Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Fast Userspace Mutexes (which I call "Futexes!"). | |
3 | * (C) Rusty Russell, IBM 2002 | |
4 | * | |
5 | * Generalized futexes, futex requeueing, misc fixes by Ingo Molnar | |
6 | * (C) Copyright 2003 Red Hat Inc, All Rights Reserved | |
7 | * | |
8 | * Removed page pinning, fix privately mapped COW pages and other cleanups | |
9 | * (C) Copyright 2003, 2004 Jamie Lokier | |
10 | * | |
0771dfef IM |
11 | * Robust futex support started by Ingo Molnar |
12 | * (C) Copyright 2006 Red Hat Inc, All Rights Reserved | |
13 | * Thanks to Thomas Gleixner for suggestions, analysis and fixes. | |
14 | * | |
c87e2837 IM |
15 | * PI-futex support started by Ingo Molnar and Thomas Gleixner |
16 | * Copyright (C) 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com> | |
17 | * Copyright (C) 2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com> | |
18 | * | |
34f01cc1 ED |
19 | * PRIVATE futexes by Eric Dumazet |
20 | * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com> | |
21 | * | |
52400ba9 DH |
22 | * Requeue-PI support by Darren Hart <dvhltc@us.ibm.com> |
23 | * Copyright (C) IBM Corporation, 2009 | |
24 | * Thanks to Thomas Gleixner for conceptual design and careful reviews. | |
25 | * | |
1da177e4 LT |
26 | * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly |
27 | * enough at me, Linus for the original (flawed) idea, Matthew | |
28 | * Kirkwood for proof-of-concept implementation. | |
29 | * | |
30 | * "The futexes are also cursed." | |
31 | * "But they come in a choice of three flavours!" | |
32 | * | |
33 | * This program is free software; you can redistribute it and/or modify | |
34 | * it under the terms of the GNU General Public License as published by | |
35 | * the Free Software Foundation; either version 2 of the License, or | |
36 | * (at your option) any later version. | |
37 | * | |
38 | * This program is distributed in the hope that it will be useful, | |
39 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
40 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
41 | * GNU General Public License for more details. | |
42 | * | |
43 | * You should have received a copy of the GNU General Public License | |
44 | * along with this program; if not, write to the Free Software | |
45 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
46 | */ | |
47 | #include <linux/slab.h> | |
48 | #include <linux/poll.h> | |
49 | #include <linux/fs.h> | |
50 | #include <linux/file.h> | |
51 | #include <linux/jhash.h> | |
52 | #include <linux/init.h> | |
53 | #include <linux/futex.h> | |
54 | #include <linux/mount.h> | |
55 | #include <linux/pagemap.h> | |
56 | #include <linux/syscalls.h> | |
7ed20e1a | 57 | #include <linux/signal.h> |
9984de1a | 58 | #include <linux/export.h> |
fd5eea42 | 59 | #include <linux/magic.h> |
b488893a PE |
60 | #include <linux/pid.h> |
61 | #include <linux/nsproxy.h> | |
bdbb776f | 62 | #include <linux/ptrace.h> |
8bd75c77 | 63 | #include <linux/sched/rt.h> |
ab1842f1 | 64 | #include <linux/hugetlb.h> |
b488893a | 65 | |
4732efbe | 66 | #include <asm/futex.h> |
1da177e4 | 67 | |
c87e2837 IM |
68 | #include "rtmutex_common.h" |
69 | ||
f26c70a4 | 70 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 71 | int __read_mostly futex_cmpxchg_enabled; |
f26c70a4 | 72 | #endif |
a0c1e907 | 73 | |
1da177e4 LT |
74 | #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8) |
75 | ||
b41277dc DH |
76 | /* |
77 | * Futex flags used to encode options to functions and preserve them across | |
78 | * restarts. | |
79 | */ | |
80 | #define FLAGS_SHARED 0x01 | |
81 | #define FLAGS_CLOCKRT 0x02 | |
82 | #define FLAGS_HAS_TIMEOUT 0x04 | |
83 | ||
c87e2837 IM |
84 | /* |
85 | * Priority Inheritance state: | |
86 | */ | |
87 | struct futex_pi_state { | |
88 | /* | |
89 | * list of 'owned' pi_state instances - these have to be | |
90 | * cleaned up in do_exit() if the task exits prematurely: | |
91 | */ | |
92 | struct list_head list; | |
93 | ||
94 | /* | |
95 | * The PI object: | |
96 | */ | |
97 | struct rt_mutex pi_mutex; | |
98 | ||
99 | struct task_struct *owner; | |
100 | atomic_t refcount; | |
101 | ||
102 | union futex_key key; | |
103 | }; | |
104 | ||
d8d88fbb DH |
105 | /** |
106 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 107 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
108 | * @task: the task waiting on the futex |
109 | * @lock_ptr: the hash bucket lock | |
110 | * @key: the key the futex is hashed on | |
111 | * @pi_state: optional priority inheritance state | |
112 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
113 | * @requeue_pi_key: the requeue_pi target futex key | |
114 | * @bitset: bitset for the optional bitmasked wakeup | |
115 | * | |
116 | * We use this hashed waitqueue, instead of a normal wait_queue_t, so | |
1da177e4 LT |
117 | * we can wake only the relevant ones (hashed queues may be shared). |
118 | * | |
119 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 120 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 121 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
122 | * the second. |
123 | * | |
124 | * PI futexes are typically woken before they are removed from the hash list via | |
125 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
126 | */ |
127 | struct futex_q { | |
ec92d082 | 128 | struct plist_node list; |
1da177e4 | 129 | |
d8d88fbb | 130 | struct task_struct *task; |
1da177e4 | 131 | spinlock_t *lock_ptr; |
1da177e4 | 132 | union futex_key key; |
c87e2837 | 133 | struct futex_pi_state *pi_state; |
52400ba9 | 134 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 135 | union futex_key *requeue_pi_key; |
cd689985 | 136 | u32 bitset; |
1da177e4 LT |
137 | }; |
138 | ||
5bdb05f9 DH |
139 | static const struct futex_q futex_q_init = { |
140 | /* list gets initialized in queue_me()*/ | |
141 | .key = FUTEX_KEY_INIT, | |
142 | .bitset = FUTEX_BITSET_MATCH_ANY | |
143 | }; | |
144 | ||
1da177e4 | 145 | /* |
b2d0994b DH |
146 | * Hash buckets are shared by all the futex_keys that hash to the same |
147 | * location. Each key may have multiple futex_q structures, one for each task | |
148 | * waiting on a futex. | |
1da177e4 LT |
149 | */ |
150 | struct futex_hash_bucket { | |
ec92d082 PP |
151 | spinlock_t lock; |
152 | struct plist_head chain; | |
1da177e4 LT |
153 | }; |
154 | ||
155 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; | |
156 | ||
1da177e4 LT |
157 | /* |
158 | * We hash on the keys returned from get_futex_key (see below). | |
159 | */ | |
160 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
161 | { | |
162 | u32 hash = jhash2((u32*)&key->both.word, | |
163 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
164 | key->both.offset); | |
165 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; | |
166 | } | |
167 | ||
168 | /* | |
169 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
170 | */ | |
171 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
172 | { | |
2bc87203 DH |
173 | return (key1 && key2 |
174 | && key1->both.word == key2->both.word | |
1da177e4 LT |
175 | && key1->both.ptr == key2->both.ptr |
176 | && key1->both.offset == key2->both.offset); | |
177 | } | |
178 | ||
38d47c1b PZ |
179 | /* |
180 | * Take a reference to the resource addressed by a key. | |
181 | * Can be called while holding spinlocks. | |
182 | * | |
183 | */ | |
184 | static void get_futex_key_refs(union futex_key *key) | |
185 | { | |
186 | if (!key->both.ptr) | |
187 | return; | |
188 | ||
189 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
190 | case FUT_OFF_INODE: | |
7de9c6ee | 191 | ihold(key->shared.inode); |
38d47c1b PZ |
192 | break; |
193 | case FUT_OFF_MMSHARED: | |
194 | atomic_inc(&key->private.mm->mm_count); | |
195 | break; | |
196 | } | |
197 | } | |
198 | ||
199 | /* | |
200 | * Drop a reference to the resource addressed by a key. | |
201 | * The hash bucket spinlock must not be held. | |
202 | */ | |
203 | static void drop_futex_key_refs(union futex_key *key) | |
204 | { | |
90621c40 DH |
205 | if (!key->both.ptr) { |
206 | /* If we're here then we tried to put a key we failed to get */ | |
207 | WARN_ON_ONCE(1); | |
38d47c1b | 208 | return; |
90621c40 | 209 | } |
38d47c1b PZ |
210 | |
211 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
212 | case FUT_OFF_INODE: | |
213 | iput(key->shared.inode); | |
214 | break; | |
215 | case FUT_OFF_MMSHARED: | |
216 | mmdrop(key->private.mm); | |
217 | break; | |
218 | } | |
219 | } | |
220 | ||
34f01cc1 | 221 | /** |
d96ee56c DH |
222 | * get_futex_key() - Get parameters which are the keys for a futex |
223 | * @uaddr: virtual address of the futex | |
224 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED | |
225 | * @key: address where result is stored. | |
9ea71503 SB |
226 | * @rw: mapping needs to be read/write (values: VERIFY_READ, |
227 | * VERIFY_WRITE) | |
34f01cc1 | 228 | * |
6c23cbbd RD |
229 | * Return: a negative error code or 0 |
230 | * | |
34f01cc1 | 231 | * The key words are stored in *key on success. |
1da177e4 | 232 | * |
6131ffaa | 233 | * For shared mappings, it's (page->index, file_inode(vma->vm_file), |
1da177e4 LT |
234 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
235 | * We can usually work out the index without swapping in the page. | |
236 | * | |
b2d0994b | 237 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 238 | */ |
64d1304a | 239 | static int |
9ea71503 | 240 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) |
1da177e4 | 241 | { |
e2970f2f | 242 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 243 | struct mm_struct *mm = current->mm; |
a5b338f2 | 244 | struct page *page, *page_head; |
9ea71503 | 245 | int err, ro = 0; |
1da177e4 LT |
246 | |
247 | /* | |
248 | * The futex address must be "naturally" aligned. | |
249 | */ | |
e2970f2f | 250 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 251 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 252 | return -EINVAL; |
e2970f2f | 253 | address -= key->both.offset; |
1da177e4 | 254 | |
34f01cc1 ED |
255 | /* |
256 | * PROCESS_PRIVATE futexes are fast. | |
257 | * As the mm cannot disappear under us and the 'key' only needs | |
258 | * virtual address, we dont even have to find the underlying vma. | |
259 | * Note : We do have to check 'uaddr' is a valid user address, | |
260 | * but access_ok() should be faster than find_vma() | |
261 | */ | |
262 | if (!fshared) { | |
7485d0d3 | 263 | if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32)))) |
34f01cc1 ED |
264 | return -EFAULT; |
265 | key->private.mm = mm; | |
266 | key->private.address = address; | |
42569c39 | 267 | get_futex_key_refs(key); |
34f01cc1 ED |
268 | return 0; |
269 | } | |
1da177e4 | 270 | |
38d47c1b | 271 | again: |
7485d0d3 | 272 | err = get_user_pages_fast(address, 1, 1, &page); |
9ea71503 SB |
273 | /* |
274 | * If write access is not required (eg. FUTEX_WAIT), try | |
275 | * and get read-only access. | |
276 | */ | |
277 | if (err == -EFAULT && rw == VERIFY_READ) { | |
278 | err = get_user_pages_fast(address, 1, 0, &page); | |
279 | ro = 1; | |
280 | } | |
38d47c1b PZ |
281 | if (err < 0) |
282 | return err; | |
9ea71503 SB |
283 | else |
284 | err = 0; | |
38d47c1b | 285 | |
a5b338f2 AA |
286 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
287 | page_head = page; | |
288 | if (unlikely(PageTail(page))) { | |
38d47c1b | 289 | put_page(page); |
a5b338f2 AA |
290 | /* serialize against __split_huge_page_splitting() */ |
291 | local_irq_disable(); | |
13bb709c | 292 | if (likely(__get_user_pages_fast(address, 1, !ro, &page) == 1)) { |
a5b338f2 AA |
293 | page_head = compound_head(page); |
294 | /* | |
295 | * page_head is valid pointer but we must pin | |
296 | * it before taking the PG_lock and/or | |
297 | * PG_compound_lock. The moment we re-enable | |
298 | * irqs __split_huge_page_splitting() can | |
299 | * return and the head page can be freed from | |
300 | * under us. We can't take the PG_lock and/or | |
301 | * PG_compound_lock on a page that could be | |
302 | * freed from under us. | |
303 | */ | |
304 | if (page != page_head) { | |
305 | get_page(page_head); | |
306 | put_page(page); | |
307 | } | |
308 | local_irq_enable(); | |
309 | } else { | |
310 | local_irq_enable(); | |
311 | goto again; | |
312 | } | |
313 | } | |
314 | #else | |
315 | page_head = compound_head(page); | |
316 | if (page != page_head) { | |
317 | get_page(page_head); | |
318 | put_page(page); | |
319 | } | |
320 | #endif | |
321 | ||
322 | lock_page(page_head); | |
e6780f72 HD |
323 | |
324 | /* | |
325 | * If page_head->mapping is NULL, then it cannot be a PageAnon | |
326 | * page; but it might be the ZERO_PAGE or in the gate area or | |
327 | * in a special mapping (all cases which we are happy to fail); | |
328 | * or it may have been a good file page when get_user_pages_fast | |
329 | * found it, but truncated or holepunched or subjected to | |
330 | * invalidate_complete_page2 before we got the page lock (also | |
331 | * cases which we are happy to fail). And we hold a reference, | |
332 | * so refcount care in invalidate_complete_page's remove_mapping | |
333 | * prevents drop_caches from setting mapping to NULL beneath us. | |
334 | * | |
335 | * The case we do have to guard against is when memory pressure made | |
336 | * shmem_writepage move it from filecache to swapcache beneath us: | |
337 | * an unlikely race, but we do need to retry for page_head->mapping. | |
338 | */ | |
a5b338f2 | 339 | if (!page_head->mapping) { |
e6780f72 | 340 | int shmem_swizzled = PageSwapCache(page_head); |
a5b338f2 AA |
341 | unlock_page(page_head); |
342 | put_page(page_head); | |
e6780f72 HD |
343 | if (shmem_swizzled) |
344 | goto again; | |
345 | return -EFAULT; | |
38d47c1b | 346 | } |
1da177e4 LT |
347 | |
348 | /* | |
349 | * Private mappings are handled in a simple way. | |
350 | * | |
351 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
352 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 353 | * the object not the particular process. |
1da177e4 | 354 | */ |
a5b338f2 | 355 | if (PageAnon(page_head)) { |
9ea71503 SB |
356 | /* |
357 | * A RO anonymous page will never change and thus doesn't make | |
358 | * sense for futex operations. | |
359 | */ | |
360 | if (ro) { | |
361 | err = -EFAULT; | |
362 | goto out; | |
363 | } | |
364 | ||
38d47c1b | 365 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 366 | key->private.mm = mm; |
e2970f2f | 367 | key->private.address = address; |
38d47c1b PZ |
368 | } else { |
369 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ | |
a5b338f2 | 370 | key->shared.inode = page_head->mapping->host; |
ab1842f1 | 371 | key->shared.pgoff = basepage_index(page); |
1da177e4 LT |
372 | } |
373 | ||
38d47c1b | 374 | get_futex_key_refs(key); |
1da177e4 | 375 | |
9ea71503 | 376 | out: |
a5b338f2 AA |
377 | unlock_page(page_head); |
378 | put_page(page_head); | |
9ea71503 | 379 | return err; |
1da177e4 LT |
380 | } |
381 | ||
ae791a2d | 382 | static inline void put_futex_key(union futex_key *key) |
1da177e4 | 383 | { |
38d47c1b | 384 | drop_futex_key_refs(key); |
1da177e4 LT |
385 | } |
386 | ||
d96ee56c DH |
387 | /** |
388 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
389 | * @uaddr: pointer to faulting user space address |
390 | * | |
391 | * Slow path to fixup the fault we just took in the atomic write | |
392 | * access to @uaddr. | |
393 | * | |
fb62db2b | 394 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
395 | * user address. We know that we faulted in the atomic pagefault |
396 | * disabled section so we can as well avoid the #PF overhead by | |
397 | * calling get_user_pages() right away. | |
398 | */ | |
399 | static int fault_in_user_writeable(u32 __user *uaddr) | |
400 | { | |
722d0172 AK |
401 | struct mm_struct *mm = current->mm; |
402 | int ret; | |
403 | ||
404 | down_read(&mm->mmap_sem); | |
2efaca92 BH |
405 | ret = fixup_user_fault(current, mm, (unsigned long)uaddr, |
406 | FAULT_FLAG_WRITE); | |
722d0172 AK |
407 | up_read(&mm->mmap_sem); |
408 | ||
d0725992 TG |
409 | return ret < 0 ? ret : 0; |
410 | } | |
411 | ||
4b1c486b DH |
412 | /** |
413 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
414 | * @hb: the hash bucket the futex_q's reside in |
415 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
416 | * |
417 | * Must be called with the hb lock held. | |
418 | */ | |
419 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
420 | union futex_key *key) | |
421 | { | |
422 | struct futex_q *this; | |
423 | ||
424 | plist_for_each_entry(this, &hb->chain, list) { | |
425 | if (match_futex(&this->key, key)) | |
426 | return this; | |
427 | } | |
428 | return NULL; | |
429 | } | |
430 | ||
37a9d912 ML |
431 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
432 | u32 uval, u32 newval) | |
36cf3b5c | 433 | { |
37a9d912 | 434 | int ret; |
36cf3b5c TG |
435 | |
436 | pagefault_disable(); | |
37a9d912 | 437 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
438 | pagefault_enable(); |
439 | ||
37a9d912 | 440 | return ret; |
36cf3b5c TG |
441 | } |
442 | ||
443 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
444 | { |
445 | int ret; | |
446 | ||
a866374a | 447 | pagefault_disable(); |
e2970f2f | 448 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 449 | pagefault_enable(); |
1da177e4 LT |
450 | |
451 | return ret ? -EFAULT : 0; | |
452 | } | |
453 | ||
c87e2837 IM |
454 | |
455 | /* | |
456 | * PI code: | |
457 | */ | |
458 | static int refill_pi_state_cache(void) | |
459 | { | |
460 | struct futex_pi_state *pi_state; | |
461 | ||
462 | if (likely(current->pi_state_cache)) | |
463 | return 0; | |
464 | ||
4668edc3 | 465 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
466 | |
467 | if (!pi_state) | |
468 | return -ENOMEM; | |
469 | ||
c87e2837 IM |
470 | INIT_LIST_HEAD(&pi_state->list); |
471 | /* pi_mutex gets initialized later */ | |
472 | pi_state->owner = NULL; | |
473 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 474 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
475 | |
476 | current->pi_state_cache = pi_state; | |
477 | ||
478 | return 0; | |
479 | } | |
480 | ||
481 | static struct futex_pi_state * alloc_pi_state(void) | |
482 | { | |
483 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
484 | ||
485 | WARN_ON(!pi_state); | |
486 | current->pi_state_cache = NULL; | |
487 | ||
488 | return pi_state; | |
489 | } | |
490 | ||
491 | static void free_pi_state(struct futex_pi_state *pi_state) | |
492 | { | |
493 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
494 | return; | |
495 | ||
496 | /* | |
497 | * If pi_state->owner is NULL, the owner is most probably dying | |
498 | * and has cleaned up the pi_state already | |
499 | */ | |
500 | if (pi_state->owner) { | |
1d615482 | 501 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
c87e2837 | 502 | list_del_init(&pi_state->list); |
1d615482 | 503 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
c87e2837 IM |
504 | |
505 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
506 | } | |
507 | ||
508 | if (current->pi_state_cache) | |
509 | kfree(pi_state); | |
510 | else { | |
511 | /* | |
512 | * pi_state->list is already empty. | |
513 | * clear pi_state->owner. | |
514 | * refcount is at 0 - put it back to 1. | |
515 | */ | |
516 | pi_state->owner = NULL; | |
517 | atomic_set(&pi_state->refcount, 1); | |
518 | current->pi_state_cache = pi_state; | |
519 | } | |
520 | } | |
521 | ||
522 | /* | |
523 | * Look up the task based on what TID userspace gave us. | |
524 | * We dont trust it. | |
525 | */ | |
526 | static struct task_struct * futex_find_get_task(pid_t pid) | |
527 | { | |
528 | struct task_struct *p; | |
529 | ||
d359b549 | 530 | rcu_read_lock(); |
228ebcbe | 531 | p = find_task_by_vpid(pid); |
7a0ea09a MH |
532 | if (p) |
533 | get_task_struct(p); | |
a06381fe | 534 | |
d359b549 | 535 | rcu_read_unlock(); |
c87e2837 IM |
536 | |
537 | return p; | |
538 | } | |
539 | ||
540 | /* | |
541 | * This task is holding PI mutexes at exit time => bad. | |
542 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
543 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
544 | */ | |
545 | void exit_pi_state_list(struct task_struct *curr) | |
546 | { | |
c87e2837 IM |
547 | struct list_head *next, *head = &curr->pi_state_list; |
548 | struct futex_pi_state *pi_state; | |
627371d7 | 549 | struct futex_hash_bucket *hb; |
38d47c1b | 550 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 551 | |
a0c1e907 TG |
552 | if (!futex_cmpxchg_enabled) |
553 | return; | |
c87e2837 IM |
554 | /* |
555 | * We are a ZOMBIE and nobody can enqueue itself on | |
556 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 557 | * versus waiters unqueueing themselves: |
c87e2837 | 558 | */ |
1d615482 | 559 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 IM |
560 | while (!list_empty(head)) { |
561 | ||
562 | next = head->next; | |
563 | pi_state = list_entry(next, struct futex_pi_state, list); | |
564 | key = pi_state->key; | |
627371d7 | 565 | hb = hash_futex(&key); |
1d615482 | 566 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 567 | |
c87e2837 IM |
568 | spin_lock(&hb->lock); |
569 | ||
1d615482 | 570 | raw_spin_lock_irq(&curr->pi_lock); |
627371d7 IM |
571 | /* |
572 | * We dropped the pi-lock, so re-check whether this | |
573 | * task still owns the PI-state: | |
574 | */ | |
c87e2837 IM |
575 | if (head->next != next) { |
576 | spin_unlock(&hb->lock); | |
577 | continue; | |
578 | } | |
579 | ||
c87e2837 | 580 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
581 | WARN_ON(list_empty(&pi_state->list)); |
582 | list_del_init(&pi_state->list); | |
c87e2837 | 583 | pi_state->owner = NULL; |
1d615482 | 584 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
585 | |
586 | rt_mutex_unlock(&pi_state->pi_mutex); | |
587 | ||
588 | spin_unlock(&hb->lock); | |
589 | ||
1d615482 | 590 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 591 | } |
1d615482 | 592 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
593 | } |
594 | ||
595 | static int | |
d0aa7a70 | 596 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
cabef9fe TG |
597 | union futex_key *key, struct futex_pi_state **ps, |
598 | struct task_struct *task) | |
c87e2837 IM |
599 | { |
600 | struct futex_pi_state *pi_state = NULL; | |
601 | struct futex_q *this, *next; | |
ec92d082 | 602 | struct plist_head *head; |
c87e2837 | 603 | struct task_struct *p; |
778e9a9c | 604 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 IM |
605 | |
606 | head = &hb->chain; | |
607 | ||
ec92d082 | 608 | plist_for_each_entry_safe(this, next, head, list) { |
d0aa7a70 | 609 | if (match_futex(&this->key, key)) { |
c87e2837 IM |
610 | /* |
611 | * Another waiter already exists - bump up | |
612 | * the refcount and return its pi_state: | |
613 | */ | |
614 | pi_state = this->pi_state; | |
06a9ec29 | 615 | /* |
fb62db2b | 616 | * Userspace might have messed up non-PI and PI futexes |
06a9ec29 TG |
617 | */ |
618 | if (unlikely(!pi_state)) | |
619 | return -EINVAL; | |
620 | ||
627371d7 | 621 | WARN_ON(!atomic_read(&pi_state->refcount)); |
59647b6a TG |
622 | |
623 | /* | |
624 | * When pi_state->owner is NULL then the owner died | |
625 | * and another waiter is on the fly. pi_state->owner | |
626 | * is fixed up by the task which acquires | |
627 | * pi_state->rt_mutex. | |
628 | * | |
629 | * We do not check for pid == 0 which can happen when | |
630 | * the owner died and robust_list_exit() cleared the | |
631 | * TID. | |
632 | */ | |
633 | if (pid && pi_state->owner) { | |
634 | /* | |
635 | * Bail out if user space manipulated the | |
636 | * futex value. | |
637 | */ | |
638 | if (pid != task_pid_vnr(pi_state->owner)) | |
639 | return -EINVAL; | |
640 | } | |
627371d7 | 641 | |
cabef9fe TG |
642 | /* |
643 | * Protect against a corrupted uval. If uval | |
644 | * is 0x80000000 then pid is 0 and the waiter | |
645 | * bit is set. So the deadlock check in the | |
646 | * calling code has failed and we did not fall | |
647 | * into the check above due to !pid. | |
648 | */ | |
649 | if (task && pi_state->owner == task) | |
650 | return -EDEADLK; | |
651 | ||
c87e2837 | 652 | atomic_inc(&pi_state->refcount); |
d0aa7a70 | 653 | *ps = pi_state; |
c87e2837 IM |
654 | |
655 | return 0; | |
656 | } | |
657 | } | |
658 | ||
659 | /* | |
e3f2ddea | 660 | * We are the first waiter - try to look up the real owner and attach |
778e9a9c | 661 | * the new pi_state to it, but bail out when TID = 0 |
c87e2837 | 662 | */ |
778e9a9c | 663 | if (!pid) |
e3f2ddea | 664 | return -ESRCH; |
c87e2837 | 665 | p = futex_find_get_task(pid); |
7a0ea09a MH |
666 | if (!p) |
667 | return -ESRCH; | |
778e9a9c | 668 | |
452d7fea TG |
669 | if (!p->mm) { |
670 | put_task_struct(p); | |
671 | return -EPERM; | |
672 | } | |
673 | ||
778e9a9c AK |
674 | /* |
675 | * We need to look at the task state flags to figure out, | |
676 | * whether the task is exiting. To protect against the do_exit | |
677 | * change of the task flags, we do this protected by | |
678 | * p->pi_lock: | |
679 | */ | |
1d615482 | 680 | raw_spin_lock_irq(&p->pi_lock); |
778e9a9c AK |
681 | if (unlikely(p->flags & PF_EXITING)) { |
682 | /* | |
683 | * The task is on the way out. When PF_EXITPIDONE is | |
684 | * set, we know that the task has finished the | |
685 | * cleanup: | |
686 | */ | |
687 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
688 | ||
1d615482 | 689 | raw_spin_unlock_irq(&p->pi_lock); |
778e9a9c AK |
690 | put_task_struct(p); |
691 | return ret; | |
692 | } | |
c87e2837 IM |
693 | |
694 | pi_state = alloc_pi_state(); | |
695 | ||
696 | /* | |
697 | * Initialize the pi_mutex in locked state and make 'p' | |
698 | * the owner of it: | |
699 | */ | |
700 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
701 | ||
702 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 703 | pi_state->key = *key; |
c87e2837 | 704 | |
627371d7 | 705 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
706 | list_add(&pi_state->list, &p->pi_state_list); |
707 | pi_state->owner = p; | |
1d615482 | 708 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
709 | |
710 | put_task_struct(p); | |
711 | ||
d0aa7a70 | 712 | *ps = pi_state; |
c87e2837 IM |
713 | |
714 | return 0; | |
715 | } | |
716 | ||
1a52084d | 717 | /** |
d96ee56c | 718 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
719 | * @uaddr: the pi futex user address |
720 | * @hb: the pi futex hash bucket | |
721 | * @key: the futex key associated with uaddr and hb | |
722 | * @ps: the pi_state pointer where we store the result of the | |
723 | * lookup | |
724 | * @task: the task to perform the atomic lock work for. This will | |
725 | * be "current" except in the case of requeue pi. | |
726 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d | 727 | * |
6c23cbbd RD |
728 | * Return: |
729 | * 0 - ready to wait; | |
730 | * 1 - acquired the lock; | |
1a52084d DH |
731 | * <0 - error |
732 | * | |
733 | * The hb->lock and futex_key refs shall be held by the caller. | |
734 | */ | |
735 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
736 | union futex_key *key, | |
737 | struct futex_pi_state **ps, | |
bab5bc9e | 738 | struct task_struct *task, int set_waiters) |
1a52084d | 739 | { |
59fa6245 | 740 | int lock_taken, ret, force_take = 0; |
c0c9ed15 | 741 | u32 uval, newval, curval, vpid = task_pid_vnr(task); |
1a52084d DH |
742 | |
743 | retry: | |
744 | ret = lock_taken = 0; | |
745 | ||
746 | /* | |
747 | * To avoid races, we attempt to take the lock here again | |
748 | * (by doing a 0 -> TID atomic cmpxchg), while holding all | |
749 | * the locks. It will most likely not succeed. | |
750 | */ | |
c0c9ed15 | 751 | newval = vpid; |
bab5bc9e DH |
752 | if (set_waiters) |
753 | newval |= FUTEX_WAITERS; | |
1a52084d | 754 | |
37a9d912 | 755 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval))) |
1a52084d DH |
756 | return -EFAULT; |
757 | ||
758 | /* | |
759 | * Detect deadlocks. | |
760 | */ | |
c0c9ed15 | 761 | if ((unlikely((curval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
762 | return -EDEADLK; |
763 | ||
764 | /* | |
63d6ad59 | 765 | * Surprise - we got the lock, but we do not trust user space at all. |
1a52084d | 766 | */ |
63d6ad59 TG |
767 | if (unlikely(!curval)) { |
768 | /* | |
769 | * We verify whether there is kernel state for this | |
770 | * futex. If not, we can safely assume, that the 0 -> | |
771 | * TID transition is correct. If state exists, we do | |
772 | * not bother to fixup the user space state as it was | |
773 | * corrupted already. | |
774 | */ | |
775 | return futex_top_waiter(hb, key) ? -EINVAL : 1; | |
776 | } | |
1a52084d DH |
777 | |
778 | uval = curval; | |
779 | ||
780 | /* | |
781 | * Set the FUTEX_WAITERS flag, so the owner will know it has someone | |
782 | * to wake at the next unlock. | |
783 | */ | |
784 | newval = curval | FUTEX_WAITERS; | |
785 | ||
786 | /* | |
59fa6245 | 787 | * Should we force take the futex? See below. |
1a52084d | 788 | */ |
59fa6245 TG |
789 | if (unlikely(force_take)) { |
790 | /* | |
791 | * Keep the OWNER_DIED and the WAITERS bit and set the | |
792 | * new TID value. | |
793 | */ | |
c0c9ed15 | 794 | newval = (curval & ~FUTEX_TID_MASK) | vpid; |
59fa6245 | 795 | force_take = 0; |
1a52084d DH |
796 | lock_taken = 1; |
797 | } | |
798 | ||
37a9d912 | 799 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))) |
1a52084d DH |
800 | return -EFAULT; |
801 | if (unlikely(curval != uval)) | |
802 | goto retry; | |
803 | ||
804 | /* | |
59fa6245 | 805 | * We took the lock due to forced take over. |
1a52084d DH |
806 | */ |
807 | if (unlikely(lock_taken)) | |
808 | return 1; | |
809 | ||
810 | /* | |
811 | * We dont have the lock. Look up the PI state (or create it if | |
812 | * we are the first waiter): | |
813 | */ | |
cabef9fe | 814 | ret = lookup_pi_state(uval, hb, key, ps, task); |
1a52084d DH |
815 | |
816 | if (unlikely(ret)) { | |
817 | switch (ret) { | |
818 | case -ESRCH: | |
819 | /* | |
59fa6245 TG |
820 | * We failed to find an owner for this |
821 | * futex. So we have no pi_state to block | |
822 | * on. This can happen in two cases: | |
823 | * | |
824 | * 1) The owner died | |
825 | * 2) A stale FUTEX_WAITERS bit | |
826 | * | |
827 | * Re-read the futex value. | |
1a52084d DH |
828 | */ |
829 | if (get_futex_value_locked(&curval, uaddr)) | |
830 | return -EFAULT; | |
831 | ||
832 | /* | |
59fa6245 TG |
833 | * If the owner died or we have a stale |
834 | * WAITERS bit the owner TID in the user space | |
835 | * futex is 0. | |
1a52084d | 836 | */ |
59fa6245 TG |
837 | if (!(curval & FUTEX_TID_MASK)) { |
838 | force_take = 1; | |
1a52084d DH |
839 | goto retry; |
840 | } | |
841 | default: | |
842 | break; | |
843 | } | |
844 | } | |
845 | ||
846 | return ret; | |
847 | } | |
848 | ||
2e12978a LJ |
849 | /** |
850 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
851 | * @q: The futex_q to unqueue | |
852 | * | |
853 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
854 | */ | |
855 | static void __unqueue_futex(struct futex_q *q) | |
856 | { | |
857 | struct futex_hash_bucket *hb; | |
858 | ||
29096202 SR |
859 | if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr)) |
860 | || WARN_ON(plist_node_empty(&q->list))) | |
2e12978a LJ |
861 | return; |
862 | ||
863 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
864 | plist_del(&q->list, &hb->chain); | |
865 | } | |
866 | ||
1da177e4 LT |
867 | /* |
868 | * The hash bucket lock must be held when this is called. | |
869 | * Afterwards, the futex_q must not be accessed. | |
870 | */ | |
871 | static void wake_futex(struct futex_q *q) | |
872 | { | |
f1a11e05 TG |
873 | struct task_struct *p = q->task; |
874 | ||
aa10990e DH |
875 | if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n")) |
876 | return; | |
877 | ||
1da177e4 | 878 | /* |
f1a11e05 | 879 | * We set q->lock_ptr = NULL _before_ we wake up the task. If |
fb62db2b RD |
880 | * a non-futex wake up happens on another CPU then the task |
881 | * might exit and p would dereference a non-existing task | |
f1a11e05 TG |
882 | * struct. Prevent this by holding a reference on p across the |
883 | * wake up. | |
1da177e4 | 884 | */ |
f1a11e05 TG |
885 | get_task_struct(p); |
886 | ||
2e12978a | 887 | __unqueue_futex(q); |
1da177e4 | 888 | /* |
f1a11e05 TG |
889 | * The waiting task can free the futex_q as soon as |
890 | * q->lock_ptr = NULL is written, without taking any locks. A | |
891 | * memory barrier is required here to prevent the following | |
892 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 893 | */ |
ccdea2f8 | 894 | smp_wmb(); |
1da177e4 | 895 | q->lock_ptr = NULL; |
f1a11e05 TG |
896 | |
897 | wake_up_state(p, TASK_NORMAL); | |
898 | put_task_struct(p); | |
1da177e4 LT |
899 | } |
900 | ||
c87e2837 IM |
901 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
902 | { | |
903 | struct task_struct *new_owner; | |
904 | struct futex_pi_state *pi_state = this->pi_state; | |
7cfdaf38 | 905 | u32 uninitialized_var(curval), newval; |
9ad5dabd | 906 | int ret = 0; |
c87e2837 IM |
907 | |
908 | if (!pi_state) | |
909 | return -EINVAL; | |
910 | ||
51246bfd TG |
911 | /* |
912 | * If current does not own the pi_state then the futex is | |
913 | * inconsistent and user space fiddled with the futex value. | |
914 | */ | |
915 | if (pi_state->owner != current) | |
916 | return -EINVAL; | |
917 | ||
d209d74d | 918 | raw_spin_lock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
919 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
920 | ||
921 | /* | |
f123c98e SR |
922 | * It is possible that the next waiter (the one that brought |
923 | * this owner to the kernel) timed out and is no longer | |
924 | * waiting on the lock. | |
c87e2837 IM |
925 | */ |
926 | if (!new_owner) | |
927 | new_owner = this->task; | |
928 | ||
929 | /* | |
9ad5dabd TG |
930 | * We pass it to the next owner. The WAITERS bit is always |
931 | * kept enabled while there is PI state around. We cleanup the | |
932 | * owner died bit, because we are the owner. | |
c87e2837 | 933 | */ |
9ad5dabd | 934 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 935 | |
9ad5dabd TG |
936 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
937 | ret = -EFAULT; | |
938 | else if (curval != uval) | |
939 | ret = -EINVAL; | |
940 | if (ret) { | |
941 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); | |
942 | return ret; | |
e3f2ddea | 943 | } |
c87e2837 | 944 | |
1d615482 | 945 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
627371d7 IM |
946 | WARN_ON(list_empty(&pi_state->list)); |
947 | list_del_init(&pi_state->list); | |
1d615482 | 948 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
627371d7 | 949 | |
1d615482 | 950 | raw_spin_lock_irq(&new_owner->pi_lock); |
627371d7 | 951 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
952 | list_add(&pi_state->list, &new_owner->pi_state_list); |
953 | pi_state->owner = new_owner; | |
1d615482 | 954 | raw_spin_unlock_irq(&new_owner->pi_lock); |
627371d7 | 955 | |
d209d74d | 956 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
957 | rt_mutex_unlock(&pi_state->pi_mutex); |
958 | ||
959 | return 0; | |
960 | } | |
961 | ||
962 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) | |
963 | { | |
7cfdaf38 | 964 | u32 uninitialized_var(oldval); |
c87e2837 IM |
965 | |
966 | /* | |
967 | * There is no waiter, so we unlock the futex. The owner died | |
968 | * bit has not to be preserved here. We are the owner: | |
969 | */ | |
37a9d912 ML |
970 | if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0)) |
971 | return -EFAULT; | |
c87e2837 IM |
972 | if (oldval != uval) |
973 | return -EAGAIN; | |
974 | ||
975 | return 0; | |
976 | } | |
977 | ||
8b8f319f IM |
978 | /* |
979 | * Express the locking dependencies for lockdep: | |
980 | */ | |
981 | static inline void | |
982 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
983 | { | |
984 | if (hb1 <= hb2) { | |
985 | spin_lock(&hb1->lock); | |
986 | if (hb1 < hb2) | |
987 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
988 | } else { /* hb1 > hb2 */ | |
989 | spin_lock(&hb2->lock); | |
990 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
991 | } | |
992 | } | |
993 | ||
5eb3dc62 DH |
994 | static inline void |
995 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
996 | { | |
f061d351 | 997 | spin_unlock(&hb1->lock); |
88f502fe IM |
998 | if (hb1 != hb2) |
999 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
1000 | } |
1001 | ||
1da177e4 | 1002 | /* |
b2d0994b | 1003 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 1004 | */ |
b41277dc DH |
1005 | static int |
1006 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 1007 | { |
e2970f2f | 1008 | struct futex_hash_bucket *hb; |
1da177e4 | 1009 | struct futex_q *this, *next; |
ec92d082 | 1010 | struct plist_head *head; |
38d47c1b | 1011 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 LT |
1012 | int ret; |
1013 | ||
cd689985 TG |
1014 | if (!bitset) |
1015 | return -EINVAL; | |
1016 | ||
9ea71503 | 1017 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ); |
1da177e4 LT |
1018 | if (unlikely(ret != 0)) |
1019 | goto out; | |
1020 | ||
e2970f2f IM |
1021 | hb = hash_futex(&key); |
1022 | spin_lock(&hb->lock); | |
1023 | head = &hb->chain; | |
1da177e4 | 1024 | |
ec92d082 | 1025 | plist_for_each_entry_safe(this, next, head, list) { |
1da177e4 | 1026 | if (match_futex (&this->key, &key)) { |
52400ba9 | 1027 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
1028 | ret = -EINVAL; |
1029 | break; | |
1030 | } | |
cd689985 TG |
1031 | |
1032 | /* Check if one of the bits is set in both bitsets */ | |
1033 | if (!(this->bitset & bitset)) | |
1034 | continue; | |
1035 | ||
1da177e4 LT |
1036 | wake_futex(this); |
1037 | if (++ret >= nr_wake) | |
1038 | break; | |
1039 | } | |
1040 | } | |
1041 | ||
e2970f2f | 1042 | spin_unlock(&hb->lock); |
ae791a2d | 1043 | put_futex_key(&key); |
42d35d48 | 1044 | out: |
1da177e4 LT |
1045 | return ret; |
1046 | } | |
1047 | ||
4732efbe JJ |
1048 | /* |
1049 | * Wake up all waiters hashed on the physical page that is mapped | |
1050 | * to this virtual address: | |
1051 | */ | |
e2970f2f | 1052 | static int |
b41277dc | 1053 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 1054 | int nr_wake, int nr_wake2, int op) |
4732efbe | 1055 | { |
38d47c1b | 1056 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 1057 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 1058 | struct plist_head *head; |
4732efbe | 1059 | struct futex_q *this, *next; |
e4dc5b7a | 1060 | int ret, op_ret; |
4732efbe | 1061 | |
e4dc5b7a | 1062 | retry: |
9ea71503 | 1063 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
4732efbe JJ |
1064 | if (unlikely(ret != 0)) |
1065 | goto out; | |
9ea71503 | 1066 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
4732efbe | 1067 | if (unlikely(ret != 0)) |
42d35d48 | 1068 | goto out_put_key1; |
4732efbe | 1069 | |
e2970f2f IM |
1070 | hb1 = hash_futex(&key1); |
1071 | hb2 = hash_futex(&key2); | |
4732efbe | 1072 | |
e4dc5b7a | 1073 | retry_private: |
eaaea803 | 1074 | double_lock_hb(hb1, hb2); |
e2970f2f | 1075 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1076 | if (unlikely(op_ret < 0)) { |
4732efbe | 1077 | |
5eb3dc62 | 1078 | double_unlock_hb(hb1, hb2); |
4732efbe | 1079 | |
7ee1dd3f | 1080 | #ifndef CONFIG_MMU |
e2970f2f IM |
1081 | /* |
1082 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
1083 | * but we might get them from range checking | |
1084 | */ | |
7ee1dd3f | 1085 | ret = op_ret; |
42d35d48 | 1086 | goto out_put_keys; |
7ee1dd3f DH |
1087 | #endif |
1088 | ||
796f8d9b DG |
1089 | if (unlikely(op_ret != -EFAULT)) { |
1090 | ret = op_ret; | |
42d35d48 | 1091 | goto out_put_keys; |
796f8d9b DG |
1092 | } |
1093 | ||
d0725992 | 1094 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 1095 | if (ret) |
de87fcc1 | 1096 | goto out_put_keys; |
4732efbe | 1097 | |
b41277dc | 1098 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1099 | goto retry_private; |
1100 | ||
ae791a2d TG |
1101 | put_futex_key(&key2); |
1102 | put_futex_key(&key1); | |
e4dc5b7a | 1103 | goto retry; |
4732efbe JJ |
1104 | } |
1105 | ||
e2970f2f | 1106 | head = &hb1->chain; |
4732efbe | 1107 | |
ec92d082 | 1108 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe | 1109 | if (match_futex (&this->key, &key1)) { |
aa10990e DH |
1110 | if (this->pi_state || this->rt_waiter) { |
1111 | ret = -EINVAL; | |
1112 | goto out_unlock; | |
1113 | } | |
4732efbe JJ |
1114 | wake_futex(this); |
1115 | if (++ret >= nr_wake) | |
1116 | break; | |
1117 | } | |
1118 | } | |
1119 | ||
1120 | if (op_ret > 0) { | |
e2970f2f | 1121 | head = &hb2->chain; |
4732efbe JJ |
1122 | |
1123 | op_ret = 0; | |
ec92d082 | 1124 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe | 1125 | if (match_futex (&this->key, &key2)) { |
aa10990e DH |
1126 | if (this->pi_state || this->rt_waiter) { |
1127 | ret = -EINVAL; | |
1128 | goto out_unlock; | |
1129 | } | |
4732efbe JJ |
1130 | wake_futex(this); |
1131 | if (++op_ret >= nr_wake2) | |
1132 | break; | |
1133 | } | |
1134 | } | |
1135 | ret += op_ret; | |
1136 | } | |
1137 | ||
aa10990e | 1138 | out_unlock: |
5eb3dc62 | 1139 | double_unlock_hb(hb1, hb2); |
42d35d48 | 1140 | out_put_keys: |
ae791a2d | 1141 | put_futex_key(&key2); |
42d35d48 | 1142 | out_put_key1: |
ae791a2d | 1143 | put_futex_key(&key1); |
42d35d48 | 1144 | out: |
4732efbe JJ |
1145 | return ret; |
1146 | } | |
1147 | ||
9121e478 DH |
1148 | /** |
1149 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1150 | * @q: the futex_q to requeue | |
1151 | * @hb1: the source hash_bucket | |
1152 | * @hb2: the target hash_bucket | |
1153 | * @key2: the new key for the requeued futex_q | |
1154 | */ | |
1155 | static inline | |
1156 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1157 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1158 | { | |
1159 | ||
1160 | /* | |
1161 | * If key1 and key2 hash to the same bucket, no need to | |
1162 | * requeue. | |
1163 | */ | |
1164 | if (likely(&hb1->chain != &hb2->chain)) { | |
1165 | plist_del(&q->list, &hb1->chain); | |
1166 | plist_add(&q->list, &hb2->chain); | |
1167 | q->lock_ptr = &hb2->lock; | |
9121e478 DH |
1168 | } |
1169 | get_futex_key_refs(key2); | |
1170 | q->key = *key2; | |
1171 | } | |
1172 | ||
52400ba9 DH |
1173 | /** |
1174 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1175 | * @q: the futex_q |
1176 | * @key: the key of the requeue target futex | |
1177 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1178 | * |
1179 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1180 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1181 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1182 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1183 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1184 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1185 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1186 | */ |
1187 | static inline | |
beda2c7e DH |
1188 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1189 | struct futex_hash_bucket *hb) | |
52400ba9 | 1190 | { |
52400ba9 DH |
1191 | get_futex_key_refs(key); |
1192 | q->key = *key; | |
1193 | ||
2e12978a | 1194 | __unqueue_futex(q); |
52400ba9 DH |
1195 | |
1196 | WARN_ON(!q->rt_waiter); | |
1197 | q->rt_waiter = NULL; | |
1198 | ||
beda2c7e | 1199 | q->lock_ptr = &hb->lock; |
beda2c7e | 1200 | |
f1a11e05 | 1201 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1202 | } |
1203 | ||
1204 | /** | |
1205 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1206 | * @pifutex: the user address of the to futex |
1207 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1208 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1209 | * @key1: the from futex key | |
1210 | * @key2: the to futex key | |
1211 | * @ps: address to store the pi_state pointer | |
1212 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1213 | * |
1214 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1215 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1216 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1217 | * hb1 and hb2 must be held by the caller. | |
52400ba9 | 1218 | * |
6c23cbbd RD |
1219 | * Return: |
1220 | * 0 - failed to acquire the lock atomically; | |
cabef9fe | 1221 | * >0 - acquired the lock, return value is vpid of the top_waiter |
52400ba9 DH |
1222 | * <0 - error |
1223 | */ | |
1224 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1225 | struct futex_hash_bucket *hb1, | |
1226 | struct futex_hash_bucket *hb2, | |
1227 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1228 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1229 | { |
bab5bc9e | 1230 | struct futex_q *top_waiter = NULL; |
52400ba9 | 1231 | u32 curval; |
cabef9fe | 1232 | int ret, vpid; |
52400ba9 DH |
1233 | |
1234 | if (get_futex_value_locked(&curval, pifutex)) | |
1235 | return -EFAULT; | |
1236 | ||
bab5bc9e DH |
1237 | /* |
1238 | * Find the top_waiter and determine if there are additional waiters. | |
1239 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1240 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1241 | * as we have means to handle the possible fault. If not, don't set | |
1242 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1243 | * the kernel. | |
1244 | */ | |
52400ba9 DH |
1245 | top_waiter = futex_top_waiter(hb1, key1); |
1246 | ||
1247 | /* There are no waiters, nothing for us to do. */ | |
1248 | if (!top_waiter) | |
1249 | return 0; | |
1250 | ||
84bc4af5 DH |
1251 | /* Ensure we requeue to the expected futex. */ |
1252 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1253 | return -EINVAL; | |
1254 | ||
52400ba9 | 1255 | /* |
bab5bc9e DH |
1256 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1257 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1258 | * in ps in contended cases. | |
52400ba9 | 1259 | */ |
cabef9fe | 1260 | vpid = task_pid_vnr(top_waiter->task); |
bab5bc9e DH |
1261 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1262 | set_waiters); | |
cabef9fe | 1263 | if (ret == 1) { |
beda2c7e | 1264 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
cabef9fe TG |
1265 | return vpid; |
1266 | } | |
52400ba9 DH |
1267 | return ret; |
1268 | } | |
1269 | ||
1270 | /** | |
1271 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1272 | * @uaddr1: source futex user address |
b41277dc | 1273 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1274 | * @uaddr2: target futex user address |
1275 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1276 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1277 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1278 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1279 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1280 | * |
1281 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1282 | * uaddr2 atomically on behalf of the top waiter. | |
1283 | * | |
6c23cbbd RD |
1284 | * Return: |
1285 | * >=0 - on success, the number of tasks requeued or woken; | |
52400ba9 | 1286 | * <0 - on error |
1da177e4 | 1287 | */ |
b41277dc DH |
1288 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
1289 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
1290 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 1291 | { |
38d47c1b | 1292 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1293 | int drop_count = 0, task_count = 0, ret; |
1294 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1295 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 1296 | struct plist_head *head1; |
1da177e4 | 1297 | struct futex_q *this, *next; |
52400ba9 DH |
1298 | |
1299 | if (requeue_pi) { | |
b58623fb TG |
1300 | /* |
1301 | * Requeue PI only works on two distinct uaddrs. This | |
1302 | * check is only valid for private futexes. See below. | |
1303 | */ | |
1304 | if (uaddr1 == uaddr2) | |
1305 | return -EINVAL; | |
1306 | ||
52400ba9 DH |
1307 | /* |
1308 | * requeue_pi requires a pi_state, try to allocate it now | |
1309 | * without any locks in case it fails. | |
1310 | */ | |
1311 | if (refill_pi_state_cache()) | |
1312 | return -ENOMEM; | |
1313 | /* | |
1314 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1315 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1316 | * returning to userspace, so as to not leave the rt_mutex with | |
1317 | * waiters and no owner. However, second and third wake-ups | |
1318 | * cannot be predicted as they involve race conditions with the | |
1319 | * first wake and a fault while looking up the pi_state. Both | |
1320 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1321 | * use nr_wake=1. | |
1322 | */ | |
1323 | if (nr_wake != 1) | |
1324 | return -EINVAL; | |
1325 | } | |
1da177e4 | 1326 | |
42d35d48 | 1327 | retry: |
52400ba9 DH |
1328 | if (pi_state != NULL) { |
1329 | /* | |
1330 | * We will have to lookup the pi_state again, so free this one | |
1331 | * to keep the accounting correct. | |
1332 | */ | |
1333 | free_pi_state(pi_state); | |
1334 | pi_state = NULL; | |
1335 | } | |
1336 | ||
9ea71503 | 1337 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
1da177e4 LT |
1338 | if (unlikely(ret != 0)) |
1339 | goto out; | |
9ea71503 SB |
1340 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, |
1341 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1342 | if (unlikely(ret != 0)) |
42d35d48 | 1343 | goto out_put_key1; |
1da177e4 | 1344 | |
b58623fb TG |
1345 | /* |
1346 | * The check above which compares uaddrs is not sufficient for | |
1347 | * shared futexes. We need to compare the keys: | |
1348 | */ | |
1349 | if (requeue_pi && match_futex(&key1, &key2)) { | |
1350 | ret = -EINVAL; | |
1351 | goto out_put_keys; | |
1352 | } | |
1353 | ||
e2970f2f IM |
1354 | hb1 = hash_futex(&key1); |
1355 | hb2 = hash_futex(&key2); | |
1da177e4 | 1356 | |
e4dc5b7a | 1357 | retry_private: |
8b8f319f | 1358 | double_lock_hb(hb1, hb2); |
1da177e4 | 1359 | |
e2970f2f IM |
1360 | if (likely(cmpval != NULL)) { |
1361 | u32 curval; | |
1da177e4 | 1362 | |
e2970f2f | 1363 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1364 | |
1365 | if (unlikely(ret)) { | |
5eb3dc62 | 1366 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1367 | |
e2970f2f | 1368 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1369 | if (ret) |
1370 | goto out_put_keys; | |
1da177e4 | 1371 | |
b41277dc | 1372 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 1373 | goto retry_private; |
1da177e4 | 1374 | |
ae791a2d TG |
1375 | put_futex_key(&key2); |
1376 | put_futex_key(&key1); | |
e4dc5b7a | 1377 | goto retry; |
1da177e4 | 1378 | } |
e2970f2f | 1379 | if (curval != *cmpval) { |
1da177e4 LT |
1380 | ret = -EAGAIN; |
1381 | goto out_unlock; | |
1382 | } | |
1383 | } | |
1384 | ||
52400ba9 | 1385 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1386 | /* |
1387 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1388 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1389 | * bit. We force this here where we are able to easily handle | |
1390 | * faults rather in the requeue loop below. | |
1391 | */ | |
52400ba9 | 1392 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1393 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1394 | |
1395 | /* | |
1396 | * At this point the top_waiter has either taken uaddr2 or is | |
1397 | * waiting on it. If the former, then the pi_state will not | |
1398 | * exist yet, look it up one more time to ensure we have a | |
cabef9fe TG |
1399 | * reference to it. If the lock was taken, ret contains the |
1400 | * vpid of the top waiter task. | |
52400ba9 | 1401 | */ |
cabef9fe | 1402 | if (ret > 0) { |
52400ba9 | 1403 | WARN_ON(pi_state); |
89061d3d | 1404 | drop_count++; |
52400ba9 | 1405 | task_count++; |
cabef9fe TG |
1406 | /* |
1407 | * If we acquired the lock, then the user | |
1408 | * space value of uaddr2 should be vpid. It | |
1409 | * cannot be changed by the top waiter as it | |
1410 | * is blocked on hb2 lock if it tries to do | |
1411 | * so. If something fiddled with it behind our | |
1412 | * back the pi state lookup might unearth | |
1413 | * it. So we rather use the known value than | |
1414 | * rereading and handing potential crap to | |
1415 | * lookup_pi_state. | |
1416 | */ | |
1417 | ret = lookup_pi_state(ret, hb2, &key2, &pi_state, NULL); | |
52400ba9 DH |
1418 | } |
1419 | ||
1420 | switch (ret) { | |
1421 | case 0: | |
1422 | break; | |
1423 | case -EFAULT: | |
1424 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1425 | put_futex_key(&key2); |
1426 | put_futex_key(&key1); | |
d0725992 | 1427 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
1428 | if (!ret) |
1429 | goto retry; | |
1430 | goto out; | |
1431 | case -EAGAIN: | |
1432 | /* The owner was exiting, try again. */ | |
1433 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1434 | put_futex_key(&key2); |
1435 | put_futex_key(&key1); | |
52400ba9 DH |
1436 | cond_resched(); |
1437 | goto retry; | |
1438 | default: | |
1439 | goto out_unlock; | |
1440 | } | |
1441 | } | |
1442 | ||
e2970f2f | 1443 | head1 = &hb1->chain; |
ec92d082 | 1444 | plist_for_each_entry_safe(this, next, head1, list) { |
52400ba9 DH |
1445 | if (task_count - nr_wake >= nr_requeue) |
1446 | break; | |
1447 | ||
1448 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1449 | continue; |
52400ba9 | 1450 | |
392741e0 DH |
1451 | /* |
1452 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
1453 | * be paired with each other and no other futex ops. | |
aa10990e DH |
1454 | * |
1455 | * We should never be requeueing a futex_q with a pi_state, | |
1456 | * which is awaiting a futex_unlock_pi(). | |
392741e0 DH |
1457 | */ |
1458 | if ((requeue_pi && !this->rt_waiter) || | |
aa10990e DH |
1459 | (!requeue_pi && this->rt_waiter) || |
1460 | this->pi_state) { | |
392741e0 DH |
1461 | ret = -EINVAL; |
1462 | break; | |
1463 | } | |
52400ba9 DH |
1464 | |
1465 | /* | |
1466 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1467 | * lock, we already woke the top_waiter. If not, it will be | |
1468 | * woken by futex_unlock_pi(). | |
1469 | */ | |
1470 | if (++task_count <= nr_wake && !requeue_pi) { | |
1da177e4 | 1471 | wake_futex(this); |
52400ba9 DH |
1472 | continue; |
1473 | } | |
1da177e4 | 1474 | |
84bc4af5 DH |
1475 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
1476 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
1477 | ret = -EINVAL; | |
1478 | break; | |
1479 | } | |
1480 | ||
52400ba9 DH |
1481 | /* |
1482 | * Requeue nr_requeue waiters and possibly one more in the case | |
1483 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1484 | */ | |
1485 | if (requeue_pi) { | |
1486 | /* Prepare the waiter to take the rt_mutex. */ | |
1487 | atomic_inc(&pi_state->refcount); | |
1488 | this->pi_state = pi_state; | |
1489 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1490 | this->rt_waiter, | |
1491 | this->task, 1); | |
1492 | if (ret == 1) { | |
1493 | /* We got the lock. */ | |
beda2c7e | 1494 | requeue_pi_wake_futex(this, &key2, hb2); |
89061d3d | 1495 | drop_count++; |
52400ba9 DH |
1496 | continue; |
1497 | } else if (ret) { | |
1498 | /* -EDEADLK */ | |
1499 | this->pi_state = NULL; | |
1500 | free_pi_state(pi_state); | |
1501 | goto out_unlock; | |
1502 | } | |
1da177e4 | 1503 | } |
52400ba9 DH |
1504 | requeue_futex(this, hb1, hb2, &key2); |
1505 | drop_count++; | |
1da177e4 LT |
1506 | } |
1507 | ||
1508 | out_unlock: | |
5eb3dc62 | 1509 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1510 | |
cd84a42f DH |
1511 | /* |
1512 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1513 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1514 | * one at key2 and updated their key pointer. We no longer need to | |
1515 | * hold the references to key1. | |
1516 | */ | |
1da177e4 | 1517 | while (--drop_count >= 0) |
9adef58b | 1518 | drop_futex_key_refs(&key1); |
1da177e4 | 1519 | |
42d35d48 | 1520 | out_put_keys: |
ae791a2d | 1521 | put_futex_key(&key2); |
42d35d48 | 1522 | out_put_key1: |
ae791a2d | 1523 | put_futex_key(&key1); |
42d35d48 | 1524 | out: |
52400ba9 DH |
1525 | if (pi_state != NULL) |
1526 | free_pi_state(pi_state); | |
1527 | return ret ? ret : task_count; | |
1da177e4 LT |
1528 | } |
1529 | ||
1530 | /* The key must be already stored in q->key. */ | |
82af7aca | 1531 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 1532 | __acquires(&hb->lock) |
1da177e4 | 1533 | { |
e2970f2f | 1534 | struct futex_hash_bucket *hb; |
1da177e4 | 1535 | |
e2970f2f IM |
1536 | hb = hash_futex(&q->key); |
1537 | q->lock_ptr = &hb->lock; | |
1da177e4 | 1538 | |
e2970f2f IM |
1539 | spin_lock(&hb->lock); |
1540 | return hb; | |
1da177e4 LT |
1541 | } |
1542 | ||
d40d65c8 DH |
1543 | static inline void |
1544 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) | |
15e408cd | 1545 | __releases(&hb->lock) |
d40d65c8 DH |
1546 | { |
1547 | spin_unlock(&hb->lock); | |
d40d65c8 DH |
1548 | } |
1549 | ||
1550 | /** | |
1551 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
1552 | * @q: The futex_q to enqueue | |
1553 | * @hb: The destination hash bucket | |
1554 | * | |
1555 | * The hb->lock must be held by the caller, and is released here. A call to | |
1556 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
1557 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
1558 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
1559 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
1560 | * an example). | |
1561 | */ | |
82af7aca | 1562 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
15e408cd | 1563 | __releases(&hb->lock) |
1da177e4 | 1564 | { |
ec92d082 PP |
1565 | int prio; |
1566 | ||
1567 | /* | |
1568 | * The priority used to register this element is | |
1569 | * - either the real thread-priority for the real-time threads | |
1570 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
1571 | * - or MAX_RT_PRIO for non-RT threads. | |
1572 | * Thus, all RT-threads are woken first in priority order, and | |
1573 | * the others are woken last, in FIFO order. | |
1574 | */ | |
1575 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1576 | ||
1577 | plist_node_init(&q->list, prio); | |
ec92d082 | 1578 | plist_add(&q->list, &hb->chain); |
c87e2837 | 1579 | q->task = current; |
e2970f2f | 1580 | spin_unlock(&hb->lock); |
1da177e4 LT |
1581 | } |
1582 | ||
d40d65c8 DH |
1583 | /** |
1584 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
1585 | * @q: The futex_q to unqueue | |
1586 | * | |
1587 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
1588 | * be paired with exactly one earlier call to queue_me(). | |
1589 | * | |
6c23cbbd RD |
1590 | * Return: |
1591 | * 1 - if the futex_q was still queued (and we removed unqueued it); | |
d40d65c8 | 1592 | * 0 - if the futex_q was already removed by the waking thread |
1da177e4 | 1593 | */ |
1da177e4 LT |
1594 | static int unqueue_me(struct futex_q *q) |
1595 | { | |
1da177e4 | 1596 | spinlock_t *lock_ptr; |
e2970f2f | 1597 | int ret = 0; |
1da177e4 LT |
1598 | |
1599 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 1600 | retry: |
1da177e4 | 1601 | lock_ptr = q->lock_ptr; |
e91467ec | 1602 | barrier(); |
c80544dc | 1603 | if (lock_ptr != NULL) { |
1da177e4 LT |
1604 | spin_lock(lock_ptr); |
1605 | /* | |
1606 | * q->lock_ptr can change between reading it and | |
1607 | * spin_lock(), causing us to take the wrong lock. This | |
1608 | * corrects the race condition. | |
1609 | * | |
1610 | * Reasoning goes like this: if we have the wrong lock, | |
1611 | * q->lock_ptr must have changed (maybe several times) | |
1612 | * between reading it and the spin_lock(). It can | |
1613 | * change again after the spin_lock() but only if it was | |
1614 | * already changed before the spin_lock(). It cannot, | |
1615 | * however, change back to the original value. Therefore | |
1616 | * we can detect whether we acquired the correct lock. | |
1617 | */ | |
1618 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
1619 | spin_unlock(lock_ptr); | |
1620 | goto retry; | |
1621 | } | |
2e12978a | 1622 | __unqueue_futex(q); |
c87e2837 IM |
1623 | |
1624 | BUG_ON(q->pi_state); | |
1625 | ||
1da177e4 LT |
1626 | spin_unlock(lock_ptr); |
1627 | ret = 1; | |
1628 | } | |
1629 | ||
9adef58b | 1630 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1631 | return ret; |
1632 | } | |
1633 | ||
c87e2837 IM |
1634 | /* |
1635 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
1636 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
1637 | * and dropped here. | |
c87e2837 | 1638 | */ |
d0aa7a70 | 1639 | static void unqueue_me_pi(struct futex_q *q) |
15e408cd | 1640 | __releases(q->lock_ptr) |
c87e2837 | 1641 | { |
2e12978a | 1642 | __unqueue_futex(q); |
c87e2837 IM |
1643 | |
1644 | BUG_ON(!q->pi_state); | |
1645 | free_pi_state(q->pi_state); | |
1646 | q->pi_state = NULL; | |
1647 | ||
d0aa7a70 | 1648 | spin_unlock(q->lock_ptr); |
c87e2837 IM |
1649 | } |
1650 | ||
d0aa7a70 | 1651 | /* |
cdf71a10 | 1652 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 1653 | * |
778e9a9c AK |
1654 | * Must be called with hash bucket lock held and mm->sem held for non |
1655 | * private futexes. | |
d0aa7a70 | 1656 | */ |
778e9a9c | 1657 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
ae791a2d | 1658 | struct task_struct *newowner) |
d0aa7a70 | 1659 | { |
cdf71a10 | 1660 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 1661 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 1662 | struct task_struct *oldowner = pi_state->owner; |
7cfdaf38 | 1663 | u32 uval, uninitialized_var(curval), newval; |
e4dc5b7a | 1664 | int ret; |
d0aa7a70 PP |
1665 | |
1666 | /* Owner died? */ | |
1b7558e4 TG |
1667 | if (!pi_state->owner) |
1668 | newtid |= FUTEX_OWNER_DIED; | |
1669 | ||
1670 | /* | |
1671 | * We are here either because we stole the rtmutex from the | |
8161239a LJ |
1672 | * previous highest priority waiter or we are the highest priority |
1673 | * waiter but failed to get the rtmutex the first time. | |
1674 | * We have to replace the newowner TID in the user space variable. | |
1675 | * This must be atomic as we have to preserve the owner died bit here. | |
1b7558e4 | 1676 | * |
b2d0994b DH |
1677 | * Note: We write the user space value _before_ changing the pi_state |
1678 | * because we can fault here. Imagine swapped out pages or a fork | |
1679 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
1680 | * |
1681 | * Modifying pi_state _before_ the user space value would | |
1682 | * leave the pi_state in an inconsistent state when we fault | |
1683 | * here, because we need to drop the hash bucket lock to | |
1684 | * handle the fault. This might be observed in the PID check | |
1685 | * in lookup_pi_state. | |
1686 | */ | |
1687 | retry: | |
1688 | if (get_futex_value_locked(&uval, uaddr)) | |
1689 | goto handle_fault; | |
1690 | ||
1691 | while (1) { | |
1692 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
1693 | ||
37a9d912 | 1694 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1b7558e4 TG |
1695 | goto handle_fault; |
1696 | if (curval == uval) | |
1697 | break; | |
1698 | uval = curval; | |
1699 | } | |
1700 | ||
1701 | /* | |
1702 | * We fixed up user space. Now we need to fix the pi_state | |
1703 | * itself. | |
1704 | */ | |
d0aa7a70 | 1705 | if (pi_state->owner != NULL) { |
1d615482 | 1706 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
d0aa7a70 PP |
1707 | WARN_ON(list_empty(&pi_state->list)); |
1708 | list_del_init(&pi_state->list); | |
1d615482 | 1709 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
1b7558e4 | 1710 | } |
d0aa7a70 | 1711 | |
cdf71a10 | 1712 | pi_state->owner = newowner; |
d0aa7a70 | 1713 | |
1d615482 | 1714 | raw_spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 1715 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 | 1716 | list_add(&pi_state->list, &newowner->pi_state_list); |
1d615482 | 1717 | raw_spin_unlock_irq(&newowner->pi_lock); |
1b7558e4 | 1718 | return 0; |
d0aa7a70 | 1719 | |
d0aa7a70 | 1720 | /* |
1b7558e4 | 1721 | * To handle the page fault we need to drop the hash bucket |
8161239a LJ |
1722 | * lock here. That gives the other task (either the highest priority |
1723 | * waiter itself or the task which stole the rtmutex) the | |
1b7558e4 TG |
1724 | * chance to try the fixup of the pi_state. So once we are |
1725 | * back from handling the fault we need to check the pi_state | |
1726 | * after reacquiring the hash bucket lock and before trying to | |
1727 | * do another fixup. When the fixup has been done already we | |
1728 | * simply return. | |
d0aa7a70 | 1729 | */ |
1b7558e4 TG |
1730 | handle_fault: |
1731 | spin_unlock(q->lock_ptr); | |
778e9a9c | 1732 | |
d0725992 | 1733 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 1734 | |
1b7558e4 | 1735 | spin_lock(q->lock_ptr); |
778e9a9c | 1736 | |
1b7558e4 TG |
1737 | /* |
1738 | * Check if someone else fixed it for us: | |
1739 | */ | |
1740 | if (pi_state->owner != oldowner) | |
1741 | return 0; | |
1742 | ||
1743 | if (ret) | |
1744 | return ret; | |
1745 | ||
1746 | goto retry; | |
d0aa7a70 PP |
1747 | } |
1748 | ||
72c1bbf3 | 1749 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 1750 | |
dd973998 DH |
1751 | /** |
1752 | * fixup_owner() - Post lock pi_state and corner case management | |
1753 | * @uaddr: user address of the futex | |
dd973998 DH |
1754 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
1755 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
1756 | * | |
1757 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
1758 | * the pi_state owner as well as handle race conditions that may allow us to | |
1759 | * acquire the lock. Must be called with the hb lock held. | |
1760 | * | |
6c23cbbd RD |
1761 | * Return: |
1762 | * 1 - success, lock taken; | |
1763 | * 0 - success, lock not taken; | |
dd973998 DH |
1764 | * <0 - on error (-EFAULT) |
1765 | */ | |
ae791a2d | 1766 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 DH |
1767 | { |
1768 | struct task_struct *owner; | |
1769 | int ret = 0; | |
1770 | ||
1771 | if (locked) { | |
1772 | /* | |
1773 | * Got the lock. We might not be the anticipated owner if we | |
1774 | * did a lock-steal - fix up the PI-state in that case: | |
1775 | */ | |
1776 | if (q->pi_state->owner != current) | |
ae791a2d | 1777 | ret = fixup_pi_state_owner(uaddr, q, current); |
dd973998 DH |
1778 | goto out; |
1779 | } | |
1780 | ||
1781 | /* | |
1782 | * Catch the rare case, where the lock was released when we were on the | |
1783 | * way back before we locked the hash bucket. | |
1784 | */ | |
1785 | if (q->pi_state->owner == current) { | |
1786 | /* | |
1787 | * Try to get the rt_mutex now. This might fail as some other | |
1788 | * task acquired the rt_mutex after we removed ourself from the | |
1789 | * rt_mutex waiters list. | |
1790 | */ | |
1791 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
1792 | locked = 1; | |
1793 | goto out; | |
1794 | } | |
1795 | ||
1796 | /* | |
1797 | * pi_state is incorrect, some other task did a lock steal and | |
1798 | * we returned due to timeout or signal without taking the | |
8161239a | 1799 | * rt_mutex. Too late. |
dd973998 | 1800 | */ |
8161239a | 1801 | raw_spin_lock(&q->pi_state->pi_mutex.wait_lock); |
dd973998 | 1802 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); |
8161239a LJ |
1803 | if (!owner) |
1804 | owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); | |
1805 | raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock); | |
ae791a2d | 1806 | ret = fixup_pi_state_owner(uaddr, q, owner); |
dd973998 DH |
1807 | goto out; |
1808 | } | |
1809 | ||
1810 | /* | |
1811 | * Paranoia check. If we did not take the lock, then we should not be | |
8161239a | 1812 | * the owner of the rt_mutex. |
dd973998 DH |
1813 | */ |
1814 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
1815 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
1816 | "pi-state %p\n", ret, | |
1817 | q->pi_state->pi_mutex.owner, | |
1818 | q->pi_state->owner); | |
1819 | ||
1820 | out: | |
1821 | return ret ? ret : locked; | |
1822 | } | |
1823 | ||
ca5f9524 DH |
1824 | /** |
1825 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
1826 | * @hb: the futex hash bucket, must be locked by the caller | |
1827 | * @q: the futex_q to queue up on | |
1828 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
1829 | */ |
1830 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 1831 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 1832 | { |
9beba3c5 DH |
1833 | /* |
1834 | * The task state is guaranteed to be set before another task can | |
1835 | * wake it. set_current_state() is implemented using set_mb() and | |
1836 | * queue_me() calls spin_unlock() upon completion, both serializing | |
1837 | * access to the hash list and forcing another memory barrier. | |
1838 | */ | |
f1a11e05 | 1839 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 1840 | queue_me(q, hb); |
ca5f9524 DH |
1841 | |
1842 | /* Arm the timer */ | |
1843 | if (timeout) { | |
1844 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); | |
1845 | if (!hrtimer_active(&timeout->timer)) | |
1846 | timeout->task = NULL; | |
1847 | } | |
1848 | ||
1849 | /* | |
0729e196 DH |
1850 | * If we have been removed from the hash list, then another task |
1851 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
1852 | */ |
1853 | if (likely(!plist_node_empty(&q->list))) { | |
1854 | /* | |
1855 | * If the timer has already expired, current will already be | |
1856 | * flagged for rescheduling. Only call schedule if there | |
1857 | * is no timeout, or if it has yet to expire. | |
1858 | */ | |
1859 | if (!timeout || timeout->task) | |
1860 | schedule(); | |
1861 | } | |
1862 | __set_current_state(TASK_RUNNING); | |
1863 | } | |
1864 | ||
f801073f DH |
1865 | /** |
1866 | * futex_wait_setup() - Prepare to wait on a futex | |
1867 | * @uaddr: the futex userspace address | |
1868 | * @val: the expected value | |
b41277dc | 1869 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
1870 | * @q: the associated futex_q |
1871 | * @hb: storage for hash_bucket pointer to be returned to caller | |
1872 | * | |
1873 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
1874 | * compare it with the expected value. Handle atomic faults internally. | |
1875 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
1876 | * with no q.key reference on failure. | |
1877 | * | |
6c23cbbd RD |
1878 | * Return: |
1879 | * 0 - uaddr contains val and hb has been locked; | |
ca4a04cf | 1880 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked |
f801073f | 1881 | */ |
b41277dc | 1882 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 1883 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 1884 | { |
e2970f2f IM |
1885 | u32 uval; |
1886 | int ret; | |
1da177e4 | 1887 | |
1da177e4 | 1888 | /* |
b2d0994b | 1889 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
1890 | * Order is important: |
1891 | * | |
1892 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
1893 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
1894 | * | |
1895 | * The basic logical guarantee of a futex is that it blocks ONLY | |
1896 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
1897 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
1898 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
1899 | * cond(var) false, which would violate the guarantee. |
1900 | * | |
8fe8f545 ML |
1901 | * On the other hand, we insert q and release the hash-bucket only |
1902 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
1903 | * absorb a wakeup if *uaddr does not match the desired values | |
1904 | * while the syscall executes. | |
1da177e4 | 1905 | */ |
f801073f | 1906 | retry: |
9ea71503 | 1907 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ); |
f801073f | 1908 | if (unlikely(ret != 0)) |
a5a2a0c7 | 1909 | return ret; |
f801073f DH |
1910 | |
1911 | retry_private: | |
1912 | *hb = queue_lock(q); | |
1913 | ||
e2970f2f | 1914 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 1915 | |
f801073f DH |
1916 | if (ret) { |
1917 | queue_unlock(q, *hb); | |
1da177e4 | 1918 | |
e2970f2f | 1919 | ret = get_user(uval, uaddr); |
e4dc5b7a | 1920 | if (ret) |
f801073f | 1921 | goto out; |
1da177e4 | 1922 | |
b41277dc | 1923 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1924 | goto retry_private; |
1925 | ||
ae791a2d | 1926 | put_futex_key(&q->key); |
e4dc5b7a | 1927 | goto retry; |
1da177e4 | 1928 | } |
ca5f9524 | 1929 | |
f801073f DH |
1930 | if (uval != val) { |
1931 | queue_unlock(q, *hb); | |
1932 | ret = -EWOULDBLOCK; | |
2fff78c7 | 1933 | } |
1da177e4 | 1934 | |
f801073f DH |
1935 | out: |
1936 | if (ret) | |
ae791a2d | 1937 | put_futex_key(&q->key); |
f801073f DH |
1938 | return ret; |
1939 | } | |
1940 | ||
b41277dc DH |
1941 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
1942 | ktime_t *abs_time, u32 bitset) | |
f801073f DH |
1943 | { |
1944 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
1945 | struct restart_block *restart; |
1946 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 1947 | struct futex_q q = futex_q_init; |
f801073f DH |
1948 | int ret; |
1949 | ||
1950 | if (!bitset) | |
1951 | return -EINVAL; | |
f801073f DH |
1952 | q.bitset = bitset; |
1953 | ||
1954 | if (abs_time) { | |
1955 | to = &timeout; | |
1956 | ||
b41277dc DH |
1957 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
1958 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
1959 | HRTIMER_MODE_ABS); | |
f801073f DH |
1960 | hrtimer_init_sleeper(to, current); |
1961 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
1962 | current->timer_slack_ns); | |
1963 | } | |
1964 | ||
d58e6576 | 1965 | retry: |
7ada876a DH |
1966 | /* |
1967 | * Prepare to wait on uaddr. On success, holds hb lock and increments | |
1968 | * q.key refs. | |
1969 | */ | |
b41277dc | 1970 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
1971 | if (ret) |
1972 | goto out; | |
1973 | ||
ca5f9524 | 1974 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 1975 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
1976 | |
1977 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 1978 | ret = 0; |
7ada876a | 1979 | /* unqueue_me() drops q.key ref */ |
1da177e4 | 1980 | if (!unqueue_me(&q)) |
7ada876a | 1981 | goto out; |
2fff78c7 | 1982 | ret = -ETIMEDOUT; |
ca5f9524 | 1983 | if (to && !to->task) |
7ada876a | 1984 | goto out; |
72c1bbf3 | 1985 | |
e2970f2f | 1986 | /* |
d58e6576 TG |
1987 | * We expect signal_pending(current), but we might be the |
1988 | * victim of a spurious wakeup as well. | |
e2970f2f | 1989 | */ |
7ada876a | 1990 | if (!signal_pending(current)) |
d58e6576 | 1991 | goto retry; |
d58e6576 | 1992 | |
2fff78c7 | 1993 | ret = -ERESTARTSYS; |
c19384b5 | 1994 | if (!abs_time) |
7ada876a | 1995 | goto out; |
1da177e4 | 1996 | |
2fff78c7 PZ |
1997 | restart = ¤t_thread_info()->restart_block; |
1998 | restart->fn = futex_wait_restart; | |
a3c74c52 | 1999 | restart->futex.uaddr = uaddr; |
2fff78c7 PZ |
2000 | restart->futex.val = val; |
2001 | restart->futex.time = abs_time->tv64; | |
2002 | restart->futex.bitset = bitset; | |
0cd9c649 | 2003 | restart->futex.flags = flags | FLAGS_HAS_TIMEOUT; |
42d35d48 | 2004 | |
2fff78c7 PZ |
2005 | ret = -ERESTART_RESTARTBLOCK; |
2006 | ||
42d35d48 | 2007 | out: |
ca5f9524 DH |
2008 | if (to) { |
2009 | hrtimer_cancel(&to->timer); | |
2010 | destroy_hrtimer_on_stack(&to->timer); | |
2011 | } | |
c87e2837 IM |
2012 | return ret; |
2013 | } | |
2014 | ||
72c1bbf3 NP |
2015 | |
2016 | static long futex_wait_restart(struct restart_block *restart) | |
2017 | { | |
a3c74c52 | 2018 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 2019 | ktime_t t, *tp = NULL; |
72c1bbf3 | 2020 | |
a72188d8 DH |
2021 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
2022 | t.tv64 = restart->futex.time; | |
2023 | tp = &t; | |
2024 | } | |
72c1bbf3 | 2025 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
2026 | |
2027 | return (long)futex_wait(uaddr, restart->futex.flags, | |
2028 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
2029 | } |
2030 | ||
2031 | ||
c87e2837 IM |
2032 | /* |
2033 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
2034 | * and failed. The kernel side here does the whole locking operation: | |
2035 | * if there are waiters then it will block, it does PI, etc. (Due to | |
2036 | * races the kernel might see a 0 value of the futex too.) | |
2037 | */ | |
b41277dc DH |
2038 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect, |
2039 | ktime_t *time, int trylock) | |
c87e2837 | 2040 | { |
c5780e97 | 2041 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 2042 | struct futex_hash_bucket *hb; |
5bdb05f9 | 2043 | struct futex_q q = futex_q_init; |
dd973998 | 2044 | int res, ret; |
c87e2837 IM |
2045 | |
2046 | if (refill_pi_state_cache()) | |
2047 | return -ENOMEM; | |
2048 | ||
c19384b5 | 2049 | if (time) { |
c5780e97 | 2050 | to = &timeout; |
237fc6e7 TG |
2051 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
2052 | HRTIMER_MODE_ABS); | |
c5780e97 | 2053 | hrtimer_init_sleeper(to, current); |
cc584b21 | 2054 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
2055 | } |
2056 | ||
42d35d48 | 2057 | retry: |
9ea71503 | 2058 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE); |
c87e2837 | 2059 | if (unlikely(ret != 0)) |
42d35d48 | 2060 | goto out; |
c87e2837 | 2061 | |
e4dc5b7a | 2062 | retry_private: |
82af7aca | 2063 | hb = queue_lock(&q); |
c87e2837 | 2064 | |
bab5bc9e | 2065 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 2066 | if (unlikely(ret)) { |
778e9a9c | 2067 | switch (ret) { |
1a52084d DH |
2068 | case 1: |
2069 | /* We got the lock. */ | |
2070 | ret = 0; | |
2071 | goto out_unlock_put_key; | |
2072 | case -EFAULT: | |
2073 | goto uaddr_faulted; | |
778e9a9c AK |
2074 | case -EAGAIN: |
2075 | /* | |
2076 | * Task is exiting and we just wait for the | |
2077 | * exit to complete. | |
2078 | */ | |
2079 | queue_unlock(&q, hb); | |
ae791a2d | 2080 | put_futex_key(&q.key); |
778e9a9c AK |
2081 | cond_resched(); |
2082 | goto retry; | |
778e9a9c | 2083 | default: |
42d35d48 | 2084 | goto out_unlock_put_key; |
c87e2837 | 2085 | } |
c87e2837 IM |
2086 | } |
2087 | ||
2088 | /* | |
2089 | * Only actually queue now that the atomic ops are done: | |
2090 | */ | |
82af7aca | 2091 | queue_me(&q, hb); |
c87e2837 | 2092 | |
c87e2837 IM |
2093 | WARN_ON(!q.pi_state); |
2094 | /* | |
2095 | * Block on the PI mutex: | |
2096 | */ | |
2097 | if (!trylock) | |
2098 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); | |
2099 | else { | |
2100 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); | |
2101 | /* Fixup the trylock return value: */ | |
2102 | ret = ret ? 0 : -EWOULDBLOCK; | |
2103 | } | |
2104 | ||
a99e4e41 | 2105 | spin_lock(q.lock_ptr); |
dd973998 DH |
2106 | /* |
2107 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2108 | * haven't already. | |
2109 | */ | |
ae791a2d | 2110 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 DH |
2111 | /* |
2112 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
2113 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
2114 | */ | |
2115 | if (res) | |
2116 | ret = (res < 0) ? res : 0; | |
c87e2837 | 2117 | |
e8f6386c | 2118 | /* |
dd973998 DH |
2119 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
2120 | * it and return the fault to userspace. | |
e8f6386c DH |
2121 | */ |
2122 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
2123 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
2124 | ||
778e9a9c AK |
2125 | /* Unqueue and drop the lock */ |
2126 | unqueue_me_pi(&q); | |
c87e2837 | 2127 | |
5ecb01cf | 2128 | goto out_put_key; |
c87e2837 | 2129 | |
42d35d48 | 2130 | out_unlock_put_key: |
c87e2837 IM |
2131 | queue_unlock(&q, hb); |
2132 | ||
42d35d48 | 2133 | out_put_key: |
ae791a2d | 2134 | put_futex_key(&q.key); |
42d35d48 | 2135 | out: |
237fc6e7 TG |
2136 | if (to) |
2137 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 2138 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 2139 | |
42d35d48 | 2140 | uaddr_faulted: |
778e9a9c AK |
2141 | queue_unlock(&q, hb); |
2142 | ||
d0725992 | 2143 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
2144 | if (ret) |
2145 | goto out_put_key; | |
c87e2837 | 2146 | |
b41277dc | 2147 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2148 | goto retry_private; |
2149 | ||
ae791a2d | 2150 | put_futex_key(&q.key); |
e4dc5b7a | 2151 | goto retry; |
c87e2837 IM |
2152 | } |
2153 | ||
c87e2837 IM |
2154 | /* |
2155 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
2156 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
2157 | * and do the rt-mutex unlock. | |
2158 | */ | |
b41277dc | 2159 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 IM |
2160 | { |
2161 | struct futex_hash_bucket *hb; | |
2162 | struct futex_q *this, *next; | |
ec92d082 | 2163 | struct plist_head *head; |
38d47c1b | 2164 | union futex_key key = FUTEX_KEY_INIT; |
c0c9ed15 | 2165 | u32 uval, vpid = task_pid_vnr(current); |
e4dc5b7a | 2166 | int ret; |
c87e2837 IM |
2167 | |
2168 | retry: | |
2169 | if (get_user(uval, uaddr)) | |
2170 | return -EFAULT; | |
2171 | /* | |
2172 | * We release only a lock we actually own: | |
2173 | */ | |
c0c9ed15 | 2174 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 2175 | return -EPERM; |
c87e2837 | 2176 | |
9ea71503 | 2177 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE); |
c87e2837 IM |
2178 | if (unlikely(ret != 0)) |
2179 | goto out; | |
2180 | ||
2181 | hb = hash_futex(&key); | |
2182 | spin_lock(&hb->lock); | |
2183 | ||
c87e2837 IM |
2184 | /* |
2185 | * To avoid races, try to do the TID -> 0 atomic transition | |
2186 | * again. If it succeeds then we can return without waking | |
9ad5dabd TG |
2187 | * anyone else up. We only try this if neither the waiters nor |
2188 | * the owner died bit are set. | |
c87e2837 | 2189 | */ |
9ad5dabd | 2190 | if (!(uval & ~FUTEX_TID_MASK) && |
37a9d912 | 2191 | cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0)) |
c87e2837 IM |
2192 | goto pi_faulted; |
2193 | /* | |
2194 | * Rare case: we managed to release the lock atomically, | |
2195 | * no need to wake anyone else up: | |
2196 | */ | |
c0c9ed15 | 2197 | if (unlikely(uval == vpid)) |
c87e2837 IM |
2198 | goto out_unlock; |
2199 | ||
2200 | /* | |
2201 | * Ok, other tasks may need to be woken up - check waiters | |
2202 | * and do the wakeup if necessary: | |
2203 | */ | |
2204 | head = &hb->chain; | |
2205 | ||
ec92d082 | 2206 | plist_for_each_entry_safe(this, next, head, list) { |
c87e2837 IM |
2207 | if (!match_futex (&this->key, &key)) |
2208 | continue; | |
2209 | ret = wake_futex_pi(uaddr, uval, this); | |
2210 | /* | |
2211 | * The atomic access to the futex value | |
2212 | * generated a pagefault, so retry the | |
2213 | * user-access and the wakeup: | |
2214 | */ | |
2215 | if (ret == -EFAULT) | |
2216 | goto pi_faulted; | |
2217 | goto out_unlock; | |
2218 | } | |
2219 | /* | |
2220 | * No waiters - kernel unlocks the futex: | |
2221 | */ | |
9ad5dabd TG |
2222 | ret = unlock_futex_pi(uaddr, uval); |
2223 | if (ret == -EFAULT) | |
2224 | goto pi_faulted; | |
c87e2837 IM |
2225 | |
2226 | out_unlock: | |
2227 | spin_unlock(&hb->lock); | |
ae791a2d | 2228 | put_futex_key(&key); |
c87e2837 | 2229 | |
42d35d48 | 2230 | out: |
c87e2837 IM |
2231 | return ret; |
2232 | ||
2233 | pi_faulted: | |
778e9a9c | 2234 | spin_unlock(&hb->lock); |
ae791a2d | 2235 | put_futex_key(&key); |
c87e2837 | 2236 | |
d0725992 | 2237 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2238 | if (!ret) |
c87e2837 IM |
2239 | goto retry; |
2240 | ||
1da177e4 LT |
2241 | return ret; |
2242 | } | |
2243 | ||
52400ba9 DH |
2244 | /** |
2245 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2246 | * @hb: the hash_bucket futex_q was original enqueued on | |
2247 | * @q: the futex_q woken while waiting to be requeued | |
2248 | * @key2: the futex_key of the requeue target futex | |
2249 | * @timeout: the timeout associated with the wait (NULL if none) | |
2250 | * | |
2251 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2252 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2253 | * the wakeup and return the appropriate error code to the caller. Must be | |
2254 | * called with the hb lock held. | |
2255 | * | |
6c23cbbd RD |
2256 | * Return: |
2257 | * 0 = no early wakeup detected; | |
2258 | * <0 = -ETIMEDOUT or -ERESTARTNOINTR | |
52400ba9 DH |
2259 | */ |
2260 | static inline | |
2261 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2262 | struct futex_q *q, union futex_key *key2, | |
2263 | struct hrtimer_sleeper *timeout) | |
2264 | { | |
2265 | int ret = 0; | |
2266 | ||
2267 | /* | |
2268 | * With the hb lock held, we avoid races while we process the wakeup. | |
2269 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2270 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2271 | * It can't be requeued from uaddr2 to something else since we don't | |
2272 | * support a PI aware source futex for requeue. | |
2273 | */ | |
2274 | if (!match_futex(&q->key, key2)) { | |
2275 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2276 | /* | |
2277 | * We were woken prior to requeue by a timeout or a signal. | |
2278 | * Unqueue the futex_q and determine which it was. | |
2279 | */ | |
2e12978a | 2280 | plist_del(&q->list, &hb->chain); |
52400ba9 | 2281 | |
d58e6576 | 2282 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 2283 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2284 | if (timeout && !timeout->task) |
2285 | ret = -ETIMEDOUT; | |
d58e6576 | 2286 | else if (signal_pending(current)) |
1c840c14 | 2287 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
2288 | } |
2289 | return ret; | |
2290 | } | |
2291 | ||
2292 | /** | |
2293 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 2294 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 2295 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
52400ba9 DH |
2296 | * the same type, no requeueing from private to shared, etc. |
2297 | * @val: the expected value of uaddr | |
2298 | * @abs_time: absolute timeout | |
56ec1607 | 2299 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
2300 | * @uaddr2: the pi futex we will take prior to returning to user-space |
2301 | * | |
2302 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
6f7b0a2a DH |
2303 | * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake |
2304 | * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to | |
2305 | * userspace. This ensures the rt_mutex maintains an owner when it has waiters; | |
2306 | * without one, the pi logic would not know which task to boost/deboost, if | |
2307 | * there was a need to. | |
52400ba9 DH |
2308 | * |
2309 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
6c23cbbd | 2310 | * via the following-- |
52400ba9 | 2311 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() |
cc6db4e6 DH |
2312 | * 2) wakeup on uaddr2 after a requeue |
2313 | * 3) signal | |
2314 | * 4) timeout | |
52400ba9 | 2315 | * |
cc6db4e6 | 2316 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
2317 | * |
2318 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2319 | * 5) successful lock | |
2320 | * 6) signal | |
2321 | * 7) timeout | |
2322 | * 8) other lock acquisition failure | |
2323 | * | |
cc6db4e6 | 2324 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
2325 | * |
2326 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2327 | * | |
6c23cbbd RD |
2328 | * Return: |
2329 | * 0 - On success; | |
52400ba9 DH |
2330 | * <0 - On error |
2331 | */ | |
b41277dc | 2332 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 2333 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 2334 | u32 __user *uaddr2) |
52400ba9 DH |
2335 | { |
2336 | struct hrtimer_sleeper timeout, *to = NULL; | |
2337 | struct rt_mutex_waiter rt_waiter; | |
2338 | struct rt_mutex *pi_mutex = NULL; | |
52400ba9 | 2339 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
2340 | union futex_key key2 = FUTEX_KEY_INIT; |
2341 | struct futex_q q = futex_q_init; | |
52400ba9 | 2342 | int res, ret; |
52400ba9 | 2343 | |
6f7b0a2a DH |
2344 | if (uaddr == uaddr2) |
2345 | return -EINVAL; | |
2346 | ||
52400ba9 DH |
2347 | if (!bitset) |
2348 | return -EINVAL; | |
2349 | ||
2350 | if (abs_time) { | |
2351 | to = &timeout; | |
b41277dc DH |
2352 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2353 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2354 | HRTIMER_MODE_ABS); | |
52400ba9 DH |
2355 | hrtimer_init_sleeper(to, current); |
2356 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2357 | current->timer_slack_ns); | |
2358 | } | |
2359 | ||
2360 | /* | |
2361 | * The waiter is allocated on our stack, manipulated by the requeue | |
2362 | * code while we sleep on uaddr. | |
2363 | */ | |
2364 | debug_rt_mutex_init_waiter(&rt_waiter); | |
2365 | rt_waiter.task = NULL; | |
2366 | ||
9ea71503 | 2367 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
52400ba9 DH |
2368 | if (unlikely(ret != 0)) |
2369 | goto out; | |
2370 | ||
84bc4af5 DH |
2371 | q.bitset = bitset; |
2372 | q.rt_waiter = &rt_waiter; | |
2373 | q.requeue_pi_key = &key2; | |
2374 | ||
7ada876a DH |
2375 | /* |
2376 | * Prepare to wait on uaddr. On success, increments q.key (key1) ref | |
2377 | * count. | |
2378 | */ | |
b41277dc | 2379 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 TG |
2380 | if (ret) |
2381 | goto out_key2; | |
52400ba9 | 2382 | |
b58623fb TG |
2383 | /* |
2384 | * The check above which compares uaddrs is not sufficient for | |
2385 | * shared futexes. We need to compare the keys: | |
2386 | */ | |
2387 | if (match_futex(&q.key, &key2)) { | |
2388 | ret = -EINVAL; | |
2389 | goto out_put_keys; | |
2390 | } | |
2391 | ||
52400ba9 | 2392 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ |
f1a11e05 | 2393 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2394 | |
2395 | spin_lock(&hb->lock); | |
2396 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2397 | spin_unlock(&hb->lock); | |
2398 | if (ret) | |
2399 | goto out_put_keys; | |
2400 | ||
2401 | /* | |
2402 | * In order for us to be here, we know our q.key == key2, and since | |
2403 | * we took the hb->lock above, we also know that futex_requeue() has | |
2404 | * completed and we no longer have to concern ourselves with a wakeup | |
7ada876a DH |
2405 | * race with the atomic proxy lock acquisition by the requeue code. The |
2406 | * futex_requeue dropped our key1 reference and incremented our key2 | |
2407 | * reference count. | |
52400ba9 DH |
2408 | */ |
2409 | ||
2410 | /* Check if the requeue code acquired the second futex for us. */ | |
2411 | if (!q.rt_waiter) { | |
2412 | /* | |
2413 | * Got the lock. We might not be the anticipated owner if we | |
2414 | * did a lock-steal - fix up the PI-state in that case. | |
2415 | */ | |
2416 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2417 | spin_lock(q.lock_ptr); | |
ae791a2d | 2418 | ret = fixup_pi_state_owner(uaddr2, &q, current); |
52400ba9 DH |
2419 | spin_unlock(q.lock_ptr); |
2420 | } | |
2421 | } else { | |
2422 | /* | |
2423 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2424 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2425 | * the pi_state. | |
2426 | */ | |
f27071cb | 2427 | WARN_ON(!q.pi_state); |
52400ba9 DH |
2428 | pi_mutex = &q.pi_state->pi_mutex; |
2429 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1); | |
2430 | debug_rt_mutex_free_waiter(&rt_waiter); | |
2431 | ||
2432 | spin_lock(q.lock_ptr); | |
2433 | /* | |
2434 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2435 | * haven't already. | |
2436 | */ | |
ae791a2d | 2437 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 DH |
2438 | /* |
2439 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 2440 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
2441 | */ |
2442 | if (res) | |
2443 | ret = (res < 0) ? res : 0; | |
2444 | ||
2445 | /* Unqueue and drop the lock. */ | |
2446 | unqueue_me_pi(&q); | |
2447 | } | |
2448 | ||
2449 | /* | |
2450 | * If fixup_pi_state_owner() faulted and was unable to handle the | |
2451 | * fault, unlock the rt_mutex and return the fault to userspace. | |
2452 | */ | |
2453 | if (ret == -EFAULT) { | |
b6070a8d | 2454 | if (pi_mutex && rt_mutex_owner(pi_mutex) == current) |
52400ba9 DH |
2455 | rt_mutex_unlock(pi_mutex); |
2456 | } else if (ret == -EINTR) { | |
52400ba9 | 2457 | /* |
cc6db4e6 DH |
2458 | * We've already been requeued, but cannot restart by calling |
2459 | * futex_lock_pi() directly. We could restart this syscall, but | |
2460 | * it would detect that the user space "val" changed and return | |
2461 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
2462 | * -EWOULDBLOCK directly. | |
52400ba9 | 2463 | */ |
2070887f | 2464 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2465 | } |
2466 | ||
2467 | out_put_keys: | |
ae791a2d | 2468 | put_futex_key(&q.key); |
c8b15a70 | 2469 | out_key2: |
ae791a2d | 2470 | put_futex_key(&key2); |
52400ba9 DH |
2471 | |
2472 | out: | |
2473 | if (to) { | |
2474 | hrtimer_cancel(&to->timer); | |
2475 | destroy_hrtimer_on_stack(&to->timer); | |
2476 | } | |
2477 | return ret; | |
2478 | } | |
2479 | ||
0771dfef IM |
2480 | /* |
2481 | * Support for robust futexes: the kernel cleans up held futexes at | |
2482 | * thread exit time. | |
2483 | * | |
2484 | * Implementation: user-space maintains a per-thread list of locks it | |
2485 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2486 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2487 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2488 | * always manipulated with the lock held, so the list is private and |
2489 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2490 | * field, to allow the kernel to clean up if the thread dies after | |
2491 | * acquiring the lock, but just before it could have added itself to | |
2492 | * the list. There can only be one such pending lock. | |
2493 | */ | |
2494 | ||
2495 | /** | |
d96ee56c DH |
2496 | * sys_set_robust_list() - Set the robust-futex list head of a task |
2497 | * @head: pointer to the list-head | |
2498 | * @len: length of the list-head, as userspace expects | |
0771dfef | 2499 | */ |
836f92ad HC |
2500 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2501 | size_t, len) | |
0771dfef | 2502 | { |
a0c1e907 TG |
2503 | if (!futex_cmpxchg_enabled) |
2504 | return -ENOSYS; | |
0771dfef IM |
2505 | /* |
2506 | * The kernel knows only one size for now: | |
2507 | */ | |
2508 | if (unlikely(len != sizeof(*head))) | |
2509 | return -EINVAL; | |
2510 | ||
2511 | current->robust_list = head; | |
2512 | ||
2513 | return 0; | |
2514 | } | |
2515 | ||
2516 | /** | |
d96ee56c DH |
2517 | * sys_get_robust_list() - Get the robust-futex list head of a task |
2518 | * @pid: pid of the process [zero for current task] | |
2519 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
2520 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 2521 | */ |
836f92ad HC |
2522 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
2523 | struct robust_list_head __user * __user *, head_ptr, | |
2524 | size_t __user *, len_ptr) | |
0771dfef | 2525 | { |
ba46df98 | 2526 | struct robust_list_head __user *head; |
0771dfef | 2527 | unsigned long ret; |
bdbb776f | 2528 | struct task_struct *p; |
0771dfef | 2529 | |
a0c1e907 TG |
2530 | if (!futex_cmpxchg_enabled) |
2531 | return -ENOSYS; | |
2532 | ||
bdbb776f KC |
2533 | rcu_read_lock(); |
2534 | ||
2535 | ret = -ESRCH; | |
0771dfef | 2536 | if (!pid) |
bdbb776f | 2537 | p = current; |
0771dfef | 2538 | else { |
228ebcbe | 2539 | p = find_task_by_vpid(pid); |
0771dfef IM |
2540 | if (!p) |
2541 | goto err_unlock; | |
0771dfef IM |
2542 | } |
2543 | ||
bdbb776f KC |
2544 | ret = -EPERM; |
2545 | if (!ptrace_may_access(p, PTRACE_MODE_READ)) | |
2546 | goto err_unlock; | |
2547 | ||
2548 | head = p->robust_list; | |
2549 | rcu_read_unlock(); | |
2550 | ||
0771dfef IM |
2551 | if (put_user(sizeof(*head), len_ptr)) |
2552 | return -EFAULT; | |
2553 | return put_user(head, head_ptr); | |
2554 | ||
2555 | err_unlock: | |
aaa2a97e | 2556 | rcu_read_unlock(); |
0771dfef IM |
2557 | |
2558 | return ret; | |
2559 | } | |
2560 | ||
2561 | /* | |
2562 | * Process a futex-list entry, check whether it's owned by the | |
2563 | * dying task, and do notification if so: | |
2564 | */ | |
e3f2ddea | 2565 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 2566 | { |
7cfdaf38 | 2567 | u32 uval, uninitialized_var(nval), mval; |
0771dfef | 2568 | |
8f17d3a5 IM |
2569 | retry: |
2570 | if (get_user(uval, uaddr)) | |
0771dfef IM |
2571 | return -1; |
2572 | ||
b488893a | 2573 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
2574 | /* |
2575 | * Ok, this dying thread is truly holding a futex | |
2576 | * of interest. Set the OWNER_DIED bit atomically | |
2577 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
2578 | * set, wake up a waiter (if any). (We have to do a | |
2579 | * futex_wake() even if OWNER_DIED is already set - | |
2580 | * to handle the rare but possible case of recursive | |
2581 | * thread-death.) The rest of the cleanup is done in | |
2582 | * userspace. | |
2583 | */ | |
e3f2ddea | 2584 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
6e0aa9f8 TG |
2585 | /* |
2586 | * We are not holding a lock here, but we want to have | |
2587 | * the pagefault_disable/enable() protection because | |
2588 | * we want to handle the fault gracefully. If the | |
2589 | * access fails we try to fault in the futex with R/W | |
2590 | * verification via get_user_pages. get_user() above | |
2591 | * does not guarantee R/W access. If that fails we | |
2592 | * give up and leave the futex locked. | |
2593 | */ | |
2594 | if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) { | |
2595 | if (fault_in_user_writeable(uaddr)) | |
2596 | return -1; | |
2597 | goto retry; | |
2598 | } | |
c87e2837 | 2599 | if (nval != uval) |
8f17d3a5 | 2600 | goto retry; |
0771dfef | 2601 | |
e3f2ddea IM |
2602 | /* |
2603 | * Wake robust non-PI futexes here. The wakeup of | |
2604 | * PI futexes happens in exit_pi_state(): | |
2605 | */ | |
36cf3b5c | 2606 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 2607 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
2608 | } |
2609 | return 0; | |
2610 | } | |
2611 | ||
e3f2ddea IM |
2612 | /* |
2613 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
2614 | */ | |
2615 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 2616 | struct robust_list __user * __user *head, |
1dcc41bb | 2617 | unsigned int *pi) |
e3f2ddea IM |
2618 | { |
2619 | unsigned long uentry; | |
2620 | ||
ba46df98 | 2621 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
2622 | return -EFAULT; |
2623 | ||
ba46df98 | 2624 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
2625 | *pi = uentry & 1; |
2626 | ||
2627 | return 0; | |
2628 | } | |
2629 | ||
0771dfef IM |
2630 | /* |
2631 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
2632 | * and mark any locks found there dead, and notify any waiters. | |
2633 | * | |
2634 | * We silently return on any sign of list-walking problem. | |
2635 | */ | |
2636 | void exit_robust_list(struct task_struct *curr) | |
2637 | { | |
2638 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 2639 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 DH |
2640 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
2641 | unsigned int uninitialized_var(next_pi); | |
0771dfef | 2642 | unsigned long futex_offset; |
9f96cb1e | 2643 | int rc; |
0771dfef | 2644 | |
a0c1e907 TG |
2645 | if (!futex_cmpxchg_enabled) |
2646 | return; | |
2647 | ||
0771dfef IM |
2648 | /* |
2649 | * Fetch the list head (which was registered earlier, via | |
2650 | * sys_set_robust_list()): | |
2651 | */ | |
e3f2ddea | 2652 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
2653 | return; |
2654 | /* | |
2655 | * Fetch the relative futex offset: | |
2656 | */ | |
2657 | if (get_user(futex_offset, &head->futex_offset)) | |
2658 | return; | |
2659 | /* | |
2660 | * Fetch any possibly pending lock-add first, and handle it | |
2661 | * if it exists: | |
2662 | */ | |
e3f2ddea | 2663 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 2664 | return; |
e3f2ddea | 2665 | |
9f96cb1e | 2666 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 2667 | while (entry != &head->list) { |
9f96cb1e MS |
2668 | /* |
2669 | * Fetch the next entry in the list before calling | |
2670 | * handle_futex_death: | |
2671 | */ | |
2672 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
2673 | /* |
2674 | * A pending lock might already be on the list, so | |
c87e2837 | 2675 | * don't process it twice: |
0771dfef IM |
2676 | */ |
2677 | if (entry != pending) | |
ba46df98 | 2678 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 2679 | curr, pi)) |
0771dfef | 2680 | return; |
9f96cb1e | 2681 | if (rc) |
0771dfef | 2682 | return; |
9f96cb1e MS |
2683 | entry = next_entry; |
2684 | pi = next_pi; | |
0771dfef IM |
2685 | /* |
2686 | * Avoid excessively long or circular lists: | |
2687 | */ | |
2688 | if (!--limit) | |
2689 | break; | |
2690 | ||
2691 | cond_resched(); | |
2692 | } | |
9f96cb1e MS |
2693 | |
2694 | if (pending) | |
2695 | handle_futex_death((void __user *)pending + futex_offset, | |
2696 | curr, pip); | |
0771dfef IM |
2697 | } |
2698 | ||
c19384b5 | 2699 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 2700 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 2701 | { |
81b40539 | 2702 | int cmd = op & FUTEX_CMD_MASK; |
b41277dc | 2703 | unsigned int flags = 0; |
34f01cc1 ED |
2704 | |
2705 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 2706 | flags |= FLAGS_SHARED; |
1da177e4 | 2707 | |
b41277dc DH |
2708 | if (op & FUTEX_CLOCK_REALTIME) { |
2709 | flags |= FLAGS_CLOCKRT; | |
2710 | if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI) | |
2711 | return -ENOSYS; | |
2712 | } | |
1da177e4 | 2713 | |
59263b51 TG |
2714 | switch (cmd) { |
2715 | case FUTEX_LOCK_PI: | |
2716 | case FUTEX_UNLOCK_PI: | |
2717 | case FUTEX_TRYLOCK_PI: | |
2718 | case FUTEX_WAIT_REQUEUE_PI: | |
2719 | case FUTEX_CMP_REQUEUE_PI: | |
2720 | if (!futex_cmpxchg_enabled) | |
2721 | return -ENOSYS; | |
2722 | } | |
2723 | ||
34f01cc1 | 2724 | switch (cmd) { |
1da177e4 | 2725 | case FUTEX_WAIT: |
cd689985 TG |
2726 | val3 = FUTEX_BITSET_MATCH_ANY; |
2727 | case FUTEX_WAIT_BITSET: | |
81b40539 | 2728 | return futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 | 2729 | case FUTEX_WAKE: |
cd689985 TG |
2730 | val3 = FUTEX_BITSET_MATCH_ANY; |
2731 | case FUTEX_WAKE_BITSET: | |
81b40539 | 2732 | return futex_wake(uaddr, flags, val, val3); |
1da177e4 | 2733 | case FUTEX_REQUEUE: |
81b40539 | 2734 | return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 | 2735 | case FUTEX_CMP_REQUEUE: |
81b40539 | 2736 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
4732efbe | 2737 | case FUTEX_WAKE_OP: |
81b40539 | 2738 | return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
c87e2837 | 2739 | case FUTEX_LOCK_PI: |
81b40539 | 2740 | return futex_lock_pi(uaddr, flags, val, timeout, 0); |
c87e2837 | 2741 | case FUTEX_UNLOCK_PI: |
81b40539 | 2742 | return futex_unlock_pi(uaddr, flags); |
c87e2837 | 2743 | case FUTEX_TRYLOCK_PI: |
81b40539 | 2744 | return futex_lock_pi(uaddr, flags, 0, timeout, 1); |
52400ba9 DH |
2745 | case FUTEX_WAIT_REQUEUE_PI: |
2746 | val3 = FUTEX_BITSET_MATCH_ANY; | |
81b40539 TG |
2747 | return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
2748 | uaddr2); | |
52400ba9 | 2749 | case FUTEX_CMP_REQUEUE_PI: |
81b40539 | 2750 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
1da177e4 | 2751 | } |
81b40539 | 2752 | return -ENOSYS; |
1da177e4 LT |
2753 | } |
2754 | ||
2755 | ||
17da2bd9 HC |
2756 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
2757 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
2758 | u32, val3) | |
1da177e4 | 2759 | { |
c19384b5 PP |
2760 | struct timespec ts; |
2761 | ktime_t t, *tp = NULL; | |
e2970f2f | 2762 | u32 val2 = 0; |
34f01cc1 | 2763 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 2764 | |
cd689985 | 2765 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
2766 | cmd == FUTEX_WAIT_BITSET || |
2767 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
c19384b5 | 2768 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 2769 | return -EFAULT; |
c19384b5 | 2770 | if (!timespec_valid(&ts)) |
9741ef96 | 2771 | return -EINVAL; |
c19384b5 PP |
2772 | |
2773 | t = timespec_to_ktime(ts); | |
34f01cc1 | 2774 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 2775 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 2776 | tp = &t; |
1da177e4 LT |
2777 | } |
2778 | /* | |
52400ba9 | 2779 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 2780 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 2781 | */ |
f54f0986 | 2782 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 2783 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 2784 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 2785 | |
c19384b5 | 2786 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
2787 | } |
2788 | ||
f26c70a4 | 2789 | static void __init futex_detect_cmpxchg(void) |
1da177e4 | 2790 | { |
f26c70a4 | 2791 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 2792 | u32 curval; |
95362fa9 | 2793 | |
a0c1e907 TG |
2794 | /* |
2795 | * This will fail and we want it. Some arch implementations do | |
2796 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
2797 | * functionality. We want to know that before we call in any | |
2798 | * of the complex code paths. Also we want to prevent | |
2799 | * registration of robust lists in that case. NULL is | |
2800 | * guaranteed to fault and we get -EFAULT on functional | |
fb62db2b | 2801 | * implementation, the non-functional ones will return |
a0c1e907 TG |
2802 | * -ENOSYS. |
2803 | */ | |
37a9d912 | 2804 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) |
a0c1e907 | 2805 | futex_cmpxchg_enabled = 1; |
f26c70a4 HC |
2806 | #endif |
2807 | } | |
2808 | ||
2809 | static int __init futex_init(void) | |
2810 | { | |
2811 | int i; | |
2812 | ||
2813 | futex_detect_cmpxchg(); | |
a0c1e907 | 2814 | |
3e4ab747 | 2815 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { |
732375c6 | 2816 | plist_head_init(&futex_queues[i].chain); |
3e4ab747 TG |
2817 | spin_lock_init(&futex_queues[i].lock); |
2818 | } | |
2819 | ||
1da177e4 LT |
2820 | return 0; |
2821 | } | |
f6d107fb | 2822 | __initcall(futex_init); |