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> |
6fa3eb70 | 65 | #include <linux/freezer.h> |
b488893a | 66 | |
4732efbe | 67 | #include <asm/futex.h> |
1da177e4 | 68 | |
c87e2837 IM |
69 | #include "rtmutex_common.h" |
70 | ||
f26c70a4 | 71 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 72 | int __read_mostly futex_cmpxchg_enabled; |
f26c70a4 | 73 | #endif |
a0c1e907 | 74 | |
1da177e4 LT |
75 | #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8) |
76 | ||
b41277dc DH |
77 | /* |
78 | * Futex flags used to encode options to functions and preserve them across | |
79 | * restarts. | |
80 | */ | |
81 | #define FLAGS_SHARED 0x01 | |
82 | #define FLAGS_CLOCKRT 0x02 | |
83 | #define FLAGS_HAS_TIMEOUT 0x04 | |
84 | ||
c87e2837 IM |
85 | /* |
86 | * Priority Inheritance state: | |
87 | */ | |
88 | struct futex_pi_state { | |
89 | /* | |
90 | * list of 'owned' pi_state instances - these have to be | |
91 | * cleaned up in do_exit() if the task exits prematurely: | |
92 | */ | |
93 | struct list_head list; | |
94 | ||
95 | /* | |
96 | * The PI object: | |
97 | */ | |
98 | struct rt_mutex pi_mutex; | |
99 | ||
100 | struct task_struct *owner; | |
101 | atomic_t refcount; | |
102 | ||
103 | union futex_key key; | |
104 | }; | |
105 | ||
d8d88fbb DH |
106 | /** |
107 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 108 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
109 | * @task: the task waiting on the futex |
110 | * @lock_ptr: the hash bucket lock | |
111 | * @key: the key the futex is hashed on | |
112 | * @pi_state: optional priority inheritance state | |
113 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
114 | * @requeue_pi_key: the requeue_pi target futex key | |
115 | * @bitset: bitset for the optional bitmasked wakeup | |
116 | * | |
117 | * We use this hashed waitqueue, instead of a normal wait_queue_t, so | |
1da177e4 LT |
118 | * we can wake only the relevant ones (hashed queues may be shared). |
119 | * | |
120 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 121 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 122 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
123 | * the second. |
124 | * | |
125 | * PI futexes are typically woken before they are removed from the hash list via | |
126 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
127 | */ |
128 | struct futex_q { | |
ec92d082 | 129 | struct plist_node list; |
1da177e4 | 130 | |
d8d88fbb | 131 | struct task_struct *task; |
1da177e4 | 132 | spinlock_t *lock_ptr; |
1da177e4 | 133 | union futex_key key; |
c87e2837 | 134 | struct futex_pi_state *pi_state; |
52400ba9 | 135 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 136 | union futex_key *requeue_pi_key; |
cd689985 | 137 | u32 bitset; |
1da177e4 LT |
138 | }; |
139 | ||
5bdb05f9 DH |
140 | static const struct futex_q futex_q_init = { |
141 | /* list gets initialized in queue_me()*/ | |
142 | .key = FUTEX_KEY_INIT, | |
143 | .bitset = FUTEX_BITSET_MATCH_ANY | |
144 | }; | |
145 | ||
1da177e4 | 146 | /* |
b2d0994b DH |
147 | * Hash buckets are shared by all the futex_keys that hash to the same |
148 | * location. Each key may have multiple futex_q structures, one for each task | |
149 | * waiting on a futex. | |
1da177e4 LT |
150 | */ |
151 | struct futex_hash_bucket { | |
ec92d082 PP |
152 | spinlock_t lock; |
153 | struct plist_head chain; | |
1da177e4 LT |
154 | }; |
155 | ||
156 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; | |
157 | ||
1da177e4 LT |
158 | /* |
159 | * We hash on the keys returned from get_futex_key (see below). | |
160 | */ | |
161 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
162 | { | |
163 | u32 hash = jhash2((u32*)&key->both.word, | |
164 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
165 | key->both.offset); | |
166 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; | |
167 | } | |
168 | ||
169 | /* | |
170 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
171 | */ | |
172 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
173 | { | |
2bc87203 DH |
174 | return (key1 && key2 |
175 | && key1->both.word == key2->both.word | |
1da177e4 LT |
176 | && key1->both.ptr == key2->both.ptr |
177 | && key1->both.offset == key2->both.offset); | |
178 | } | |
179 | ||
38d47c1b PZ |
180 | /* |
181 | * Take a reference to the resource addressed by a key. | |
182 | * Can be called while holding spinlocks. | |
183 | * | |
184 | */ | |
185 | static void get_futex_key_refs(union futex_key *key) | |
186 | { | |
187 | if (!key->both.ptr) | |
188 | return; | |
189 | ||
190 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
191 | case FUT_OFF_INODE: | |
7de9c6ee | 192 | ihold(key->shared.inode); |
38d47c1b PZ |
193 | break; |
194 | case FUT_OFF_MMSHARED: | |
195 | atomic_inc(&key->private.mm->mm_count); | |
196 | break; | |
197 | } | |
198 | } | |
199 | ||
200 | /* | |
201 | * Drop a reference to the resource addressed by a key. | |
202 | * The hash bucket spinlock must not be held. | |
203 | */ | |
204 | static void drop_futex_key_refs(union futex_key *key) | |
205 | { | |
90621c40 DH |
206 | if (!key->both.ptr) { |
207 | /* If we're here then we tried to put a key we failed to get */ | |
208 | WARN_ON_ONCE(1); | |
38d47c1b | 209 | return; |
90621c40 | 210 | } |
38d47c1b PZ |
211 | |
212 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
213 | case FUT_OFF_INODE: | |
214 | iput(key->shared.inode); | |
215 | break; | |
216 | case FUT_OFF_MMSHARED: | |
217 | mmdrop(key->private.mm); | |
218 | break; | |
219 | } | |
220 | } | |
221 | ||
34f01cc1 | 222 | /** |
d96ee56c DH |
223 | * get_futex_key() - Get parameters which are the keys for a futex |
224 | * @uaddr: virtual address of the futex | |
225 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED | |
226 | * @key: address where result is stored. | |
9ea71503 SB |
227 | * @rw: mapping needs to be read/write (values: VERIFY_READ, |
228 | * VERIFY_WRITE) | |
34f01cc1 | 229 | * |
6c23cbbd RD |
230 | * Return: a negative error code or 0 |
231 | * | |
34f01cc1 | 232 | * The key words are stored in *key on success. |
1da177e4 | 233 | * |
6131ffaa | 234 | * For shared mappings, it's (page->index, file_inode(vma->vm_file), |
1da177e4 LT |
235 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
236 | * We can usually work out the index without swapping in the page. | |
237 | * | |
b2d0994b | 238 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 239 | */ |
64d1304a | 240 | static int |
9ea71503 | 241 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) |
1da177e4 | 242 | { |
e2970f2f | 243 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 244 | struct mm_struct *mm = current->mm; |
a5b338f2 | 245 | struct page *page, *page_head; |
9ea71503 | 246 | int err, ro = 0; |
1da177e4 LT |
247 | |
248 | /* | |
249 | * The futex address must be "naturally" aligned. | |
250 | */ | |
e2970f2f | 251 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 252 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 253 | return -EINVAL; |
e2970f2f | 254 | address -= key->both.offset; |
1da177e4 | 255 | |
34f01cc1 ED |
256 | /* |
257 | * PROCESS_PRIVATE futexes are fast. | |
258 | * As the mm cannot disappear under us and the 'key' only needs | |
259 | * virtual address, we dont even have to find the underlying vma. | |
260 | * Note : We do have to check 'uaddr' is a valid user address, | |
261 | * but access_ok() should be faster than find_vma() | |
262 | */ | |
263 | if (!fshared) { | |
7485d0d3 | 264 | if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32)))) |
34f01cc1 ED |
265 | return -EFAULT; |
266 | key->private.mm = mm; | |
267 | key->private.address = address; | |
42569c39 | 268 | get_futex_key_refs(key); |
34f01cc1 ED |
269 | return 0; |
270 | } | |
1da177e4 | 271 | |
38d47c1b | 272 | again: |
7485d0d3 | 273 | err = get_user_pages_fast(address, 1, 1, &page); |
9ea71503 SB |
274 | /* |
275 | * If write access is not required (eg. FUTEX_WAIT), try | |
276 | * and get read-only access. | |
277 | */ | |
278 | if (err == -EFAULT && rw == VERIFY_READ) { | |
279 | err = get_user_pages_fast(address, 1, 0, &page); | |
280 | ro = 1; | |
281 | } | |
38d47c1b PZ |
282 | if (err < 0) |
283 | return err; | |
9ea71503 SB |
284 | else |
285 | err = 0; | |
38d47c1b | 286 | |
a5b338f2 AA |
287 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
288 | page_head = page; | |
289 | if (unlikely(PageTail(page))) { | |
38d47c1b | 290 | put_page(page); |
a5b338f2 AA |
291 | /* serialize against __split_huge_page_splitting() */ |
292 | local_irq_disable(); | |
13bb709c | 293 | if (likely(__get_user_pages_fast(address, 1, !ro, &page) == 1)) { |
a5b338f2 AA |
294 | page_head = compound_head(page); |
295 | /* | |
296 | * page_head is valid pointer but we must pin | |
297 | * it before taking the PG_lock and/or | |
298 | * PG_compound_lock. The moment we re-enable | |
299 | * irqs __split_huge_page_splitting() can | |
300 | * return and the head page can be freed from | |
301 | * under us. We can't take the PG_lock and/or | |
302 | * PG_compound_lock on a page that could be | |
303 | * freed from under us. | |
304 | */ | |
305 | if (page != page_head) { | |
306 | get_page(page_head); | |
307 | put_page(page); | |
308 | } | |
309 | local_irq_enable(); | |
310 | } else { | |
311 | local_irq_enable(); | |
312 | goto again; | |
313 | } | |
314 | } | |
315 | #else | |
316 | page_head = compound_head(page); | |
317 | if (page != page_head) { | |
318 | get_page(page_head); | |
319 | put_page(page); | |
320 | } | |
321 | #endif | |
322 | ||
323 | lock_page(page_head); | |
e6780f72 HD |
324 | |
325 | /* | |
326 | * If page_head->mapping is NULL, then it cannot be a PageAnon | |
327 | * page; but it might be the ZERO_PAGE or in the gate area or | |
328 | * in a special mapping (all cases which we are happy to fail); | |
329 | * or it may have been a good file page when get_user_pages_fast | |
330 | * found it, but truncated or holepunched or subjected to | |
331 | * invalidate_complete_page2 before we got the page lock (also | |
332 | * cases which we are happy to fail). And we hold a reference, | |
333 | * so refcount care in invalidate_complete_page's remove_mapping | |
334 | * prevents drop_caches from setting mapping to NULL beneath us. | |
335 | * | |
336 | * The case we do have to guard against is when memory pressure made | |
337 | * shmem_writepage move it from filecache to swapcache beneath us: | |
338 | * an unlikely race, but we do need to retry for page_head->mapping. | |
339 | */ | |
a5b338f2 | 340 | if (!page_head->mapping) { |
e6780f72 | 341 | int shmem_swizzled = PageSwapCache(page_head); |
a5b338f2 AA |
342 | unlock_page(page_head); |
343 | put_page(page_head); | |
e6780f72 HD |
344 | if (shmem_swizzled) |
345 | goto again; | |
346 | return -EFAULT; | |
38d47c1b | 347 | } |
1da177e4 LT |
348 | |
349 | /* | |
350 | * Private mappings are handled in a simple way. | |
351 | * | |
352 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
353 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 354 | * the object not the particular process. |
1da177e4 | 355 | */ |
a5b338f2 | 356 | if (PageAnon(page_head)) { |
9ea71503 SB |
357 | /* |
358 | * A RO anonymous page will never change and thus doesn't make | |
359 | * sense for futex operations. | |
360 | */ | |
361 | if (ro) { | |
362 | err = -EFAULT; | |
363 | goto out; | |
364 | } | |
365 | ||
38d47c1b | 366 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 367 | key->private.mm = mm; |
e2970f2f | 368 | key->private.address = address; |
38d47c1b PZ |
369 | } else { |
370 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ | |
a5b338f2 | 371 | key->shared.inode = page_head->mapping->host; |
ab1842f1 | 372 | key->shared.pgoff = basepage_index(page); |
1da177e4 LT |
373 | } |
374 | ||
38d47c1b | 375 | get_futex_key_refs(key); |
1da177e4 | 376 | |
9ea71503 | 377 | out: |
a5b338f2 AA |
378 | unlock_page(page_head); |
379 | put_page(page_head); | |
9ea71503 | 380 | return err; |
1da177e4 LT |
381 | } |
382 | ||
ae791a2d | 383 | static inline void put_futex_key(union futex_key *key) |
1da177e4 | 384 | { |
38d47c1b | 385 | drop_futex_key_refs(key); |
1da177e4 LT |
386 | } |
387 | ||
d96ee56c DH |
388 | /** |
389 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
390 | * @uaddr: pointer to faulting user space address |
391 | * | |
392 | * Slow path to fixup the fault we just took in the atomic write | |
393 | * access to @uaddr. | |
394 | * | |
fb62db2b | 395 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
396 | * user address. We know that we faulted in the atomic pagefault |
397 | * disabled section so we can as well avoid the #PF overhead by | |
398 | * calling get_user_pages() right away. | |
399 | */ | |
400 | static int fault_in_user_writeable(u32 __user *uaddr) | |
401 | { | |
722d0172 AK |
402 | struct mm_struct *mm = current->mm; |
403 | int ret; | |
404 | ||
405 | down_read(&mm->mmap_sem); | |
2efaca92 BH |
406 | ret = fixup_user_fault(current, mm, (unsigned long)uaddr, |
407 | FAULT_FLAG_WRITE); | |
722d0172 AK |
408 | up_read(&mm->mmap_sem); |
409 | ||
d0725992 TG |
410 | return ret < 0 ? ret : 0; |
411 | } | |
412 | ||
4b1c486b DH |
413 | /** |
414 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
415 | * @hb: the hash bucket the futex_q's reside in |
416 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
417 | * |
418 | * Must be called with the hb lock held. | |
419 | */ | |
420 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
421 | union futex_key *key) | |
422 | { | |
423 | struct futex_q *this; | |
424 | ||
425 | plist_for_each_entry(this, &hb->chain, list) { | |
426 | if (match_futex(&this->key, key)) | |
427 | return this; | |
428 | } | |
429 | return NULL; | |
430 | } | |
431 | ||
37a9d912 ML |
432 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
433 | u32 uval, u32 newval) | |
36cf3b5c | 434 | { |
37a9d912 | 435 | int ret; |
36cf3b5c TG |
436 | |
437 | pagefault_disable(); | |
37a9d912 | 438 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
439 | pagefault_enable(); |
440 | ||
37a9d912 | 441 | return ret; |
36cf3b5c TG |
442 | } |
443 | ||
444 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
445 | { |
446 | int ret; | |
447 | ||
a866374a | 448 | pagefault_disable(); |
e2970f2f | 449 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 450 | pagefault_enable(); |
1da177e4 LT |
451 | |
452 | return ret ? -EFAULT : 0; | |
453 | } | |
454 | ||
c87e2837 IM |
455 | |
456 | /* | |
457 | * PI code: | |
458 | */ | |
459 | static int refill_pi_state_cache(void) | |
460 | { | |
461 | struct futex_pi_state *pi_state; | |
462 | ||
463 | if (likely(current->pi_state_cache)) | |
464 | return 0; | |
465 | ||
4668edc3 | 466 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
467 | |
468 | if (!pi_state) | |
469 | return -ENOMEM; | |
470 | ||
c87e2837 IM |
471 | INIT_LIST_HEAD(&pi_state->list); |
472 | /* pi_mutex gets initialized later */ | |
473 | pi_state->owner = NULL; | |
474 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 475 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
476 | |
477 | current->pi_state_cache = pi_state; | |
478 | ||
479 | return 0; | |
480 | } | |
481 | ||
482 | static struct futex_pi_state * alloc_pi_state(void) | |
483 | { | |
484 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
485 | ||
486 | WARN_ON(!pi_state); | |
487 | current->pi_state_cache = NULL; | |
488 | ||
489 | return pi_state; | |
490 | } | |
491 | ||
492 | static void free_pi_state(struct futex_pi_state *pi_state) | |
493 | { | |
494 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
495 | return; | |
496 | ||
497 | /* | |
498 | * If pi_state->owner is NULL, the owner is most probably dying | |
499 | * and has cleaned up the pi_state already | |
500 | */ | |
501 | if (pi_state->owner) { | |
1d615482 | 502 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
c87e2837 | 503 | list_del_init(&pi_state->list); |
1d615482 | 504 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
c87e2837 IM |
505 | |
506 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
507 | } | |
508 | ||
509 | if (current->pi_state_cache) | |
510 | kfree(pi_state); | |
511 | else { | |
512 | /* | |
513 | * pi_state->list is already empty. | |
514 | * clear pi_state->owner. | |
515 | * refcount is at 0 - put it back to 1. | |
516 | */ | |
517 | pi_state->owner = NULL; | |
518 | atomic_set(&pi_state->refcount, 1); | |
519 | current->pi_state_cache = pi_state; | |
520 | } | |
521 | } | |
522 | ||
523 | /* | |
524 | * Look up the task based on what TID userspace gave us. | |
525 | * We dont trust it. | |
526 | */ | |
527 | static struct task_struct * futex_find_get_task(pid_t pid) | |
528 | { | |
529 | struct task_struct *p; | |
530 | ||
d359b549 | 531 | rcu_read_lock(); |
228ebcbe | 532 | p = find_task_by_vpid(pid); |
7a0ea09a MH |
533 | if (p) |
534 | get_task_struct(p); | |
a06381fe | 535 | |
d359b549 | 536 | rcu_read_unlock(); |
c87e2837 IM |
537 | |
538 | return p; | |
539 | } | |
540 | ||
541 | /* | |
542 | * This task is holding PI mutexes at exit time => bad. | |
543 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
544 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
545 | */ | |
546 | void exit_pi_state_list(struct task_struct *curr) | |
547 | { | |
c87e2837 IM |
548 | struct list_head *next, *head = &curr->pi_state_list; |
549 | struct futex_pi_state *pi_state; | |
627371d7 | 550 | struct futex_hash_bucket *hb; |
38d47c1b | 551 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 552 | |
a0c1e907 TG |
553 | if (!futex_cmpxchg_enabled) |
554 | return; | |
c87e2837 IM |
555 | /* |
556 | * We are a ZOMBIE and nobody can enqueue itself on | |
557 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 558 | * versus waiters unqueueing themselves: |
c87e2837 | 559 | */ |
1d615482 | 560 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 IM |
561 | while (!list_empty(head)) { |
562 | ||
563 | next = head->next; | |
564 | pi_state = list_entry(next, struct futex_pi_state, list); | |
565 | key = pi_state->key; | |
627371d7 | 566 | hb = hash_futex(&key); |
1d615482 | 567 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 568 | |
c87e2837 IM |
569 | spin_lock(&hb->lock); |
570 | ||
1d615482 | 571 | raw_spin_lock_irq(&curr->pi_lock); |
627371d7 IM |
572 | /* |
573 | * We dropped the pi-lock, so re-check whether this | |
574 | * task still owns the PI-state: | |
575 | */ | |
c87e2837 IM |
576 | if (head->next != next) { |
577 | spin_unlock(&hb->lock); | |
578 | continue; | |
579 | } | |
580 | ||
c87e2837 | 581 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
582 | WARN_ON(list_empty(&pi_state->list)); |
583 | list_del_init(&pi_state->list); | |
c87e2837 | 584 | pi_state->owner = NULL; |
1d615482 | 585 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
586 | |
587 | rt_mutex_unlock(&pi_state->pi_mutex); | |
588 | ||
589 | spin_unlock(&hb->lock); | |
590 | ||
1d615482 | 591 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 592 | } |
1d615482 | 593 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
594 | } |
595 | ||
efccdcdb TG |
596 | /* |
597 | * We need to check the following states: | |
598 | * | |
599 | * Waiter | pi_state | pi->owner | uTID | uODIED | ? | |
600 | * | |
601 | * [1] NULL | --- | --- | 0 | 0/1 | Valid | |
602 | * [2] NULL | --- | --- | >0 | 0/1 | Valid | |
603 | * | |
604 | * [3] Found | NULL | -- | Any | 0/1 | Invalid | |
605 | * | |
606 | * [4] Found | Found | NULL | 0 | 1 | Valid | |
607 | * [5] Found | Found | NULL | >0 | 1 | Invalid | |
608 | * | |
609 | * [6] Found | Found | task | 0 | 1 | Valid | |
610 | * | |
611 | * [7] Found | Found | NULL | Any | 0 | Invalid | |
612 | * | |
613 | * [8] Found | Found | task | ==taskTID | 0/1 | Valid | |
614 | * [9] Found | Found | task | 0 | 0 | Invalid | |
615 | * [10] Found | Found | task | !=taskTID | 0/1 | Invalid | |
616 | * | |
617 | * [1] Indicates that the kernel can acquire the futex atomically. We | |
618 | * came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. | |
619 | * | |
620 | * [2] Valid, if TID does not belong to a kernel thread. If no matching | |
621 | * thread is found then it indicates that the owner TID has died. | |
622 | * | |
623 | * [3] Invalid. The waiter is queued on a non PI futex | |
624 | * | |
625 | * [4] Valid state after exit_robust_list(), which sets the user space | |
626 | * value to FUTEX_WAITERS | FUTEX_OWNER_DIED. | |
627 | * | |
628 | * [5] The user space value got manipulated between exit_robust_list() | |
629 | * and exit_pi_state_list() | |
630 | * | |
631 | * [6] Valid state after exit_pi_state_list() which sets the new owner in | |
632 | * the pi_state but cannot access the user space value. | |
633 | * | |
634 | * [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. | |
635 | * | |
636 | * [8] Owner and user space value match | |
637 | * | |
638 | * [9] There is no transient state which sets the user space TID to 0 | |
639 | * except exit_robust_list(), but this is indicated by the | |
640 | * FUTEX_OWNER_DIED bit. See [4] | |
641 | * | |
642 | * [10] There is no transient state which leaves owner and user space | |
643 | * TID out of sync. | |
644 | */ | |
c87e2837 | 645 | static int |
d0aa7a70 | 646 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
efccdcdb | 647 | union futex_key *key, struct futex_pi_state **ps) |
c87e2837 IM |
648 | { |
649 | struct futex_pi_state *pi_state = NULL; | |
650 | struct futex_q *this, *next; | |
ec92d082 | 651 | struct plist_head *head; |
c87e2837 | 652 | struct task_struct *p; |
778e9a9c | 653 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 IM |
654 | |
655 | head = &hb->chain; | |
656 | ||
ec92d082 | 657 | plist_for_each_entry_safe(this, next, head, list) { |
d0aa7a70 | 658 | if (match_futex(&this->key, key)) { |
c87e2837 | 659 | /* |
efccdcdb TG |
660 | * Sanity check the waiter before increasing |
661 | * the refcount and attaching to it. | |
c87e2837 IM |
662 | */ |
663 | pi_state = this->pi_state; | |
06a9ec29 | 664 | /* |
efccdcdb TG |
665 | * Userspace might have messed up non-PI and |
666 | * PI futexes [3] | |
06a9ec29 TG |
667 | */ |
668 | if (unlikely(!pi_state)) | |
669 | return -EINVAL; | |
670 | ||
627371d7 | 671 | WARN_ON(!atomic_read(&pi_state->refcount)); |
59647b6a TG |
672 | |
673 | /* | |
efccdcdb | 674 | * Handle the owner died case: |
59647b6a | 675 | */ |
efccdcdb | 676 | if (uval & FUTEX_OWNER_DIED) { |
59647b6a | 677 | /* |
efccdcdb TG |
678 | * exit_pi_state_list sets owner to NULL and |
679 | * wakes the topmost waiter. The task which | |
680 | * acquires the pi_state->rt_mutex will fixup | |
681 | * owner. | |
59647b6a | 682 | */ |
efccdcdb TG |
683 | if (!pi_state->owner) { |
684 | /* | |
685 | * No pi state owner, but the user | |
686 | * space TID is not 0. Inconsistent | |
687 | * state. [5] | |
688 | */ | |
689 | if (pid) | |
690 | return -EINVAL; | |
691 | /* | |
692 | * Take a ref on the state and | |
693 | * return. [4] | |
694 | */ | |
695 | goto out_state; | |
696 | } | |
697 | ||
698 | /* | |
699 | * If TID is 0, then either the dying owner | |
700 | * has not yet executed exit_pi_state_list() | |
701 | * or some waiter acquired the rtmutex in the | |
702 | * pi state, but did not yet fixup the TID in | |
703 | * user space. | |
704 | * | |
705 | * Take a ref on the state and return. [6] | |
706 | */ | |
707 | if (!pid) | |
708 | goto out_state; | |
709 | } else { | |
710 | /* | |
711 | * If the owner died bit is not set, | |
712 | * then the pi_state must have an | |
713 | * owner. [7] | |
714 | */ | |
715 | if (!pi_state->owner) | |
59647b6a TG |
716 | return -EINVAL; |
717 | } | |
627371d7 | 718 | |
cabef9fe | 719 | /* |
efccdcdb TG |
720 | * Bail out if user space manipulated the |
721 | * futex value. If pi state exists then the | |
722 | * owner TID must be the same as the user | |
723 | * space TID. [9/10] | |
cabef9fe | 724 | */ |
efccdcdb TG |
725 | if (pid != task_pid_vnr(pi_state->owner)) |
726 | return -EINVAL; | |
cabef9fe | 727 | |
efccdcdb | 728 | out_state: |
c87e2837 | 729 | atomic_inc(&pi_state->refcount); |
d0aa7a70 | 730 | *ps = pi_state; |
c87e2837 IM |
731 | return 0; |
732 | } | |
733 | } | |
734 | ||
735 | /* | |
e3f2ddea | 736 | * We are the first waiter - try to look up the real owner and attach |
efccdcdb | 737 | * the new pi_state to it, but bail out when TID = 0 [1] |
c87e2837 | 738 | */ |
778e9a9c | 739 | if (!pid) |
e3f2ddea | 740 | return -ESRCH; |
c87e2837 | 741 | p = futex_find_get_task(pid); |
7a0ea09a MH |
742 | if (!p) |
743 | return -ESRCH; | |
778e9a9c | 744 | |
452d7fea TG |
745 | if (!p->mm) { |
746 | put_task_struct(p); | |
747 | return -EPERM; | |
748 | } | |
749 | ||
778e9a9c AK |
750 | /* |
751 | * We need to look at the task state flags to figure out, | |
752 | * whether the task is exiting. To protect against the do_exit | |
753 | * change of the task flags, we do this protected by | |
754 | * p->pi_lock: | |
755 | */ | |
1d615482 | 756 | raw_spin_lock_irq(&p->pi_lock); |
778e9a9c AK |
757 | if (unlikely(p->flags & PF_EXITING)) { |
758 | /* | |
759 | * The task is on the way out. When PF_EXITPIDONE is | |
760 | * set, we know that the task has finished the | |
761 | * cleanup: | |
762 | */ | |
763 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
764 | ||
1d615482 | 765 | raw_spin_unlock_irq(&p->pi_lock); |
778e9a9c AK |
766 | put_task_struct(p); |
767 | return ret; | |
768 | } | |
c87e2837 | 769 | |
efccdcdb TG |
770 | /* |
771 | * No existing pi state. First waiter. [2] | |
772 | */ | |
c87e2837 IM |
773 | pi_state = alloc_pi_state(); |
774 | ||
775 | /* | |
776 | * Initialize the pi_mutex in locked state and make 'p' | |
777 | * the owner of it: | |
778 | */ | |
779 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
780 | ||
781 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 782 | pi_state->key = *key; |
c87e2837 | 783 | |
627371d7 | 784 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
785 | list_add(&pi_state->list, &p->pi_state_list); |
786 | pi_state->owner = p; | |
1d615482 | 787 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
788 | |
789 | put_task_struct(p); | |
790 | ||
d0aa7a70 | 791 | *ps = pi_state; |
c87e2837 IM |
792 | |
793 | return 0; | |
794 | } | |
795 | ||
1a52084d | 796 | /** |
d96ee56c | 797 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
798 | * @uaddr: the pi futex user address |
799 | * @hb: the pi futex hash bucket | |
800 | * @key: the futex key associated with uaddr and hb | |
801 | * @ps: the pi_state pointer where we store the result of the | |
802 | * lookup | |
803 | * @task: the task to perform the atomic lock work for. This will | |
804 | * be "current" except in the case of requeue pi. | |
805 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d | 806 | * |
6c23cbbd RD |
807 | * Return: |
808 | * 0 - ready to wait; | |
809 | * 1 - acquired the lock; | |
1a52084d DH |
810 | * <0 - error |
811 | * | |
812 | * The hb->lock and futex_key refs shall be held by the caller. | |
813 | */ | |
814 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
815 | union futex_key *key, | |
816 | struct futex_pi_state **ps, | |
bab5bc9e | 817 | struct task_struct *task, int set_waiters) |
1a52084d | 818 | { |
59fa6245 | 819 | int lock_taken, ret, force_take = 0; |
c0c9ed15 | 820 | u32 uval, newval, curval, vpid = task_pid_vnr(task); |
1a52084d DH |
821 | |
822 | retry: | |
823 | ret = lock_taken = 0; | |
824 | ||
825 | /* | |
826 | * To avoid races, we attempt to take the lock here again | |
827 | * (by doing a 0 -> TID atomic cmpxchg), while holding all | |
828 | * the locks. It will most likely not succeed. | |
829 | */ | |
c0c9ed15 | 830 | newval = vpid; |
bab5bc9e DH |
831 | if (set_waiters) |
832 | newval |= FUTEX_WAITERS; | |
1a52084d | 833 | |
37a9d912 | 834 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval))) |
1a52084d DH |
835 | return -EFAULT; |
836 | ||
837 | /* | |
838 | * Detect deadlocks. | |
839 | */ | |
c0c9ed15 | 840 | if ((unlikely((curval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
841 | return -EDEADLK; |
842 | ||
843 | /* | |
63d6ad59 | 844 | * Surprise - we got the lock, but we do not trust user space at all. |
1a52084d | 845 | */ |
63d6ad59 TG |
846 | if (unlikely(!curval)) { |
847 | /* | |
848 | * We verify whether there is kernel state for this | |
849 | * futex. If not, we can safely assume, that the 0 -> | |
850 | * TID transition is correct. If state exists, we do | |
851 | * not bother to fixup the user space state as it was | |
852 | * corrupted already. | |
853 | */ | |
854 | return futex_top_waiter(hb, key) ? -EINVAL : 1; | |
855 | } | |
1a52084d DH |
856 | |
857 | uval = curval; | |
858 | ||
859 | /* | |
860 | * Set the FUTEX_WAITERS flag, so the owner will know it has someone | |
861 | * to wake at the next unlock. | |
862 | */ | |
863 | newval = curval | FUTEX_WAITERS; | |
864 | ||
865 | /* | |
59fa6245 | 866 | * Should we force take the futex? See below. |
1a52084d | 867 | */ |
59fa6245 TG |
868 | if (unlikely(force_take)) { |
869 | /* | |
870 | * Keep the OWNER_DIED and the WAITERS bit and set the | |
871 | * new TID value. | |
872 | */ | |
c0c9ed15 | 873 | newval = (curval & ~FUTEX_TID_MASK) | vpid; |
59fa6245 | 874 | force_take = 0; |
1a52084d DH |
875 | lock_taken = 1; |
876 | } | |
877 | ||
37a9d912 | 878 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))) |
1a52084d DH |
879 | return -EFAULT; |
880 | if (unlikely(curval != uval)) | |
881 | goto retry; | |
882 | ||
883 | /* | |
59fa6245 | 884 | * We took the lock due to forced take over. |
1a52084d DH |
885 | */ |
886 | if (unlikely(lock_taken)) | |
887 | return 1; | |
888 | ||
889 | /* | |
890 | * We dont have the lock. Look up the PI state (or create it if | |
891 | * we are the first waiter): | |
892 | */ | |
efccdcdb | 893 | ret = lookup_pi_state(uval, hb, key, ps); |
1a52084d DH |
894 | |
895 | if (unlikely(ret)) { | |
896 | switch (ret) { | |
897 | case -ESRCH: | |
898 | /* | |
59fa6245 TG |
899 | * We failed to find an owner for this |
900 | * futex. So we have no pi_state to block | |
901 | * on. This can happen in two cases: | |
902 | * | |
903 | * 1) The owner died | |
904 | * 2) A stale FUTEX_WAITERS bit | |
905 | * | |
906 | * Re-read the futex value. | |
1a52084d DH |
907 | */ |
908 | if (get_futex_value_locked(&curval, uaddr)) | |
909 | return -EFAULT; | |
910 | ||
911 | /* | |
59fa6245 TG |
912 | * If the owner died or we have a stale |
913 | * WAITERS bit the owner TID in the user space | |
914 | * futex is 0. | |
1a52084d | 915 | */ |
59fa6245 TG |
916 | if (!(curval & FUTEX_TID_MASK)) { |
917 | force_take = 1; | |
1a52084d DH |
918 | goto retry; |
919 | } | |
920 | default: | |
921 | break; | |
922 | } | |
923 | } | |
924 | ||
925 | return ret; | |
926 | } | |
927 | ||
2e12978a LJ |
928 | /** |
929 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
930 | * @q: The futex_q to unqueue | |
931 | * | |
932 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
933 | */ | |
934 | static void __unqueue_futex(struct futex_q *q) | |
935 | { | |
936 | struct futex_hash_bucket *hb; | |
937 | ||
29096202 SR |
938 | if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr)) |
939 | || WARN_ON(plist_node_empty(&q->list))) | |
2e12978a LJ |
940 | return; |
941 | ||
942 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
943 | plist_del(&q->list, &hb->chain); | |
944 | } | |
945 | ||
1da177e4 LT |
946 | /* |
947 | * The hash bucket lock must be held when this is called. | |
948 | * Afterwards, the futex_q must not be accessed. | |
949 | */ | |
950 | static void wake_futex(struct futex_q *q) | |
951 | { | |
f1a11e05 TG |
952 | struct task_struct *p = q->task; |
953 | ||
aa10990e DH |
954 | if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n")) |
955 | return; | |
956 | ||
1da177e4 | 957 | /* |
f1a11e05 | 958 | * We set q->lock_ptr = NULL _before_ we wake up the task. If |
fb62db2b RD |
959 | * a non-futex wake up happens on another CPU then the task |
960 | * might exit and p would dereference a non-existing task | |
f1a11e05 TG |
961 | * struct. Prevent this by holding a reference on p across the |
962 | * wake up. | |
1da177e4 | 963 | */ |
f1a11e05 TG |
964 | get_task_struct(p); |
965 | ||
2e12978a | 966 | __unqueue_futex(q); |
1da177e4 | 967 | /* |
f1a11e05 TG |
968 | * The waiting task can free the futex_q as soon as |
969 | * q->lock_ptr = NULL is written, without taking any locks. A | |
970 | * memory barrier is required here to prevent the following | |
971 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 972 | */ |
ccdea2f8 | 973 | smp_wmb(); |
1da177e4 | 974 | q->lock_ptr = NULL; |
f1a11e05 TG |
975 | |
976 | wake_up_state(p, TASK_NORMAL); | |
977 | put_task_struct(p); | |
1da177e4 LT |
978 | } |
979 | ||
c87e2837 IM |
980 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
981 | { | |
982 | struct task_struct *new_owner; | |
983 | struct futex_pi_state *pi_state = this->pi_state; | |
7cfdaf38 | 984 | u32 uninitialized_var(curval), newval; |
9ad5dabd | 985 | int ret = 0; |
c87e2837 IM |
986 | |
987 | if (!pi_state) | |
988 | return -EINVAL; | |
989 | ||
51246bfd TG |
990 | /* |
991 | * If current does not own the pi_state then the futex is | |
992 | * inconsistent and user space fiddled with the futex value. | |
993 | */ | |
994 | if (pi_state->owner != current) | |
995 | return -EINVAL; | |
996 | ||
d209d74d | 997 | raw_spin_lock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
998 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
999 | ||
1000 | /* | |
f123c98e SR |
1001 | * It is possible that the next waiter (the one that brought |
1002 | * this owner to the kernel) timed out and is no longer | |
1003 | * waiting on the lock. | |
c87e2837 IM |
1004 | */ |
1005 | if (!new_owner) | |
1006 | new_owner = this->task; | |
1007 | ||
1008 | /* | |
9ad5dabd TG |
1009 | * We pass it to the next owner. The WAITERS bit is always |
1010 | * kept enabled while there is PI state around. We cleanup the | |
1011 | * owner died bit, because we are the owner. | |
c87e2837 | 1012 | */ |
9ad5dabd | 1013 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 1014 | |
9ad5dabd TG |
1015 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1016 | ret = -EFAULT; | |
1017 | else if (curval != uval) | |
1018 | ret = -EINVAL; | |
1019 | if (ret) { | |
1020 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); | |
1021 | return ret; | |
e3f2ddea | 1022 | } |
c87e2837 | 1023 | |
1d615482 | 1024 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
627371d7 IM |
1025 | WARN_ON(list_empty(&pi_state->list)); |
1026 | list_del_init(&pi_state->list); | |
1d615482 | 1027 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
627371d7 | 1028 | |
1d615482 | 1029 | raw_spin_lock_irq(&new_owner->pi_lock); |
627371d7 | 1030 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
1031 | list_add(&pi_state->list, &new_owner->pi_state_list); |
1032 | pi_state->owner = new_owner; | |
1d615482 | 1033 | raw_spin_unlock_irq(&new_owner->pi_lock); |
627371d7 | 1034 | |
d209d74d | 1035 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
1036 | rt_mutex_unlock(&pi_state->pi_mutex); |
1037 | ||
1038 | return 0; | |
1039 | } | |
1040 | ||
1041 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) | |
1042 | { | |
7cfdaf38 | 1043 | u32 uninitialized_var(oldval); |
c87e2837 IM |
1044 | |
1045 | /* | |
1046 | * There is no waiter, so we unlock the futex. The owner died | |
1047 | * bit has not to be preserved here. We are the owner: | |
1048 | */ | |
37a9d912 ML |
1049 | if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0)) |
1050 | return -EFAULT; | |
c87e2837 IM |
1051 | if (oldval != uval) |
1052 | return -EAGAIN; | |
1053 | ||
1054 | return 0; | |
1055 | } | |
1056 | ||
8b8f319f IM |
1057 | /* |
1058 | * Express the locking dependencies for lockdep: | |
1059 | */ | |
1060 | static inline void | |
1061 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1062 | { | |
1063 | if (hb1 <= hb2) { | |
1064 | spin_lock(&hb1->lock); | |
1065 | if (hb1 < hb2) | |
1066 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
1067 | } else { /* hb1 > hb2 */ | |
1068 | spin_lock(&hb2->lock); | |
1069 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
1070 | } | |
1071 | } | |
1072 | ||
5eb3dc62 DH |
1073 | static inline void |
1074 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1075 | { | |
f061d351 | 1076 | spin_unlock(&hb1->lock); |
88f502fe IM |
1077 | if (hb1 != hb2) |
1078 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
1079 | } |
1080 | ||
1da177e4 | 1081 | /* |
b2d0994b | 1082 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 1083 | */ |
b41277dc DH |
1084 | static int |
1085 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 1086 | { |
e2970f2f | 1087 | struct futex_hash_bucket *hb; |
1da177e4 | 1088 | struct futex_q *this, *next; |
ec92d082 | 1089 | struct plist_head *head; |
38d47c1b | 1090 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 LT |
1091 | int ret; |
1092 | ||
cd689985 TG |
1093 | if (!bitset) |
1094 | return -EINVAL; | |
1095 | ||
9ea71503 | 1096 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ); |
1da177e4 LT |
1097 | if (unlikely(ret != 0)) |
1098 | goto out; | |
1099 | ||
e2970f2f IM |
1100 | hb = hash_futex(&key); |
1101 | spin_lock(&hb->lock); | |
1102 | head = &hb->chain; | |
1da177e4 | 1103 | |
ec92d082 | 1104 | plist_for_each_entry_safe(this, next, head, list) { |
1da177e4 | 1105 | if (match_futex (&this->key, &key)) { |
52400ba9 | 1106 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
1107 | ret = -EINVAL; |
1108 | break; | |
1109 | } | |
cd689985 TG |
1110 | |
1111 | /* Check if one of the bits is set in both bitsets */ | |
1112 | if (!(this->bitset & bitset)) | |
1113 | continue; | |
1114 | ||
1da177e4 LT |
1115 | wake_futex(this); |
1116 | if (++ret >= nr_wake) | |
1117 | break; | |
1118 | } | |
1119 | } | |
1120 | ||
e2970f2f | 1121 | spin_unlock(&hb->lock); |
ae791a2d | 1122 | put_futex_key(&key); |
42d35d48 | 1123 | out: |
1da177e4 LT |
1124 | return ret; |
1125 | } | |
1126 | ||
4732efbe JJ |
1127 | /* |
1128 | * Wake up all waiters hashed on the physical page that is mapped | |
1129 | * to this virtual address: | |
1130 | */ | |
e2970f2f | 1131 | static int |
b41277dc | 1132 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 1133 | int nr_wake, int nr_wake2, int op) |
4732efbe | 1134 | { |
38d47c1b | 1135 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 1136 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 1137 | struct plist_head *head; |
4732efbe | 1138 | struct futex_q *this, *next; |
e4dc5b7a | 1139 | int ret, op_ret; |
4732efbe | 1140 | |
e4dc5b7a | 1141 | retry: |
9ea71503 | 1142 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
4732efbe JJ |
1143 | if (unlikely(ret != 0)) |
1144 | goto out; | |
9ea71503 | 1145 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
4732efbe | 1146 | if (unlikely(ret != 0)) |
42d35d48 | 1147 | goto out_put_key1; |
4732efbe | 1148 | |
e2970f2f IM |
1149 | hb1 = hash_futex(&key1); |
1150 | hb2 = hash_futex(&key2); | |
4732efbe | 1151 | |
e4dc5b7a | 1152 | retry_private: |
eaaea803 | 1153 | double_lock_hb(hb1, hb2); |
e2970f2f | 1154 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1155 | if (unlikely(op_ret < 0)) { |
4732efbe | 1156 | |
5eb3dc62 | 1157 | double_unlock_hb(hb1, hb2); |
4732efbe | 1158 | |
7ee1dd3f | 1159 | #ifndef CONFIG_MMU |
e2970f2f IM |
1160 | /* |
1161 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
1162 | * but we might get them from range checking | |
1163 | */ | |
7ee1dd3f | 1164 | ret = op_ret; |
42d35d48 | 1165 | goto out_put_keys; |
7ee1dd3f DH |
1166 | #endif |
1167 | ||
796f8d9b DG |
1168 | if (unlikely(op_ret != -EFAULT)) { |
1169 | ret = op_ret; | |
42d35d48 | 1170 | goto out_put_keys; |
796f8d9b DG |
1171 | } |
1172 | ||
d0725992 | 1173 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 1174 | if (ret) |
de87fcc1 | 1175 | goto out_put_keys; |
4732efbe | 1176 | |
b41277dc | 1177 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1178 | goto retry_private; |
1179 | ||
ae791a2d TG |
1180 | put_futex_key(&key2); |
1181 | put_futex_key(&key1); | |
e4dc5b7a | 1182 | goto retry; |
4732efbe JJ |
1183 | } |
1184 | ||
e2970f2f | 1185 | head = &hb1->chain; |
4732efbe | 1186 | |
ec92d082 | 1187 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe | 1188 | if (match_futex (&this->key, &key1)) { |
aa10990e DH |
1189 | if (this->pi_state || this->rt_waiter) { |
1190 | ret = -EINVAL; | |
1191 | goto out_unlock; | |
1192 | } | |
4732efbe JJ |
1193 | wake_futex(this); |
1194 | if (++ret >= nr_wake) | |
1195 | break; | |
1196 | } | |
1197 | } | |
1198 | ||
1199 | if (op_ret > 0) { | |
e2970f2f | 1200 | head = &hb2->chain; |
4732efbe JJ |
1201 | |
1202 | op_ret = 0; | |
ec92d082 | 1203 | plist_for_each_entry_safe(this, next, head, list) { |
4732efbe | 1204 | if (match_futex (&this->key, &key2)) { |
aa10990e DH |
1205 | if (this->pi_state || this->rt_waiter) { |
1206 | ret = -EINVAL; | |
1207 | goto out_unlock; | |
1208 | } | |
4732efbe JJ |
1209 | wake_futex(this); |
1210 | if (++op_ret >= nr_wake2) | |
1211 | break; | |
1212 | } | |
1213 | } | |
1214 | ret += op_ret; | |
1215 | } | |
1216 | ||
aa10990e | 1217 | out_unlock: |
5eb3dc62 | 1218 | double_unlock_hb(hb1, hb2); |
42d35d48 | 1219 | out_put_keys: |
ae791a2d | 1220 | put_futex_key(&key2); |
42d35d48 | 1221 | out_put_key1: |
ae791a2d | 1222 | put_futex_key(&key1); |
42d35d48 | 1223 | out: |
4732efbe JJ |
1224 | return ret; |
1225 | } | |
1226 | ||
9121e478 DH |
1227 | /** |
1228 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1229 | * @q: the futex_q to requeue | |
1230 | * @hb1: the source hash_bucket | |
1231 | * @hb2: the target hash_bucket | |
1232 | * @key2: the new key for the requeued futex_q | |
1233 | */ | |
1234 | static inline | |
1235 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1236 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1237 | { | |
1238 | ||
1239 | /* | |
1240 | * If key1 and key2 hash to the same bucket, no need to | |
1241 | * requeue. | |
1242 | */ | |
1243 | if (likely(&hb1->chain != &hb2->chain)) { | |
1244 | plist_del(&q->list, &hb1->chain); | |
1245 | plist_add(&q->list, &hb2->chain); | |
1246 | q->lock_ptr = &hb2->lock; | |
9121e478 DH |
1247 | } |
1248 | get_futex_key_refs(key2); | |
1249 | q->key = *key2; | |
1250 | } | |
1251 | ||
52400ba9 DH |
1252 | /** |
1253 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1254 | * @q: the futex_q |
1255 | * @key: the key of the requeue target futex | |
1256 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1257 | * |
1258 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1259 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1260 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1261 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1262 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1263 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1264 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1265 | */ |
1266 | static inline | |
beda2c7e DH |
1267 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1268 | struct futex_hash_bucket *hb) | |
52400ba9 | 1269 | { |
52400ba9 DH |
1270 | get_futex_key_refs(key); |
1271 | q->key = *key; | |
1272 | ||
2e12978a | 1273 | __unqueue_futex(q); |
52400ba9 DH |
1274 | |
1275 | WARN_ON(!q->rt_waiter); | |
1276 | q->rt_waiter = NULL; | |
1277 | ||
beda2c7e | 1278 | q->lock_ptr = &hb->lock; |
beda2c7e | 1279 | |
f1a11e05 | 1280 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1281 | } |
1282 | ||
1283 | /** | |
1284 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1285 | * @pifutex: the user address of the to futex |
1286 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1287 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1288 | * @key1: the from futex key | |
1289 | * @key2: the to futex key | |
1290 | * @ps: address to store the pi_state pointer | |
1291 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1292 | * |
1293 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1294 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1295 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1296 | * hb1 and hb2 must be held by the caller. | |
52400ba9 | 1297 | * |
6c23cbbd RD |
1298 | * Return: |
1299 | * 0 - failed to acquire the lock atomically; | |
cabef9fe | 1300 | * >0 - acquired the lock, return value is vpid of the top_waiter |
52400ba9 DH |
1301 | * <0 - error |
1302 | */ | |
1303 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1304 | struct futex_hash_bucket *hb1, | |
1305 | struct futex_hash_bucket *hb2, | |
1306 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1307 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1308 | { |
bab5bc9e | 1309 | struct futex_q *top_waiter = NULL; |
52400ba9 | 1310 | u32 curval; |
cabef9fe | 1311 | int ret, vpid; |
52400ba9 DH |
1312 | |
1313 | if (get_futex_value_locked(&curval, pifutex)) | |
1314 | return -EFAULT; | |
1315 | ||
bab5bc9e DH |
1316 | /* |
1317 | * Find the top_waiter and determine if there are additional waiters. | |
1318 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1319 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1320 | * as we have means to handle the possible fault. If not, don't set | |
1321 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1322 | * the kernel. | |
1323 | */ | |
52400ba9 DH |
1324 | top_waiter = futex_top_waiter(hb1, key1); |
1325 | ||
1326 | /* There are no waiters, nothing for us to do. */ | |
1327 | if (!top_waiter) | |
1328 | return 0; | |
1329 | ||
84bc4af5 DH |
1330 | /* Ensure we requeue to the expected futex. */ |
1331 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1332 | return -EINVAL; | |
1333 | ||
52400ba9 | 1334 | /* |
bab5bc9e DH |
1335 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1336 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1337 | * in ps in contended cases. | |
52400ba9 | 1338 | */ |
cabef9fe | 1339 | vpid = task_pid_vnr(top_waiter->task); |
bab5bc9e DH |
1340 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1341 | set_waiters); | |
cabef9fe | 1342 | if (ret == 1) { |
beda2c7e | 1343 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
cabef9fe TG |
1344 | return vpid; |
1345 | } | |
52400ba9 DH |
1346 | return ret; |
1347 | } | |
1348 | ||
1349 | /** | |
1350 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1351 | * @uaddr1: source futex user address |
b41277dc | 1352 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1353 | * @uaddr2: target futex user address |
1354 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1355 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1356 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1357 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1358 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1359 | * |
1360 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1361 | * uaddr2 atomically on behalf of the top waiter. | |
1362 | * | |
6c23cbbd RD |
1363 | * Return: |
1364 | * >=0 - on success, the number of tasks requeued or woken; | |
52400ba9 | 1365 | * <0 - on error |
1da177e4 | 1366 | */ |
b41277dc DH |
1367 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
1368 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
1369 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 1370 | { |
38d47c1b | 1371 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1372 | int drop_count = 0, task_count = 0, ret; |
1373 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1374 | struct futex_hash_bucket *hb1, *hb2; |
ec92d082 | 1375 | struct plist_head *head1; |
1da177e4 | 1376 | struct futex_q *this, *next; |
52400ba9 DH |
1377 | |
1378 | if (requeue_pi) { | |
b58623fb TG |
1379 | /* |
1380 | * Requeue PI only works on two distinct uaddrs. This | |
1381 | * check is only valid for private futexes. See below. | |
1382 | */ | |
1383 | if (uaddr1 == uaddr2) | |
1384 | return -EINVAL; | |
1385 | ||
52400ba9 DH |
1386 | /* |
1387 | * requeue_pi requires a pi_state, try to allocate it now | |
1388 | * without any locks in case it fails. | |
1389 | */ | |
1390 | if (refill_pi_state_cache()) | |
1391 | return -ENOMEM; | |
1392 | /* | |
1393 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1394 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1395 | * returning to userspace, so as to not leave the rt_mutex with | |
1396 | * waiters and no owner. However, second and third wake-ups | |
1397 | * cannot be predicted as they involve race conditions with the | |
1398 | * first wake and a fault while looking up the pi_state. Both | |
1399 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1400 | * use nr_wake=1. | |
1401 | */ | |
1402 | if (nr_wake != 1) | |
1403 | return -EINVAL; | |
1404 | } | |
1da177e4 | 1405 | |
42d35d48 | 1406 | retry: |
52400ba9 DH |
1407 | if (pi_state != NULL) { |
1408 | /* | |
1409 | * We will have to lookup the pi_state again, so free this one | |
1410 | * to keep the accounting correct. | |
1411 | */ | |
1412 | free_pi_state(pi_state); | |
1413 | pi_state = NULL; | |
1414 | } | |
1415 | ||
9ea71503 | 1416 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
1da177e4 LT |
1417 | if (unlikely(ret != 0)) |
1418 | goto out; | |
9ea71503 SB |
1419 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, |
1420 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1421 | if (unlikely(ret != 0)) |
42d35d48 | 1422 | goto out_put_key1; |
1da177e4 | 1423 | |
b58623fb TG |
1424 | /* |
1425 | * The check above which compares uaddrs is not sufficient for | |
1426 | * shared futexes. We need to compare the keys: | |
1427 | */ | |
1428 | if (requeue_pi && match_futex(&key1, &key2)) { | |
1429 | ret = -EINVAL; | |
1430 | goto out_put_keys; | |
1431 | } | |
1432 | ||
e2970f2f IM |
1433 | hb1 = hash_futex(&key1); |
1434 | hb2 = hash_futex(&key2); | |
1da177e4 | 1435 | |
e4dc5b7a | 1436 | retry_private: |
8b8f319f | 1437 | double_lock_hb(hb1, hb2); |
1da177e4 | 1438 | |
e2970f2f IM |
1439 | if (likely(cmpval != NULL)) { |
1440 | u32 curval; | |
1da177e4 | 1441 | |
e2970f2f | 1442 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1443 | |
1444 | if (unlikely(ret)) { | |
5eb3dc62 | 1445 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1446 | |
e2970f2f | 1447 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1448 | if (ret) |
1449 | goto out_put_keys; | |
1da177e4 | 1450 | |
b41277dc | 1451 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 1452 | goto retry_private; |
1da177e4 | 1453 | |
ae791a2d TG |
1454 | put_futex_key(&key2); |
1455 | put_futex_key(&key1); | |
e4dc5b7a | 1456 | goto retry; |
1da177e4 | 1457 | } |
e2970f2f | 1458 | if (curval != *cmpval) { |
1da177e4 LT |
1459 | ret = -EAGAIN; |
1460 | goto out_unlock; | |
1461 | } | |
1462 | } | |
1463 | ||
52400ba9 | 1464 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1465 | /* |
1466 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1467 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1468 | * bit. We force this here where we are able to easily handle | |
1469 | * faults rather in the requeue loop below. | |
1470 | */ | |
52400ba9 | 1471 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1472 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1473 | |
1474 | /* | |
1475 | * At this point the top_waiter has either taken uaddr2 or is | |
1476 | * waiting on it. If the former, then the pi_state will not | |
1477 | * exist yet, look it up one more time to ensure we have a | |
cabef9fe TG |
1478 | * reference to it. If the lock was taken, ret contains the |
1479 | * vpid of the top waiter task. | |
52400ba9 | 1480 | */ |
cabef9fe | 1481 | if (ret > 0) { |
52400ba9 | 1482 | WARN_ON(pi_state); |
89061d3d | 1483 | drop_count++; |
52400ba9 | 1484 | task_count++; |
cabef9fe TG |
1485 | /* |
1486 | * If we acquired the lock, then the user | |
1487 | * space value of uaddr2 should be vpid. It | |
1488 | * cannot be changed by the top waiter as it | |
1489 | * is blocked on hb2 lock if it tries to do | |
1490 | * so. If something fiddled with it behind our | |
1491 | * back the pi state lookup might unearth | |
1492 | * it. So we rather use the known value than | |
1493 | * rereading and handing potential crap to | |
1494 | * lookup_pi_state. | |
1495 | */ | |
efccdcdb | 1496 | ret = lookup_pi_state(ret, hb2, &key2, &pi_state); |
52400ba9 DH |
1497 | } |
1498 | ||
1499 | switch (ret) { | |
1500 | case 0: | |
1501 | break; | |
1502 | case -EFAULT: | |
1503 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1504 | put_futex_key(&key2); |
1505 | put_futex_key(&key1); | |
d0725992 | 1506 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
1507 | if (!ret) |
1508 | goto retry; | |
1509 | goto out; | |
1510 | case -EAGAIN: | |
1511 | /* The owner was exiting, try again. */ | |
1512 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1513 | put_futex_key(&key2); |
1514 | put_futex_key(&key1); | |
52400ba9 DH |
1515 | cond_resched(); |
1516 | goto retry; | |
1517 | default: | |
1518 | goto out_unlock; | |
1519 | } | |
1520 | } | |
1521 | ||
e2970f2f | 1522 | head1 = &hb1->chain; |
ec92d082 | 1523 | plist_for_each_entry_safe(this, next, head1, list) { |
52400ba9 DH |
1524 | if (task_count - nr_wake >= nr_requeue) |
1525 | break; | |
1526 | ||
1527 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1528 | continue; |
52400ba9 | 1529 | |
392741e0 DH |
1530 | /* |
1531 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
1532 | * be paired with each other and no other futex ops. | |
aa10990e DH |
1533 | * |
1534 | * We should never be requeueing a futex_q with a pi_state, | |
1535 | * which is awaiting a futex_unlock_pi(). | |
392741e0 DH |
1536 | */ |
1537 | if ((requeue_pi && !this->rt_waiter) || | |
aa10990e DH |
1538 | (!requeue_pi && this->rt_waiter) || |
1539 | this->pi_state) { | |
392741e0 DH |
1540 | ret = -EINVAL; |
1541 | break; | |
1542 | } | |
52400ba9 DH |
1543 | |
1544 | /* | |
1545 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1546 | * lock, we already woke the top_waiter. If not, it will be | |
1547 | * woken by futex_unlock_pi(). | |
1548 | */ | |
1549 | if (++task_count <= nr_wake && !requeue_pi) { | |
1da177e4 | 1550 | wake_futex(this); |
52400ba9 DH |
1551 | continue; |
1552 | } | |
1da177e4 | 1553 | |
84bc4af5 DH |
1554 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
1555 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
1556 | ret = -EINVAL; | |
1557 | break; | |
1558 | } | |
1559 | ||
52400ba9 DH |
1560 | /* |
1561 | * Requeue nr_requeue waiters and possibly one more in the case | |
1562 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1563 | */ | |
1564 | if (requeue_pi) { | |
1565 | /* Prepare the waiter to take the rt_mutex. */ | |
1566 | atomic_inc(&pi_state->refcount); | |
1567 | this->pi_state = pi_state; | |
1568 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1569 | this->rt_waiter, | |
1570 | this->task, 1); | |
1571 | if (ret == 1) { | |
1572 | /* We got the lock. */ | |
beda2c7e | 1573 | requeue_pi_wake_futex(this, &key2, hb2); |
89061d3d | 1574 | drop_count++; |
52400ba9 DH |
1575 | continue; |
1576 | } else if (ret) { | |
1577 | /* -EDEADLK */ | |
1578 | this->pi_state = NULL; | |
1579 | free_pi_state(pi_state); | |
1580 | goto out_unlock; | |
1581 | } | |
1da177e4 | 1582 | } |
52400ba9 DH |
1583 | requeue_futex(this, hb1, hb2, &key2); |
1584 | drop_count++; | |
1da177e4 LT |
1585 | } |
1586 | ||
1587 | out_unlock: | |
5eb3dc62 | 1588 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1589 | |
cd84a42f DH |
1590 | /* |
1591 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1592 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1593 | * one at key2 and updated their key pointer. We no longer need to | |
1594 | * hold the references to key1. | |
1595 | */ | |
1da177e4 | 1596 | while (--drop_count >= 0) |
9adef58b | 1597 | drop_futex_key_refs(&key1); |
1da177e4 | 1598 | |
42d35d48 | 1599 | out_put_keys: |
ae791a2d | 1600 | put_futex_key(&key2); |
42d35d48 | 1601 | out_put_key1: |
ae791a2d | 1602 | put_futex_key(&key1); |
42d35d48 | 1603 | out: |
52400ba9 DH |
1604 | if (pi_state != NULL) |
1605 | free_pi_state(pi_state); | |
1606 | return ret ? ret : task_count; | |
1da177e4 LT |
1607 | } |
1608 | ||
1609 | /* The key must be already stored in q->key. */ | |
82af7aca | 1610 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 1611 | __acquires(&hb->lock) |
1da177e4 | 1612 | { |
e2970f2f | 1613 | struct futex_hash_bucket *hb; |
1da177e4 | 1614 | |
e2970f2f IM |
1615 | hb = hash_futex(&q->key); |
1616 | q->lock_ptr = &hb->lock; | |
1da177e4 | 1617 | |
e2970f2f IM |
1618 | spin_lock(&hb->lock); |
1619 | return hb; | |
1da177e4 LT |
1620 | } |
1621 | ||
d40d65c8 DH |
1622 | static inline void |
1623 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) | |
15e408cd | 1624 | __releases(&hb->lock) |
d40d65c8 DH |
1625 | { |
1626 | spin_unlock(&hb->lock); | |
d40d65c8 DH |
1627 | } |
1628 | ||
1629 | /** | |
1630 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
1631 | * @q: The futex_q to enqueue | |
1632 | * @hb: The destination hash bucket | |
1633 | * | |
1634 | * The hb->lock must be held by the caller, and is released here. A call to | |
1635 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
1636 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
1637 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
1638 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
1639 | * an example). | |
1640 | */ | |
82af7aca | 1641 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
15e408cd | 1642 | __releases(&hb->lock) |
1da177e4 | 1643 | { |
ec92d082 PP |
1644 | int prio; |
1645 | ||
1646 | /* | |
1647 | * The priority used to register this element is | |
1648 | * - either the real thread-priority for the real-time threads | |
1649 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
1650 | * - or MAX_RT_PRIO for non-RT threads. | |
1651 | * Thus, all RT-threads are woken first in priority order, and | |
1652 | * the others are woken last, in FIFO order. | |
1653 | */ | |
1654 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1655 | ||
1656 | plist_node_init(&q->list, prio); | |
ec92d082 | 1657 | plist_add(&q->list, &hb->chain); |
c87e2837 | 1658 | q->task = current; |
e2970f2f | 1659 | spin_unlock(&hb->lock); |
1da177e4 LT |
1660 | } |
1661 | ||
d40d65c8 DH |
1662 | /** |
1663 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
1664 | * @q: The futex_q to unqueue | |
1665 | * | |
1666 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
1667 | * be paired with exactly one earlier call to queue_me(). | |
1668 | * | |
6c23cbbd RD |
1669 | * Return: |
1670 | * 1 - if the futex_q was still queued (and we removed unqueued it); | |
d40d65c8 | 1671 | * 0 - if the futex_q was already removed by the waking thread |
1da177e4 | 1672 | */ |
1da177e4 LT |
1673 | static int unqueue_me(struct futex_q *q) |
1674 | { | |
1da177e4 | 1675 | spinlock_t *lock_ptr; |
e2970f2f | 1676 | int ret = 0; |
1da177e4 LT |
1677 | |
1678 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 1679 | retry: |
1da177e4 | 1680 | lock_ptr = q->lock_ptr; |
e91467ec | 1681 | barrier(); |
c80544dc | 1682 | if (lock_ptr != NULL) { |
1da177e4 LT |
1683 | spin_lock(lock_ptr); |
1684 | /* | |
1685 | * q->lock_ptr can change between reading it and | |
1686 | * spin_lock(), causing us to take the wrong lock. This | |
1687 | * corrects the race condition. | |
1688 | * | |
1689 | * Reasoning goes like this: if we have the wrong lock, | |
1690 | * q->lock_ptr must have changed (maybe several times) | |
1691 | * between reading it and the spin_lock(). It can | |
1692 | * change again after the spin_lock() but only if it was | |
1693 | * already changed before the spin_lock(). It cannot, | |
1694 | * however, change back to the original value. Therefore | |
1695 | * we can detect whether we acquired the correct lock. | |
1696 | */ | |
1697 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
1698 | spin_unlock(lock_ptr); | |
1699 | goto retry; | |
1700 | } | |
2e12978a | 1701 | __unqueue_futex(q); |
c87e2837 IM |
1702 | |
1703 | BUG_ON(q->pi_state); | |
1704 | ||
1da177e4 LT |
1705 | spin_unlock(lock_ptr); |
1706 | ret = 1; | |
1707 | } | |
1708 | ||
9adef58b | 1709 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1710 | return ret; |
1711 | } | |
1712 | ||
c87e2837 IM |
1713 | /* |
1714 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
1715 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
1716 | * and dropped here. | |
c87e2837 | 1717 | */ |
d0aa7a70 | 1718 | static void unqueue_me_pi(struct futex_q *q) |
15e408cd | 1719 | __releases(q->lock_ptr) |
c87e2837 | 1720 | { |
2e12978a | 1721 | __unqueue_futex(q); |
c87e2837 IM |
1722 | |
1723 | BUG_ON(!q->pi_state); | |
1724 | free_pi_state(q->pi_state); | |
1725 | q->pi_state = NULL; | |
1726 | ||
d0aa7a70 | 1727 | spin_unlock(q->lock_ptr); |
c87e2837 IM |
1728 | } |
1729 | ||
d0aa7a70 | 1730 | /* |
cdf71a10 | 1731 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 1732 | * |
778e9a9c AK |
1733 | * Must be called with hash bucket lock held and mm->sem held for non |
1734 | * private futexes. | |
d0aa7a70 | 1735 | */ |
778e9a9c | 1736 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
ae791a2d | 1737 | struct task_struct *newowner) |
d0aa7a70 | 1738 | { |
cdf71a10 | 1739 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 1740 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 1741 | struct task_struct *oldowner = pi_state->owner; |
7cfdaf38 | 1742 | u32 uval, uninitialized_var(curval), newval; |
e4dc5b7a | 1743 | int ret; |
d0aa7a70 PP |
1744 | |
1745 | /* Owner died? */ | |
1b7558e4 TG |
1746 | if (!pi_state->owner) |
1747 | newtid |= FUTEX_OWNER_DIED; | |
1748 | ||
1749 | /* | |
1750 | * We are here either because we stole the rtmutex from the | |
8161239a LJ |
1751 | * previous highest priority waiter or we are the highest priority |
1752 | * waiter but failed to get the rtmutex the first time. | |
1753 | * We have to replace the newowner TID in the user space variable. | |
1754 | * This must be atomic as we have to preserve the owner died bit here. | |
1b7558e4 | 1755 | * |
b2d0994b DH |
1756 | * Note: We write the user space value _before_ changing the pi_state |
1757 | * because we can fault here. Imagine swapped out pages or a fork | |
1758 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
1759 | * |
1760 | * Modifying pi_state _before_ the user space value would | |
1761 | * leave the pi_state in an inconsistent state when we fault | |
1762 | * here, because we need to drop the hash bucket lock to | |
1763 | * handle the fault. This might be observed in the PID check | |
1764 | * in lookup_pi_state. | |
1765 | */ | |
1766 | retry: | |
1767 | if (get_futex_value_locked(&uval, uaddr)) | |
1768 | goto handle_fault; | |
1769 | ||
1770 | while (1) { | |
1771 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
1772 | ||
37a9d912 | 1773 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1b7558e4 TG |
1774 | goto handle_fault; |
1775 | if (curval == uval) | |
1776 | break; | |
1777 | uval = curval; | |
1778 | } | |
1779 | ||
1780 | /* | |
1781 | * We fixed up user space. Now we need to fix the pi_state | |
1782 | * itself. | |
1783 | */ | |
d0aa7a70 | 1784 | if (pi_state->owner != NULL) { |
1d615482 | 1785 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
d0aa7a70 PP |
1786 | WARN_ON(list_empty(&pi_state->list)); |
1787 | list_del_init(&pi_state->list); | |
1d615482 | 1788 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
1b7558e4 | 1789 | } |
d0aa7a70 | 1790 | |
cdf71a10 | 1791 | pi_state->owner = newowner; |
d0aa7a70 | 1792 | |
1d615482 | 1793 | raw_spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 1794 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 | 1795 | list_add(&pi_state->list, &newowner->pi_state_list); |
1d615482 | 1796 | raw_spin_unlock_irq(&newowner->pi_lock); |
1b7558e4 | 1797 | return 0; |
d0aa7a70 | 1798 | |
d0aa7a70 | 1799 | /* |
1b7558e4 | 1800 | * To handle the page fault we need to drop the hash bucket |
8161239a LJ |
1801 | * lock here. That gives the other task (either the highest priority |
1802 | * waiter itself or the task which stole the rtmutex) the | |
1b7558e4 TG |
1803 | * chance to try the fixup of the pi_state. So once we are |
1804 | * back from handling the fault we need to check the pi_state | |
1805 | * after reacquiring the hash bucket lock and before trying to | |
1806 | * do another fixup. When the fixup has been done already we | |
1807 | * simply return. | |
d0aa7a70 | 1808 | */ |
1b7558e4 TG |
1809 | handle_fault: |
1810 | spin_unlock(q->lock_ptr); | |
778e9a9c | 1811 | |
d0725992 | 1812 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 1813 | |
1b7558e4 | 1814 | spin_lock(q->lock_ptr); |
778e9a9c | 1815 | |
1b7558e4 TG |
1816 | /* |
1817 | * Check if someone else fixed it for us: | |
1818 | */ | |
1819 | if (pi_state->owner != oldowner) | |
1820 | return 0; | |
1821 | ||
1822 | if (ret) | |
1823 | return ret; | |
1824 | ||
1825 | goto retry; | |
d0aa7a70 PP |
1826 | } |
1827 | ||
72c1bbf3 | 1828 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 1829 | |
dd973998 DH |
1830 | /** |
1831 | * fixup_owner() - Post lock pi_state and corner case management | |
1832 | * @uaddr: user address of the futex | |
dd973998 DH |
1833 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
1834 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
1835 | * | |
1836 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
1837 | * the pi_state owner as well as handle race conditions that may allow us to | |
1838 | * acquire the lock. Must be called with the hb lock held. | |
1839 | * | |
6c23cbbd RD |
1840 | * Return: |
1841 | * 1 - success, lock taken; | |
1842 | * 0 - success, lock not taken; | |
dd973998 DH |
1843 | * <0 - on error (-EFAULT) |
1844 | */ | |
ae791a2d | 1845 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 DH |
1846 | { |
1847 | struct task_struct *owner; | |
1848 | int ret = 0; | |
1849 | ||
1850 | if (locked) { | |
1851 | /* | |
1852 | * Got the lock. We might not be the anticipated owner if we | |
1853 | * did a lock-steal - fix up the PI-state in that case: | |
1854 | */ | |
1855 | if (q->pi_state->owner != current) | |
ae791a2d | 1856 | ret = fixup_pi_state_owner(uaddr, q, current); |
dd973998 DH |
1857 | goto out; |
1858 | } | |
1859 | ||
1860 | /* | |
1861 | * Catch the rare case, where the lock was released when we were on the | |
1862 | * way back before we locked the hash bucket. | |
1863 | */ | |
1864 | if (q->pi_state->owner == current) { | |
1865 | /* | |
1866 | * Try to get the rt_mutex now. This might fail as some other | |
1867 | * task acquired the rt_mutex after we removed ourself from the | |
1868 | * rt_mutex waiters list. | |
1869 | */ | |
1870 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
1871 | locked = 1; | |
1872 | goto out; | |
1873 | } | |
1874 | ||
1875 | /* | |
1876 | * pi_state is incorrect, some other task did a lock steal and | |
1877 | * we returned due to timeout or signal without taking the | |
8161239a | 1878 | * rt_mutex. Too late. |
dd973998 | 1879 | */ |
8161239a | 1880 | raw_spin_lock(&q->pi_state->pi_mutex.wait_lock); |
dd973998 | 1881 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); |
8161239a LJ |
1882 | if (!owner) |
1883 | owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); | |
1884 | raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock); | |
ae791a2d | 1885 | ret = fixup_pi_state_owner(uaddr, q, owner); |
dd973998 DH |
1886 | goto out; |
1887 | } | |
1888 | ||
1889 | /* | |
1890 | * Paranoia check. If we did not take the lock, then we should not be | |
8161239a | 1891 | * the owner of the rt_mutex. |
dd973998 DH |
1892 | */ |
1893 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
1894 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
1895 | "pi-state %p\n", ret, | |
1896 | q->pi_state->pi_mutex.owner, | |
1897 | q->pi_state->owner); | |
1898 | ||
1899 | out: | |
1900 | return ret ? ret : locked; | |
1901 | } | |
1902 | ||
ca5f9524 DH |
1903 | /** |
1904 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
1905 | * @hb: the futex hash bucket, must be locked by the caller | |
1906 | * @q: the futex_q to queue up on | |
1907 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
1908 | */ |
1909 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 1910 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 1911 | { |
9beba3c5 DH |
1912 | /* |
1913 | * The task state is guaranteed to be set before another task can | |
1914 | * wake it. set_current_state() is implemented using set_mb() and | |
1915 | * queue_me() calls spin_unlock() upon completion, both serializing | |
1916 | * access to the hash list and forcing another memory barrier. | |
1917 | */ | |
f1a11e05 | 1918 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 1919 | queue_me(q, hb); |
ca5f9524 DH |
1920 | |
1921 | /* Arm the timer */ | |
1922 | if (timeout) { | |
1923 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); | |
1924 | if (!hrtimer_active(&timeout->timer)) | |
1925 | timeout->task = NULL; | |
1926 | } | |
1927 | ||
1928 | /* | |
0729e196 DH |
1929 | * If we have been removed from the hash list, then another task |
1930 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
1931 | */ |
1932 | if (likely(!plist_node_empty(&q->list))) { | |
1933 | /* | |
1934 | * If the timer has already expired, current will already be | |
1935 | * flagged for rescheduling. Only call schedule if there | |
1936 | * is no timeout, or if it has yet to expire. | |
1937 | */ | |
1938 | if (!timeout || timeout->task) | |
6fa3eb70 | 1939 | freezable_schedule(); |
ca5f9524 DH |
1940 | } |
1941 | __set_current_state(TASK_RUNNING); | |
1942 | } | |
1943 | ||
f801073f DH |
1944 | /** |
1945 | * futex_wait_setup() - Prepare to wait on a futex | |
1946 | * @uaddr: the futex userspace address | |
1947 | * @val: the expected value | |
b41277dc | 1948 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
1949 | * @q: the associated futex_q |
1950 | * @hb: storage for hash_bucket pointer to be returned to caller | |
1951 | * | |
1952 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
1953 | * compare it with the expected value. Handle atomic faults internally. | |
1954 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
1955 | * with no q.key reference on failure. | |
1956 | * | |
6c23cbbd RD |
1957 | * Return: |
1958 | * 0 - uaddr contains val and hb has been locked; | |
ca4a04cf | 1959 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked |
f801073f | 1960 | */ |
b41277dc | 1961 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 1962 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 1963 | { |
e2970f2f IM |
1964 | u32 uval; |
1965 | int ret; | |
1da177e4 | 1966 | |
1da177e4 | 1967 | /* |
b2d0994b | 1968 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
1969 | * Order is important: |
1970 | * | |
1971 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
1972 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
1973 | * | |
1974 | * The basic logical guarantee of a futex is that it blocks ONLY | |
1975 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
1976 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
1977 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
1978 | * cond(var) false, which would violate the guarantee. |
1979 | * | |
8fe8f545 ML |
1980 | * On the other hand, we insert q and release the hash-bucket only |
1981 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
1982 | * absorb a wakeup if *uaddr does not match the desired values | |
1983 | * while the syscall executes. | |
1da177e4 | 1984 | */ |
f801073f | 1985 | retry: |
9ea71503 | 1986 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ); |
f801073f | 1987 | if (unlikely(ret != 0)) |
a5a2a0c7 | 1988 | return ret; |
f801073f DH |
1989 | |
1990 | retry_private: | |
1991 | *hb = queue_lock(q); | |
1992 | ||
e2970f2f | 1993 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 1994 | |
f801073f DH |
1995 | if (ret) { |
1996 | queue_unlock(q, *hb); | |
1da177e4 | 1997 | |
e2970f2f | 1998 | ret = get_user(uval, uaddr); |
e4dc5b7a | 1999 | if (ret) |
f801073f | 2000 | goto out; |
1da177e4 | 2001 | |
b41277dc | 2002 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2003 | goto retry_private; |
2004 | ||
ae791a2d | 2005 | put_futex_key(&q->key); |
e4dc5b7a | 2006 | goto retry; |
1da177e4 | 2007 | } |
ca5f9524 | 2008 | |
f801073f DH |
2009 | if (uval != val) { |
2010 | queue_unlock(q, *hb); | |
2011 | ret = -EWOULDBLOCK; | |
2fff78c7 | 2012 | } |
1da177e4 | 2013 | |
f801073f DH |
2014 | out: |
2015 | if (ret) | |
ae791a2d | 2016 | put_futex_key(&q->key); |
f801073f DH |
2017 | return ret; |
2018 | } | |
2019 | ||
b41277dc DH |
2020 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
2021 | ktime_t *abs_time, u32 bitset) | |
f801073f DH |
2022 | { |
2023 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
2024 | struct restart_block *restart; |
2025 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 2026 | struct futex_q q = futex_q_init; |
f801073f DH |
2027 | int ret; |
2028 | ||
2029 | if (!bitset) | |
2030 | return -EINVAL; | |
f801073f DH |
2031 | q.bitset = bitset; |
2032 | ||
2033 | if (abs_time) { | |
2034 | to = &timeout; | |
2035 | ||
b41277dc DH |
2036 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2037 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2038 | HRTIMER_MODE_ABS); | |
f801073f DH |
2039 | hrtimer_init_sleeper(to, current); |
2040 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2041 | current->timer_slack_ns); | |
2042 | } | |
2043 | ||
d58e6576 | 2044 | retry: |
7ada876a DH |
2045 | /* |
2046 | * Prepare to wait on uaddr. On success, holds hb lock and increments | |
2047 | * q.key refs. | |
2048 | */ | |
b41277dc | 2049 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
2050 | if (ret) |
2051 | goto out; | |
2052 | ||
ca5f9524 | 2053 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 2054 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
2055 | |
2056 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 2057 | ret = 0; |
7ada876a | 2058 | /* unqueue_me() drops q.key ref */ |
1da177e4 | 2059 | if (!unqueue_me(&q)) |
7ada876a | 2060 | goto out; |
2fff78c7 | 2061 | ret = -ETIMEDOUT; |
ca5f9524 | 2062 | if (to && !to->task) |
7ada876a | 2063 | goto out; |
72c1bbf3 | 2064 | |
e2970f2f | 2065 | /* |
d58e6576 TG |
2066 | * We expect signal_pending(current), but we might be the |
2067 | * victim of a spurious wakeup as well. | |
e2970f2f | 2068 | */ |
7ada876a | 2069 | if (!signal_pending(current)) |
d58e6576 | 2070 | goto retry; |
d58e6576 | 2071 | |
2fff78c7 | 2072 | ret = -ERESTARTSYS; |
c19384b5 | 2073 | if (!abs_time) |
7ada876a | 2074 | goto out; |
1da177e4 | 2075 | |
2fff78c7 PZ |
2076 | restart = ¤t_thread_info()->restart_block; |
2077 | restart->fn = futex_wait_restart; | |
a3c74c52 | 2078 | restart->futex.uaddr = uaddr; |
2fff78c7 PZ |
2079 | restart->futex.val = val; |
2080 | restart->futex.time = abs_time->tv64; | |
2081 | restart->futex.bitset = bitset; | |
0cd9c649 | 2082 | restart->futex.flags = flags | FLAGS_HAS_TIMEOUT; |
42d35d48 | 2083 | |
2fff78c7 PZ |
2084 | ret = -ERESTART_RESTARTBLOCK; |
2085 | ||
42d35d48 | 2086 | out: |
ca5f9524 DH |
2087 | if (to) { |
2088 | hrtimer_cancel(&to->timer); | |
2089 | destroy_hrtimer_on_stack(&to->timer); | |
2090 | } | |
c87e2837 IM |
2091 | return ret; |
2092 | } | |
2093 | ||
72c1bbf3 NP |
2094 | |
2095 | static long futex_wait_restart(struct restart_block *restart) | |
2096 | { | |
a3c74c52 | 2097 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 2098 | ktime_t t, *tp = NULL; |
72c1bbf3 | 2099 | |
a72188d8 DH |
2100 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
2101 | t.tv64 = restart->futex.time; | |
2102 | tp = &t; | |
2103 | } | |
72c1bbf3 | 2104 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
2105 | |
2106 | return (long)futex_wait(uaddr, restart->futex.flags, | |
2107 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
2108 | } |
2109 | ||
2110 | ||
c87e2837 IM |
2111 | /* |
2112 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
2113 | * and failed. The kernel side here does the whole locking operation: | |
2114 | * if there are waiters then it will block, it does PI, etc. (Due to | |
2115 | * races the kernel might see a 0 value of the futex too.) | |
2116 | */ | |
b41277dc DH |
2117 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect, |
2118 | ktime_t *time, int trylock) | |
c87e2837 | 2119 | { |
c5780e97 | 2120 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 2121 | struct futex_hash_bucket *hb; |
5bdb05f9 | 2122 | struct futex_q q = futex_q_init; |
dd973998 | 2123 | int res, ret; |
c87e2837 IM |
2124 | |
2125 | if (refill_pi_state_cache()) | |
2126 | return -ENOMEM; | |
2127 | ||
c19384b5 | 2128 | if (time) { |
c5780e97 | 2129 | to = &timeout; |
237fc6e7 TG |
2130 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
2131 | HRTIMER_MODE_ABS); | |
c5780e97 | 2132 | hrtimer_init_sleeper(to, current); |
cc584b21 | 2133 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
2134 | } |
2135 | ||
42d35d48 | 2136 | retry: |
9ea71503 | 2137 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE); |
c87e2837 | 2138 | if (unlikely(ret != 0)) |
42d35d48 | 2139 | goto out; |
c87e2837 | 2140 | |
e4dc5b7a | 2141 | retry_private: |
82af7aca | 2142 | hb = queue_lock(&q); |
c87e2837 | 2143 | |
bab5bc9e | 2144 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 2145 | if (unlikely(ret)) { |
778e9a9c | 2146 | switch (ret) { |
1a52084d DH |
2147 | case 1: |
2148 | /* We got the lock. */ | |
2149 | ret = 0; | |
2150 | goto out_unlock_put_key; | |
2151 | case -EFAULT: | |
2152 | goto uaddr_faulted; | |
778e9a9c AK |
2153 | case -EAGAIN: |
2154 | /* | |
2155 | * Task is exiting and we just wait for the | |
2156 | * exit to complete. | |
2157 | */ | |
2158 | queue_unlock(&q, hb); | |
ae791a2d | 2159 | put_futex_key(&q.key); |
778e9a9c AK |
2160 | cond_resched(); |
2161 | goto retry; | |
778e9a9c | 2162 | default: |
42d35d48 | 2163 | goto out_unlock_put_key; |
c87e2837 | 2164 | } |
c87e2837 IM |
2165 | } |
2166 | ||
2167 | /* | |
2168 | * Only actually queue now that the atomic ops are done: | |
2169 | */ | |
82af7aca | 2170 | queue_me(&q, hb); |
c87e2837 | 2171 | |
c87e2837 IM |
2172 | WARN_ON(!q.pi_state); |
2173 | /* | |
2174 | * Block on the PI mutex: | |
2175 | */ | |
2176 | if (!trylock) | |
2177 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); | |
2178 | else { | |
2179 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); | |
2180 | /* Fixup the trylock return value: */ | |
2181 | ret = ret ? 0 : -EWOULDBLOCK; | |
2182 | } | |
2183 | ||
a99e4e41 | 2184 | spin_lock(q.lock_ptr); |
dd973998 DH |
2185 | /* |
2186 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2187 | * haven't already. | |
2188 | */ | |
ae791a2d | 2189 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 DH |
2190 | /* |
2191 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
2192 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
2193 | */ | |
2194 | if (res) | |
2195 | ret = (res < 0) ? res : 0; | |
c87e2837 | 2196 | |
e8f6386c | 2197 | /* |
dd973998 DH |
2198 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
2199 | * it and return the fault to userspace. | |
e8f6386c DH |
2200 | */ |
2201 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
2202 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
2203 | ||
778e9a9c AK |
2204 | /* Unqueue and drop the lock */ |
2205 | unqueue_me_pi(&q); | |
c87e2837 | 2206 | |
5ecb01cf | 2207 | goto out_put_key; |
c87e2837 | 2208 | |
42d35d48 | 2209 | out_unlock_put_key: |
c87e2837 IM |
2210 | queue_unlock(&q, hb); |
2211 | ||
42d35d48 | 2212 | out_put_key: |
ae791a2d | 2213 | put_futex_key(&q.key); |
42d35d48 | 2214 | out: |
237fc6e7 TG |
2215 | if (to) |
2216 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 2217 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 2218 | |
42d35d48 | 2219 | uaddr_faulted: |
778e9a9c AK |
2220 | queue_unlock(&q, hb); |
2221 | ||
d0725992 | 2222 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
2223 | if (ret) |
2224 | goto out_put_key; | |
c87e2837 | 2225 | |
b41277dc | 2226 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2227 | goto retry_private; |
2228 | ||
ae791a2d | 2229 | put_futex_key(&q.key); |
e4dc5b7a | 2230 | goto retry; |
c87e2837 IM |
2231 | } |
2232 | ||
c87e2837 IM |
2233 | /* |
2234 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
2235 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
2236 | * and do the rt-mutex unlock. | |
2237 | */ | |
b41277dc | 2238 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 IM |
2239 | { |
2240 | struct futex_hash_bucket *hb; | |
2241 | struct futex_q *this, *next; | |
ec92d082 | 2242 | struct plist_head *head; |
38d47c1b | 2243 | union futex_key key = FUTEX_KEY_INIT; |
c0c9ed15 | 2244 | u32 uval, vpid = task_pid_vnr(current); |
e4dc5b7a | 2245 | int ret; |
c87e2837 IM |
2246 | |
2247 | retry: | |
2248 | if (get_user(uval, uaddr)) | |
2249 | return -EFAULT; | |
2250 | /* | |
2251 | * We release only a lock we actually own: | |
2252 | */ | |
c0c9ed15 | 2253 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 2254 | return -EPERM; |
c87e2837 | 2255 | |
9ea71503 | 2256 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE); |
c87e2837 IM |
2257 | if (unlikely(ret != 0)) |
2258 | goto out; | |
2259 | ||
2260 | hb = hash_futex(&key); | |
2261 | spin_lock(&hb->lock); | |
2262 | ||
c87e2837 IM |
2263 | /* |
2264 | * To avoid races, try to do the TID -> 0 atomic transition | |
2265 | * again. If it succeeds then we can return without waking | |
9ad5dabd TG |
2266 | * anyone else up. We only try this if neither the waiters nor |
2267 | * the owner died bit are set. | |
c87e2837 | 2268 | */ |
9ad5dabd | 2269 | if (!(uval & ~FUTEX_TID_MASK) && |
37a9d912 | 2270 | cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0)) |
c87e2837 IM |
2271 | goto pi_faulted; |
2272 | /* | |
2273 | * Rare case: we managed to release the lock atomically, | |
2274 | * no need to wake anyone else up: | |
2275 | */ | |
c0c9ed15 | 2276 | if (unlikely(uval == vpid)) |
c87e2837 IM |
2277 | goto out_unlock; |
2278 | ||
2279 | /* | |
2280 | * Ok, other tasks may need to be woken up - check waiters | |
2281 | * and do the wakeup if necessary: | |
2282 | */ | |
2283 | head = &hb->chain; | |
2284 | ||
ec92d082 | 2285 | plist_for_each_entry_safe(this, next, head, list) { |
c87e2837 IM |
2286 | if (!match_futex (&this->key, &key)) |
2287 | continue; | |
2288 | ret = wake_futex_pi(uaddr, uval, this); | |
2289 | /* | |
2290 | * The atomic access to the futex value | |
2291 | * generated a pagefault, so retry the | |
2292 | * user-access and the wakeup: | |
2293 | */ | |
2294 | if (ret == -EFAULT) | |
2295 | goto pi_faulted; | |
2296 | goto out_unlock; | |
2297 | } | |
2298 | /* | |
2299 | * No waiters - kernel unlocks the futex: | |
2300 | */ | |
9ad5dabd TG |
2301 | ret = unlock_futex_pi(uaddr, uval); |
2302 | if (ret == -EFAULT) | |
2303 | goto pi_faulted; | |
c87e2837 IM |
2304 | |
2305 | out_unlock: | |
2306 | spin_unlock(&hb->lock); | |
ae791a2d | 2307 | put_futex_key(&key); |
c87e2837 | 2308 | |
42d35d48 | 2309 | out: |
c87e2837 IM |
2310 | return ret; |
2311 | ||
2312 | pi_faulted: | |
778e9a9c | 2313 | spin_unlock(&hb->lock); |
ae791a2d | 2314 | put_futex_key(&key); |
c87e2837 | 2315 | |
d0725992 | 2316 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2317 | if (!ret) |
c87e2837 IM |
2318 | goto retry; |
2319 | ||
1da177e4 LT |
2320 | return ret; |
2321 | } | |
2322 | ||
52400ba9 DH |
2323 | /** |
2324 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2325 | * @hb: the hash_bucket futex_q was original enqueued on | |
2326 | * @q: the futex_q woken while waiting to be requeued | |
2327 | * @key2: the futex_key of the requeue target futex | |
2328 | * @timeout: the timeout associated with the wait (NULL if none) | |
2329 | * | |
2330 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2331 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2332 | * the wakeup and return the appropriate error code to the caller. Must be | |
2333 | * called with the hb lock held. | |
2334 | * | |
6c23cbbd RD |
2335 | * Return: |
2336 | * 0 = no early wakeup detected; | |
2337 | * <0 = -ETIMEDOUT or -ERESTARTNOINTR | |
52400ba9 DH |
2338 | */ |
2339 | static inline | |
2340 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2341 | struct futex_q *q, union futex_key *key2, | |
2342 | struct hrtimer_sleeper *timeout) | |
2343 | { | |
2344 | int ret = 0; | |
2345 | ||
2346 | /* | |
2347 | * With the hb lock held, we avoid races while we process the wakeup. | |
2348 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2349 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2350 | * It can't be requeued from uaddr2 to something else since we don't | |
2351 | * support a PI aware source futex for requeue. | |
2352 | */ | |
2353 | if (!match_futex(&q->key, key2)) { | |
2354 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2355 | /* | |
2356 | * We were woken prior to requeue by a timeout or a signal. | |
2357 | * Unqueue the futex_q and determine which it was. | |
2358 | */ | |
2e12978a | 2359 | plist_del(&q->list, &hb->chain); |
52400ba9 | 2360 | |
d58e6576 | 2361 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 2362 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2363 | if (timeout && !timeout->task) |
2364 | ret = -ETIMEDOUT; | |
d58e6576 | 2365 | else if (signal_pending(current)) |
1c840c14 | 2366 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
2367 | } |
2368 | return ret; | |
2369 | } | |
2370 | ||
2371 | /** | |
2372 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 2373 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 2374 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
52400ba9 DH |
2375 | * the same type, no requeueing from private to shared, etc. |
2376 | * @val: the expected value of uaddr | |
2377 | * @abs_time: absolute timeout | |
56ec1607 | 2378 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
2379 | * @uaddr2: the pi futex we will take prior to returning to user-space |
2380 | * | |
2381 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
6f7b0a2a DH |
2382 | * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake |
2383 | * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to | |
2384 | * userspace. This ensures the rt_mutex maintains an owner when it has waiters; | |
2385 | * without one, the pi logic would not know which task to boost/deboost, if | |
2386 | * there was a need to. | |
52400ba9 DH |
2387 | * |
2388 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
6c23cbbd | 2389 | * via the following-- |
52400ba9 | 2390 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() |
cc6db4e6 DH |
2391 | * 2) wakeup on uaddr2 after a requeue |
2392 | * 3) signal | |
2393 | * 4) timeout | |
52400ba9 | 2394 | * |
cc6db4e6 | 2395 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
2396 | * |
2397 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2398 | * 5) successful lock | |
2399 | * 6) signal | |
2400 | * 7) timeout | |
2401 | * 8) other lock acquisition failure | |
2402 | * | |
cc6db4e6 | 2403 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
2404 | * |
2405 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2406 | * | |
6c23cbbd RD |
2407 | * Return: |
2408 | * 0 - On success; | |
52400ba9 DH |
2409 | * <0 - On error |
2410 | */ | |
b41277dc | 2411 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 2412 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 2413 | u32 __user *uaddr2) |
52400ba9 DH |
2414 | { |
2415 | struct hrtimer_sleeper timeout, *to = NULL; | |
2416 | struct rt_mutex_waiter rt_waiter; | |
52400ba9 | 2417 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
2418 | union futex_key key2 = FUTEX_KEY_INIT; |
2419 | struct futex_q q = futex_q_init; | |
52400ba9 | 2420 | int res, ret; |
52400ba9 | 2421 | |
6f7b0a2a DH |
2422 | if (uaddr == uaddr2) |
2423 | return -EINVAL; | |
2424 | ||
52400ba9 DH |
2425 | if (!bitset) |
2426 | return -EINVAL; | |
2427 | ||
2428 | if (abs_time) { | |
2429 | to = &timeout; | |
b41277dc DH |
2430 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2431 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2432 | HRTIMER_MODE_ABS); | |
52400ba9 DH |
2433 | hrtimer_init_sleeper(to, current); |
2434 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2435 | current->timer_slack_ns); | |
2436 | } | |
2437 | ||
2438 | /* | |
2439 | * The waiter is allocated on our stack, manipulated by the requeue | |
2440 | * code while we sleep on uaddr. | |
2441 | */ | |
2442 | debug_rt_mutex_init_waiter(&rt_waiter); | |
2443 | rt_waiter.task = NULL; | |
2444 | ||
9ea71503 | 2445 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
52400ba9 DH |
2446 | if (unlikely(ret != 0)) |
2447 | goto out; | |
2448 | ||
84bc4af5 DH |
2449 | q.bitset = bitset; |
2450 | q.rt_waiter = &rt_waiter; | |
2451 | q.requeue_pi_key = &key2; | |
2452 | ||
7ada876a DH |
2453 | /* |
2454 | * Prepare to wait on uaddr. On success, increments q.key (key1) ref | |
2455 | * count. | |
2456 | */ | |
b41277dc | 2457 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 TG |
2458 | if (ret) |
2459 | goto out_key2; | |
52400ba9 | 2460 | |
b58623fb TG |
2461 | /* |
2462 | * The check above which compares uaddrs is not sufficient for | |
2463 | * shared futexes. We need to compare the keys: | |
2464 | */ | |
2465 | if (match_futex(&q.key, &key2)) { | |
2466 | ret = -EINVAL; | |
2467 | goto out_put_keys; | |
2468 | } | |
2469 | ||
52400ba9 | 2470 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ |
f1a11e05 | 2471 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2472 | |
2473 | spin_lock(&hb->lock); | |
2474 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2475 | spin_unlock(&hb->lock); | |
2476 | if (ret) | |
2477 | goto out_put_keys; | |
2478 | ||
2479 | /* | |
2480 | * In order for us to be here, we know our q.key == key2, and since | |
2481 | * we took the hb->lock above, we also know that futex_requeue() has | |
2482 | * completed and we no longer have to concern ourselves with a wakeup | |
7ada876a DH |
2483 | * race with the atomic proxy lock acquisition by the requeue code. The |
2484 | * futex_requeue dropped our key1 reference and incremented our key2 | |
2485 | * reference count. | |
52400ba9 DH |
2486 | */ |
2487 | ||
2488 | /* Check if the requeue code acquired the second futex for us. */ | |
2489 | if (!q.rt_waiter) { | |
2490 | /* | |
2491 | * Got the lock. We might not be the anticipated owner if we | |
2492 | * did a lock-steal - fix up the PI-state in that case. | |
2493 | */ | |
2494 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2495 | spin_lock(q.lock_ptr); | |
ae791a2d | 2496 | ret = fixup_pi_state_owner(uaddr2, &q, current); |
4401c71d PZ |
2497 | if (ret && rt_mutex_owner(&q.pi_state->pi_mutex) == current) |
2498 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
a8957067 TG |
2499 | /* |
2500 | * Drop the reference to the pi state which | |
2501 | * the requeue_pi() code acquired for us. | |
2502 | */ | |
2503 | free_pi_state(q.pi_state); | |
52400ba9 DH |
2504 | spin_unlock(q.lock_ptr); |
2505 | } | |
2506 | } else { | |
8f4a52d2 PZ |
2507 | struct rt_mutex *pi_mutex; |
2508 | ||
52400ba9 DH |
2509 | /* |
2510 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2511 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2512 | * the pi_state. | |
2513 | */ | |
f27071cb | 2514 | WARN_ON(!q.pi_state); |
52400ba9 DH |
2515 | pi_mutex = &q.pi_state->pi_mutex; |
2516 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1); | |
2517 | debug_rt_mutex_free_waiter(&rt_waiter); | |
2518 | ||
2519 | spin_lock(q.lock_ptr); | |
2520 | /* | |
2521 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2522 | * haven't already. | |
2523 | */ | |
ae791a2d | 2524 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 DH |
2525 | /* |
2526 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 2527 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
2528 | */ |
2529 | if (res) | |
2530 | ret = (res < 0) ? res : 0; | |
2531 | ||
8f4a52d2 PZ |
2532 | /* |
2533 | * If fixup_pi_state_owner() faulted and was unable to handle | |
2534 | * the fault, unlock the rt_mutex and return the fault to | |
2535 | * userspace. | |
2536 | */ | |
2537 | if (ret && rt_mutex_owner(pi_mutex) == current) | |
2538 | rt_mutex_unlock(pi_mutex); | |
2539 | ||
52400ba9 DH |
2540 | /* Unqueue and drop the lock. */ |
2541 | unqueue_me_pi(&q); | |
2542 | } | |
2543 | ||
8f4a52d2 | 2544 | if (ret == -EINTR) { |
52400ba9 | 2545 | /* |
cc6db4e6 DH |
2546 | * We've already been requeued, but cannot restart by calling |
2547 | * futex_lock_pi() directly. We could restart this syscall, but | |
2548 | * it would detect that the user space "val" changed and return | |
2549 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
2550 | * -EWOULDBLOCK directly. | |
52400ba9 | 2551 | */ |
2070887f | 2552 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2553 | } |
2554 | ||
2555 | out_put_keys: | |
ae791a2d | 2556 | put_futex_key(&q.key); |
c8b15a70 | 2557 | out_key2: |
ae791a2d | 2558 | put_futex_key(&key2); |
52400ba9 DH |
2559 | |
2560 | out: | |
2561 | if (to) { | |
2562 | hrtimer_cancel(&to->timer); | |
2563 | destroy_hrtimer_on_stack(&to->timer); | |
2564 | } | |
2565 | return ret; | |
2566 | } | |
2567 | ||
0771dfef IM |
2568 | /* |
2569 | * Support for robust futexes: the kernel cleans up held futexes at | |
2570 | * thread exit time. | |
2571 | * | |
2572 | * Implementation: user-space maintains a per-thread list of locks it | |
2573 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2574 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2575 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2576 | * always manipulated with the lock held, so the list is private and |
2577 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2578 | * field, to allow the kernel to clean up if the thread dies after | |
2579 | * acquiring the lock, but just before it could have added itself to | |
2580 | * the list. There can only be one such pending lock. | |
2581 | */ | |
2582 | ||
2583 | /** | |
d96ee56c DH |
2584 | * sys_set_robust_list() - Set the robust-futex list head of a task |
2585 | * @head: pointer to the list-head | |
2586 | * @len: length of the list-head, as userspace expects | |
0771dfef | 2587 | */ |
836f92ad HC |
2588 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2589 | size_t, len) | |
0771dfef | 2590 | { |
a0c1e907 TG |
2591 | if (!futex_cmpxchg_enabled) |
2592 | return -ENOSYS; | |
0771dfef IM |
2593 | /* |
2594 | * The kernel knows only one size for now: | |
2595 | */ | |
2596 | if (unlikely(len != sizeof(*head))) | |
2597 | return -EINVAL; | |
2598 | ||
2599 | current->robust_list = head; | |
2600 | ||
2601 | return 0; | |
2602 | } | |
2603 | ||
2604 | /** | |
d96ee56c DH |
2605 | * sys_get_robust_list() - Get the robust-futex list head of a task |
2606 | * @pid: pid of the process [zero for current task] | |
2607 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
2608 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 2609 | */ |
836f92ad HC |
2610 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
2611 | struct robust_list_head __user * __user *, head_ptr, | |
2612 | size_t __user *, len_ptr) | |
0771dfef | 2613 | { |
ba46df98 | 2614 | struct robust_list_head __user *head; |
0771dfef | 2615 | unsigned long ret; |
bdbb776f | 2616 | struct task_struct *p; |
0771dfef | 2617 | |
a0c1e907 TG |
2618 | if (!futex_cmpxchg_enabled) |
2619 | return -ENOSYS; | |
2620 | ||
bdbb776f KC |
2621 | rcu_read_lock(); |
2622 | ||
2623 | ret = -ESRCH; | |
0771dfef | 2624 | if (!pid) |
bdbb776f | 2625 | p = current; |
0771dfef | 2626 | else { |
228ebcbe | 2627 | p = find_task_by_vpid(pid); |
0771dfef IM |
2628 | if (!p) |
2629 | goto err_unlock; | |
0771dfef IM |
2630 | } |
2631 | ||
bdbb776f | 2632 | ret = -EPERM; |
414f6fbc | 2633 | if (!ptrace_may_access(p, PTRACE_MODE_READ_REALCREDS)) |
bdbb776f KC |
2634 | goto err_unlock; |
2635 | ||
2636 | head = p->robust_list; | |
2637 | rcu_read_unlock(); | |
2638 | ||
0771dfef IM |
2639 | if (put_user(sizeof(*head), len_ptr)) |
2640 | return -EFAULT; | |
2641 | return put_user(head, head_ptr); | |
2642 | ||
2643 | err_unlock: | |
aaa2a97e | 2644 | rcu_read_unlock(); |
0771dfef IM |
2645 | |
2646 | return ret; | |
2647 | } | |
2648 | ||
2649 | /* | |
2650 | * Process a futex-list entry, check whether it's owned by the | |
2651 | * dying task, and do notification if so: | |
2652 | */ | |
e3f2ddea | 2653 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 2654 | { |
7cfdaf38 | 2655 | u32 uval, uninitialized_var(nval), mval; |
0771dfef | 2656 | |
8f17d3a5 IM |
2657 | retry: |
2658 | if (get_user(uval, uaddr)) | |
0771dfef IM |
2659 | return -1; |
2660 | ||
b488893a | 2661 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
2662 | /* |
2663 | * Ok, this dying thread is truly holding a futex | |
2664 | * of interest. Set the OWNER_DIED bit atomically | |
2665 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
2666 | * set, wake up a waiter (if any). (We have to do a | |
2667 | * futex_wake() even if OWNER_DIED is already set - | |
2668 | * to handle the rare but possible case of recursive | |
2669 | * thread-death.) The rest of the cleanup is done in | |
2670 | * userspace. | |
2671 | */ | |
e3f2ddea | 2672 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
6e0aa9f8 TG |
2673 | /* |
2674 | * We are not holding a lock here, but we want to have | |
2675 | * the pagefault_disable/enable() protection because | |
2676 | * we want to handle the fault gracefully. If the | |
2677 | * access fails we try to fault in the futex with R/W | |
2678 | * verification via get_user_pages. get_user() above | |
2679 | * does not guarantee R/W access. If that fails we | |
2680 | * give up and leave the futex locked. | |
2681 | */ | |
2682 | if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) { | |
2683 | if (fault_in_user_writeable(uaddr)) | |
2684 | return -1; | |
2685 | goto retry; | |
2686 | } | |
c87e2837 | 2687 | if (nval != uval) |
8f17d3a5 | 2688 | goto retry; |
0771dfef | 2689 | |
e3f2ddea IM |
2690 | /* |
2691 | * Wake robust non-PI futexes here. The wakeup of | |
2692 | * PI futexes happens in exit_pi_state(): | |
2693 | */ | |
36cf3b5c | 2694 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 2695 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
2696 | } |
2697 | return 0; | |
2698 | } | |
2699 | ||
e3f2ddea IM |
2700 | /* |
2701 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
2702 | */ | |
2703 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 2704 | struct robust_list __user * __user *head, |
1dcc41bb | 2705 | unsigned int *pi) |
e3f2ddea IM |
2706 | { |
2707 | unsigned long uentry; | |
2708 | ||
ba46df98 | 2709 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
2710 | return -EFAULT; |
2711 | ||
ba46df98 | 2712 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
2713 | *pi = uentry & 1; |
2714 | ||
2715 | return 0; | |
2716 | } | |
2717 | ||
0771dfef IM |
2718 | /* |
2719 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
2720 | * and mark any locks found there dead, and notify any waiters. | |
2721 | * | |
2722 | * We silently return on any sign of list-walking problem. | |
2723 | */ | |
2724 | void exit_robust_list(struct task_struct *curr) | |
2725 | { | |
2726 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 2727 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 DH |
2728 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
2729 | unsigned int uninitialized_var(next_pi); | |
0771dfef | 2730 | unsigned long futex_offset; |
9f96cb1e | 2731 | int rc; |
0771dfef | 2732 | |
a0c1e907 TG |
2733 | if (!futex_cmpxchg_enabled) |
2734 | return; | |
2735 | ||
0771dfef IM |
2736 | /* |
2737 | * Fetch the list head (which was registered earlier, via | |
2738 | * sys_set_robust_list()): | |
2739 | */ | |
e3f2ddea | 2740 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
2741 | return; |
2742 | /* | |
2743 | * Fetch the relative futex offset: | |
2744 | */ | |
2745 | if (get_user(futex_offset, &head->futex_offset)) | |
2746 | return; | |
2747 | /* | |
2748 | * Fetch any possibly pending lock-add first, and handle it | |
2749 | * if it exists: | |
2750 | */ | |
e3f2ddea | 2751 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 2752 | return; |
e3f2ddea | 2753 | |
9f96cb1e | 2754 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 2755 | while (entry != &head->list) { |
9f96cb1e MS |
2756 | /* |
2757 | * Fetch the next entry in the list before calling | |
2758 | * handle_futex_death: | |
2759 | */ | |
2760 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
2761 | /* |
2762 | * A pending lock might already be on the list, so | |
c87e2837 | 2763 | * don't process it twice: |
0771dfef IM |
2764 | */ |
2765 | if (entry != pending) | |
ba46df98 | 2766 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 2767 | curr, pi)) |
0771dfef | 2768 | return; |
9f96cb1e | 2769 | if (rc) |
0771dfef | 2770 | return; |
9f96cb1e MS |
2771 | entry = next_entry; |
2772 | pi = next_pi; | |
0771dfef IM |
2773 | /* |
2774 | * Avoid excessively long or circular lists: | |
2775 | */ | |
2776 | if (!--limit) | |
2777 | break; | |
2778 | ||
2779 | cond_resched(); | |
2780 | } | |
9f96cb1e MS |
2781 | |
2782 | if (pending) | |
2783 | handle_futex_death((void __user *)pending + futex_offset, | |
2784 | curr, pip); | |
0771dfef IM |
2785 | } |
2786 | ||
c19384b5 | 2787 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 2788 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 2789 | { |
81b40539 | 2790 | int cmd = op & FUTEX_CMD_MASK; |
b41277dc | 2791 | unsigned int flags = 0; |
34f01cc1 ED |
2792 | |
2793 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 2794 | flags |= FLAGS_SHARED; |
1da177e4 | 2795 | |
b41277dc DH |
2796 | if (op & FUTEX_CLOCK_REALTIME) { |
2797 | flags |= FLAGS_CLOCKRT; | |
2798 | if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI) | |
2799 | return -ENOSYS; | |
2800 | } | |
1da177e4 | 2801 | |
59263b51 TG |
2802 | switch (cmd) { |
2803 | case FUTEX_LOCK_PI: | |
2804 | case FUTEX_UNLOCK_PI: | |
2805 | case FUTEX_TRYLOCK_PI: | |
2806 | case FUTEX_WAIT_REQUEUE_PI: | |
2807 | case FUTEX_CMP_REQUEUE_PI: | |
2808 | if (!futex_cmpxchg_enabled) | |
2809 | return -ENOSYS; | |
2810 | } | |
2811 | ||
34f01cc1 | 2812 | switch (cmd) { |
1da177e4 | 2813 | case FUTEX_WAIT: |
cd689985 TG |
2814 | val3 = FUTEX_BITSET_MATCH_ANY; |
2815 | case FUTEX_WAIT_BITSET: | |
81b40539 | 2816 | return futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 | 2817 | case FUTEX_WAKE: |
cd689985 TG |
2818 | val3 = FUTEX_BITSET_MATCH_ANY; |
2819 | case FUTEX_WAKE_BITSET: | |
81b40539 | 2820 | return futex_wake(uaddr, flags, val, val3); |
1da177e4 | 2821 | case FUTEX_REQUEUE: |
81b40539 | 2822 | return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 | 2823 | case FUTEX_CMP_REQUEUE: |
81b40539 | 2824 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
4732efbe | 2825 | case FUTEX_WAKE_OP: |
81b40539 | 2826 | return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
c87e2837 | 2827 | case FUTEX_LOCK_PI: |
81b40539 | 2828 | return futex_lock_pi(uaddr, flags, val, timeout, 0); |
c87e2837 | 2829 | case FUTEX_UNLOCK_PI: |
81b40539 | 2830 | return futex_unlock_pi(uaddr, flags); |
c87e2837 | 2831 | case FUTEX_TRYLOCK_PI: |
81b40539 | 2832 | return futex_lock_pi(uaddr, flags, 0, timeout, 1); |
52400ba9 DH |
2833 | case FUTEX_WAIT_REQUEUE_PI: |
2834 | val3 = FUTEX_BITSET_MATCH_ANY; | |
81b40539 TG |
2835 | return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
2836 | uaddr2); | |
52400ba9 | 2837 | case FUTEX_CMP_REQUEUE_PI: |
81b40539 | 2838 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
1da177e4 | 2839 | } |
81b40539 | 2840 | return -ENOSYS; |
1da177e4 LT |
2841 | } |
2842 | ||
2843 | ||
17da2bd9 HC |
2844 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
2845 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
2846 | u32, val3) | |
1da177e4 | 2847 | { |
c19384b5 PP |
2848 | struct timespec ts; |
2849 | ktime_t t, *tp = NULL; | |
e2970f2f | 2850 | u32 val2 = 0; |
34f01cc1 | 2851 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 2852 | |
cd689985 | 2853 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
2854 | cmd == FUTEX_WAIT_BITSET || |
2855 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
c19384b5 | 2856 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 2857 | return -EFAULT; |
c19384b5 | 2858 | if (!timespec_valid(&ts)) |
9741ef96 | 2859 | return -EINVAL; |
c19384b5 PP |
2860 | |
2861 | t = timespec_to_ktime(ts); | |
34f01cc1 | 2862 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 2863 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 2864 | tp = &t; |
1da177e4 LT |
2865 | } |
2866 | /* | |
52400ba9 | 2867 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 2868 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 2869 | */ |
f54f0986 | 2870 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 2871 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 2872 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 2873 | |
c19384b5 | 2874 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
2875 | } |
2876 | ||
f26c70a4 | 2877 | static void __init futex_detect_cmpxchg(void) |
1da177e4 | 2878 | { |
f26c70a4 | 2879 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 2880 | u32 curval; |
95362fa9 | 2881 | |
a0c1e907 TG |
2882 | /* |
2883 | * This will fail and we want it. Some arch implementations do | |
2884 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
2885 | * functionality. We want to know that before we call in any | |
2886 | * of the complex code paths. Also we want to prevent | |
2887 | * registration of robust lists in that case. NULL is | |
2888 | * guaranteed to fault and we get -EFAULT on functional | |
fb62db2b | 2889 | * implementation, the non-functional ones will return |
a0c1e907 TG |
2890 | * -ENOSYS. |
2891 | */ | |
37a9d912 | 2892 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) |
a0c1e907 | 2893 | futex_cmpxchg_enabled = 1; |
f26c70a4 HC |
2894 | #endif |
2895 | } | |
2896 | ||
2897 | static int __init futex_init(void) | |
2898 | { | |
2899 | int i; | |
2900 | ||
2901 | futex_detect_cmpxchg(); | |
a0c1e907 | 2902 | |
3e4ab747 | 2903 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { |
732375c6 | 2904 | plist_head_init(&futex_queues[i].chain); |
3e4ab747 TG |
2905 | spin_lock_init(&futex_queues[i].lock); |
2906 | } | |
2907 | ||
1da177e4 LT |
2908 | return 0; |
2909 | } | |
bbd20686 | 2910 | core_initcall(futex_init); |