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> |
b488893a | 63 | |
4732efbe | 64 | #include <asm/futex.h> |
1da177e4 | 65 | |
c87e2837 IM |
66 | #include "rtmutex_common.h" |
67 | ||
a0c1e907 TG |
68 | int __read_mostly futex_cmpxchg_enabled; |
69 | ||
1da177e4 LT |
70 | #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8) |
71 | ||
b41277dc DH |
72 | /* |
73 | * Futex flags used to encode options to functions and preserve them across | |
74 | * restarts. | |
75 | */ | |
76 | #define FLAGS_SHARED 0x01 | |
77 | #define FLAGS_CLOCKRT 0x02 | |
78 | #define FLAGS_HAS_TIMEOUT 0x04 | |
79 | ||
c87e2837 IM |
80 | /* |
81 | * Priority Inheritance state: | |
82 | */ | |
83 | struct futex_pi_state { | |
84 | /* | |
85 | * list of 'owned' pi_state instances - these have to be | |
86 | * cleaned up in do_exit() if the task exits prematurely: | |
87 | */ | |
88 | struct list_head list; | |
89 | ||
90 | /* | |
91 | * The PI object: | |
92 | */ | |
93 | struct rt_mutex pi_mutex; | |
94 | ||
95 | struct task_struct *owner; | |
96 | atomic_t refcount; | |
97 | ||
98 | union futex_key key; | |
99 | }; | |
100 | ||
d8d88fbb DH |
101 | /** |
102 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 103 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
104 | * @task: the task waiting on the futex |
105 | * @lock_ptr: the hash bucket lock | |
106 | * @key: the key the futex is hashed on | |
107 | * @pi_state: optional priority inheritance state | |
108 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
109 | * @requeue_pi_key: the requeue_pi target futex key | |
110 | * @bitset: bitset for the optional bitmasked wakeup | |
111 | * | |
112 | * We use this hashed waitqueue, instead of a normal wait_queue_t, so | |
1da177e4 LT |
113 | * we can wake only the relevant ones (hashed queues may be shared). |
114 | * | |
115 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 116 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 117 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
118 | * the second. |
119 | * | |
120 | * PI futexes are typically woken before they are removed from the hash list via | |
121 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
122 | */ |
123 | struct futex_q { | |
ec92d082 | 124 | struct plist_node list; |
1da177e4 | 125 | |
d8d88fbb | 126 | struct task_struct *task; |
1da177e4 | 127 | spinlock_t *lock_ptr; |
1da177e4 | 128 | union futex_key key; |
c87e2837 | 129 | struct futex_pi_state *pi_state; |
52400ba9 | 130 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 131 | union futex_key *requeue_pi_key; |
cd689985 | 132 | u32 bitset; |
1da177e4 LT |
133 | }; |
134 | ||
5bdb05f9 DH |
135 | static const struct futex_q futex_q_init = { |
136 | /* list gets initialized in queue_me()*/ | |
137 | .key = FUTEX_KEY_INIT, | |
138 | .bitset = FUTEX_BITSET_MATCH_ANY | |
139 | }; | |
140 | ||
1da177e4 | 141 | /* |
b2d0994b DH |
142 | * Hash buckets are shared by all the futex_keys that hash to the same |
143 | * location. Each key may have multiple futex_q structures, one for each task | |
144 | * waiting on a futex. | |
1da177e4 LT |
145 | */ |
146 | struct futex_hash_bucket { | |
ec92d082 PP |
147 | spinlock_t lock; |
148 | struct plist_head chain; | |
1da177e4 LT |
149 | }; |
150 | ||
151 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; | |
152 | ||
1da177e4 LT |
153 | /* |
154 | * We hash on the keys returned from get_futex_key (see below). | |
155 | */ | |
156 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
157 | { | |
158 | u32 hash = jhash2((u32*)&key->both.word, | |
159 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
160 | key->both.offset); | |
161 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; | |
162 | } | |
163 | ||
164 | /* | |
165 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
166 | */ | |
167 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
168 | { | |
2bc87203 DH |
169 | return (key1 && key2 |
170 | && key1->both.word == key2->both.word | |
1da177e4 LT |
171 | && key1->both.ptr == key2->both.ptr |
172 | && key1->both.offset == key2->both.offset); | |
173 | } | |
174 | ||
38d47c1b PZ |
175 | /* |
176 | * Take a reference to the resource addressed by a key. | |
177 | * Can be called while holding spinlocks. | |
178 | * | |
179 | */ | |
180 | static void get_futex_key_refs(union futex_key *key) | |
181 | { | |
182 | if (!key->both.ptr) | |
183 | return; | |
184 | ||
185 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
186 | case FUT_OFF_INODE: | |
7de9c6ee | 187 | ihold(key->shared.inode); |
38d47c1b PZ |
188 | break; |
189 | case FUT_OFF_MMSHARED: | |
190 | atomic_inc(&key->private.mm->mm_count); | |
191 | break; | |
192 | } | |
193 | } | |
194 | ||
195 | /* | |
196 | * Drop a reference to the resource addressed by a key. | |
197 | * The hash bucket spinlock must not be held. | |
198 | */ | |
199 | static void drop_futex_key_refs(union futex_key *key) | |
200 | { | |
90621c40 DH |
201 | if (!key->both.ptr) { |
202 | /* If we're here then we tried to put a key we failed to get */ | |
203 | WARN_ON_ONCE(1); | |
38d47c1b | 204 | return; |
90621c40 | 205 | } |
38d47c1b PZ |
206 | |
207 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
208 | case FUT_OFF_INODE: | |
209 | iput(key->shared.inode); | |
210 | break; | |
211 | case FUT_OFF_MMSHARED: | |
212 | mmdrop(key->private.mm); | |
213 | break; | |
214 | } | |
215 | } | |
216 | ||
34f01cc1 | 217 | /** |
d96ee56c DH |
218 | * get_futex_key() - Get parameters which are the keys for a futex |
219 | * @uaddr: virtual address of the futex | |
220 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED | |
221 | * @key: address where result is stored. | |
9ea71503 SB |
222 | * @rw: mapping needs to be read/write (values: VERIFY_READ, |
223 | * VERIFY_WRITE) | |
34f01cc1 ED |
224 | * |
225 | * Returns a negative error code or 0 | |
226 | * The key words are stored in *key on success. | |
1da177e4 | 227 | * |
f3a43f3f | 228 | * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode, |
1da177e4 LT |
229 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
230 | * We can usually work out the index without swapping in the page. | |
231 | * | |
b2d0994b | 232 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 233 | */ |
64d1304a | 234 | static int |
9ea71503 | 235 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) |
1da177e4 | 236 | { |
e2970f2f | 237 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 238 | struct mm_struct *mm = current->mm; |
a5b338f2 | 239 | struct page *page, *page_head; |
9ea71503 | 240 | int err, ro = 0; |
1da177e4 LT |
241 | |
242 | /* | |
243 | * The futex address must be "naturally" aligned. | |
244 | */ | |
e2970f2f | 245 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 246 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 247 | return -EINVAL; |
e2970f2f | 248 | address -= key->both.offset; |
1da177e4 | 249 | |
34f01cc1 ED |
250 | /* |
251 | * PROCESS_PRIVATE futexes are fast. | |
252 | * As the mm cannot disappear under us and the 'key' only needs | |
253 | * virtual address, we dont even have to find the underlying vma. | |
254 | * Note : We do have to check 'uaddr' is a valid user address, | |
255 | * but access_ok() should be faster than find_vma() | |
256 | */ | |
257 | if (!fshared) { | |
7485d0d3 | 258 | if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32)))) |
34f01cc1 ED |
259 | return -EFAULT; |
260 | key->private.mm = mm; | |
261 | key->private.address = address; | |
42569c39 | 262 | get_futex_key_refs(key); |
34f01cc1 ED |
263 | return 0; |
264 | } | |
1da177e4 | 265 | |
38d47c1b | 266 | again: |
7485d0d3 | 267 | err = get_user_pages_fast(address, 1, 1, &page); |
9ea71503 SB |
268 | /* |
269 | * If write access is not required (eg. FUTEX_WAIT), try | |
270 | * and get read-only access. | |
271 | */ | |
272 | if (err == -EFAULT && rw == VERIFY_READ) { | |
273 | err = get_user_pages_fast(address, 1, 0, &page); | |
274 | ro = 1; | |
275 | } | |
38d47c1b PZ |
276 | if (err < 0) |
277 | return err; | |
9ea71503 SB |
278 | else |
279 | err = 0; | |
38d47c1b | 280 | |
a5b338f2 AA |
281 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
282 | page_head = page; | |
283 | if (unlikely(PageTail(page))) { | |
38d47c1b | 284 | put_page(page); |
a5b338f2 AA |
285 | /* serialize against __split_huge_page_splitting() */ |
286 | local_irq_disable(); | |
287 | if (likely(__get_user_pages_fast(address, 1, 1, &page) == 1)) { | |
288 | page_head = compound_head(page); | |
289 | /* | |
290 | * page_head is valid pointer but we must pin | |
291 | * it before taking the PG_lock and/or | |
292 | * PG_compound_lock. The moment we re-enable | |
293 | * irqs __split_huge_page_splitting() can | |
294 | * return and the head page can be freed from | |
295 | * under us. We can't take the PG_lock and/or | |
296 | * PG_compound_lock on a page that could be | |
297 | * freed from under us. | |
298 | */ | |
299 | if (page != page_head) { | |
300 | get_page(page_head); | |
301 | put_page(page); | |
302 | } | |
303 | local_irq_enable(); | |
304 | } else { | |
305 | local_irq_enable(); | |
306 | goto again; | |
307 | } | |
308 | } | |
309 | #else | |
310 | page_head = compound_head(page); | |
311 | if (page != page_head) { | |
312 | get_page(page_head); | |
313 | put_page(page); | |
314 | } | |
315 | #endif | |
316 | ||
317 | lock_page(page_head); | |
e6780f72 HD |
318 | |
319 | /* | |
320 | * If page_head->mapping is NULL, then it cannot be a PageAnon | |
321 | * page; but it might be the ZERO_PAGE or in the gate area or | |
322 | * in a special mapping (all cases which we are happy to fail); | |
323 | * or it may have been a good file page when get_user_pages_fast | |
324 | * found it, but truncated or holepunched or subjected to | |
325 | * invalidate_complete_page2 before we got the page lock (also | |
326 | * cases which we are happy to fail). And we hold a reference, | |
327 | * so refcount care in invalidate_complete_page's remove_mapping | |
328 | * prevents drop_caches from setting mapping to NULL beneath us. | |
329 | * | |
330 | * The case we do have to guard against is when memory pressure made | |
331 | * shmem_writepage move it from filecache to swapcache beneath us: | |
332 | * an unlikely race, but we do need to retry for page_head->mapping. | |
333 | */ | |
a5b338f2 | 334 | if (!page_head->mapping) { |
e6780f72 | 335 | int shmem_swizzled = PageSwapCache(page_head); |
a5b338f2 AA |
336 | unlock_page(page_head); |
337 | put_page(page_head); | |
e6780f72 HD |
338 | if (shmem_swizzled) |
339 | goto again; | |
340 | return -EFAULT; | |
38d47c1b | 341 | } |
1da177e4 LT |
342 | |
343 | /* | |
344 | * Private mappings are handled in a simple way. | |
345 | * | |
346 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
347 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 348 | * the object not the particular process. |
1da177e4 | 349 | */ |
a5b338f2 | 350 | if (PageAnon(page_head)) { |
9ea71503 SB |
351 | /* |
352 | * A RO anonymous page will never change and thus doesn't make | |
353 | * sense for futex operations. | |
354 | */ | |
355 | if (ro) { | |
356 | err = -EFAULT; | |
357 | goto out; | |
358 | } | |
359 | ||
38d47c1b | 360 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 361 | key->private.mm = mm; |
e2970f2f | 362 | key->private.address = address; |
38d47c1b PZ |
363 | } else { |
364 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ | |
a5b338f2 AA |
365 | key->shared.inode = page_head->mapping->host; |
366 | key->shared.pgoff = page_head->index; | |
1da177e4 LT |
367 | } |
368 | ||
38d47c1b | 369 | get_futex_key_refs(key); |
1da177e4 | 370 | |
9ea71503 | 371 | out: |
a5b338f2 AA |
372 | unlock_page(page_head); |
373 | put_page(page_head); | |
9ea71503 | 374 | return err; |
1da177e4 LT |
375 | } |
376 | ||
ae791a2d | 377 | static inline void put_futex_key(union futex_key *key) |
1da177e4 | 378 | { |
38d47c1b | 379 | drop_futex_key_refs(key); |
1da177e4 LT |
380 | } |
381 | ||
d96ee56c DH |
382 | /** |
383 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
384 | * @uaddr: pointer to faulting user space address |
385 | * | |
386 | * Slow path to fixup the fault we just took in the atomic write | |
387 | * access to @uaddr. | |
388 | * | |
fb62db2b | 389 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
390 | * user address. We know that we faulted in the atomic pagefault |
391 | * disabled section so we can as well avoid the #PF overhead by | |
392 | * calling get_user_pages() right away. | |
393 | */ | |
394 | static int fault_in_user_writeable(u32 __user *uaddr) | |
395 | { | |
722d0172 AK |
396 | struct mm_struct *mm = current->mm; |
397 | int ret; | |
398 | ||
399 | down_read(&mm->mmap_sem); | |
2efaca92 BH |
400 | ret = fixup_user_fault(current, mm, (unsigned long)uaddr, |
401 | FAULT_FLAG_WRITE); | |
722d0172 AK |
402 | up_read(&mm->mmap_sem); |
403 | ||
d0725992 TG |
404 | return ret < 0 ? ret : 0; |
405 | } | |
406 | ||
4b1c486b DH |
407 | /** |
408 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
409 | * @hb: the hash bucket the futex_q's reside in |
410 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
411 | * |
412 | * Must be called with the hb lock held. | |
413 | */ | |
414 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
415 | union futex_key *key) | |
416 | { | |
417 | struct futex_q *this; | |
418 | ||
419 | plist_for_each_entry(this, &hb->chain, list) { | |
420 | if (match_futex(&this->key, key)) | |
421 | return this; | |
422 | } | |
423 | return NULL; | |
424 | } | |
425 | ||
37a9d912 ML |
426 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
427 | u32 uval, u32 newval) | |
36cf3b5c | 428 | { |
37a9d912 | 429 | int ret; |
36cf3b5c TG |
430 | |
431 | pagefault_disable(); | |
37a9d912 | 432 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
433 | pagefault_enable(); |
434 | ||
37a9d912 | 435 | return ret; |
36cf3b5c TG |
436 | } |
437 | ||
438 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
439 | { |
440 | int ret; | |
441 | ||
a866374a | 442 | pagefault_disable(); |
e2970f2f | 443 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 444 | pagefault_enable(); |
1da177e4 LT |
445 | |
446 | return ret ? -EFAULT : 0; | |
447 | } | |
448 | ||
c87e2837 IM |
449 | |
450 | /* | |
451 | * PI code: | |
452 | */ | |
453 | static int refill_pi_state_cache(void) | |
454 | { | |
455 | struct futex_pi_state *pi_state; | |
456 | ||
457 | if (likely(current->pi_state_cache)) | |
458 | return 0; | |
459 | ||
4668edc3 | 460 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
461 | |
462 | if (!pi_state) | |
463 | return -ENOMEM; | |
464 | ||
c87e2837 IM |
465 | INIT_LIST_HEAD(&pi_state->list); |
466 | /* pi_mutex gets initialized later */ | |
467 | pi_state->owner = NULL; | |
468 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 469 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
470 | |
471 | current->pi_state_cache = pi_state; | |
472 | ||
473 | return 0; | |
474 | } | |
475 | ||
476 | static struct futex_pi_state * alloc_pi_state(void) | |
477 | { | |
478 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
479 | ||
480 | WARN_ON(!pi_state); | |
481 | current->pi_state_cache = NULL; | |
482 | ||
483 | return pi_state; | |
484 | } | |
485 | ||
486 | static void free_pi_state(struct futex_pi_state *pi_state) | |
487 | { | |
488 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
489 | return; | |
490 | ||
491 | /* | |
492 | * If pi_state->owner is NULL, the owner is most probably dying | |
493 | * and has cleaned up the pi_state already | |
494 | */ | |
495 | if (pi_state->owner) { | |
1d615482 | 496 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
c87e2837 | 497 | list_del_init(&pi_state->list); |
1d615482 | 498 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
c87e2837 IM |
499 | |
500 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
501 | } | |
502 | ||
503 | if (current->pi_state_cache) | |
504 | kfree(pi_state); | |
505 | else { | |
506 | /* | |
507 | * pi_state->list is already empty. | |
508 | * clear pi_state->owner. | |
509 | * refcount is at 0 - put it back to 1. | |
510 | */ | |
511 | pi_state->owner = NULL; | |
512 | atomic_set(&pi_state->refcount, 1); | |
513 | current->pi_state_cache = pi_state; | |
514 | } | |
515 | } | |
516 | ||
517 | /* | |
518 | * Look up the task based on what TID userspace gave us. | |
519 | * We dont trust it. | |
520 | */ | |
521 | static struct task_struct * futex_find_get_task(pid_t pid) | |
522 | { | |
523 | struct task_struct *p; | |
524 | ||
d359b549 | 525 | rcu_read_lock(); |
228ebcbe | 526 | p = find_task_by_vpid(pid); |
7a0ea09a MH |
527 | if (p) |
528 | get_task_struct(p); | |
a06381fe | 529 | |
d359b549 | 530 | rcu_read_unlock(); |
c87e2837 IM |
531 | |
532 | return p; | |
533 | } | |
534 | ||
535 | /* | |
536 | * This task is holding PI mutexes at exit time => bad. | |
537 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
538 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
539 | */ | |
540 | void exit_pi_state_list(struct task_struct *curr) | |
541 | { | |
c87e2837 IM |
542 | struct list_head *next, *head = &curr->pi_state_list; |
543 | struct futex_pi_state *pi_state; | |
627371d7 | 544 | struct futex_hash_bucket *hb; |
38d47c1b | 545 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 546 | |
a0c1e907 TG |
547 | if (!futex_cmpxchg_enabled) |
548 | return; | |
c87e2837 IM |
549 | /* |
550 | * We are a ZOMBIE and nobody can enqueue itself on | |
551 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 552 | * versus waiters unqueueing themselves: |
c87e2837 | 553 | */ |
1d615482 | 554 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 IM |
555 | while (!list_empty(head)) { |
556 | ||
557 | next = head->next; | |
558 | pi_state = list_entry(next, struct futex_pi_state, list); | |
559 | key = pi_state->key; | |
627371d7 | 560 | hb = hash_futex(&key); |
1d615482 | 561 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 562 | |
c87e2837 IM |
563 | spin_lock(&hb->lock); |
564 | ||
1d615482 | 565 | raw_spin_lock_irq(&curr->pi_lock); |
627371d7 IM |
566 | /* |
567 | * We dropped the pi-lock, so re-check whether this | |
568 | * task still owns the PI-state: | |
569 | */ | |
c87e2837 IM |
570 | if (head->next != next) { |
571 | spin_unlock(&hb->lock); | |
572 | continue; | |
573 | } | |
574 | ||
c87e2837 | 575 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
576 | WARN_ON(list_empty(&pi_state->list)); |
577 | list_del_init(&pi_state->list); | |
c87e2837 | 578 | pi_state->owner = NULL; |
1d615482 | 579 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
580 | |
581 | rt_mutex_unlock(&pi_state->pi_mutex); | |
582 | ||
583 | spin_unlock(&hb->lock); | |
584 | ||
1d615482 | 585 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 586 | } |
1d615482 | 587 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
588 | } |
589 | ||
590 | static int | |
d0aa7a70 PP |
591 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
592 | union futex_key *key, struct futex_pi_state **ps) | |
c87e2837 IM |
593 | { |
594 | struct futex_pi_state *pi_state = NULL; | |
595 | struct futex_q *this, *next; | |
ec92d082 | 596 | struct plist_head *head; |
c87e2837 | 597 | struct task_struct *p; |
778e9a9c | 598 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 IM |
599 | |
600 | head = &hb->chain; | |
601 | ||
ec92d082 | 602 | plist_for_each_entry_safe(this, next, head, list) { |
d0aa7a70 | 603 | if (match_futex(&this->key, key)) { |
c87e2837 IM |
604 | /* |
605 | * Another waiter already exists - bump up | |
606 | * the refcount and return its pi_state: | |
607 | */ | |
608 | pi_state = this->pi_state; | |
06a9ec29 | 609 | /* |
fb62db2b | 610 | * Userspace might have messed up non-PI and PI futexes |
06a9ec29 TG |
611 | */ |
612 | if (unlikely(!pi_state)) | |
613 | return -EINVAL; | |
614 | ||
627371d7 | 615 | WARN_ON(!atomic_read(&pi_state->refcount)); |
59647b6a TG |
616 | |
617 | /* | |
618 | * When pi_state->owner is NULL then the owner died | |
619 | * and another waiter is on the fly. pi_state->owner | |
620 | * is fixed up by the task which acquires | |
621 | * pi_state->rt_mutex. | |
622 | * | |
623 | * We do not check for pid == 0 which can happen when | |
624 | * the owner died and robust_list_exit() cleared the | |
625 | * TID. | |
626 | */ | |
627 | if (pid && pi_state->owner) { | |
628 | /* | |
629 | * Bail out if user space manipulated the | |
630 | * futex value. | |
631 | */ | |
632 | if (pid != task_pid_vnr(pi_state->owner)) | |
633 | return -EINVAL; | |
634 | } | |
627371d7 | 635 | |
c87e2837 | 636 | atomic_inc(&pi_state->refcount); |
d0aa7a70 | 637 | *ps = pi_state; |
c87e2837 IM |
638 | |
639 | return 0; | |
640 | } | |
641 | } | |
642 | ||
643 | /* | |
e3f2ddea | 644 | * We are the first waiter - try to look up the real owner and attach |
778e9a9c | 645 | * the new pi_state to it, but bail out when TID = 0 |
c87e2837 | 646 | */ |
778e9a9c | 647 | if (!pid) |
e3f2ddea | 648 | return -ESRCH; |
c87e2837 | 649 | p = futex_find_get_task(pid); |
7a0ea09a MH |
650 | if (!p) |
651 | return -ESRCH; | |
778e9a9c AK |
652 | |
653 | /* | |
654 | * We need to look at the task state flags to figure out, | |
655 | * whether the task is exiting. To protect against the do_exit | |
656 | * change of the task flags, we do this protected by | |
657 | * p->pi_lock: | |
658 | */ | |
1d615482 | 659 | raw_spin_lock_irq(&p->pi_lock); |
778e9a9c AK |
660 | if (unlikely(p->flags & PF_EXITING)) { |
661 | /* | |
662 | * The task is on the way out. When PF_EXITPIDONE is | |
663 | * set, we know that the task has finished the | |
664 | * cleanup: | |
665 | */ | |
666 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
667 | ||
1d615482 | 668 | raw_spin_unlock_irq(&p->pi_lock); |
778e9a9c AK |
669 | put_task_struct(p); |
670 | return ret; | |
671 | } | |
c87e2837 IM |
672 | |
673 | pi_state = alloc_pi_state(); | |
674 | ||
675 | /* | |
676 | * Initialize the pi_mutex in locked state and make 'p' | |
677 | * the owner of it: | |
678 | */ | |
679 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
680 | ||
681 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 682 | pi_state->key = *key; |
c87e2837 | 683 | |
627371d7 | 684 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
685 | list_add(&pi_state->list, &p->pi_state_list); |
686 | pi_state->owner = p; | |
1d615482 | 687 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
688 | |
689 | put_task_struct(p); | |
690 | ||
d0aa7a70 | 691 | *ps = pi_state; |
c87e2837 IM |
692 | |
693 | return 0; | |
694 | } | |
695 | ||
1a52084d | 696 | /** |
d96ee56c | 697 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
698 | * @uaddr: the pi futex user address |
699 | * @hb: the pi futex hash bucket | |
700 | * @key: the futex key associated with uaddr and hb | |
701 | * @ps: the pi_state pointer where we store the result of the | |
702 | * lookup | |
703 | * @task: the task to perform the atomic lock work for. This will | |
704 | * be "current" except in the case of requeue pi. | |
705 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d DH |
706 | * |
707 | * Returns: | |
708 | * 0 - ready to wait | |
709 | * 1 - acquired the lock | |
710 | * <0 - error | |
711 | * | |
712 | * The hb->lock and futex_key refs shall be held by the caller. | |
713 | */ | |
714 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
715 | union futex_key *key, | |
716 | struct futex_pi_state **ps, | |
bab5bc9e | 717 | struct task_struct *task, int set_waiters) |
1a52084d | 718 | { |
59fa6245 | 719 | int lock_taken, ret, force_take = 0; |
c0c9ed15 | 720 | u32 uval, newval, curval, vpid = task_pid_vnr(task); |
1a52084d DH |
721 | |
722 | retry: | |
723 | ret = lock_taken = 0; | |
724 | ||
725 | /* | |
726 | * To avoid races, we attempt to take the lock here again | |
727 | * (by doing a 0 -> TID atomic cmpxchg), while holding all | |
728 | * the locks. It will most likely not succeed. | |
729 | */ | |
c0c9ed15 | 730 | newval = vpid; |
bab5bc9e DH |
731 | if (set_waiters) |
732 | newval |= FUTEX_WAITERS; | |
1a52084d | 733 | |
37a9d912 | 734 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval))) |
1a52084d DH |
735 | return -EFAULT; |
736 | ||
737 | /* | |
738 | * Detect deadlocks. | |
739 | */ | |
c0c9ed15 | 740 | if ((unlikely((curval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
741 | return -EDEADLK; |
742 | ||
743 | /* | |
744 | * Surprise - we got the lock. Just return to userspace: | |
745 | */ | |
746 | if (unlikely(!curval)) | |
747 | return 1; | |
748 | ||
749 | uval = curval; | |
750 | ||
751 | /* | |
752 | * Set the FUTEX_WAITERS flag, so the owner will know it has someone | |
753 | * to wake at the next unlock. | |
754 | */ | |
755 | newval = curval | FUTEX_WAITERS; | |
756 | ||
757 | /* | |
59fa6245 | 758 | * Should we force take the futex? See below. |
1a52084d | 759 | */ |
59fa6245 TG |
760 | if (unlikely(force_take)) { |
761 | /* | |
762 | * Keep the OWNER_DIED and the WAITERS bit and set the | |
763 | * new TID value. | |
764 | */ | |
c0c9ed15 | 765 | newval = (curval & ~FUTEX_TID_MASK) | vpid; |
59fa6245 | 766 | force_take = 0; |
1a52084d DH |
767 | lock_taken = 1; |
768 | } | |
769 | ||
37a9d912 | 770 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))) |
1a52084d DH |
771 | return -EFAULT; |
772 | if (unlikely(curval != uval)) | |
773 | goto retry; | |
774 | ||
775 | /* | |
59fa6245 | 776 | * We took the lock due to forced take over. |
1a52084d DH |
777 | */ |
778 | if (unlikely(lock_taken)) | |
779 | return 1; | |
780 | ||
781 | /* | |
782 | * We dont have the lock. Look up the PI state (or create it if | |
783 | * we are the first waiter): | |
784 | */ | |
785 | ret = lookup_pi_state(uval, hb, key, ps); | |
786 | ||
787 | if (unlikely(ret)) { | |
788 | switch (ret) { | |
789 | case -ESRCH: | |
790 | /* | |
59fa6245 TG |
791 | * We failed to find an owner for this |
792 | * futex. So we have no pi_state to block | |
793 | * on. This can happen in two cases: | |
794 | * | |
795 | * 1) The owner died | |
796 | * 2) A stale FUTEX_WAITERS bit | |
797 | * | |
798 | * Re-read the futex value. | |
1a52084d DH |
799 | */ |
800 | if (get_futex_value_locked(&curval, uaddr)) | |
801 | return -EFAULT; | |
802 | ||
803 | /* | |
59fa6245 TG |
804 | * If the owner died or we have a stale |
805 | * WAITERS bit the owner TID in the user space | |
806 | * futex is 0. | |
1a52084d | 807 | */ |
59fa6245 TG |
808 | if (!(curval & FUTEX_TID_MASK)) { |
809 | force_take = 1; | |
1a52084d DH |
810 | goto retry; |
811 | } | |
812 | default: | |
813 | break; | |
814 | } | |
815 | } | |
816 | ||
817 | return ret; | |
818 | } | |
819 | ||
2e12978a LJ |
820 | /** |
821 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
822 | * @q: The futex_q to unqueue | |
823 | * | |
824 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
825 | */ | |
826 | static void __unqueue_futex(struct futex_q *q) | |
827 | { | |
828 | struct futex_hash_bucket *hb; | |
829 | ||
29096202 SR |
830 | if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr)) |
831 | || WARN_ON(plist_node_empty(&q->list))) | |
2e12978a LJ |
832 | return; |
833 | ||
834 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
835 | plist_del(&q->list, &hb->chain); | |
836 | } | |
837 | ||
1da177e4 LT |
838 | /* |
839 | * The hash bucket lock must be held when this is called. | |
840 | * Afterwards, the futex_q must not be accessed. | |
841 | */ | |
842 | static void wake_futex(struct futex_q *q) | |
843 | { | |
f1a11e05 TG |
844 | struct task_struct *p = q->task; |
845 | ||
1da177e4 | 846 | /* |
f1a11e05 | 847 | * We set q->lock_ptr = NULL _before_ we wake up the task. If |
fb62db2b RD |
848 | * a non-futex wake up happens on another CPU then the task |
849 | * might exit and p would dereference a non-existing task | |
f1a11e05 TG |
850 | * struct. Prevent this by holding a reference on p across the |
851 | * wake up. | |
1da177e4 | 852 | */ |
f1a11e05 TG |
853 | get_task_struct(p); |
854 | ||
2e12978a | 855 | __unqueue_futex(q); |
1da177e4 | 856 | /* |
f1a11e05 TG |
857 | * The waiting task can free the futex_q as soon as |
858 | * q->lock_ptr = NULL is written, without taking any locks. A | |
859 | * memory barrier is required here to prevent the following | |
860 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 861 | */ |
ccdea2f8 | 862 | smp_wmb(); |
1da177e4 | 863 | q->lock_ptr = NULL; |
f1a11e05 TG |
864 | |
865 | wake_up_state(p, TASK_NORMAL); | |
866 | put_task_struct(p); | |
1da177e4 LT |
867 | } |
868 | ||
c87e2837 IM |
869 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
870 | { | |
871 | struct task_struct *new_owner; | |
872 | struct futex_pi_state *pi_state = this->pi_state; | |
7cfdaf38 | 873 | u32 uninitialized_var(curval), newval; |
c87e2837 IM |
874 | |
875 | if (!pi_state) | |
876 | return -EINVAL; | |
877 | ||
51246bfd TG |
878 | /* |
879 | * If current does not own the pi_state then the futex is | |
880 | * inconsistent and user space fiddled with the futex value. | |
881 | */ | |
882 | if (pi_state->owner != current) | |
883 | return -EINVAL; | |
884 | ||
d209d74d | 885 | raw_spin_lock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
886 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
887 | ||
888 | /* | |
f123c98e SR |
889 | * It is possible that the next waiter (the one that brought |
890 | * this owner to the kernel) timed out and is no longer | |
891 | * waiting on the lock. | |
c87e2837 IM |
892 | */ |
893 | if (!new_owner) | |
894 | new_owner = this->task; | |
895 | ||
896 | /* | |
897 | * We pass it to the next owner. (The WAITERS bit is always | |
898 | * kept enabled while there is PI state around. We must also | |
899 | * preserve the owner died bit.) | |
900 | */ | |
e3f2ddea | 901 | if (!(uval & FUTEX_OWNER_DIED)) { |
778e9a9c AK |
902 | int ret = 0; |
903 | ||
b488893a | 904 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 905 | |
37a9d912 | 906 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
778e9a9c | 907 | ret = -EFAULT; |
cde898fa | 908 | else if (curval != uval) |
778e9a9c AK |
909 | ret = -EINVAL; |
910 | if (ret) { | |
d209d74d | 911 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
778e9a9c AK |
912 | return ret; |
913 | } | |
e3f2ddea | 914 | } |
c87e2837 | 915 | |
1d615482 | 916 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
627371d7 IM |
917 | WARN_ON(list_empty(&pi_state->list)); |
918 | list_del_init(&pi_state->list); | |
1d615482 | 919 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
627371d7 | 920 | |
1d615482 | 921 | raw_spin_lock_irq(&new_owner->pi_lock); |
627371d7 | 922 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
923 | list_add(&pi_state->list, &new_owner->pi_state_list); |
924 | pi_state->owner = new_owner; | |
1d615482 | 925 | raw_spin_unlock_irq(&new_owner->pi_lock); |
627371d7 | 926 | |
d209d74d | 927 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
928 | rt_mutex_unlock(&pi_state->pi_mutex); |
929 | ||
930 | return 0; | |
931 | } | |
932 | ||
933 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) | |
934 | { | |
7cfdaf38 | 935 | u32 uninitialized_var(oldval); |
c87e2837 IM |
936 | |
937 | /* | |
938 | * There is no waiter, so we unlock the futex. The owner died | |
939 | * bit has not to be preserved here. We are the owner: | |
940 | */ | |
37a9d912 ML |
941 | if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0)) |
942 | return -EFAULT; | |
c87e2837 IM |
943 | if (oldval != uval) |
944 | return -EAGAIN; | |
945 | ||
946 | return 0; | |
947 | } | |
948 | ||
8b8f319f IM |
949 | /* |
950 | * Express the locking dependencies for lockdep: | |
951 | */ | |
952 | static inline void | |
953 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
954 | { | |
955 | if (hb1 <= hb2) { | |
956 | spin_lock(&hb1->lock); | |
957 | if (hb1 < hb2) | |
958 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
959 | } else { /* hb1 > hb2 */ | |
960 | spin_lock(&hb2->lock); | |
961 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
962 | } | |
963 | } | |
964 | ||
5eb3dc62 DH |
965 | static inline void |
966 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
967 | { | |
f061d351 | 968 | spin_unlock(&hb1->lock); |
88f502fe IM |
969 | if (hb1 != hb2) |
970 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
971 | } |
972 | ||
1da177e4 | 973 | /* |
b2d0994b | 974 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 975 | */ |
b41277dc DH |
976 | static int |
977 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 978 | { |
e2970f2f | 979 | struct futex_hash_bucket *hb; |
1da177e4 | 980 | struct futex_q *this, *next; |
ec92d082 | 981 | struct plist_head *head; |
38d47c1b | 982 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 LT |
983 | int ret; |
984 | ||
cd689985 TG |
985 | if (!bitset) |
986 | return -EINVAL; | |
987 | ||
9ea71503 | 988 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ); |
1da177e4 LT |
989 | if (unlikely(ret != 0)) |
990 | goto out; | |
991 | ||
e2970f2f IM |
992 | hb = hash_futex(&key); |
993 | spin_lock(&hb->lock); | |
994 | head = &hb->chain; | |
1da177e4 | 995 | |
ec92d082 | 996 | plist_for_each_entry_safe(this, next, head, list) { |
1da177e4 | 997 | if (match_futex (&this->key, &key)) { |
52400ba9 | 998 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
999 | ret = -EINVAL; |
1000 | break; | |
1001 | } | |
cd689985 TG |
1002 | |
1003 | /* Check if one of the bits is set in both bitsets */ | |
1004 | if (!(this->bitset & bitset)) | |
1005 | continue; | |
1006 | ||
1da177e4 LT |
1007 | wake_futex(this); |
1008 | if (++ret >= nr_wake) | |
1009 | break; | |
1010 | } | |
1011 | } | |
1012 | ||
e2970f2f | 1013 | spin_unlock(&hb->lock); |
ae791a2d | 1014 | put_futex_key(&key); |
42d35d48 | 1015 | out: |
1da177e4 LT |
1016 | return ret; |
1017 | } | |
1018 | ||
4732efbe JJ |
1019 | /* |
1020 | * Wake up all waiters hashed on the physical page that is mapped | |
1021 | * to this virtual address: | |
1022 | */ | |
e2970f2f | 1023 | static int |
b41277dc | 1024 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 1025 | int nr_wake, int nr_wake2, int op) |
4732efbe | 1026 | { |
38d47c1b | 1027 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 1028 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 1029 | struct plist_head *head; |
4732efbe | 1030 | struct futex_q *this, *next; |
e4dc5b7a | 1031 | int ret, op_ret; |
4732efbe | 1032 | |
e4dc5b7a | 1033 | retry: |
9ea71503 | 1034 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
4732efbe JJ |
1035 | if (unlikely(ret != 0)) |
1036 | goto out; | |
9ea71503 | 1037 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
4732efbe | 1038 | if (unlikely(ret != 0)) |
42d35d48 | 1039 | goto out_put_key1; |
4732efbe | 1040 | |
e2970f2f IM |
1041 | hb1 = hash_futex(&key1); |
1042 | hb2 = hash_futex(&key2); | |
4732efbe | 1043 | |
e4dc5b7a | 1044 | retry_private: |
eaaea803 | 1045 | double_lock_hb(hb1, hb2); |
e2970f2f | 1046 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1047 | if (unlikely(op_ret < 0)) { |
4732efbe | 1048 | |
5eb3dc62 | 1049 | double_unlock_hb(hb1, hb2); |
4732efbe | 1050 | |
7ee1dd3f | 1051 | #ifndef CONFIG_MMU |
e2970f2f IM |
1052 | /* |
1053 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
1054 | * but we might get them from range checking | |
1055 | */ | |
7ee1dd3f | 1056 | ret = op_ret; |
42d35d48 | 1057 | goto out_put_keys; |
7ee1dd3f DH |
1058 | #endif |
1059 | ||
796f8d9b DG |
1060 | if (unlikely(op_ret != -EFAULT)) { |
1061 | ret = op_ret; | |
42d35d48 | 1062 | goto out_put_keys; |
796f8d9b DG |
1063 | } |
1064 | ||
d0725992 | 1065 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 1066 | if (ret) |
de87fcc1 | 1067 | goto out_put_keys; |
4732efbe | 1068 | |
b41277dc | 1069 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1070 | goto retry_private; |
1071 | ||
ae791a2d TG |
1072 | put_futex_key(&key2); |
1073 | put_futex_key(&key1); | |
e4dc5b7a | 1074 | goto retry; |
4732efbe JJ |
1075 | } |
1076 | ||
e2970f2f | 1077 | head = &hb1->chain; |
4732efbe | 1078 | |
ec92d082 | 1079 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
1080 | if (match_futex (&this->key, &key1)) { |
1081 | wake_futex(this); | |
1082 | if (++ret >= nr_wake) | |
1083 | break; | |
1084 | } | |
1085 | } | |
1086 | ||
1087 | if (op_ret > 0) { | |
e2970f2f | 1088 | head = &hb2->chain; |
4732efbe JJ |
1089 | |
1090 | op_ret = 0; | |
ec92d082 | 1091 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe JJ |
1092 | if (match_futex (&this->key, &key2)) { |
1093 | wake_futex(this); | |
1094 | if (++op_ret >= nr_wake2) | |
1095 | break; | |
1096 | } | |
1097 | } | |
1098 | ret += op_ret; | |
1099 | } | |
1100 | ||
5eb3dc62 | 1101 | double_unlock_hb(hb1, hb2); |
42d35d48 | 1102 | out_put_keys: |
ae791a2d | 1103 | put_futex_key(&key2); |
42d35d48 | 1104 | out_put_key1: |
ae791a2d | 1105 | put_futex_key(&key1); |
42d35d48 | 1106 | out: |
4732efbe JJ |
1107 | return ret; |
1108 | } | |
1109 | ||
9121e478 DH |
1110 | /** |
1111 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1112 | * @q: the futex_q to requeue | |
1113 | * @hb1: the source hash_bucket | |
1114 | * @hb2: the target hash_bucket | |
1115 | * @key2: the new key for the requeued futex_q | |
1116 | */ | |
1117 | static inline | |
1118 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1119 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1120 | { | |
1121 | ||
1122 | /* | |
1123 | * If key1 and key2 hash to the same bucket, no need to | |
1124 | * requeue. | |
1125 | */ | |
1126 | if (likely(&hb1->chain != &hb2->chain)) { | |
1127 | plist_del(&q->list, &hb1->chain); | |
1128 | plist_add(&q->list, &hb2->chain); | |
1129 | q->lock_ptr = &hb2->lock; | |
9121e478 DH |
1130 | } |
1131 | get_futex_key_refs(key2); | |
1132 | q->key = *key2; | |
1133 | } | |
1134 | ||
52400ba9 DH |
1135 | /** |
1136 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1137 | * @q: the futex_q |
1138 | * @key: the key of the requeue target futex | |
1139 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1140 | * |
1141 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1142 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1143 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1144 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1145 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1146 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1147 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1148 | */ |
1149 | static inline | |
beda2c7e DH |
1150 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1151 | struct futex_hash_bucket *hb) | |
52400ba9 | 1152 | { |
52400ba9 DH |
1153 | get_futex_key_refs(key); |
1154 | q->key = *key; | |
1155 | ||
2e12978a | 1156 | __unqueue_futex(q); |
52400ba9 DH |
1157 | |
1158 | WARN_ON(!q->rt_waiter); | |
1159 | q->rt_waiter = NULL; | |
1160 | ||
beda2c7e | 1161 | q->lock_ptr = &hb->lock; |
beda2c7e | 1162 | |
f1a11e05 | 1163 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1164 | } |
1165 | ||
1166 | /** | |
1167 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1168 | * @pifutex: the user address of the to futex |
1169 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1170 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1171 | * @key1: the from futex key | |
1172 | * @key2: the to futex key | |
1173 | * @ps: address to store the pi_state pointer | |
1174 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1175 | * |
1176 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1177 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1178 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1179 | * hb1 and hb2 must be held by the caller. | |
52400ba9 DH |
1180 | * |
1181 | * Returns: | |
1182 | * 0 - failed to acquire the lock atomicly | |
1183 | * 1 - acquired the lock | |
1184 | * <0 - error | |
1185 | */ | |
1186 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1187 | struct futex_hash_bucket *hb1, | |
1188 | struct futex_hash_bucket *hb2, | |
1189 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1190 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1191 | { |
bab5bc9e | 1192 | struct futex_q *top_waiter = NULL; |
52400ba9 DH |
1193 | u32 curval; |
1194 | int ret; | |
1195 | ||
1196 | if (get_futex_value_locked(&curval, pifutex)) | |
1197 | return -EFAULT; | |
1198 | ||
bab5bc9e DH |
1199 | /* |
1200 | * Find the top_waiter and determine if there are additional waiters. | |
1201 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1202 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1203 | * as we have means to handle the possible fault. If not, don't set | |
1204 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1205 | * the kernel. | |
1206 | */ | |
52400ba9 DH |
1207 | top_waiter = futex_top_waiter(hb1, key1); |
1208 | ||
1209 | /* There are no waiters, nothing for us to do. */ | |
1210 | if (!top_waiter) | |
1211 | return 0; | |
1212 | ||
84bc4af5 DH |
1213 | /* Ensure we requeue to the expected futex. */ |
1214 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1215 | return -EINVAL; | |
1216 | ||
52400ba9 | 1217 | /* |
bab5bc9e DH |
1218 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1219 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1220 | * in ps in contended cases. | |
52400ba9 | 1221 | */ |
bab5bc9e DH |
1222 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1223 | set_waiters); | |
52400ba9 | 1224 | if (ret == 1) |
beda2c7e | 1225 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
52400ba9 DH |
1226 | |
1227 | return ret; | |
1228 | } | |
1229 | ||
1230 | /** | |
1231 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1232 | * @uaddr1: source futex user address |
b41277dc | 1233 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1234 | * @uaddr2: target futex user address |
1235 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1236 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1237 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1238 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1239 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1240 | * |
1241 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1242 | * uaddr2 atomically on behalf of the top waiter. | |
1243 | * | |
1244 | * Returns: | |
1245 | * >=0 - on success, the number of tasks requeued or woken | |
1246 | * <0 - on error | |
1da177e4 | 1247 | */ |
b41277dc DH |
1248 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
1249 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
1250 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 1251 | { |
38d47c1b | 1252 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1253 | int drop_count = 0, task_count = 0, ret; |
1254 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1255 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 1256 | struct plist_head *head1; |
1da177e4 | 1257 | struct futex_q *this, *next; |
52400ba9 DH |
1258 | u32 curval2; |
1259 | ||
1260 | if (requeue_pi) { | |
1261 | /* | |
1262 | * requeue_pi requires a pi_state, try to allocate it now | |
1263 | * without any locks in case it fails. | |
1264 | */ | |
1265 | if (refill_pi_state_cache()) | |
1266 | return -ENOMEM; | |
1267 | /* | |
1268 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1269 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1270 | * returning to userspace, so as to not leave the rt_mutex with | |
1271 | * waiters and no owner. However, second and third wake-ups | |
1272 | * cannot be predicted as they involve race conditions with the | |
1273 | * first wake and a fault while looking up the pi_state. Both | |
1274 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1275 | * use nr_wake=1. | |
1276 | */ | |
1277 | if (nr_wake != 1) | |
1278 | return -EINVAL; | |
1279 | } | |
1da177e4 | 1280 | |
42d35d48 | 1281 | retry: |
52400ba9 DH |
1282 | if (pi_state != NULL) { |
1283 | /* | |
1284 | * We will have to lookup the pi_state again, so free this one | |
1285 | * to keep the accounting correct. | |
1286 | */ | |
1287 | free_pi_state(pi_state); | |
1288 | pi_state = NULL; | |
1289 | } | |
1290 | ||
9ea71503 | 1291 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
1da177e4 LT |
1292 | if (unlikely(ret != 0)) |
1293 | goto out; | |
9ea71503 SB |
1294 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, |
1295 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1296 | if (unlikely(ret != 0)) |
42d35d48 | 1297 | goto out_put_key1; |
1da177e4 | 1298 | |
e2970f2f IM |
1299 | hb1 = hash_futex(&key1); |
1300 | hb2 = hash_futex(&key2); | |
1da177e4 | 1301 | |
e4dc5b7a | 1302 | retry_private: |
8b8f319f | 1303 | double_lock_hb(hb1, hb2); |
1da177e4 | 1304 | |
e2970f2f IM |
1305 | if (likely(cmpval != NULL)) { |
1306 | u32 curval; | |
1da177e4 | 1307 | |
e2970f2f | 1308 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1309 | |
1310 | if (unlikely(ret)) { | |
5eb3dc62 | 1311 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1312 | |
e2970f2f | 1313 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1314 | if (ret) |
1315 | goto out_put_keys; | |
1da177e4 | 1316 | |
b41277dc | 1317 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 1318 | goto retry_private; |
1da177e4 | 1319 | |
ae791a2d TG |
1320 | put_futex_key(&key2); |
1321 | put_futex_key(&key1); | |
e4dc5b7a | 1322 | goto retry; |
1da177e4 | 1323 | } |
e2970f2f | 1324 | if (curval != *cmpval) { |
1da177e4 LT |
1325 | ret = -EAGAIN; |
1326 | goto out_unlock; | |
1327 | } | |
1328 | } | |
1329 | ||
52400ba9 | 1330 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1331 | /* |
1332 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1333 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1334 | * bit. We force this here where we are able to easily handle | |
1335 | * faults rather in the requeue loop below. | |
1336 | */ | |
52400ba9 | 1337 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1338 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1339 | |
1340 | /* | |
1341 | * At this point the top_waiter has either taken uaddr2 or is | |
1342 | * waiting on it. If the former, then the pi_state will not | |
1343 | * exist yet, look it up one more time to ensure we have a | |
1344 | * reference to it. | |
1345 | */ | |
1346 | if (ret == 1) { | |
1347 | WARN_ON(pi_state); | |
89061d3d | 1348 | drop_count++; |
52400ba9 DH |
1349 | task_count++; |
1350 | ret = get_futex_value_locked(&curval2, uaddr2); | |
1351 | if (!ret) | |
1352 | ret = lookup_pi_state(curval2, hb2, &key2, | |
1353 | &pi_state); | |
1354 | } | |
1355 | ||
1356 | switch (ret) { | |
1357 | case 0: | |
1358 | break; | |
1359 | case -EFAULT: | |
1360 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1361 | put_futex_key(&key2); |
1362 | put_futex_key(&key1); | |
d0725992 | 1363 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
1364 | if (!ret) |
1365 | goto retry; | |
1366 | goto out; | |
1367 | case -EAGAIN: | |
1368 | /* The owner was exiting, try again. */ | |
1369 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1370 | put_futex_key(&key2); |
1371 | put_futex_key(&key1); | |
52400ba9 DH |
1372 | cond_resched(); |
1373 | goto retry; | |
1374 | default: | |
1375 | goto out_unlock; | |
1376 | } | |
1377 | } | |
1378 | ||
e2970f2f | 1379 | head1 = &hb1->chain; |
ec92d082 | 1380 | plist_for_each_entry_safe(this, next, head1, list) { |
52400ba9 DH |
1381 | if (task_count - nr_wake >= nr_requeue) |
1382 | break; | |
1383 | ||
1384 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1385 | continue; |
52400ba9 | 1386 | |
392741e0 DH |
1387 | /* |
1388 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
1389 | * be paired with each other and no other futex ops. | |
1390 | */ | |
1391 | if ((requeue_pi && !this->rt_waiter) || | |
1392 | (!requeue_pi && this->rt_waiter)) { | |
1393 | ret = -EINVAL; | |
1394 | break; | |
1395 | } | |
52400ba9 DH |
1396 | |
1397 | /* | |
1398 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1399 | * lock, we already woke the top_waiter. If not, it will be | |
1400 | * woken by futex_unlock_pi(). | |
1401 | */ | |
1402 | if (++task_count <= nr_wake && !requeue_pi) { | |
1da177e4 | 1403 | wake_futex(this); |
52400ba9 DH |
1404 | continue; |
1405 | } | |
1da177e4 | 1406 | |
84bc4af5 DH |
1407 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
1408 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
1409 | ret = -EINVAL; | |
1410 | break; | |
1411 | } | |
1412 | ||
52400ba9 DH |
1413 | /* |
1414 | * Requeue nr_requeue waiters and possibly one more in the case | |
1415 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1416 | */ | |
1417 | if (requeue_pi) { | |
1418 | /* Prepare the waiter to take the rt_mutex. */ | |
1419 | atomic_inc(&pi_state->refcount); | |
1420 | this->pi_state = pi_state; | |
1421 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1422 | this->rt_waiter, | |
1423 | this->task, 1); | |
1424 | if (ret == 1) { | |
1425 | /* We got the lock. */ | |
beda2c7e | 1426 | requeue_pi_wake_futex(this, &key2, hb2); |
89061d3d | 1427 | drop_count++; |
52400ba9 DH |
1428 | continue; |
1429 | } else if (ret) { | |
1430 | /* -EDEADLK */ | |
1431 | this->pi_state = NULL; | |
1432 | free_pi_state(pi_state); | |
1433 | goto out_unlock; | |
1434 | } | |
1da177e4 | 1435 | } |
52400ba9 DH |
1436 | requeue_futex(this, hb1, hb2, &key2); |
1437 | drop_count++; | |
1da177e4 LT |
1438 | } |
1439 | ||
1440 | out_unlock: | |
5eb3dc62 | 1441 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1442 | |
cd84a42f DH |
1443 | /* |
1444 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1445 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1446 | * one at key2 and updated their key pointer. We no longer need to | |
1447 | * hold the references to key1. | |
1448 | */ | |
1da177e4 | 1449 | while (--drop_count >= 0) |
9adef58b | 1450 | drop_futex_key_refs(&key1); |
1da177e4 | 1451 | |
42d35d48 | 1452 | out_put_keys: |
ae791a2d | 1453 | put_futex_key(&key2); |
42d35d48 | 1454 | out_put_key1: |
ae791a2d | 1455 | put_futex_key(&key1); |
42d35d48 | 1456 | out: |
52400ba9 DH |
1457 | if (pi_state != NULL) |
1458 | free_pi_state(pi_state); | |
1459 | return ret ? ret : task_count; | |
1da177e4 LT |
1460 | } |
1461 | ||
1462 | /* The key must be already stored in q->key. */ | |
82af7aca | 1463 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 1464 | __acquires(&hb->lock) |
1da177e4 | 1465 | { |
e2970f2f | 1466 | struct futex_hash_bucket *hb; |
1da177e4 | 1467 | |
e2970f2f IM |
1468 | hb = hash_futex(&q->key); |
1469 | q->lock_ptr = &hb->lock; | |
1da177e4 | 1470 | |
e2970f2f IM |
1471 | spin_lock(&hb->lock); |
1472 | return hb; | |
1da177e4 LT |
1473 | } |
1474 | ||
d40d65c8 DH |
1475 | static inline void |
1476 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) | |
15e408cd | 1477 | __releases(&hb->lock) |
d40d65c8 DH |
1478 | { |
1479 | spin_unlock(&hb->lock); | |
d40d65c8 DH |
1480 | } |
1481 | ||
1482 | /** | |
1483 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
1484 | * @q: The futex_q to enqueue | |
1485 | * @hb: The destination hash bucket | |
1486 | * | |
1487 | * The hb->lock must be held by the caller, and is released here. A call to | |
1488 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
1489 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
1490 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
1491 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
1492 | * an example). | |
1493 | */ | |
82af7aca | 1494 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
15e408cd | 1495 | __releases(&hb->lock) |
1da177e4 | 1496 | { |
ec92d082 PP |
1497 | int prio; |
1498 | ||
1499 | /* | |
1500 | * The priority used to register this element is | |
1501 | * - either the real thread-priority for the real-time threads | |
1502 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
1503 | * - or MAX_RT_PRIO for non-RT threads. | |
1504 | * Thus, all RT-threads are woken first in priority order, and | |
1505 | * the others are woken last, in FIFO order. | |
1506 | */ | |
1507 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1508 | ||
1509 | plist_node_init(&q->list, prio); | |
ec92d082 | 1510 | plist_add(&q->list, &hb->chain); |
c87e2837 | 1511 | q->task = current; |
e2970f2f | 1512 | spin_unlock(&hb->lock); |
1da177e4 LT |
1513 | } |
1514 | ||
d40d65c8 DH |
1515 | /** |
1516 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
1517 | * @q: The futex_q to unqueue | |
1518 | * | |
1519 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
1520 | * be paired with exactly one earlier call to queue_me(). | |
1521 | * | |
1522 | * Returns: | |
1523 | * 1 - if the futex_q was still queued (and we removed unqueued it) | |
1524 | * 0 - if the futex_q was already removed by the waking thread | |
1da177e4 | 1525 | */ |
1da177e4 LT |
1526 | static int unqueue_me(struct futex_q *q) |
1527 | { | |
1da177e4 | 1528 | spinlock_t *lock_ptr; |
e2970f2f | 1529 | int ret = 0; |
1da177e4 LT |
1530 | |
1531 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 1532 | retry: |
1da177e4 | 1533 | lock_ptr = q->lock_ptr; |
e91467ec | 1534 | barrier(); |
c80544dc | 1535 | if (lock_ptr != NULL) { |
1da177e4 LT |
1536 | spin_lock(lock_ptr); |
1537 | /* | |
1538 | * q->lock_ptr can change between reading it and | |
1539 | * spin_lock(), causing us to take the wrong lock. This | |
1540 | * corrects the race condition. | |
1541 | * | |
1542 | * Reasoning goes like this: if we have the wrong lock, | |
1543 | * q->lock_ptr must have changed (maybe several times) | |
1544 | * between reading it and the spin_lock(). It can | |
1545 | * change again after the spin_lock() but only if it was | |
1546 | * already changed before the spin_lock(). It cannot, | |
1547 | * however, change back to the original value. Therefore | |
1548 | * we can detect whether we acquired the correct lock. | |
1549 | */ | |
1550 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
1551 | spin_unlock(lock_ptr); | |
1552 | goto retry; | |
1553 | } | |
2e12978a | 1554 | __unqueue_futex(q); |
c87e2837 IM |
1555 | |
1556 | BUG_ON(q->pi_state); | |
1557 | ||
1da177e4 LT |
1558 | spin_unlock(lock_ptr); |
1559 | ret = 1; | |
1560 | } | |
1561 | ||
9adef58b | 1562 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1563 | return ret; |
1564 | } | |
1565 | ||
c87e2837 IM |
1566 | /* |
1567 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
1568 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
1569 | * and dropped here. | |
c87e2837 | 1570 | */ |
d0aa7a70 | 1571 | static void unqueue_me_pi(struct futex_q *q) |
15e408cd | 1572 | __releases(q->lock_ptr) |
c87e2837 | 1573 | { |
2e12978a | 1574 | __unqueue_futex(q); |
c87e2837 IM |
1575 | |
1576 | BUG_ON(!q->pi_state); | |
1577 | free_pi_state(q->pi_state); | |
1578 | q->pi_state = NULL; | |
1579 | ||
d0aa7a70 | 1580 | spin_unlock(q->lock_ptr); |
c87e2837 IM |
1581 | } |
1582 | ||
d0aa7a70 | 1583 | /* |
cdf71a10 | 1584 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 1585 | * |
778e9a9c AK |
1586 | * Must be called with hash bucket lock held and mm->sem held for non |
1587 | * private futexes. | |
d0aa7a70 | 1588 | */ |
778e9a9c | 1589 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
ae791a2d | 1590 | struct task_struct *newowner) |
d0aa7a70 | 1591 | { |
cdf71a10 | 1592 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 1593 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 1594 | struct task_struct *oldowner = pi_state->owner; |
7cfdaf38 | 1595 | u32 uval, uninitialized_var(curval), newval; |
e4dc5b7a | 1596 | int ret; |
d0aa7a70 PP |
1597 | |
1598 | /* Owner died? */ | |
1b7558e4 TG |
1599 | if (!pi_state->owner) |
1600 | newtid |= FUTEX_OWNER_DIED; | |
1601 | ||
1602 | /* | |
1603 | * We are here either because we stole the rtmutex from the | |
8161239a LJ |
1604 | * previous highest priority waiter or we are the highest priority |
1605 | * waiter but failed to get the rtmutex the first time. | |
1606 | * We have to replace the newowner TID in the user space variable. | |
1607 | * This must be atomic as we have to preserve the owner died bit here. | |
1b7558e4 | 1608 | * |
b2d0994b DH |
1609 | * Note: We write the user space value _before_ changing the pi_state |
1610 | * because we can fault here. Imagine swapped out pages or a fork | |
1611 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
1612 | * |
1613 | * Modifying pi_state _before_ the user space value would | |
1614 | * leave the pi_state in an inconsistent state when we fault | |
1615 | * here, because we need to drop the hash bucket lock to | |
1616 | * handle the fault. This might be observed in the PID check | |
1617 | * in lookup_pi_state. | |
1618 | */ | |
1619 | retry: | |
1620 | if (get_futex_value_locked(&uval, uaddr)) | |
1621 | goto handle_fault; | |
1622 | ||
1623 | while (1) { | |
1624 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
1625 | ||
37a9d912 | 1626 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1b7558e4 TG |
1627 | goto handle_fault; |
1628 | if (curval == uval) | |
1629 | break; | |
1630 | uval = curval; | |
1631 | } | |
1632 | ||
1633 | /* | |
1634 | * We fixed up user space. Now we need to fix the pi_state | |
1635 | * itself. | |
1636 | */ | |
d0aa7a70 | 1637 | if (pi_state->owner != NULL) { |
1d615482 | 1638 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
d0aa7a70 PP |
1639 | WARN_ON(list_empty(&pi_state->list)); |
1640 | list_del_init(&pi_state->list); | |
1d615482 | 1641 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
1b7558e4 | 1642 | } |
d0aa7a70 | 1643 | |
cdf71a10 | 1644 | pi_state->owner = newowner; |
d0aa7a70 | 1645 | |
1d615482 | 1646 | raw_spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 1647 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 | 1648 | list_add(&pi_state->list, &newowner->pi_state_list); |
1d615482 | 1649 | raw_spin_unlock_irq(&newowner->pi_lock); |
1b7558e4 | 1650 | return 0; |
d0aa7a70 | 1651 | |
d0aa7a70 | 1652 | /* |
1b7558e4 | 1653 | * To handle the page fault we need to drop the hash bucket |
8161239a LJ |
1654 | * lock here. That gives the other task (either the highest priority |
1655 | * waiter itself or the task which stole the rtmutex) the | |
1b7558e4 TG |
1656 | * chance to try the fixup of the pi_state. So once we are |
1657 | * back from handling the fault we need to check the pi_state | |
1658 | * after reacquiring the hash bucket lock and before trying to | |
1659 | * do another fixup. When the fixup has been done already we | |
1660 | * simply return. | |
d0aa7a70 | 1661 | */ |
1b7558e4 TG |
1662 | handle_fault: |
1663 | spin_unlock(q->lock_ptr); | |
778e9a9c | 1664 | |
d0725992 | 1665 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 1666 | |
1b7558e4 | 1667 | spin_lock(q->lock_ptr); |
778e9a9c | 1668 | |
1b7558e4 TG |
1669 | /* |
1670 | * Check if someone else fixed it for us: | |
1671 | */ | |
1672 | if (pi_state->owner != oldowner) | |
1673 | return 0; | |
1674 | ||
1675 | if (ret) | |
1676 | return ret; | |
1677 | ||
1678 | goto retry; | |
d0aa7a70 PP |
1679 | } |
1680 | ||
72c1bbf3 | 1681 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 1682 | |
dd973998 DH |
1683 | /** |
1684 | * fixup_owner() - Post lock pi_state and corner case management | |
1685 | * @uaddr: user address of the futex | |
dd973998 DH |
1686 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
1687 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
1688 | * | |
1689 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
1690 | * the pi_state owner as well as handle race conditions that may allow us to | |
1691 | * acquire the lock. Must be called with the hb lock held. | |
1692 | * | |
1693 | * Returns: | |
1694 | * 1 - success, lock taken | |
1695 | * 0 - success, lock not taken | |
1696 | * <0 - on error (-EFAULT) | |
1697 | */ | |
ae791a2d | 1698 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 DH |
1699 | { |
1700 | struct task_struct *owner; | |
1701 | int ret = 0; | |
1702 | ||
1703 | if (locked) { | |
1704 | /* | |
1705 | * Got the lock. We might not be the anticipated owner if we | |
1706 | * did a lock-steal - fix up the PI-state in that case: | |
1707 | */ | |
1708 | if (q->pi_state->owner != current) | |
ae791a2d | 1709 | ret = fixup_pi_state_owner(uaddr, q, current); |
dd973998 DH |
1710 | goto out; |
1711 | } | |
1712 | ||
1713 | /* | |
1714 | * Catch the rare case, where the lock was released when we were on the | |
1715 | * way back before we locked the hash bucket. | |
1716 | */ | |
1717 | if (q->pi_state->owner == current) { | |
1718 | /* | |
1719 | * Try to get the rt_mutex now. This might fail as some other | |
1720 | * task acquired the rt_mutex after we removed ourself from the | |
1721 | * rt_mutex waiters list. | |
1722 | */ | |
1723 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
1724 | locked = 1; | |
1725 | goto out; | |
1726 | } | |
1727 | ||
1728 | /* | |
1729 | * pi_state is incorrect, some other task did a lock steal and | |
1730 | * we returned due to timeout or signal without taking the | |
8161239a | 1731 | * rt_mutex. Too late. |
dd973998 | 1732 | */ |
8161239a | 1733 | raw_spin_lock(&q->pi_state->pi_mutex.wait_lock); |
dd973998 | 1734 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); |
8161239a LJ |
1735 | if (!owner) |
1736 | owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); | |
1737 | raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock); | |
ae791a2d | 1738 | ret = fixup_pi_state_owner(uaddr, q, owner); |
dd973998 DH |
1739 | goto out; |
1740 | } | |
1741 | ||
1742 | /* | |
1743 | * Paranoia check. If we did not take the lock, then we should not be | |
8161239a | 1744 | * the owner of the rt_mutex. |
dd973998 DH |
1745 | */ |
1746 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
1747 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
1748 | "pi-state %p\n", ret, | |
1749 | q->pi_state->pi_mutex.owner, | |
1750 | q->pi_state->owner); | |
1751 | ||
1752 | out: | |
1753 | return ret ? ret : locked; | |
1754 | } | |
1755 | ||
ca5f9524 DH |
1756 | /** |
1757 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
1758 | * @hb: the futex hash bucket, must be locked by the caller | |
1759 | * @q: the futex_q to queue up on | |
1760 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
1761 | */ |
1762 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 1763 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 1764 | { |
9beba3c5 DH |
1765 | /* |
1766 | * The task state is guaranteed to be set before another task can | |
1767 | * wake it. set_current_state() is implemented using set_mb() and | |
1768 | * queue_me() calls spin_unlock() upon completion, both serializing | |
1769 | * access to the hash list and forcing another memory barrier. | |
1770 | */ | |
f1a11e05 | 1771 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 1772 | queue_me(q, hb); |
ca5f9524 DH |
1773 | |
1774 | /* Arm the timer */ | |
1775 | if (timeout) { | |
1776 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); | |
1777 | if (!hrtimer_active(&timeout->timer)) | |
1778 | timeout->task = NULL; | |
1779 | } | |
1780 | ||
1781 | /* | |
0729e196 DH |
1782 | * If we have been removed from the hash list, then another task |
1783 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
1784 | */ |
1785 | if (likely(!plist_node_empty(&q->list))) { | |
1786 | /* | |
1787 | * If the timer has already expired, current will already be | |
1788 | * flagged for rescheduling. Only call schedule if there | |
1789 | * is no timeout, or if it has yet to expire. | |
1790 | */ | |
1791 | if (!timeout || timeout->task) | |
1792 | schedule(); | |
1793 | } | |
1794 | __set_current_state(TASK_RUNNING); | |
1795 | } | |
1796 | ||
f801073f DH |
1797 | /** |
1798 | * futex_wait_setup() - Prepare to wait on a futex | |
1799 | * @uaddr: the futex userspace address | |
1800 | * @val: the expected value | |
b41277dc | 1801 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
1802 | * @q: the associated futex_q |
1803 | * @hb: storage for hash_bucket pointer to be returned to caller | |
1804 | * | |
1805 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
1806 | * compare it with the expected value. Handle atomic faults internally. | |
1807 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
1808 | * with no q.key reference on failure. | |
1809 | * | |
1810 | * Returns: | |
1811 | * 0 - uaddr contains val and hb has been locked | |
ca4a04cf | 1812 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked |
f801073f | 1813 | */ |
b41277dc | 1814 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 1815 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 1816 | { |
e2970f2f IM |
1817 | u32 uval; |
1818 | int ret; | |
1da177e4 | 1819 | |
1da177e4 | 1820 | /* |
b2d0994b | 1821 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
1822 | * Order is important: |
1823 | * | |
1824 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
1825 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
1826 | * | |
1827 | * The basic logical guarantee of a futex is that it blocks ONLY | |
1828 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
1829 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
1830 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
1831 | * cond(var) false, which would violate the guarantee. |
1832 | * | |
8fe8f545 ML |
1833 | * On the other hand, we insert q and release the hash-bucket only |
1834 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
1835 | * absorb a wakeup if *uaddr does not match the desired values | |
1836 | * while the syscall executes. | |
1da177e4 | 1837 | */ |
f801073f | 1838 | retry: |
9ea71503 | 1839 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ); |
f801073f | 1840 | if (unlikely(ret != 0)) |
a5a2a0c7 | 1841 | return ret; |
f801073f DH |
1842 | |
1843 | retry_private: | |
1844 | *hb = queue_lock(q); | |
1845 | ||
e2970f2f | 1846 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 1847 | |
f801073f DH |
1848 | if (ret) { |
1849 | queue_unlock(q, *hb); | |
1da177e4 | 1850 | |
e2970f2f | 1851 | ret = get_user(uval, uaddr); |
e4dc5b7a | 1852 | if (ret) |
f801073f | 1853 | goto out; |
1da177e4 | 1854 | |
b41277dc | 1855 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1856 | goto retry_private; |
1857 | ||
ae791a2d | 1858 | put_futex_key(&q->key); |
e4dc5b7a | 1859 | goto retry; |
1da177e4 | 1860 | } |
ca5f9524 | 1861 | |
f801073f DH |
1862 | if (uval != val) { |
1863 | queue_unlock(q, *hb); | |
1864 | ret = -EWOULDBLOCK; | |
2fff78c7 | 1865 | } |
1da177e4 | 1866 | |
f801073f DH |
1867 | out: |
1868 | if (ret) | |
ae791a2d | 1869 | put_futex_key(&q->key); |
f801073f DH |
1870 | return ret; |
1871 | } | |
1872 | ||
b41277dc DH |
1873 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
1874 | ktime_t *abs_time, u32 bitset) | |
f801073f DH |
1875 | { |
1876 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
1877 | struct restart_block *restart; |
1878 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 1879 | struct futex_q q = futex_q_init; |
f801073f DH |
1880 | int ret; |
1881 | ||
1882 | if (!bitset) | |
1883 | return -EINVAL; | |
f801073f DH |
1884 | q.bitset = bitset; |
1885 | ||
1886 | if (abs_time) { | |
1887 | to = &timeout; | |
1888 | ||
b41277dc DH |
1889 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
1890 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
1891 | HRTIMER_MODE_ABS); | |
f801073f DH |
1892 | hrtimer_init_sleeper(to, current); |
1893 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
1894 | current->timer_slack_ns); | |
1895 | } | |
1896 | ||
d58e6576 | 1897 | retry: |
7ada876a DH |
1898 | /* |
1899 | * Prepare to wait on uaddr. On success, holds hb lock and increments | |
1900 | * q.key refs. | |
1901 | */ | |
b41277dc | 1902 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
1903 | if (ret) |
1904 | goto out; | |
1905 | ||
ca5f9524 | 1906 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 1907 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
1908 | |
1909 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 1910 | ret = 0; |
7ada876a | 1911 | /* unqueue_me() drops q.key ref */ |
1da177e4 | 1912 | if (!unqueue_me(&q)) |
7ada876a | 1913 | goto out; |
2fff78c7 | 1914 | ret = -ETIMEDOUT; |
ca5f9524 | 1915 | if (to && !to->task) |
7ada876a | 1916 | goto out; |
72c1bbf3 | 1917 | |
e2970f2f | 1918 | /* |
d58e6576 TG |
1919 | * We expect signal_pending(current), but we might be the |
1920 | * victim of a spurious wakeup as well. | |
e2970f2f | 1921 | */ |
7ada876a | 1922 | if (!signal_pending(current)) |
d58e6576 | 1923 | goto retry; |
d58e6576 | 1924 | |
2fff78c7 | 1925 | ret = -ERESTARTSYS; |
c19384b5 | 1926 | if (!abs_time) |
7ada876a | 1927 | goto out; |
1da177e4 | 1928 | |
2fff78c7 PZ |
1929 | restart = ¤t_thread_info()->restart_block; |
1930 | restart->fn = futex_wait_restart; | |
a3c74c52 | 1931 | restart->futex.uaddr = uaddr; |
2fff78c7 PZ |
1932 | restart->futex.val = val; |
1933 | restart->futex.time = abs_time->tv64; | |
1934 | restart->futex.bitset = bitset; | |
0cd9c649 | 1935 | restart->futex.flags = flags | FLAGS_HAS_TIMEOUT; |
42d35d48 | 1936 | |
2fff78c7 PZ |
1937 | ret = -ERESTART_RESTARTBLOCK; |
1938 | ||
42d35d48 | 1939 | out: |
ca5f9524 DH |
1940 | if (to) { |
1941 | hrtimer_cancel(&to->timer); | |
1942 | destroy_hrtimer_on_stack(&to->timer); | |
1943 | } | |
c87e2837 IM |
1944 | return ret; |
1945 | } | |
1946 | ||
72c1bbf3 NP |
1947 | |
1948 | static long futex_wait_restart(struct restart_block *restart) | |
1949 | { | |
a3c74c52 | 1950 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 1951 | ktime_t t, *tp = NULL; |
72c1bbf3 | 1952 | |
a72188d8 DH |
1953 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
1954 | t.tv64 = restart->futex.time; | |
1955 | tp = &t; | |
1956 | } | |
72c1bbf3 | 1957 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
1958 | |
1959 | return (long)futex_wait(uaddr, restart->futex.flags, | |
1960 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
1961 | } |
1962 | ||
1963 | ||
c87e2837 IM |
1964 | /* |
1965 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
1966 | * and failed. The kernel side here does the whole locking operation: | |
1967 | * if there are waiters then it will block, it does PI, etc. (Due to | |
1968 | * races the kernel might see a 0 value of the futex too.) | |
1969 | */ | |
b41277dc DH |
1970 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect, |
1971 | ktime_t *time, int trylock) | |
c87e2837 | 1972 | { |
c5780e97 | 1973 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 1974 | struct futex_hash_bucket *hb; |
5bdb05f9 | 1975 | struct futex_q q = futex_q_init; |
dd973998 | 1976 | int res, ret; |
c87e2837 IM |
1977 | |
1978 | if (refill_pi_state_cache()) | |
1979 | return -ENOMEM; | |
1980 | ||
c19384b5 | 1981 | if (time) { |
c5780e97 | 1982 | to = &timeout; |
237fc6e7 TG |
1983 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
1984 | HRTIMER_MODE_ABS); | |
c5780e97 | 1985 | hrtimer_init_sleeper(to, current); |
cc584b21 | 1986 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
1987 | } |
1988 | ||
42d35d48 | 1989 | retry: |
9ea71503 | 1990 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE); |
c87e2837 | 1991 | if (unlikely(ret != 0)) |
42d35d48 | 1992 | goto out; |
c87e2837 | 1993 | |
e4dc5b7a | 1994 | retry_private: |
82af7aca | 1995 | hb = queue_lock(&q); |
c87e2837 | 1996 | |
bab5bc9e | 1997 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 1998 | if (unlikely(ret)) { |
778e9a9c | 1999 | switch (ret) { |
1a52084d DH |
2000 | case 1: |
2001 | /* We got the lock. */ | |
2002 | ret = 0; | |
2003 | goto out_unlock_put_key; | |
2004 | case -EFAULT: | |
2005 | goto uaddr_faulted; | |
778e9a9c AK |
2006 | case -EAGAIN: |
2007 | /* | |
2008 | * Task is exiting and we just wait for the | |
2009 | * exit to complete. | |
2010 | */ | |
2011 | queue_unlock(&q, hb); | |
ae791a2d | 2012 | put_futex_key(&q.key); |
778e9a9c AK |
2013 | cond_resched(); |
2014 | goto retry; | |
778e9a9c | 2015 | default: |
42d35d48 | 2016 | goto out_unlock_put_key; |
c87e2837 | 2017 | } |
c87e2837 IM |
2018 | } |
2019 | ||
2020 | /* | |
2021 | * Only actually queue now that the atomic ops are done: | |
2022 | */ | |
82af7aca | 2023 | queue_me(&q, hb); |
c87e2837 | 2024 | |
c87e2837 IM |
2025 | WARN_ON(!q.pi_state); |
2026 | /* | |
2027 | * Block on the PI mutex: | |
2028 | */ | |
2029 | if (!trylock) | |
2030 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); | |
2031 | else { | |
2032 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); | |
2033 | /* Fixup the trylock return value: */ | |
2034 | ret = ret ? 0 : -EWOULDBLOCK; | |
2035 | } | |
2036 | ||
a99e4e41 | 2037 | spin_lock(q.lock_ptr); |
dd973998 DH |
2038 | /* |
2039 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2040 | * haven't already. | |
2041 | */ | |
ae791a2d | 2042 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 DH |
2043 | /* |
2044 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
2045 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
2046 | */ | |
2047 | if (res) | |
2048 | ret = (res < 0) ? res : 0; | |
c87e2837 | 2049 | |
e8f6386c | 2050 | /* |
dd973998 DH |
2051 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
2052 | * it and return the fault to userspace. | |
e8f6386c DH |
2053 | */ |
2054 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
2055 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
2056 | ||
778e9a9c AK |
2057 | /* Unqueue and drop the lock */ |
2058 | unqueue_me_pi(&q); | |
c87e2837 | 2059 | |
5ecb01cf | 2060 | goto out_put_key; |
c87e2837 | 2061 | |
42d35d48 | 2062 | out_unlock_put_key: |
c87e2837 IM |
2063 | queue_unlock(&q, hb); |
2064 | ||
42d35d48 | 2065 | out_put_key: |
ae791a2d | 2066 | put_futex_key(&q.key); |
42d35d48 | 2067 | out: |
237fc6e7 TG |
2068 | if (to) |
2069 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 2070 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 2071 | |
42d35d48 | 2072 | uaddr_faulted: |
778e9a9c AK |
2073 | queue_unlock(&q, hb); |
2074 | ||
d0725992 | 2075 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
2076 | if (ret) |
2077 | goto out_put_key; | |
c87e2837 | 2078 | |
b41277dc | 2079 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2080 | goto retry_private; |
2081 | ||
ae791a2d | 2082 | put_futex_key(&q.key); |
e4dc5b7a | 2083 | goto retry; |
c87e2837 IM |
2084 | } |
2085 | ||
c87e2837 IM |
2086 | /* |
2087 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
2088 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
2089 | * and do the rt-mutex unlock. | |
2090 | */ | |
b41277dc | 2091 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 IM |
2092 | { |
2093 | struct futex_hash_bucket *hb; | |
2094 | struct futex_q *this, *next; | |
ec92d082 | 2095 | struct plist_head *head; |
38d47c1b | 2096 | union futex_key key = FUTEX_KEY_INIT; |
c0c9ed15 | 2097 | u32 uval, vpid = task_pid_vnr(current); |
e4dc5b7a | 2098 | int ret; |
c87e2837 IM |
2099 | |
2100 | retry: | |
2101 | if (get_user(uval, uaddr)) | |
2102 | return -EFAULT; | |
2103 | /* | |
2104 | * We release only a lock we actually own: | |
2105 | */ | |
c0c9ed15 | 2106 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 2107 | return -EPERM; |
c87e2837 | 2108 | |
9ea71503 | 2109 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE); |
c87e2837 IM |
2110 | if (unlikely(ret != 0)) |
2111 | goto out; | |
2112 | ||
2113 | hb = hash_futex(&key); | |
2114 | spin_lock(&hb->lock); | |
2115 | ||
c87e2837 IM |
2116 | /* |
2117 | * To avoid races, try to do the TID -> 0 atomic transition | |
2118 | * again. If it succeeds then we can return without waking | |
2119 | * anyone else up: | |
2120 | */ | |
37a9d912 ML |
2121 | if (!(uval & FUTEX_OWNER_DIED) && |
2122 | cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0)) | |
c87e2837 IM |
2123 | goto pi_faulted; |
2124 | /* | |
2125 | * Rare case: we managed to release the lock atomically, | |
2126 | * no need to wake anyone else up: | |
2127 | */ | |
c0c9ed15 | 2128 | if (unlikely(uval == vpid)) |
c87e2837 IM |
2129 | goto out_unlock; |
2130 | ||
2131 | /* | |
2132 | * Ok, other tasks may need to be woken up - check waiters | |
2133 | * and do the wakeup if necessary: | |
2134 | */ | |
2135 | head = &hb->chain; | |
2136 | ||
ec92d082 | 2137 | plist_for_each_entry_safe(this, next, head, list) { |
c87e2837 IM |
2138 | if (!match_futex (&this->key, &key)) |
2139 | continue; | |
2140 | ret = wake_futex_pi(uaddr, uval, this); | |
2141 | /* | |
2142 | * The atomic access to the futex value | |
2143 | * generated a pagefault, so retry the | |
2144 | * user-access and the wakeup: | |
2145 | */ | |
2146 | if (ret == -EFAULT) | |
2147 | goto pi_faulted; | |
2148 | goto out_unlock; | |
2149 | } | |
2150 | /* | |
2151 | * No waiters - kernel unlocks the futex: | |
2152 | */ | |
e3f2ddea IM |
2153 | if (!(uval & FUTEX_OWNER_DIED)) { |
2154 | ret = unlock_futex_pi(uaddr, uval); | |
2155 | if (ret == -EFAULT) | |
2156 | goto pi_faulted; | |
2157 | } | |
c87e2837 IM |
2158 | |
2159 | out_unlock: | |
2160 | spin_unlock(&hb->lock); | |
ae791a2d | 2161 | put_futex_key(&key); |
c87e2837 | 2162 | |
42d35d48 | 2163 | out: |
c87e2837 IM |
2164 | return ret; |
2165 | ||
2166 | pi_faulted: | |
778e9a9c | 2167 | spin_unlock(&hb->lock); |
ae791a2d | 2168 | put_futex_key(&key); |
c87e2837 | 2169 | |
d0725992 | 2170 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2171 | if (!ret) |
c87e2837 IM |
2172 | goto retry; |
2173 | ||
1da177e4 LT |
2174 | return ret; |
2175 | } | |
2176 | ||
52400ba9 DH |
2177 | /** |
2178 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2179 | * @hb: the hash_bucket futex_q was original enqueued on | |
2180 | * @q: the futex_q woken while waiting to be requeued | |
2181 | * @key2: the futex_key of the requeue target futex | |
2182 | * @timeout: the timeout associated with the wait (NULL if none) | |
2183 | * | |
2184 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2185 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2186 | * the wakeup and return the appropriate error code to the caller. Must be | |
2187 | * called with the hb lock held. | |
2188 | * | |
2189 | * Returns | |
2190 | * 0 - no early wakeup detected | |
1c840c14 | 2191 | * <0 - -ETIMEDOUT or -ERESTARTNOINTR |
52400ba9 DH |
2192 | */ |
2193 | static inline | |
2194 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2195 | struct futex_q *q, union futex_key *key2, | |
2196 | struct hrtimer_sleeper *timeout) | |
2197 | { | |
2198 | int ret = 0; | |
2199 | ||
2200 | /* | |
2201 | * With the hb lock held, we avoid races while we process the wakeup. | |
2202 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2203 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2204 | * It can't be requeued from uaddr2 to something else since we don't | |
2205 | * support a PI aware source futex for requeue. | |
2206 | */ | |
2207 | if (!match_futex(&q->key, key2)) { | |
2208 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2209 | /* | |
2210 | * We were woken prior to requeue by a timeout or a signal. | |
2211 | * Unqueue the futex_q and determine which it was. | |
2212 | */ | |
2e12978a | 2213 | plist_del(&q->list, &hb->chain); |
52400ba9 | 2214 | |
d58e6576 | 2215 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 2216 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2217 | if (timeout && !timeout->task) |
2218 | ret = -ETIMEDOUT; | |
d58e6576 | 2219 | else if (signal_pending(current)) |
1c840c14 | 2220 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
2221 | } |
2222 | return ret; | |
2223 | } | |
2224 | ||
2225 | /** | |
2226 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 2227 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 2228 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
52400ba9 DH |
2229 | * the same type, no requeueing from private to shared, etc. |
2230 | * @val: the expected value of uaddr | |
2231 | * @abs_time: absolute timeout | |
56ec1607 | 2232 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
2233 | * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0) |
2234 | * @uaddr2: the pi futex we will take prior to returning to user-space | |
2235 | * | |
2236 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
6f7b0a2a DH |
2237 | * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake |
2238 | * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to | |
2239 | * userspace. This ensures the rt_mutex maintains an owner when it has waiters; | |
2240 | * without one, the pi logic would not know which task to boost/deboost, if | |
2241 | * there was a need to. | |
52400ba9 DH |
2242 | * |
2243 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
2244 | * via the following: | |
2245 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() | |
cc6db4e6 DH |
2246 | * 2) wakeup on uaddr2 after a requeue |
2247 | * 3) signal | |
2248 | * 4) timeout | |
52400ba9 | 2249 | * |
cc6db4e6 | 2250 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
2251 | * |
2252 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2253 | * 5) successful lock | |
2254 | * 6) signal | |
2255 | * 7) timeout | |
2256 | * 8) other lock acquisition failure | |
2257 | * | |
cc6db4e6 | 2258 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
2259 | * |
2260 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2261 | * | |
2262 | * Returns: | |
2263 | * 0 - On success | |
2264 | * <0 - On error | |
2265 | */ | |
b41277dc | 2266 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 2267 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 2268 | u32 __user *uaddr2) |
52400ba9 DH |
2269 | { |
2270 | struct hrtimer_sleeper timeout, *to = NULL; | |
2271 | struct rt_mutex_waiter rt_waiter; | |
2272 | struct rt_mutex *pi_mutex = NULL; | |
52400ba9 | 2273 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
2274 | union futex_key key2 = FUTEX_KEY_INIT; |
2275 | struct futex_q q = futex_q_init; | |
52400ba9 | 2276 | int res, ret; |
52400ba9 | 2277 | |
6f7b0a2a DH |
2278 | if (uaddr == uaddr2) |
2279 | return -EINVAL; | |
2280 | ||
52400ba9 DH |
2281 | if (!bitset) |
2282 | return -EINVAL; | |
2283 | ||
2284 | if (abs_time) { | |
2285 | to = &timeout; | |
b41277dc DH |
2286 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2287 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2288 | HRTIMER_MODE_ABS); | |
52400ba9 DH |
2289 | hrtimer_init_sleeper(to, current); |
2290 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2291 | current->timer_slack_ns); | |
2292 | } | |
2293 | ||
2294 | /* | |
2295 | * The waiter is allocated on our stack, manipulated by the requeue | |
2296 | * code while we sleep on uaddr. | |
2297 | */ | |
2298 | debug_rt_mutex_init_waiter(&rt_waiter); | |
2299 | rt_waiter.task = NULL; | |
2300 | ||
9ea71503 | 2301 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
52400ba9 DH |
2302 | if (unlikely(ret != 0)) |
2303 | goto out; | |
2304 | ||
84bc4af5 DH |
2305 | q.bitset = bitset; |
2306 | q.rt_waiter = &rt_waiter; | |
2307 | q.requeue_pi_key = &key2; | |
2308 | ||
7ada876a DH |
2309 | /* |
2310 | * Prepare to wait on uaddr. On success, increments q.key (key1) ref | |
2311 | * count. | |
2312 | */ | |
b41277dc | 2313 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 TG |
2314 | if (ret) |
2315 | goto out_key2; | |
52400ba9 DH |
2316 | |
2317 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ | |
f1a11e05 | 2318 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2319 | |
2320 | spin_lock(&hb->lock); | |
2321 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2322 | spin_unlock(&hb->lock); | |
2323 | if (ret) | |
2324 | goto out_put_keys; | |
2325 | ||
2326 | /* | |
2327 | * In order for us to be here, we know our q.key == key2, and since | |
2328 | * we took the hb->lock above, we also know that futex_requeue() has | |
2329 | * completed and we no longer have to concern ourselves with a wakeup | |
7ada876a DH |
2330 | * race with the atomic proxy lock acquisition by the requeue code. The |
2331 | * futex_requeue dropped our key1 reference and incremented our key2 | |
2332 | * reference count. | |
52400ba9 DH |
2333 | */ |
2334 | ||
2335 | /* Check if the requeue code acquired the second futex for us. */ | |
2336 | if (!q.rt_waiter) { | |
2337 | /* | |
2338 | * Got the lock. We might not be the anticipated owner if we | |
2339 | * did a lock-steal - fix up the PI-state in that case. | |
2340 | */ | |
2341 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2342 | spin_lock(q.lock_ptr); | |
ae791a2d | 2343 | ret = fixup_pi_state_owner(uaddr2, &q, current); |
52400ba9 DH |
2344 | spin_unlock(q.lock_ptr); |
2345 | } | |
2346 | } else { | |
2347 | /* | |
2348 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2349 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2350 | * the pi_state. | |
2351 | */ | |
f27071cb | 2352 | WARN_ON(!q.pi_state); |
52400ba9 DH |
2353 | pi_mutex = &q.pi_state->pi_mutex; |
2354 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1); | |
2355 | debug_rt_mutex_free_waiter(&rt_waiter); | |
2356 | ||
2357 | spin_lock(q.lock_ptr); | |
2358 | /* | |
2359 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2360 | * haven't already. | |
2361 | */ | |
ae791a2d | 2362 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 DH |
2363 | /* |
2364 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 2365 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
2366 | */ |
2367 | if (res) | |
2368 | ret = (res < 0) ? res : 0; | |
2369 | ||
2370 | /* Unqueue and drop the lock. */ | |
2371 | unqueue_me_pi(&q); | |
2372 | } | |
2373 | ||
2374 | /* | |
2375 | * If fixup_pi_state_owner() faulted and was unable to handle the | |
2376 | * fault, unlock the rt_mutex and return the fault to userspace. | |
2377 | */ | |
2378 | if (ret == -EFAULT) { | |
b6070a8d | 2379 | if (pi_mutex && rt_mutex_owner(pi_mutex) == current) |
52400ba9 DH |
2380 | rt_mutex_unlock(pi_mutex); |
2381 | } else if (ret == -EINTR) { | |
52400ba9 | 2382 | /* |
cc6db4e6 DH |
2383 | * We've already been requeued, but cannot restart by calling |
2384 | * futex_lock_pi() directly. We could restart this syscall, but | |
2385 | * it would detect that the user space "val" changed and return | |
2386 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
2387 | * -EWOULDBLOCK directly. | |
52400ba9 | 2388 | */ |
2070887f | 2389 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2390 | } |
2391 | ||
2392 | out_put_keys: | |
ae791a2d | 2393 | put_futex_key(&q.key); |
c8b15a70 | 2394 | out_key2: |
ae791a2d | 2395 | put_futex_key(&key2); |
52400ba9 DH |
2396 | |
2397 | out: | |
2398 | if (to) { | |
2399 | hrtimer_cancel(&to->timer); | |
2400 | destroy_hrtimer_on_stack(&to->timer); | |
2401 | } | |
2402 | return ret; | |
2403 | } | |
2404 | ||
0771dfef IM |
2405 | /* |
2406 | * Support for robust futexes: the kernel cleans up held futexes at | |
2407 | * thread exit time. | |
2408 | * | |
2409 | * Implementation: user-space maintains a per-thread list of locks it | |
2410 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2411 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2412 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2413 | * always manipulated with the lock held, so the list is private and |
2414 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2415 | * field, to allow the kernel to clean up if the thread dies after | |
2416 | * acquiring the lock, but just before it could have added itself to | |
2417 | * the list. There can only be one such pending lock. | |
2418 | */ | |
2419 | ||
2420 | /** | |
d96ee56c DH |
2421 | * sys_set_robust_list() - Set the robust-futex list head of a task |
2422 | * @head: pointer to the list-head | |
2423 | * @len: length of the list-head, as userspace expects | |
0771dfef | 2424 | */ |
836f92ad HC |
2425 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2426 | size_t, len) | |
0771dfef | 2427 | { |
a0c1e907 TG |
2428 | if (!futex_cmpxchg_enabled) |
2429 | return -ENOSYS; | |
0771dfef IM |
2430 | /* |
2431 | * The kernel knows only one size for now: | |
2432 | */ | |
2433 | if (unlikely(len != sizeof(*head))) | |
2434 | return -EINVAL; | |
2435 | ||
2436 | current->robust_list = head; | |
2437 | ||
2438 | return 0; | |
2439 | } | |
2440 | ||
2441 | /** | |
d96ee56c DH |
2442 | * sys_get_robust_list() - Get the robust-futex list head of a task |
2443 | * @pid: pid of the process [zero for current task] | |
2444 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
2445 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 2446 | */ |
836f92ad HC |
2447 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
2448 | struct robust_list_head __user * __user *, head_ptr, | |
2449 | size_t __user *, len_ptr) | |
0771dfef | 2450 | { |
ba46df98 | 2451 | struct robust_list_head __user *head; |
0771dfef | 2452 | unsigned long ret; |
bdbb776f | 2453 | struct task_struct *p; |
0771dfef | 2454 | |
a0c1e907 TG |
2455 | if (!futex_cmpxchg_enabled) |
2456 | return -ENOSYS; | |
2457 | ||
ec0c4274 KC |
2458 | WARN_ONCE(1, "deprecated: get_robust_list will be deleted in 2013.\n"); |
2459 | ||
bdbb776f KC |
2460 | rcu_read_lock(); |
2461 | ||
2462 | ret = -ESRCH; | |
0771dfef | 2463 | if (!pid) |
bdbb776f | 2464 | p = current; |
0771dfef | 2465 | else { |
228ebcbe | 2466 | p = find_task_by_vpid(pid); |
0771dfef IM |
2467 | if (!p) |
2468 | goto err_unlock; | |
0771dfef IM |
2469 | } |
2470 | ||
bdbb776f KC |
2471 | ret = -EPERM; |
2472 | if (!ptrace_may_access(p, PTRACE_MODE_READ)) | |
2473 | goto err_unlock; | |
2474 | ||
2475 | head = p->robust_list; | |
2476 | rcu_read_unlock(); | |
2477 | ||
0771dfef IM |
2478 | if (put_user(sizeof(*head), len_ptr)) |
2479 | return -EFAULT; | |
2480 | return put_user(head, head_ptr); | |
2481 | ||
2482 | err_unlock: | |
aaa2a97e | 2483 | rcu_read_unlock(); |
0771dfef IM |
2484 | |
2485 | return ret; | |
2486 | } | |
2487 | ||
2488 | /* | |
2489 | * Process a futex-list entry, check whether it's owned by the | |
2490 | * dying task, and do notification if so: | |
2491 | */ | |
e3f2ddea | 2492 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 2493 | { |
7cfdaf38 | 2494 | u32 uval, uninitialized_var(nval), mval; |
0771dfef | 2495 | |
8f17d3a5 IM |
2496 | retry: |
2497 | if (get_user(uval, uaddr)) | |
0771dfef IM |
2498 | return -1; |
2499 | ||
b488893a | 2500 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
2501 | /* |
2502 | * Ok, this dying thread is truly holding a futex | |
2503 | * of interest. Set the OWNER_DIED bit atomically | |
2504 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
2505 | * set, wake up a waiter (if any). (We have to do a | |
2506 | * futex_wake() even if OWNER_DIED is already set - | |
2507 | * to handle the rare but possible case of recursive | |
2508 | * thread-death.) The rest of the cleanup is done in | |
2509 | * userspace. | |
2510 | */ | |
e3f2ddea | 2511 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
6e0aa9f8 TG |
2512 | /* |
2513 | * We are not holding a lock here, but we want to have | |
2514 | * the pagefault_disable/enable() protection because | |
2515 | * we want to handle the fault gracefully. If the | |
2516 | * access fails we try to fault in the futex with R/W | |
2517 | * verification via get_user_pages. get_user() above | |
2518 | * does not guarantee R/W access. If that fails we | |
2519 | * give up and leave the futex locked. | |
2520 | */ | |
2521 | if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) { | |
2522 | if (fault_in_user_writeable(uaddr)) | |
2523 | return -1; | |
2524 | goto retry; | |
2525 | } | |
c87e2837 | 2526 | if (nval != uval) |
8f17d3a5 | 2527 | goto retry; |
0771dfef | 2528 | |
e3f2ddea IM |
2529 | /* |
2530 | * Wake robust non-PI futexes here. The wakeup of | |
2531 | * PI futexes happens in exit_pi_state(): | |
2532 | */ | |
36cf3b5c | 2533 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 2534 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
2535 | } |
2536 | return 0; | |
2537 | } | |
2538 | ||
e3f2ddea IM |
2539 | /* |
2540 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
2541 | */ | |
2542 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 2543 | struct robust_list __user * __user *head, |
1dcc41bb | 2544 | unsigned int *pi) |
e3f2ddea IM |
2545 | { |
2546 | unsigned long uentry; | |
2547 | ||
ba46df98 | 2548 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
2549 | return -EFAULT; |
2550 | ||
ba46df98 | 2551 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
2552 | *pi = uentry & 1; |
2553 | ||
2554 | return 0; | |
2555 | } | |
2556 | ||
0771dfef IM |
2557 | /* |
2558 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
2559 | * and mark any locks found there dead, and notify any waiters. | |
2560 | * | |
2561 | * We silently return on any sign of list-walking problem. | |
2562 | */ | |
2563 | void exit_robust_list(struct task_struct *curr) | |
2564 | { | |
2565 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 2566 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 DH |
2567 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
2568 | unsigned int uninitialized_var(next_pi); | |
0771dfef | 2569 | unsigned long futex_offset; |
9f96cb1e | 2570 | int rc; |
0771dfef | 2571 | |
a0c1e907 TG |
2572 | if (!futex_cmpxchg_enabled) |
2573 | return; | |
2574 | ||
0771dfef IM |
2575 | /* |
2576 | * Fetch the list head (which was registered earlier, via | |
2577 | * sys_set_robust_list()): | |
2578 | */ | |
e3f2ddea | 2579 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
2580 | return; |
2581 | /* | |
2582 | * Fetch the relative futex offset: | |
2583 | */ | |
2584 | if (get_user(futex_offset, &head->futex_offset)) | |
2585 | return; | |
2586 | /* | |
2587 | * Fetch any possibly pending lock-add first, and handle it | |
2588 | * if it exists: | |
2589 | */ | |
e3f2ddea | 2590 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 2591 | return; |
e3f2ddea | 2592 | |
9f96cb1e | 2593 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 2594 | while (entry != &head->list) { |
9f96cb1e MS |
2595 | /* |
2596 | * Fetch the next entry in the list before calling | |
2597 | * handle_futex_death: | |
2598 | */ | |
2599 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
2600 | /* |
2601 | * A pending lock might already be on the list, so | |
c87e2837 | 2602 | * don't process it twice: |
0771dfef IM |
2603 | */ |
2604 | if (entry != pending) | |
ba46df98 | 2605 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 2606 | curr, pi)) |
0771dfef | 2607 | return; |
9f96cb1e | 2608 | if (rc) |
0771dfef | 2609 | return; |
9f96cb1e MS |
2610 | entry = next_entry; |
2611 | pi = next_pi; | |
0771dfef IM |
2612 | /* |
2613 | * Avoid excessively long or circular lists: | |
2614 | */ | |
2615 | if (!--limit) | |
2616 | break; | |
2617 | ||
2618 | cond_resched(); | |
2619 | } | |
9f96cb1e MS |
2620 | |
2621 | if (pending) | |
2622 | handle_futex_death((void __user *)pending + futex_offset, | |
2623 | curr, pip); | |
0771dfef IM |
2624 | } |
2625 | ||
c19384b5 | 2626 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 2627 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 2628 | { |
81b40539 | 2629 | int cmd = op & FUTEX_CMD_MASK; |
b41277dc | 2630 | unsigned int flags = 0; |
34f01cc1 ED |
2631 | |
2632 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 2633 | flags |= FLAGS_SHARED; |
1da177e4 | 2634 | |
b41277dc DH |
2635 | if (op & FUTEX_CLOCK_REALTIME) { |
2636 | flags |= FLAGS_CLOCKRT; | |
2637 | if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI) | |
2638 | return -ENOSYS; | |
2639 | } | |
1da177e4 | 2640 | |
59263b51 TG |
2641 | switch (cmd) { |
2642 | case FUTEX_LOCK_PI: | |
2643 | case FUTEX_UNLOCK_PI: | |
2644 | case FUTEX_TRYLOCK_PI: | |
2645 | case FUTEX_WAIT_REQUEUE_PI: | |
2646 | case FUTEX_CMP_REQUEUE_PI: | |
2647 | if (!futex_cmpxchg_enabled) | |
2648 | return -ENOSYS; | |
2649 | } | |
2650 | ||
34f01cc1 | 2651 | switch (cmd) { |
1da177e4 | 2652 | case FUTEX_WAIT: |
cd689985 TG |
2653 | val3 = FUTEX_BITSET_MATCH_ANY; |
2654 | case FUTEX_WAIT_BITSET: | |
81b40539 | 2655 | return futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 | 2656 | case FUTEX_WAKE: |
cd689985 TG |
2657 | val3 = FUTEX_BITSET_MATCH_ANY; |
2658 | case FUTEX_WAKE_BITSET: | |
81b40539 | 2659 | return futex_wake(uaddr, flags, val, val3); |
1da177e4 | 2660 | case FUTEX_REQUEUE: |
81b40539 | 2661 | return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 | 2662 | case FUTEX_CMP_REQUEUE: |
81b40539 | 2663 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
4732efbe | 2664 | case FUTEX_WAKE_OP: |
81b40539 | 2665 | return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
c87e2837 | 2666 | case FUTEX_LOCK_PI: |
81b40539 | 2667 | return futex_lock_pi(uaddr, flags, val, timeout, 0); |
c87e2837 | 2668 | case FUTEX_UNLOCK_PI: |
81b40539 | 2669 | return futex_unlock_pi(uaddr, flags); |
c87e2837 | 2670 | case FUTEX_TRYLOCK_PI: |
81b40539 | 2671 | return futex_lock_pi(uaddr, flags, 0, timeout, 1); |
52400ba9 DH |
2672 | case FUTEX_WAIT_REQUEUE_PI: |
2673 | val3 = FUTEX_BITSET_MATCH_ANY; | |
81b40539 TG |
2674 | return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
2675 | uaddr2); | |
52400ba9 | 2676 | case FUTEX_CMP_REQUEUE_PI: |
81b40539 | 2677 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
1da177e4 | 2678 | } |
81b40539 | 2679 | return -ENOSYS; |
1da177e4 LT |
2680 | } |
2681 | ||
2682 | ||
17da2bd9 HC |
2683 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
2684 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
2685 | u32, val3) | |
1da177e4 | 2686 | { |
c19384b5 PP |
2687 | struct timespec ts; |
2688 | ktime_t t, *tp = NULL; | |
e2970f2f | 2689 | u32 val2 = 0; |
34f01cc1 | 2690 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 2691 | |
cd689985 | 2692 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
2693 | cmd == FUTEX_WAIT_BITSET || |
2694 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
c19384b5 | 2695 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 2696 | return -EFAULT; |
c19384b5 | 2697 | if (!timespec_valid(&ts)) |
9741ef96 | 2698 | return -EINVAL; |
c19384b5 PP |
2699 | |
2700 | t = timespec_to_ktime(ts); | |
34f01cc1 | 2701 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 2702 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 2703 | tp = &t; |
1da177e4 LT |
2704 | } |
2705 | /* | |
52400ba9 | 2706 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 2707 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 2708 | */ |
f54f0986 | 2709 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 2710 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 2711 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 2712 | |
c19384b5 | 2713 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
2714 | } |
2715 | ||
f6d107fb | 2716 | static int __init futex_init(void) |
1da177e4 | 2717 | { |
a0c1e907 | 2718 | u32 curval; |
3e4ab747 | 2719 | int i; |
95362fa9 | 2720 | |
a0c1e907 TG |
2721 | /* |
2722 | * This will fail and we want it. Some arch implementations do | |
2723 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
2724 | * functionality. We want to know that before we call in any | |
2725 | * of the complex code paths. Also we want to prevent | |
2726 | * registration of robust lists in that case. NULL is | |
2727 | * guaranteed to fault and we get -EFAULT on functional | |
fb62db2b | 2728 | * implementation, the non-functional ones will return |
a0c1e907 TG |
2729 | * -ENOSYS. |
2730 | */ | |
37a9d912 | 2731 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) |
a0c1e907 TG |
2732 | futex_cmpxchg_enabled = 1; |
2733 | ||
3e4ab747 | 2734 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { |
732375c6 | 2735 | plist_head_init(&futex_queues[i].chain); |
3e4ab747 TG |
2736 | spin_lock_init(&futex_queues[i].lock); |
2737 | } | |
2738 | ||
1da177e4 LT |
2739 | return 0; |
2740 | } | |
f6d107fb | 2741 | __initcall(futex_init); |