2482 | /* | |
2483 | * We were woken prior to requeue by a timeout or a signal. | |
2484 | * Unqueue the futex_q and determine which it was. | |
2485 | */ | |
2e12978a | 2486 | plist_del(&q->list, &hb->chain); |
11d4616b | 2487 | hb_waiters_dec(hb); |
52400ba9 | 2488 | |
d58e6576 | 2489 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 2490 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2491 | if (timeout && !timeout->task) |
2492 | ret = -ETIMEDOUT; | |
d58e6576 | 2493 | else if (signal_pending(current)) |
1c840c14 | 2494 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
2495 | } |
2496 | return ret; | |
2497 | } | |
2498 | ||
2499 | /** | |
2500 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 2501 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 2502 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
52400ba9 DH |
2503 | * the same type, no requeueing from private to shared, etc. |
2504 | * @val: the expected value of uaddr | |
2505 | * @abs_time: absolute timeout | |
56ec1607 | 2506 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
2507 | * @uaddr2: the pi futex we will take prior to returning to user-space |
2508 | * | |
2509 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
6f7b0a2a DH |
2510 | * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake |
2511 | * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to | |
2512 | * userspace. This ensures the rt_mutex maintains an owner when it has waiters; | |
2513 | * without one, the pi logic would not know which task to boost/deboost, if | |
2514 | * there was a need to. | |
52400ba9 DH |
2515 | * |
2516 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
6c23cbbd | 2517 | * via the following-- |
52400ba9 | 2518 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() |
cc6db4e6 DH |
2519 | * 2) wakeup on uaddr2 after a requeue |
2520 | * 3) signal | |
2521 | * 4) timeout | |
52400ba9 | 2522 | * |
cc6db4e6 | 2523 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
2524 | * |
2525 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2526 | * 5) successful lock | |
2527 | * 6) signal | |
2528 | * 7) timeout | |
2529 | * 8) other lock acquisition failure | |
2530 | * | |
cc6db4e6 | 2531 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
2532 | * |
2533 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2534 | * | |
6c23cbbd RD |
2535 | * Return: |
2536 | * 0 - On success; | |
52400ba9 DH |
2537 | * <0 - On error |
2538 | */ | |
b41277dc | 2539 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 2540 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 2541 | u32 __user *uaddr2) |
52400ba9 DH |
2542 | { |
2543 | struct hrtimer_sleeper timeout, *to = NULL; | |
2544 | struct rt_mutex_waiter rt_waiter; | |
2545 | struct rt_mutex *pi_mutex = NULL; | |
52400ba9 | 2546 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
2547 | union futex_key key2 = FUTEX_KEY_INIT; |
2548 | struct futex_q q = futex_q_init; | |
52400ba9 | 2549 | int res, ret; |
52400ba9 | 2550 | |
6f7b0a2a DH |
2551 | if (uaddr == uaddr2) |
2552 | return -EINVAL; | |
2553 | ||
52400ba9 DH |
2554 | if (!bitset) |
2555 | return -EINVAL; | |
2556 | ||
2557 | if (abs_time) { | |
2558 | to = &timeout; | |
b41277dc DH |
2559 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2560 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2561 | HRTIMER_MODE_ABS); | |
52400ba9 DH |
2562 | hrtimer_init_sleeper(to, current); |
2563 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2564 | current->timer_slack_ns); | |
2565 | } | |
2566 | ||
2567 | /* | |
2568 | * The waiter is allocated on our stack, manipulated by the requeue | |
2569 | * code while we sleep on uaddr. | |
2570 | */ | |
2571 | debug_rt_mutex_init_waiter(&rt_waiter); | |
fb00aca4 PZ |
2572 | RB_CLEAR_NODE(&rt_waiter.pi_tree_entry); |
2573 | RB_CLEAR_NODE(&rt_waiter.tree_entry); | |
52400ba9 DH |
2574 | rt_waiter.task = NULL; |
2575 | ||
9ea71503 | 2576 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
52400ba9 DH |
2577 | if (unlikely(ret != 0)) |
2578 | goto out; | |
2579 | ||
84bc4af5 DH |
2580 | q.bitset = bitset; |
2581 | q.rt_waiter = &rt_waiter; | |
2582 | q.requeue_pi_key = &key2; | |
2583 | ||
7ada876a DH |
2584 | /* |
2585 | * Prepare to wait on uaddr. On success, increments q.key (key1) ref | |
2586 | * count. | |
2587 | */ | |
b41277dc | 2588 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 TG |
2589 | if (ret) |
2590 | goto out_key2; | |
52400ba9 | 2591 | |
e9c243a5 TG |
2592 | /* |
2593 | * The check above which compares uaddrs is not sufficient for | |
2594 | * shared futexes. We need to compare the keys: | |
2595 | */ | |
2596 | if (match_futex(&q.key, &key2)) { | |
13c42c2f | 2597 | queue_unlock(hb); |
e9c243a5 TG |
2598 | ret = -EINVAL; |
2599 | goto out_put_keys; | |
2600 | } | |
2601 | ||
52400ba9 | 2602 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ |
f1a11e05 | 2603 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2604 | |
2605 | spin_lock(&hb->lock); | |
2606 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2607 | spin_unlock(&hb->lock); | |
2608 | if (ret) | |
2609 | goto out_put_keys; | |
2610 | ||
2611 | /* | |
2612 | * In order for us to be here, we know our q.key == key2, and since | |
2613 | * we took the hb->lock above, we also know that futex_requeue() has | |
2614 | * completed and we no longer have to concern ourselves with a wakeup | |
7ada876a DH |
2615 | * race with the atomic proxy lock acquisition by the requeue code. The |
2616 | * futex_requeue dropped our key1 reference and incremented our key2 | |
2617 | * reference count. | |
52400ba9 DH |
2618 | */ |
2619 | ||
2620 | /* Check if the requeue code acquired the second futex for us. */ | |
2621 | if (!q.rt_waiter) { | |
2622 | /* | |
2623 | * Got the lock. We might not be the anticipated owner if we | |
2624 | * did a lock-steal - fix up the PI-state in that case. | |
2625 | */ | |
2626 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2627 | spin_lock(q.lock_ptr); | |
ae791a2d | 2628 | ret = fixup_pi_state_owner(uaddr2, &q, current); |
52400ba9 DH |
2629 | spin_unlock(q.lock_ptr); |
2630 | } | |
2631 | } else { | |
2632 | /* | |
2633 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2634 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2635 | * the pi_state. | |
2636 | */ | |
f27071cb | 2637 | WARN_ON(!q.pi_state); |
52400ba9 | 2638 | pi_mutex = &q.pi_state->pi_mutex; |
c051b21f | 2639 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter); |
52400ba9 DH |
2640 | debug_rt_mutex_free_waiter(&rt_waiter); |
2641 | ||
2642 | spin_lock(q.lock_ptr); | |
2643 | /* | |
2644 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2645 | * haven't already. | |
2646 | */ | |
ae791a2d | 2647 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 DH |
2648 | /* |
2649 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 2650 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
2651 | */ |
2652 | if (res) | |
2653 | ret = (res < 0) ? res : 0; | |
2654 | ||
2655 | /* Unqueue and drop the lock. */ | |
2656 | unqueue_me_pi(&q); | |
2657 | } | |
2658 | ||
2659 | /* | |
2660 | * If fixup_pi_state_owner() faulted and was unable to handle the | |
2661 | * fault, unlock the rt_mutex and return the fault to userspace. | |
2662 | */ | |
2663 | if (ret == -EFAULT) { | |
b6070a8d | 2664 | if (pi_mutex && rt_mutex_owner(pi_mutex) == current) |
52400ba9 DH |
2665 | rt_mutex_unlock(pi_mutex); |
2666 | } else if (ret == -EINTR) { | |
52400ba9 | 2667 | /* |
cc6db4e6 DH |
2668 | * We've already been requeued, but cannot restart by calling |
2669 | * futex_lock_pi() directly. We could restart this syscall, but | |
2670 | * it would detect that the user space "val" changed and return | |
2671 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
2672 | * -EWOULDBLOCK directly. | |
52400ba9 | 2673 | */ |
2070887f | 2674 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2675 | } |
2676 | ||
2677 | out_put_keys: | |
ae791a2d | 2678 | put_futex_key(&q.key); |
c8b15a70 | 2679 | out_key2: |
ae791a2d | 2680 | put_futex_key(&key2); |
52400ba9 DH |
2681 | |
2682 | out: | |
2683 | if (to) { | |
2684 | hrtimer_cancel(&to->timer); | |
2685 | destroy_hrtimer_on_stack(&to->timer); | |
2686 | } | |
2687 | return ret; | |
2688 | } | |
2689 | ||
0771dfef IM |
2690 | /* |
2691 | * Support for robust futexes: the kernel cleans up held futexes at | |
2692 | * thread exit time. | |
2693 | * | |
2694 | * Implementation: user-space maintains a per-thread list of locks it | |
2695 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2696 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2697 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2698 | * always manipulated with the lock held, so the list is private and |
2699 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2700 | * field, to allow the kernel to clean up if the thread dies after | |
2701 | * acquiring the lock, but just before it could have added itself to | |
2702 | * the list. There can only be one such pending lock. | |
2703 | */ | |
2704 | ||
2705 | /** | |
d96ee56c DH |
2706 | * sys_set_robust_list() - Set the robust-futex list head of a task |
2707 | * @head: pointer to the list-head | |
2708 | * @len: length of the list-head, as userspace expects | |
0771dfef | 2709 | */ |
836f92ad HC |
2710 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2711 | size_t, len) | |
0771dfef | 2712 | { |
a0c1e907 TG |
2713 | if (!futex_cmpxchg_enabled) |
2714 | return -ENOSYS; | |
0771dfef IM |
2715 | /* |
2716 | * The kernel knows only one size for now: | |
2717 | */ | |
2718 | if (unlikely(len != sizeof(*head))) | |
2719 | return -EINVAL; | |
2720 | ||
2721 | current->robust_list = head; | |
2722 | ||
2723 | return 0; | |
2724 | } | |
2725 | ||
2726 | /** | |
d96ee56c DH |
2727 | * sys_get_robust_list() - Get the robust-futex list head of a task |
2728 | * @pid: pid of the process [zero for current task] | |
2729 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
2730 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 2731 | */ |
836f92ad HC |
2732 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
2733 | struct robust_list_head __user * __user *, head_ptr, | |
2734 | size_t __user *, len_ptr) | |
0771dfef | 2735 | { |
ba46df98 | 2736 | struct robust_list_head __user *head; |
0771dfef | 2737 | unsigned long ret; |
bdbb776f | 2738 | struct task_struct *p; |
0771dfef | 2739 | |
a0c1e907 TG |
2740 | if (!futex_cmpxchg_enabled) |
2741 | return -ENOSYS; | |
2742 | ||
bdbb776f KC |
2743 | rcu_read_lock(); |
2744 | ||
2745 | ret = -ESRCH; | |
0771dfef | 2746 | if (!pid) |
bdbb776f | 2747 | p = current; |
0771dfef | 2748 | else { |
228ebcbe | 2749 | p = find_task_by_vpid(pid); |
0771dfef IM |
2750 | if (!p) |
2751 | goto err_unlock; | |
0771dfef IM |
2752 | } |
2753 | ||
bdbb776f KC |
2754 | ret = -EPERM; |
2755 | if (!ptrace_may_access(p, PTRACE_MODE_READ)) | |
2756 | goto err_unlock; | |
2757 | ||
2758 | head = p->robust_list; | |
2759 | rcu_read_unlock(); | |
2760 | ||
0771dfef IM |
2761 | if (put_user(sizeof(*head), len_ptr)) |
2762 | return -EFAULT; | |
2763 | return put_user(head, head_ptr); | |
2764 | ||
2765 | err_unlock: | |
aaa2a97e | 2766 | rcu_read_unlock(); |
0771dfef IM |
2767 | |
2768 | return ret; | |
2769 | } | |
2770 | ||
2771 | /* | |
2772 | * Process a futex-list entry, check whether it's owned by the | |
2773 | * dying task, and do notification if so: | |
2774 | */ | |
e3f2ddea | 2775 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 2776 | { |
7cfdaf38 | 2777 | u32 uval, uninitialized_var(nval), mval; |
0771dfef | 2778 | |
8f17d3a5 IM |
2779 | retry: |
2780 | if (get_user(uval, uaddr)) | |
0771dfef IM |
2781 | return -1; |
2782 | ||
b488893a | 2783 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
2784 | /* |
2785 | * Ok, this dying thread is truly holding a futex | |
2786 | * of interest. Set the OWNER_DIED bit atomically | |
2787 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
2788 | * set, wake up a waiter (if any). (We have to do a | |
2789 | * futex_wake() even if OWNER_DIED is already set - | |
2790 | * to handle the rare but possible case of recursive | |
2791 | * thread-death.) The rest of the cleanup is done in | |
2792 | * userspace. | |
2793 | */ | |
e3f2ddea | 2794 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
6e0aa9f8 TG |
2795 | /* |
2796 | * We are not holding a lock here, but we want to have | |
2797 | * the pagefault_disable/enable() protection because | |
2798 | * we want to handle the fault gracefully. If the | |
2799 | * access fails we try to fault in the futex with R/W | |
2800 | * verification via get_user_pages. get_user() above | |
2801 | * does not guarantee R/W access. If that fails we | |
2802 | * give up and leave the futex locked. | |
2803 | */ | |
2804 | if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) { | |
2805 | if (fault_in_user_writeable(uaddr)) | |
2806 | return -1; | |
2807 | goto retry; | |
2808 | } | |
c87e2837 | 2809 | if (nval != uval) |
8f17d3a5 | 2810 | goto retry; |
0771dfef | 2811 | |
e3f2ddea IM |
2812 | /* |
2813 | * Wake robust non-PI futexes here. The wakeup of | |
2814 | * PI futexes happens in exit_pi_state(): | |
2815 | */ | |
36cf3b5c | 2816 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 2817 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
2818 | } |
2819 | return 0; | |
2820 | } | |
2821 | ||
e3f2ddea IM |
2822 | /* |
2823 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
2824 | */ | |
2825 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 2826 | struct robust_list __user * __user *head, |
1dcc41bb | 2827 | unsigned int *pi) |
e3f2ddea IM |
2828 | { |
2829 | unsigned long uentry; | |
2830 | ||
ba46df98 | 2831 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
2832 | return -EFAULT; |
2833 | ||
ba46df98 | 2834 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
2835 | *pi = uentry & 1; |
2836 | ||
2837 | return 0; | |
2838 | } | |
2839 | ||
0771dfef IM |
2840 | /* |
2841 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
2842 | * and mark any locks found there dead, and notify any waiters. | |
2843 | * | |
2844 | * We silently return on any sign of list-walking problem. | |
2845 | */ | |
2846 | void exit_robust_list(struct task_struct *curr) | |
2847 | { | |
2848 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 2849 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 DH |
2850 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
2851 | unsigned int uninitialized_var(next_pi); | |
0771dfef | 2852 | unsigned long futex_offset; |
9f96cb1e | 2853 | int rc; |
0771dfef | 2854 | |
a0c1e907 TG |
2855 | if (!futex_cmpxchg_enabled) |
2856 | return; | |
2857 | ||
0771dfef IM |
2858 | /* |
2859 | * Fetch the list head (which was registered earlier, via | |
2860 | * sys_set_robust_list()): | |
2861 | */ | |
e3f2ddea | 2862 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
2863 | return; |
2864 | /* | |
2865 | * Fetch the relative futex offset: | |
2866 | */ | |
2867 | if (get_user(futex_offset, &head->futex_offset)) | |
2868 | return; | |
2869 | /* | |
2870 | * Fetch any possibly pending lock-add first, and handle it | |
2871 | * if it exists: | |
2872 | */ | |
e3f2ddea | 2873 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 2874 | return; |
e3f2ddea | 2875 | |
9f96cb1e | 2876 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 2877 | while (entry != &head->list) { |
9f96cb1e MS |
2878 | /* |
2879 | * Fetch the next entry in the list before calling | |
2880 | * handle_futex_death: | |
2881 | */ | |
2882 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
2883 | /* |
2884 | * A pending lock might already be on the list, so | |
c87e2837 | 2885 | * don't process it twice: |
0771dfef IM |
2886 | */ |
2887 | if (entry != pending) | |
ba46df98 | 2888 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 2889 | curr, pi)) |
0771dfef | 2890 | return; |
9f96cb1e | 2891 | if (rc) |
0771dfef | 2892 | return; |
9f96cb1e MS |
2893 | entry = next_entry; |
2894 | pi = next_pi; | |
0771dfef IM |
2895 | /* |
2896 | * Avoid excessively long or circular lists: | |
2897 | */ | |
2898 | if (!--limit) | |
2899 | break; | |
2900 | ||
2901 | cond_resched(); | |
2902 | } | |
9f96cb1e MS |
2903 | |
2904 | if (pending) | |
2905 | handle_futex_death((void __user *)pending + futex_offset, | |
2906 | curr, pip); | |
0771dfef IM |
2907 | } |
2908 | ||
c19384b5 | 2909 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 2910 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 2911 | { |
81b40539 | 2912 | int cmd = op & FUTEX_CMD_MASK; |
b41277dc | 2913 | unsigned int flags = 0; |
34f01cc1 ED |
2914 | |
2915 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 2916 | flags |= FLAGS_SHARED; |
1da177e4 | 2917 | |
b41277dc DH |
2918 | if (op & FUTEX_CLOCK_REALTIME) { |
2919 | flags |= FLAGS_CLOCKRT; | |
2920 | if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI) | |
2921 | return -ENOSYS; | |
2922 | } | |
1da177e4 | 2923 | |
59263b51 TG |
2924 | switch (cmd) { |
2925 | case FUTEX_LOCK_PI: | |
2926 | case FUTEX_UNLOCK_PI: | |
2927 | case FUTEX_TRYLOCK_PI: | |
2928 | case FUTEX_WAIT_REQUEUE_PI: | |
2929 | case FUTEX_CMP_REQUEUE_PI: | |
2930 | if (!futex_cmpxchg_enabled) | |
2931 | return -ENOSYS; | |
2932 | } | |
2933 | ||
34f01cc1 | 2934 | switch (cmd) { |
1da177e4 | 2935 | case FUTEX_WAIT: |
cd689985 TG |
2936 | val3 = FUTEX_BITSET_MATCH_ANY; |
2937 | case FUTEX_WAIT_BITSET: | |
81b40539 | 2938 | return futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 | 2939 | case FUTEX_WAKE: |
cd689985 TG |
2940 | val3 = FUTEX_BITSET_MATCH_ANY; |
2941 | case FUTEX_WAKE_BITSET: | |
81b40539 | 2942 | return futex_wake(uaddr, flags, val, val3); |
1da177e4 | 2943 | case FUTEX_REQUEUE: |
81b40539 | 2944 | return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 | 2945 | case FUTEX_CMP_REQUEUE: |
81b40539 | 2946 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
4732efbe | 2947 | case FUTEX_WAKE_OP: |
81b40539 | 2948 | return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
c87e2837 | 2949 | case FUTEX_LOCK_PI: |
81b40539 | 2950 | return futex_lock_pi(uaddr, flags, val, timeout, 0); |
c87e2837 | 2951 | case FUTEX_UNLOCK_PI: |
81b40539 | 2952 | return futex_unlock_pi(uaddr, flags); |
c87e2837 | 2953 | case FUTEX_TRYLOCK_PI: |
81b40539 | 2954 | return futex_lock_pi(uaddr, flags, 0, timeout, 1); |
52400ba9 DH |
2955 | case FUTEX_WAIT_REQUEUE_PI: |
2956 | val3 = FUTEX_BITSET_MATCH_ANY; | |
81b40539 TG |
2957 | return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
2958 | uaddr2); | |
52400ba9 | 2959 | case FUTEX_CMP_REQUEUE_PI: |
81b40539 | 2960 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
1da177e4 | 2961 | } |
81b40539 | 2962 | return -ENOSYS; |
1da177e4 LT |
2963 | } |
2964 | ||
2965 | ||
17da2bd9 HC |
2966 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
2967 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
2968 | u32, val3) | |
1da177e4 | 2969 | { |
c19384b5 PP |
2970 | struct timespec ts; |
2971 | ktime_t t, *tp = NULL; | |
e2970f2f | 2972 | u32 val2 = 0; |
34f01cc1 | 2973 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 2974 | |
cd689985 | 2975 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
2976 | cmd == FUTEX_WAIT_BITSET || |
2977 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
c19384b5 | 2978 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 2979 | return -EFAULT; |
c19384b5 | 2980 | if (!timespec_valid(&ts)) |
9741ef96 | 2981 | return -EINVAL; |
c19384b5 PP |
2982 | |
2983 | t = timespec_to_ktime(ts); | |
34f01cc1 | 2984 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 2985 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 2986 | tp = &t; |
1da177e4 LT |
2987 | } |
2988 | /* | |
52400ba9 | 2989 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 2990 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 2991 | */ |
f54f0986 | 2992 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 2993 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 2994 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 2995 | |
c19384b5 | 2996 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
2997 | } |
2998 | ||
03b8c7b6 | 2999 | static void __init futex_detect_cmpxchg(void) |
1da177e4 | 3000 | { |
03b8c7b6 | 3001 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 3002 | u32 curval; |
03b8c7b6 HC |
3003 | |
3004 | /* | |
3005 | * This will fail and we want it. Some arch implementations do | |
3006 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
3007 | * functionality. We want to know that before we call in any | |
3008 | * of the complex code paths. Also we want to prevent | |
3009 | * registration of robust lists in that case. NULL is | |
3010 | * guaranteed to fault and we get -EFAULT on functional | |
3011 | * implementation, the non-functional ones will return | |
3012 | * -ENOSYS. | |
3013 | */ | |
3014 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) | |
3015 | futex_cmpxchg_enabled = 1; | |
3016 | #endif | |
3017 | } | |
3018 | ||
3019 | static int __init futex_init(void) | |
3020 | { | |
63b1a816 | 3021 | unsigned int futex_shift; |
a52b89eb DB |
3022 | unsigned long i; |
3023 | ||
3024 | #if CONFIG_BASE_SMALL | |
3025 | futex_hashsize = 16; | |
3026 | #else | |
3027 | futex_hashsize = roundup_pow_of_two(256 * num_possible_cpus()); | |
3028 | #endif | |
3029 | ||
3030 | futex_queues = alloc_large_system_hash("futex", sizeof(*futex_queues), | |
3031 | futex_hashsize, 0, | |
3032 | futex_hashsize < 256 ? HASH_SMALL : 0, | |
63b1a816 HC |
3033 | &futex_shift, NULL, |
3034 | futex_hashsize, futex_hashsize); | |
3035 | futex_hashsize = 1UL << futex_shift; | |
03b8c7b6 HC |
3036 | |
3037 | futex_detect_cmpxchg(); | |
a0c1e907 | 3038 | |
a52b89eb | 3039 | for (i = 0; i < futex_hashsize; i++) { |
11d4616b | 3040 | atomic_set(&futex_queues[i].waiters, 0); |
732375c6 | 3041 | plist_head_init(&futex_queues[i].chain); |
3e4ab747 TG |
3042 | spin_lock_init(&futex_queues[i].lock); |
3043 | } | |
3044 | ||
1da177e4 LT |
3045 | return 0; |
3046 | } | |
f6d107fb | 3047 | __initcall(futex_init); |