]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Fast Userspace Mutexes (which I call "Futexes!"). | |
3 | * (C) Rusty Russell, IBM 2002 | |
4 | * | |
5 | * Generalized futexes, futex requeueing, misc fixes by Ingo Molnar | |
6 | * (C) Copyright 2003 Red Hat Inc, All Rights Reserved | |
7 | * | |
8 | * Removed page pinning, fix privately mapped COW pages and other cleanups | |
9 | * (C) Copyright 2003, 2004 Jamie Lokier | |
10 | * | |
0771dfef IM |
11 | * Robust futex support started by Ingo Molnar |
12 | * (C) Copyright 2006 Red Hat Inc, All Rights Reserved | |
13 | * Thanks to Thomas Gleixner for suggestions, analysis and fixes. | |
14 | * | |
c87e2837 IM |
15 | * PI-futex support started by Ingo Molnar and Thomas Gleixner |
16 | * Copyright (C) 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com> | |
17 | * Copyright (C) 2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com> | |
18 | * | |
34f01cc1 ED |
19 | * PRIVATE futexes by Eric Dumazet |
20 | * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com> | |
21 | * | |
52400ba9 DH |
22 | * Requeue-PI support by Darren Hart <dvhltc@us.ibm.com> |
23 | * Copyright (C) IBM Corporation, 2009 | |
24 | * Thanks to Thomas Gleixner for conceptual design and careful reviews. | |
25 | * | |
1da177e4 LT |
26 | * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly |
27 | * enough at me, Linus for the original (flawed) idea, Matthew | |
28 | * Kirkwood for proof-of-concept implementation. | |
29 | * | |
30 | * "The futexes are also cursed." | |
31 | * "But they come in a choice of three flavours!" | |
32 | * | |
33 | * This program is free software; you can redistribute it and/or modify | |
34 | * it under the terms of the GNU General Public License as published by | |
35 | * the Free Software Foundation; either version 2 of the License, or | |
36 | * (at your option) any later version. | |
37 | * | |
38 | * This program is distributed in the hope that it will be useful, | |
39 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
40 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
41 | * GNU General Public License for more details. | |
42 | * | |
43 | * You should have received a copy of the GNU General Public License | |
44 | * along with this program; if not, write to the Free Software | |
45 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
46 | */ | |
47 | #include <linux/slab.h> | |
48 | #include <linux/poll.h> | |
49 | #include <linux/fs.h> | |
50 | #include <linux/file.h> | |
51 | #include <linux/jhash.h> | |
52 | #include <linux/init.h> | |
53 | #include <linux/futex.h> | |
54 | #include <linux/mount.h> | |
55 | #include <linux/pagemap.h> | |
56 | #include <linux/syscalls.h> | |
7ed20e1a | 57 | #include <linux/signal.h> |
9984de1a | 58 | #include <linux/export.h> |
fd5eea42 | 59 | #include <linux/magic.h> |
b488893a PE |
60 | #include <linux/pid.h> |
61 | #include <linux/nsproxy.h> | |
bdbb776f | 62 | #include <linux/ptrace.h> |
8bd75c77 | 63 | #include <linux/sched/rt.h> |
13d60f4b | 64 | #include <linux/hugetlb.h> |
88c8004f | 65 | #include <linux/freezer.h> |
a52b89eb | 66 | #include <linux/bootmem.h> |
b488893a | 67 | |
4732efbe | 68 | #include <asm/futex.h> |
1da177e4 | 69 | |
1696a8be | 70 | #include "locking/rtmutex_common.h" |
c87e2837 | 71 | |
99b60ce6 TG |
72 | /* |
73 | * Basic futex operation and ordering guarantees: | |
74 | * | |
75 | * The waiter reads the futex value in user space and calls | |
76 | * futex_wait(). This function computes the hash bucket and acquires | |
77 | * the hash bucket lock. After that it reads the futex user space value | |
78 | * again and verifies that the data has not changed. If it has not | |
79 | * changed it enqueues itself into the hash bucket, releases the hash | |
80 | * bucket lock and schedules. | |
81 | * | |
82 | * The waker side modifies the user space value of the futex and calls | |
83 | * futex_wake(). This functions computes the hash bucket and acquires | |
84 | * the hash bucket lock. Then it looks for waiters on that futex in the | |
85 | * hash bucket and wakes them. | |
86 | * | |
87 | * Note that the spin_lock serializes waiters and wakers, so that the | |
88 | * following scenario is avoided: | |
89 | * | |
90 | * CPU 0 CPU 1 | |
91 | * val = *futex; | |
92 | * sys_futex(WAIT, futex, val); | |
93 | * futex_wait(futex, val); | |
94 | * uval = *futex; | |
95 | * *futex = newval; | |
96 | * sys_futex(WAKE, futex); | |
97 | * futex_wake(futex); | |
98 | * if (queue_empty()) | |
99 | * return; | |
100 | * if (uval == val) | |
101 | * lock(hash_bucket(futex)); | |
102 | * queue(); | |
103 | * unlock(hash_bucket(futex)); | |
104 | * schedule(); | |
105 | * | |
106 | * This would cause the waiter on CPU 0 to wait forever because it | |
107 | * missed the transition of the user space value from val to newval | |
108 | * and the waker did not find the waiter in the hash bucket queue. | |
109 | * The spinlock serializes that: | |
110 | * | |
111 | * CPU 0 CPU 1 | |
112 | * val = *futex; | |
113 | * sys_futex(WAIT, futex, val); | |
114 | * futex_wait(futex, val); | |
115 | * lock(hash_bucket(futex)); | |
116 | * uval = *futex; | |
117 | * *futex = newval; | |
118 | * sys_futex(WAKE, futex); | |
119 | * futex_wake(futex); | |
120 | * lock(hash_bucket(futex)); | |
121 | * if (uval == val) | |
122 | * queue(); | |
123 | * unlock(hash_bucket(futex)); | |
124 | * schedule(); if (!queue_empty()) | |
125 | * wake_waiters(futex); | |
126 | * unlock(hash_bucket(futex)); | |
127 | */ | |
128 | ||
a0c1e907 TG |
129 | int __read_mostly futex_cmpxchg_enabled; |
130 | ||
b41277dc DH |
131 | /* |
132 | * Futex flags used to encode options to functions and preserve them across | |
133 | * restarts. | |
134 | */ | |
135 | #define FLAGS_SHARED 0x01 | |
136 | #define FLAGS_CLOCKRT 0x02 | |
137 | #define FLAGS_HAS_TIMEOUT 0x04 | |
138 | ||
c87e2837 IM |
139 | /* |
140 | * Priority Inheritance state: | |
141 | */ | |
142 | struct futex_pi_state { | |
143 | /* | |
144 | * list of 'owned' pi_state instances - these have to be | |
145 | * cleaned up in do_exit() if the task exits prematurely: | |
146 | */ | |
147 | struct list_head list; | |
148 | ||
149 | /* | |
150 | * The PI object: | |
151 | */ | |
152 | struct rt_mutex pi_mutex; | |
153 | ||
154 | struct task_struct *owner; | |
155 | atomic_t refcount; | |
156 | ||
157 | union futex_key key; | |
158 | }; | |
159 | ||
d8d88fbb DH |
160 | /** |
161 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 162 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
163 | * @task: the task waiting on the futex |
164 | * @lock_ptr: the hash bucket lock | |
165 | * @key: the key the futex is hashed on | |
166 | * @pi_state: optional priority inheritance state | |
167 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
168 | * @requeue_pi_key: the requeue_pi target futex key | |
169 | * @bitset: bitset for the optional bitmasked wakeup | |
170 | * | |
171 | * We use this hashed waitqueue, instead of a normal wait_queue_t, so | |
1da177e4 LT |
172 | * we can wake only the relevant ones (hashed queues may be shared). |
173 | * | |
174 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 175 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 176 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
177 | * the second. |
178 | * | |
179 | * PI futexes are typically woken before they are removed from the hash list via | |
180 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
181 | */ |
182 | struct futex_q { | |
ec92d082 | 183 | struct plist_node list; |
1da177e4 | 184 | |
d8d88fbb | 185 | struct task_struct *task; |
1da177e4 | 186 | spinlock_t *lock_ptr; |
1da177e4 | 187 | union futex_key key; |
c87e2837 | 188 | struct futex_pi_state *pi_state; |
52400ba9 | 189 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 190 | union futex_key *requeue_pi_key; |
cd689985 | 191 | u32 bitset; |
1da177e4 LT |
192 | }; |
193 | ||
5bdb05f9 DH |
194 | static const struct futex_q futex_q_init = { |
195 | /* list gets initialized in queue_me()*/ | |
196 | .key = FUTEX_KEY_INIT, | |
197 | .bitset = FUTEX_BITSET_MATCH_ANY | |
198 | }; | |
199 | ||
1da177e4 | 200 | /* |
b2d0994b DH |
201 | * Hash buckets are shared by all the futex_keys that hash to the same |
202 | * location. Each key may have multiple futex_q structures, one for each task | |
203 | * waiting on a futex. | |
1da177e4 LT |
204 | */ |
205 | struct futex_hash_bucket { | |
ec92d082 PP |
206 | spinlock_t lock; |
207 | struct plist_head chain; | |
a52b89eb | 208 | } ____cacheline_aligned_in_smp; |
1da177e4 | 209 | |
a52b89eb DB |
210 | static unsigned long __read_mostly futex_hashsize; |
211 | ||
212 | static struct futex_hash_bucket *futex_queues; | |
1da177e4 | 213 | |
1da177e4 LT |
214 | /* |
215 | * We hash on the keys returned from get_futex_key (see below). | |
216 | */ | |
217 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
218 | { | |
219 | u32 hash = jhash2((u32*)&key->both.word, | |
220 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
221 | key->both.offset); | |
a52b89eb | 222 | return &futex_queues[hash & (futex_hashsize - 1)]; |
1da177e4 LT |
223 | } |
224 | ||
225 | /* | |
226 | * Return 1 if two futex_keys are equal, 0 otherwise. | |
227 | */ | |
228 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
229 | { | |
2bc87203 DH |
230 | return (key1 && key2 |
231 | && key1->both.word == key2->both.word | |
1da177e4 LT |
232 | && key1->both.ptr == key2->both.ptr |
233 | && key1->both.offset == key2->both.offset); | |
234 | } | |
235 | ||
38d47c1b PZ |
236 | /* |
237 | * Take a reference to the resource addressed by a key. | |
238 | * Can be called while holding spinlocks. | |
239 | * | |
240 | */ | |
241 | static void get_futex_key_refs(union futex_key *key) | |
242 | { | |
243 | if (!key->both.ptr) | |
244 | return; | |
245 | ||
246 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
247 | case FUT_OFF_INODE: | |
7de9c6ee | 248 | ihold(key->shared.inode); |
38d47c1b PZ |
249 | break; |
250 | case FUT_OFF_MMSHARED: | |
251 | atomic_inc(&key->private.mm->mm_count); | |
252 | break; | |
253 | } | |
254 | } | |
255 | ||
256 | /* | |
257 | * Drop a reference to the resource addressed by a key. | |
258 | * The hash bucket spinlock must not be held. | |
259 | */ | |
260 | static void drop_futex_key_refs(union futex_key *key) | |
261 | { | |
90621c40 DH |
262 | if (!key->both.ptr) { |
263 | /* If we're here then we tried to put a key we failed to get */ | |
264 | WARN_ON_ONCE(1); | |
38d47c1b | 265 | return; |
90621c40 | 266 | } |
38d47c1b PZ |
267 | |
268 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { | |
269 | case FUT_OFF_INODE: | |
270 | iput(key->shared.inode); | |
271 | break; | |
272 | case FUT_OFF_MMSHARED: | |
273 | mmdrop(key->private.mm); | |
274 | break; | |
275 | } | |
276 | } | |
277 | ||
34f01cc1 | 278 | /** |
d96ee56c DH |
279 | * get_futex_key() - Get parameters which are the keys for a futex |
280 | * @uaddr: virtual address of the futex | |
281 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED | |
282 | * @key: address where result is stored. | |
9ea71503 SB |
283 | * @rw: mapping needs to be read/write (values: VERIFY_READ, |
284 | * VERIFY_WRITE) | |
34f01cc1 | 285 | * |
6c23cbbd RD |
286 | * Return: a negative error code or 0 |
287 | * | |
34f01cc1 | 288 | * The key words are stored in *key on success. |
1da177e4 | 289 | * |
6131ffaa | 290 | * For shared mappings, it's (page->index, file_inode(vma->vm_file), |
1da177e4 LT |
291 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
292 | * We can usually work out the index without swapping in the page. | |
293 | * | |
b2d0994b | 294 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 295 | */ |
64d1304a | 296 | static int |
9ea71503 | 297 | get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) |
1da177e4 | 298 | { |
e2970f2f | 299 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 300 | struct mm_struct *mm = current->mm; |
a5b338f2 | 301 | struct page *page, *page_head; |
9ea71503 | 302 | int err, ro = 0; |
1da177e4 LT |
303 | |
304 | /* | |
305 | * The futex address must be "naturally" aligned. | |
306 | */ | |
e2970f2f | 307 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 308 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 309 | return -EINVAL; |
e2970f2f | 310 | address -= key->both.offset; |
1da177e4 | 311 | |
5cdec2d8 LT |
312 | if (unlikely(!access_ok(rw, uaddr, sizeof(u32)))) |
313 | return -EFAULT; | |
314 | ||
34f01cc1 ED |
315 | /* |
316 | * PROCESS_PRIVATE futexes are fast. | |
317 | * As the mm cannot disappear under us and the 'key' only needs | |
318 | * virtual address, we dont even have to find the underlying vma. | |
319 | * Note : We do have to check 'uaddr' is a valid user address, | |
320 | * but access_ok() should be faster than find_vma() | |
321 | */ | |
322 | if (!fshared) { | |
34f01cc1 ED |
323 | key->private.mm = mm; |
324 | key->private.address = address; | |
42569c39 | 325 | get_futex_key_refs(key); |
34f01cc1 ED |
326 | return 0; |
327 | } | |
1da177e4 | 328 | |
38d47c1b | 329 | again: |
7485d0d3 | 330 | err = get_user_pages_fast(address, 1, 1, &page); |
9ea71503 SB |
331 | /* |
332 | * If write access is not required (eg. FUTEX_WAIT), try | |
333 | * and get read-only access. | |
334 | */ | |
335 | if (err == -EFAULT && rw == VERIFY_READ) { | |
336 | err = get_user_pages_fast(address, 1, 0, &page); | |
337 | ro = 1; | |
338 | } | |
38d47c1b PZ |
339 | if (err < 0) |
340 | return err; | |
9ea71503 SB |
341 | else |
342 | err = 0; | |
38d47c1b | 343 | |
a5b338f2 AA |
344 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
345 | page_head = page; | |
346 | if (unlikely(PageTail(page))) { | |
38d47c1b | 347 | put_page(page); |
a5b338f2 AA |
348 | /* serialize against __split_huge_page_splitting() */ |
349 | local_irq_disable(); | |
f12d5bfc | 350 | if (likely(__get_user_pages_fast(address, 1, !ro, &page) == 1)) { |
a5b338f2 AA |
351 | page_head = compound_head(page); |
352 | /* | |
353 | * page_head is valid pointer but we must pin | |
354 | * it before taking the PG_lock and/or | |
355 | * PG_compound_lock. The moment we re-enable | |
356 | * irqs __split_huge_page_splitting() can | |
357 | * return and the head page can be freed from | |
358 | * under us. We can't take the PG_lock and/or | |
359 | * PG_compound_lock on a page that could be | |
360 | * freed from under us. | |
361 | */ | |
362 | if (page != page_head) { | |
363 | get_page(page_head); | |
364 | put_page(page); | |
365 | } | |
366 | local_irq_enable(); | |
367 | } else { | |
368 | local_irq_enable(); | |
369 | goto again; | |
370 | } | |
371 | } | |
372 | #else | |
373 | page_head = compound_head(page); | |
374 | if (page != page_head) { | |
375 | get_page(page_head); | |
376 | put_page(page); | |
377 | } | |
378 | #endif | |
379 | ||
380 | lock_page(page_head); | |
e6780f72 HD |
381 | |
382 | /* | |
383 | * If page_head->mapping is NULL, then it cannot be a PageAnon | |
384 | * page; but it might be the ZERO_PAGE or in the gate area or | |
385 | * in a special mapping (all cases which we are happy to fail); | |
386 | * or it may have been a good file page when get_user_pages_fast | |
387 | * found it, but truncated or holepunched or subjected to | |
388 | * invalidate_complete_page2 before we got the page lock (also | |
389 | * cases which we are happy to fail). And we hold a reference, | |
390 | * so refcount care in invalidate_complete_page's remove_mapping | |
391 | * prevents drop_caches from setting mapping to NULL beneath us. | |
392 | * | |
393 | * The case we do have to guard against is when memory pressure made | |
394 | * shmem_writepage move it from filecache to swapcache beneath us: | |
395 | * an unlikely race, but we do need to retry for page_head->mapping. | |
396 | */ | |
a5b338f2 | 397 | if (!page_head->mapping) { |
e6780f72 | 398 | int shmem_swizzled = PageSwapCache(page_head); |
a5b338f2 AA |
399 | unlock_page(page_head); |
400 | put_page(page_head); | |
e6780f72 HD |
401 | if (shmem_swizzled) |
402 | goto again; | |
403 | return -EFAULT; | |
38d47c1b | 404 | } |
1da177e4 LT |
405 | |
406 | /* | |
407 | * Private mappings are handled in a simple way. | |
408 | * | |
409 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if | |
410 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 411 | * the object not the particular process. |
1da177e4 | 412 | */ |
a5b338f2 | 413 | if (PageAnon(page_head)) { |
9ea71503 SB |
414 | /* |
415 | * A RO anonymous page will never change and thus doesn't make | |
416 | * sense for futex operations. | |
417 | */ | |
418 | if (ro) { | |
419 | err = -EFAULT; | |
420 | goto out; | |
421 | } | |
422 | ||
38d47c1b | 423 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 424 | key->private.mm = mm; |
e2970f2f | 425 | key->private.address = address; |
38d47c1b PZ |
426 | } else { |
427 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ | |
a5b338f2 | 428 | key->shared.inode = page_head->mapping->host; |
13d60f4b | 429 | key->shared.pgoff = basepage_index(page); |
1da177e4 LT |
430 | } |
431 | ||
38d47c1b | 432 | get_futex_key_refs(key); |
1da177e4 | 433 | |
9ea71503 | 434 | out: |
a5b338f2 AA |
435 | unlock_page(page_head); |
436 | put_page(page_head); | |
9ea71503 | 437 | return err; |
1da177e4 LT |
438 | } |
439 | ||
ae791a2d | 440 | static inline void put_futex_key(union futex_key *key) |
1da177e4 | 441 | { |
38d47c1b | 442 | drop_futex_key_refs(key); |
1da177e4 LT |
443 | } |
444 | ||
d96ee56c DH |
445 | /** |
446 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
447 | * @uaddr: pointer to faulting user space address |
448 | * | |
449 | * Slow path to fixup the fault we just took in the atomic write | |
450 | * access to @uaddr. | |
451 | * | |
fb62db2b | 452 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
453 | * user address. We know that we faulted in the atomic pagefault |
454 | * disabled section so we can as well avoid the #PF overhead by | |
455 | * calling get_user_pages() right away. | |
456 | */ | |
457 | static int fault_in_user_writeable(u32 __user *uaddr) | |
458 | { | |
722d0172 AK |
459 | struct mm_struct *mm = current->mm; |
460 | int ret; | |
461 | ||
462 | down_read(&mm->mmap_sem); | |
2efaca92 BH |
463 | ret = fixup_user_fault(current, mm, (unsigned long)uaddr, |
464 | FAULT_FLAG_WRITE); | |
722d0172 AK |
465 | up_read(&mm->mmap_sem); |
466 | ||
d0725992 TG |
467 | return ret < 0 ? ret : 0; |
468 | } | |
469 | ||
4b1c486b DH |
470 | /** |
471 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
472 | * @hb: the hash bucket the futex_q's reside in |
473 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
474 | * |
475 | * Must be called with the hb lock held. | |
476 | */ | |
477 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
478 | union futex_key *key) | |
479 | { | |
480 | struct futex_q *this; | |
481 | ||
482 | plist_for_each_entry(this, &hb->chain, list) { | |
483 | if (match_futex(&this->key, key)) | |
484 | return this; | |
485 | } | |
486 | return NULL; | |
487 | } | |
488 | ||
37a9d912 ML |
489 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
490 | u32 uval, u32 newval) | |
36cf3b5c | 491 | { |
37a9d912 | 492 | int ret; |
36cf3b5c TG |
493 | |
494 | pagefault_disable(); | |
37a9d912 | 495 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
496 | pagefault_enable(); |
497 | ||
37a9d912 | 498 | return ret; |
36cf3b5c TG |
499 | } |
500 | ||
501 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
502 | { |
503 | int ret; | |
504 | ||
a866374a | 505 | pagefault_disable(); |
e2970f2f | 506 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
a866374a | 507 | pagefault_enable(); |
1da177e4 LT |
508 | |
509 | return ret ? -EFAULT : 0; | |
510 | } | |
511 | ||
c87e2837 IM |
512 | |
513 | /* | |
514 | * PI code: | |
515 | */ | |
516 | static int refill_pi_state_cache(void) | |
517 | { | |
518 | struct futex_pi_state *pi_state; | |
519 | ||
520 | if (likely(current->pi_state_cache)) | |
521 | return 0; | |
522 | ||
4668edc3 | 523 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
524 | |
525 | if (!pi_state) | |
526 | return -ENOMEM; | |
527 | ||
c87e2837 IM |
528 | INIT_LIST_HEAD(&pi_state->list); |
529 | /* pi_mutex gets initialized later */ | |
530 | pi_state->owner = NULL; | |
531 | atomic_set(&pi_state->refcount, 1); | |
38d47c1b | 532 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
533 | |
534 | current->pi_state_cache = pi_state; | |
535 | ||
536 | return 0; | |
537 | } | |
538 | ||
539 | static struct futex_pi_state * alloc_pi_state(void) | |
540 | { | |
541 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
542 | ||
543 | WARN_ON(!pi_state); | |
544 | current->pi_state_cache = NULL; | |
545 | ||
546 | return pi_state; | |
547 | } | |
548 | ||
549 | static void free_pi_state(struct futex_pi_state *pi_state) | |
550 | { | |
551 | if (!atomic_dec_and_test(&pi_state->refcount)) | |
552 | return; | |
553 | ||
554 | /* | |
555 | * If pi_state->owner is NULL, the owner is most probably dying | |
556 | * and has cleaned up the pi_state already | |
557 | */ | |
558 | if (pi_state->owner) { | |
1d615482 | 559 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
c87e2837 | 560 | list_del_init(&pi_state->list); |
1d615482 | 561 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
c87e2837 IM |
562 | |
563 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); | |
564 | } | |
565 | ||
566 | if (current->pi_state_cache) | |
567 | kfree(pi_state); | |
568 | else { | |
569 | /* | |
570 | * pi_state->list is already empty. | |
571 | * clear pi_state->owner. | |
572 | * refcount is at 0 - put it back to 1. | |
573 | */ | |
574 | pi_state->owner = NULL; | |
575 | atomic_set(&pi_state->refcount, 1); | |
576 | current->pi_state_cache = pi_state; | |
577 | } | |
578 | } | |
579 | ||
580 | /* | |
581 | * Look up the task based on what TID userspace gave us. | |
582 | * We dont trust it. | |
583 | */ | |
584 | static struct task_struct * futex_find_get_task(pid_t pid) | |
585 | { | |
586 | struct task_struct *p; | |
587 | ||
d359b549 | 588 | rcu_read_lock(); |
228ebcbe | 589 | p = find_task_by_vpid(pid); |
7a0ea09a MH |
590 | if (p) |
591 | get_task_struct(p); | |
a06381fe | 592 | |
d359b549 | 593 | rcu_read_unlock(); |
c87e2837 IM |
594 | |
595 | return p; | |
596 | } | |
597 | ||
598 | /* | |
599 | * This task is holding PI mutexes at exit time => bad. | |
600 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
601 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
602 | */ | |
603 | void exit_pi_state_list(struct task_struct *curr) | |
604 | { | |
c87e2837 IM |
605 | struct list_head *next, *head = &curr->pi_state_list; |
606 | struct futex_pi_state *pi_state; | |
627371d7 | 607 | struct futex_hash_bucket *hb; |
38d47c1b | 608 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 609 | |
a0c1e907 TG |
610 | if (!futex_cmpxchg_enabled) |
611 | return; | |
c87e2837 IM |
612 | /* |
613 | * We are a ZOMBIE and nobody can enqueue itself on | |
614 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 615 | * versus waiters unqueueing themselves: |
c87e2837 | 616 | */ |
1d615482 | 617 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 IM |
618 | while (!list_empty(head)) { |
619 | ||
620 | next = head->next; | |
621 | pi_state = list_entry(next, struct futex_pi_state, list); | |
622 | key = pi_state->key; | |
627371d7 | 623 | hb = hash_futex(&key); |
1d615482 | 624 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 625 | |
c87e2837 IM |
626 | spin_lock(&hb->lock); |
627 | ||
1d615482 | 628 | raw_spin_lock_irq(&curr->pi_lock); |
627371d7 IM |
629 | /* |
630 | * We dropped the pi-lock, so re-check whether this | |
631 | * task still owns the PI-state: | |
632 | */ | |
c87e2837 IM |
633 | if (head->next != next) { |
634 | spin_unlock(&hb->lock); | |
635 | continue; | |
636 | } | |
637 | ||
c87e2837 | 638 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
639 | WARN_ON(list_empty(&pi_state->list)); |
640 | list_del_init(&pi_state->list); | |
c87e2837 | 641 | pi_state->owner = NULL; |
1d615482 | 642 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
643 | |
644 | rt_mutex_unlock(&pi_state->pi_mutex); | |
645 | ||
646 | spin_unlock(&hb->lock); | |
647 | ||
1d615482 | 648 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 649 | } |
1d615482 | 650 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 IM |
651 | } |
652 | ||
653 | static int | |
d0aa7a70 PP |
654 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
655 | union futex_key *key, struct futex_pi_state **ps) | |
c87e2837 IM |
656 | { |
657 | struct futex_pi_state *pi_state = NULL; | |
658 | struct futex_q *this, *next; | |
c87e2837 | 659 | struct task_struct *p; |
778e9a9c | 660 | pid_t pid = uval & FUTEX_TID_MASK; |
c87e2837 | 661 | |
0d00c7b2 | 662 | plist_for_each_entry_safe(this, next, &hb->chain, list) { |
d0aa7a70 | 663 | if (match_futex(&this->key, key)) { |
c87e2837 IM |
664 | /* |
665 | * Another waiter already exists - bump up | |
666 | * the refcount and return its pi_state: | |
667 | */ | |
668 | pi_state = this->pi_state; | |
06a9ec29 | 669 | /* |
fb62db2b | 670 | * Userspace might have messed up non-PI and PI futexes |
06a9ec29 TG |
671 | */ |
672 | if (unlikely(!pi_state)) | |
673 | return -EINVAL; | |
674 | ||
627371d7 | 675 | WARN_ON(!atomic_read(&pi_state->refcount)); |
59647b6a TG |
676 | |
677 | /* | |
678 | * When pi_state->owner is NULL then the owner died | |
679 | * and another waiter is on the fly. pi_state->owner | |
680 | * is fixed up by the task which acquires | |
681 | * pi_state->rt_mutex. | |
682 | * | |
683 | * We do not check for pid == 0 which can happen when | |
684 | * the owner died and robust_list_exit() cleared the | |
685 | * TID. | |
686 | */ | |
687 | if (pid && pi_state->owner) { | |
688 | /* | |
689 | * Bail out if user space manipulated the | |
690 | * futex value. | |
691 | */ | |
692 | if (pid != task_pid_vnr(pi_state->owner)) | |
693 | return -EINVAL; | |
694 | } | |
627371d7 | 695 | |
c87e2837 | 696 | atomic_inc(&pi_state->refcount); |
d0aa7a70 | 697 | *ps = pi_state; |
c87e2837 IM |
698 | |
699 | return 0; | |
700 | } | |
701 | } | |
702 | ||
703 | /* | |
e3f2ddea | 704 | * We are the first waiter - try to look up the real owner and attach |
778e9a9c | 705 | * the new pi_state to it, but bail out when TID = 0 |
c87e2837 | 706 | */ |
778e9a9c | 707 | if (!pid) |
e3f2ddea | 708 | return -ESRCH; |
c87e2837 | 709 | p = futex_find_get_task(pid); |
7a0ea09a MH |
710 | if (!p) |
711 | return -ESRCH; | |
778e9a9c AK |
712 | |
713 | /* | |
714 | * We need to look at the task state flags to figure out, | |
715 | * whether the task is exiting. To protect against the do_exit | |
716 | * change of the task flags, we do this protected by | |
717 | * p->pi_lock: | |
718 | */ | |
1d615482 | 719 | raw_spin_lock_irq(&p->pi_lock); |
778e9a9c AK |
720 | if (unlikely(p->flags & PF_EXITING)) { |
721 | /* | |
722 | * The task is on the way out. When PF_EXITPIDONE is | |
723 | * set, we know that the task has finished the | |
724 | * cleanup: | |
725 | */ | |
726 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; | |
727 | ||
1d615482 | 728 | raw_spin_unlock_irq(&p->pi_lock); |
778e9a9c AK |
729 | put_task_struct(p); |
730 | return ret; | |
731 | } | |
c87e2837 IM |
732 | |
733 | pi_state = alloc_pi_state(); | |
734 | ||
735 | /* | |
736 | * Initialize the pi_mutex in locked state and make 'p' | |
737 | * the owner of it: | |
738 | */ | |
739 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
740 | ||
741 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 742 | pi_state->key = *key; |
c87e2837 | 743 | |
627371d7 | 744 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
745 | list_add(&pi_state->list, &p->pi_state_list); |
746 | pi_state->owner = p; | |
1d615482 | 747 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
748 | |
749 | put_task_struct(p); | |
750 | ||
d0aa7a70 | 751 | *ps = pi_state; |
c87e2837 IM |
752 | |
753 | return 0; | |
754 | } | |
755 | ||
1a52084d | 756 | /** |
d96ee56c | 757 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
758 | * @uaddr: the pi futex user address |
759 | * @hb: the pi futex hash bucket | |
760 | * @key: the futex key associated with uaddr and hb | |
761 | * @ps: the pi_state pointer where we store the result of the | |
762 | * lookup | |
763 | * @task: the task to perform the atomic lock work for. This will | |
764 | * be "current" except in the case of requeue pi. | |
765 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
1a52084d | 766 | * |
6c23cbbd RD |
767 | * Return: |
768 | * 0 - ready to wait; | |
769 | * 1 - acquired the lock; | |
1a52084d DH |
770 | * <0 - error |
771 | * | |
772 | * The hb->lock and futex_key refs shall be held by the caller. | |
773 | */ | |
774 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
775 | union futex_key *key, | |
776 | struct futex_pi_state **ps, | |
bab5bc9e | 777 | struct task_struct *task, int set_waiters) |
1a52084d | 778 | { |
59fa6245 | 779 | int lock_taken, ret, force_take = 0; |
c0c9ed15 | 780 | u32 uval, newval, curval, vpid = task_pid_vnr(task); |
1a52084d DH |
781 | |
782 | retry: | |
783 | ret = lock_taken = 0; | |
784 | ||
785 | /* | |
786 | * To avoid races, we attempt to take the lock here again | |
787 | * (by doing a 0 -> TID atomic cmpxchg), while holding all | |
788 | * the locks. It will most likely not succeed. | |
789 | */ | |
c0c9ed15 | 790 | newval = vpid; |
bab5bc9e DH |
791 | if (set_waiters) |
792 | newval |= FUTEX_WAITERS; | |
1a52084d | 793 | |
37a9d912 | 794 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval))) |
1a52084d DH |
795 | return -EFAULT; |
796 | ||
797 | /* | |
798 | * Detect deadlocks. | |
799 | */ | |
c0c9ed15 | 800 | if ((unlikely((curval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
801 | return -EDEADLK; |
802 | ||
803 | /* | |
804 | * Surprise - we got the lock. Just return to userspace: | |
805 | */ | |
806 | if (unlikely(!curval)) | |
807 | return 1; | |
808 | ||
809 | uval = curval; | |
810 | ||
811 | /* | |
812 | * Set the FUTEX_WAITERS flag, so the owner will know it has someone | |
813 | * to wake at the next unlock. | |
814 | */ | |
815 | newval = curval | FUTEX_WAITERS; | |
816 | ||
817 | /* | |
59fa6245 | 818 | * Should we force take the futex? See below. |
1a52084d | 819 | */ |
59fa6245 TG |
820 | if (unlikely(force_take)) { |
821 | /* | |
822 | * Keep the OWNER_DIED and the WAITERS bit and set the | |
823 | * new TID value. | |
824 | */ | |
c0c9ed15 | 825 | newval = (curval & ~FUTEX_TID_MASK) | vpid; |
59fa6245 | 826 | force_take = 0; |
1a52084d DH |
827 | lock_taken = 1; |
828 | } | |
829 | ||
37a9d912 | 830 | if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))) |
1a52084d DH |
831 | return -EFAULT; |
832 | if (unlikely(curval != uval)) | |
833 | goto retry; | |
834 | ||
835 | /* | |
59fa6245 | 836 | * We took the lock due to forced take over. |
1a52084d DH |
837 | */ |
838 | if (unlikely(lock_taken)) | |
839 | return 1; | |
840 | ||
841 | /* | |
842 | * We dont have the lock. Look up the PI state (or create it if | |
843 | * we are the first waiter): | |
844 | */ | |
845 | ret = lookup_pi_state(uval, hb, key, ps); | |
846 | ||
847 | if (unlikely(ret)) { | |
848 | switch (ret) { | |
849 | case -ESRCH: | |
850 | /* | |
59fa6245 TG |
851 | * We failed to find an owner for this |
852 | * futex. So we have no pi_state to block | |
853 | * on. This can happen in two cases: | |
854 | * | |
855 | * 1) The owner died | |
856 | * 2) A stale FUTEX_WAITERS bit | |
857 | * | |
858 | * Re-read the futex value. | |
1a52084d DH |
859 | */ |
860 | if (get_futex_value_locked(&curval, uaddr)) | |
861 | return -EFAULT; | |
862 | ||
863 | /* | |
59fa6245 TG |
864 | * If the owner died or we have a stale |
865 | * WAITERS bit the owner TID in the user space | |
866 | * futex is 0. | |
1a52084d | 867 | */ |
59fa6245 TG |
868 | if (!(curval & FUTEX_TID_MASK)) { |
869 | force_take = 1; | |
1a52084d DH |
870 | goto retry; |
871 | } | |
872 | default: | |
873 | break; | |
874 | } | |
875 | } | |
876 | ||
877 | return ret; | |
878 | } | |
879 | ||
2e12978a LJ |
880 | /** |
881 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
882 | * @q: The futex_q to unqueue | |
883 | * | |
884 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
885 | */ | |
886 | static void __unqueue_futex(struct futex_q *q) | |
887 | { | |
888 | struct futex_hash_bucket *hb; | |
889 | ||
29096202 SR |
890 | if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr)) |
891 | || WARN_ON(plist_node_empty(&q->list))) | |
2e12978a LJ |
892 | return; |
893 | ||
894 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
895 | plist_del(&q->list, &hb->chain); | |
896 | } | |
897 | ||
1da177e4 LT |
898 | /* |
899 | * The hash bucket lock must be held when this is called. | |
900 | * Afterwards, the futex_q must not be accessed. | |
901 | */ | |
902 | static void wake_futex(struct futex_q *q) | |
903 | { | |
f1a11e05 TG |
904 | struct task_struct *p = q->task; |
905 | ||
aa10990e DH |
906 | if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n")) |
907 | return; | |
908 | ||
1da177e4 | 909 | /* |
f1a11e05 | 910 | * We set q->lock_ptr = NULL _before_ we wake up the task. If |
fb62db2b RD |
911 | * a non-futex wake up happens on another CPU then the task |
912 | * might exit and p would dereference a non-existing task | |
f1a11e05 TG |
913 | * struct. Prevent this by holding a reference on p across the |
914 | * wake up. | |
1da177e4 | 915 | */ |
f1a11e05 TG |
916 | get_task_struct(p); |
917 | ||
2e12978a | 918 | __unqueue_futex(q); |
1da177e4 | 919 | /* |
f1a11e05 TG |
920 | * The waiting task can free the futex_q as soon as |
921 | * q->lock_ptr = NULL is written, without taking any locks. A | |
922 | * memory barrier is required here to prevent the following | |
923 | * store to lock_ptr from getting ahead of the plist_del. | |
1da177e4 | 924 | */ |
ccdea2f8 | 925 | smp_wmb(); |
1da177e4 | 926 | q->lock_ptr = NULL; |
f1a11e05 TG |
927 | |
928 | wake_up_state(p, TASK_NORMAL); | |
929 | put_task_struct(p); | |
1da177e4 LT |
930 | } |
931 | ||
c87e2837 IM |
932 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
933 | { | |
934 | struct task_struct *new_owner; | |
935 | struct futex_pi_state *pi_state = this->pi_state; | |
7cfdaf38 | 936 | u32 uninitialized_var(curval), newval; |
c87e2837 IM |
937 | |
938 | if (!pi_state) | |
939 | return -EINVAL; | |
940 | ||
51246bfd TG |
941 | /* |
942 | * If current does not own the pi_state then the futex is | |
943 | * inconsistent and user space fiddled with the futex value. | |
944 | */ | |
945 | if (pi_state->owner != current) | |
946 | return -EINVAL; | |
947 | ||
d209d74d | 948 | raw_spin_lock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
949 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
950 | ||
951 | /* | |
f123c98e SR |
952 | * It is possible that the next waiter (the one that brought |
953 | * this owner to the kernel) timed out and is no longer | |
954 | * waiting on the lock. | |
c87e2837 IM |
955 | */ |
956 | if (!new_owner) | |
957 | new_owner = this->task; | |
958 | ||
959 | /* | |
960 | * We pass it to the next owner. (The WAITERS bit is always | |
961 | * kept enabled while there is PI state around. We must also | |
962 | * preserve the owner died bit.) | |
963 | */ | |
e3f2ddea | 964 | if (!(uval & FUTEX_OWNER_DIED)) { |
778e9a9c AK |
965 | int ret = 0; |
966 | ||
b488893a | 967 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 968 | |
37a9d912 | 969 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
778e9a9c | 970 | ret = -EFAULT; |
cde898fa | 971 | else if (curval != uval) |
778e9a9c AK |
972 | ret = -EINVAL; |
973 | if (ret) { | |
d209d74d | 974 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
778e9a9c AK |
975 | return ret; |
976 | } | |
e3f2ddea | 977 | } |
c87e2837 | 978 | |
1d615482 | 979 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
627371d7 IM |
980 | WARN_ON(list_empty(&pi_state->list)); |
981 | list_del_init(&pi_state->list); | |
1d615482 | 982 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
627371d7 | 983 | |
1d615482 | 984 | raw_spin_lock_irq(&new_owner->pi_lock); |
627371d7 | 985 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 IM |
986 | list_add(&pi_state->list, &new_owner->pi_state_list); |
987 | pi_state->owner = new_owner; | |
1d615482 | 988 | raw_spin_unlock_irq(&new_owner->pi_lock); |
627371d7 | 989 | |
d209d74d | 990 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
991 | rt_mutex_unlock(&pi_state->pi_mutex); |
992 | ||
993 | return 0; | |
994 | } | |
995 | ||
996 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) | |
997 | { | |
7cfdaf38 | 998 | u32 uninitialized_var(oldval); |
c87e2837 IM |
999 | |
1000 | /* | |
1001 | * There is no waiter, so we unlock the futex. The owner died | |
1002 | * bit has not to be preserved here. We are the owner: | |
1003 | */ | |
37a9d912 ML |
1004 | if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0)) |
1005 | return -EFAULT; | |
c87e2837 IM |
1006 | if (oldval != uval) |
1007 | return -EAGAIN; | |
1008 | ||
1009 | return 0; | |
1010 | } | |
1011 | ||
8b8f319f IM |
1012 | /* |
1013 | * Express the locking dependencies for lockdep: | |
1014 | */ | |
1015 | static inline void | |
1016 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1017 | { | |
1018 | if (hb1 <= hb2) { | |
1019 | spin_lock(&hb1->lock); | |
1020 | if (hb1 < hb2) | |
1021 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
1022 | } else { /* hb1 > hb2 */ | |
1023 | spin_lock(&hb2->lock); | |
1024 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
1025 | } | |
1026 | } | |
1027 | ||
5eb3dc62 DH |
1028 | static inline void |
1029 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1030 | { | |
f061d351 | 1031 | spin_unlock(&hb1->lock); |
88f502fe IM |
1032 | if (hb1 != hb2) |
1033 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
1034 | } |
1035 | ||
1da177e4 | 1036 | /* |
b2d0994b | 1037 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 1038 | */ |
b41277dc DH |
1039 | static int |
1040 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 1041 | { |
e2970f2f | 1042 | struct futex_hash_bucket *hb; |
1da177e4 | 1043 | struct futex_q *this, *next; |
38d47c1b | 1044 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 LT |
1045 | int ret; |
1046 | ||
cd689985 TG |
1047 | if (!bitset) |
1048 | return -EINVAL; | |
1049 | ||
9ea71503 | 1050 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ); |
1da177e4 LT |
1051 | if (unlikely(ret != 0)) |
1052 | goto out; | |
1053 | ||
e2970f2f IM |
1054 | hb = hash_futex(&key); |
1055 | spin_lock(&hb->lock); | |
1da177e4 | 1056 | |
0d00c7b2 | 1057 | plist_for_each_entry_safe(this, next, &hb->chain, list) { |
1da177e4 | 1058 | if (match_futex (&this->key, &key)) { |
52400ba9 | 1059 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
1060 | ret = -EINVAL; |
1061 | break; | |
1062 | } | |
cd689985 TG |
1063 | |
1064 | /* Check if one of the bits is set in both bitsets */ | |
1065 | if (!(this->bitset & bitset)) | |
1066 | continue; | |
1067 | ||
1da177e4 LT |
1068 | wake_futex(this); |
1069 | if (++ret >= nr_wake) | |
1070 | break; | |
1071 | } | |
1072 | } | |
1073 | ||
e2970f2f | 1074 | spin_unlock(&hb->lock); |
ae791a2d | 1075 | put_futex_key(&key); |
42d35d48 | 1076 | out: |
1da177e4 LT |
1077 | return ret; |
1078 | } | |
1079 | ||
4732efbe JJ |
1080 | /* |
1081 | * Wake up all waiters hashed on the physical page that is mapped | |
1082 | * to this virtual address: | |
1083 | */ | |
e2970f2f | 1084 | static int |
b41277dc | 1085 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 1086 | int nr_wake, int nr_wake2, int op) |
4732efbe | 1087 | { |
38d47c1b | 1088 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 1089 | struct futex_hash_bucket *hb1, *hb2; |
4732efbe | 1090 | struct futex_q *this, *next; |
e4dc5b7a | 1091 | int ret, op_ret; |
4732efbe | 1092 | |
e4dc5b7a | 1093 | retry: |
9ea71503 | 1094 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
4732efbe JJ |
1095 | if (unlikely(ret != 0)) |
1096 | goto out; | |
9ea71503 | 1097 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
4732efbe | 1098 | if (unlikely(ret != 0)) |
42d35d48 | 1099 | goto out_put_key1; |
4732efbe | 1100 | |
e2970f2f IM |
1101 | hb1 = hash_futex(&key1); |
1102 | hb2 = hash_futex(&key2); | |
4732efbe | 1103 | |
e4dc5b7a | 1104 | retry_private: |
eaaea803 | 1105 | double_lock_hb(hb1, hb2); |
e2970f2f | 1106 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1107 | if (unlikely(op_ret < 0)) { |
4732efbe | 1108 | |
5eb3dc62 | 1109 | double_unlock_hb(hb1, hb2); |
4732efbe | 1110 | |
7ee1dd3f | 1111 | #ifndef CONFIG_MMU |
e2970f2f IM |
1112 | /* |
1113 | * we don't get EFAULT from MMU faults if we don't have an MMU, | |
1114 | * but we might get them from range checking | |
1115 | */ | |
7ee1dd3f | 1116 | ret = op_ret; |
42d35d48 | 1117 | goto out_put_keys; |
7ee1dd3f DH |
1118 | #endif |
1119 | ||
796f8d9b DG |
1120 | if (unlikely(op_ret != -EFAULT)) { |
1121 | ret = op_ret; | |
42d35d48 | 1122 | goto out_put_keys; |
796f8d9b DG |
1123 | } |
1124 | ||
d0725992 | 1125 | ret = fault_in_user_writeable(uaddr2); |
4732efbe | 1126 | if (ret) |
de87fcc1 | 1127 | goto out_put_keys; |
4732efbe | 1128 | |
b41277dc | 1129 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1130 | goto retry_private; |
1131 | ||
ae791a2d TG |
1132 | put_futex_key(&key2); |
1133 | put_futex_key(&key1); | |
e4dc5b7a | 1134 | goto retry; |
4732efbe JJ |
1135 | } |
1136 | ||
0d00c7b2 | 1137 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
4732efbe | 1138 | if (match_futex (&this->key, &key1)) { |
aa10990e DH |
1139 | if (this->pi_state || this->rt_waiter) { |
1140 | ret = -EINVAL; | |
1141 | goto out_unlock; | |
1142 | } | |
4732efbe JJ |
1143 | wake_futex(this); |
1144 | if (++ret >= nr_wake) | |
1145 | break; | |
1146 | } | |
1147 | } | |
1148 | ||
1149 | if (op_ret > 0) { | |
4732efbe | 1150 | op_ret = 0; |
0d00c7b2 | 1151 | plist_for_each_entry_safe(this, next, &hb2->chain, list) { |
4732efbe | 1152 | if (match_futex (&this->key, &key2)) { |
aa10990e DH |
1153 | if (this->pi_state || this->rt_waiter) { |
1154 | ret = -EINVAL; | |
1155 | goto out_unlock; | |
1156 | } | |
4732efbe JJ |
1157 | wake_futex(this); |
1158 | if (++op_ret >= nr_wake2) | |
1159 | break; | |
1160 | } | |
1161 | } | |
1162 | ret += op_ret; | |
1163 | } | |
1164 | ||
aa10990e | 1165 | out_unlock: |
5eb3dc62 | 1166 | double_unlock_hb(hb1, hb2); |
42d35d48 | 1167 | out_put_keys: |
ae791a2d | 1168 | put_futex_key(&key2); |
42d35d48 | 1169 | out_put_key1: |
ae791a2d | 1170 | put_futex_key(&key1); |
42d35d48 | 1171 | out: |
4732efbe JJ |
1172 | return ret; |
1173 | } | |
1174 | ||
9121e478 DH |
1175 | /** |
1176 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1177 | * @q: the futex_q to requeue | |
1178 | * @hb1: the source hash_bucket | |
1179 | * @hb2: the target hash_bucket | |
1180 | * @key2: the new key for the requeued futex_q | |
1181 | */ | |
1182 | static inline | |
1183 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1184 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1185 | { | |
1186 | ||
1187 | /* | |
1188 | * If key1 and key2 hash to the same bucket, no need to | |
1189 | * requeue. | |
1190 | */ | |
1191 | if (likely(&hb1->chain != &hb2->chain)) { | |
1192 | plist_del(&q->list, &hb1->chain); | |
1193 | plist_add(&q->list, &hb2->chain); | |
1194 | q->lock_ptr = &hb2->lock; | |
9121e478 DH |
1195 | } |
1196 | get_futex_key_refs(key2); | |
1197 | q->key = *key2; | |
1198 | } | |
1199 | ||
52400ba9 DH |
1200 | /** |
1201 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1202 | * @q: the futex_q |
1203 | * @key: the key of the requeue target futex | |
1204 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1205 | * |
1206 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1207 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1208 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1209 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1210 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1211 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1212 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1213 | */ |
1214 | static inline | |
beda2c7e DH |
1215 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1216 | struct futex_hash_bucket *hb) | |
52400ba9 | 1217 | { |
52400ba9 DH |
1218 | get_futex_key_refs(key); |
1219 | q->key = *key; | |
1220 | ||
2e12978a | 1221 | __unqueue_futex(q); |
52400ba9 DH |
1222 | |
1223 | WARN_ON(!q->rt_waiter); | |
1224 | q->rt_waiter = NULL; | |
1225 | ||
beda2c7e | 1226 | q->lock_ptr = &hb->lock; |
beda2c7e | 1227 | |
f1a11e05 | 1228 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1229 | } |
1230 | ||
1231 | /** | |
1232 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1233 | * @pifutex: the user address of the to futex |
1234 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1235 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1236 | * @key1: the from futex key | |
1237 | * @key2: the to futex key | |
1238 | * @ps: address to store the pi_state pointer | |
1239 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) | |
52400ba9 DH |
1240 | * |
1241 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1242 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1243 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1244 | * hb1 and hb2 must be held by the caller. | |
52400ba9 | 1245 | * |
6c23cbbd RD |
1246 | * Return: |
1247 | * 0 - failed to acquire the lock atomically; | |
1248 | * 1 - acquired the lock; | |
52400ba9 DH |
1249 | * <0 - error |
1250 | */ | |
1251 | static int futex_proxy_trylock_atomic(u32 __user *pifutex, | |
1252 | struct futex_hash_bucket *hb1, | |
1253 | struct futex_hash_bucket *hb2, | |
1254 | union futex_key *key1, union futex_key *key2, | |
bab5bc9e | 1255 | struct futex_pi_state **ps, int set_waiters) |
52400ba9 | 1256 | { |
bab5bc9e | 1257 | struct futex_q *top_waiter = NULL; |
52400ba9 DH |
1258 | u32 curval; |
1259 | int ret; | |
1260 | ||
1261 | if (get_futex_value_locked(&curval, pifutex)) | |
1262 | return -EFAULT; | |
1263 | ||
bab5bc9e DH |
1264 | /* |
1265 | * Find the top_waiter and determine if there are additional waiters. | |
1266 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1267 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1268 | * as we have means to handle the possible fault. If not, don't set | |
1269 | * the bit unecessarily as it will force the subsequent unlock to enter | |
1270 | * the kernel. | |
1271 | */ | |
52400ba9 DH |
1272 | top_waiter = futex_top_waiter(hb1, key1); |
1273 | ||
1274 | /* There are no waiters, nothing for us to do. */ | |
1275 | if (!top_waiter) | |
1276 | return 0; | |
1277 | ||
84bc4af5 DH |
1278 | /* Ensure we requeue to the expected futex. */ |
1279 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1280 | return -EINVAL; | |
1281 | ||
52400ba9 | 1282 | /* |
bab5bc9e DH |
1283 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1284 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1285 | * in ps in contended cases. | |
52400ba9 | 1286 | */ |
bab5bc9e DH |
1287 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
1288 | set_waiters); | |
52400ba9 | 1289 | if (ret == 1) |
beda2c7e | 1290 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
52400ba9 DH |
1291 | |
1292 | return ret; | |
1293 | } | |
1294 | ||
1295 | /** | |
1296 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1297 | * @uaddr1: source futex user address |
b41277dc | 1298 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1299 | * @uaddr2: target futex user address |
1300 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1301 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1302 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1303 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1304 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1305 | * |
1306 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1307 | * uaddr2 atomically on behalf of the top waiter. | |
1308 | * | |
6c23cbbd RD |
1309 | * Return: |
1310 | * >=0 - on success, the number of tasks requeued or woken; | |
52400ba9 | 1311 | * <0 - on error |
1da177e4 | 1312 | */ |
b41277dc DH |
1313 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
1314 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
1315 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 1316 | { |
38d47c1b | 1317 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
52400ba9 DH |
1318 | int drop_count = 0, task_count = 0, ret; |
1319 | struct futex_pi_state *pi_state = NULL; | |
e2970f2f | 1320 | struct futex_hash_bucket *hb1, *hb2; |
1da177e4 | 1321 | struct futex_q *this, *next; |
52400ba9 DH |
1322 | u32 curval2; |
1323 | ||
1324 | if (requeue_pi) { | |
1325 | /* | |
1326 | * requeue_pi requires a pi_state, try to allocate it now | |
1327 | * without any locks in case it fails. | |
1328 | */ | |
1329 | if (refill_pi_state_cache()) | |
1330 | return -ENOMEM; | |
1331 | /* | |
1332 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1333 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1334 | * returning to userspace, so as to not leave the rt_mutex with | |
1335 | * waiters and no owner. However, second and third wake-ups | |
1336 | * cannot be predicted as they involve race conditions with the | |
1337 | * first wake and a fault while looking up the pi_state. Both | |
1338 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1339 | * use nr_wake=1. | |
1340 | */ | |
1341 | if (nr_wake != 1) | |
1342 | return -EINVAL; | |
1343 | } | |
1da177e4 | 1344 | |
42d35d48 | 1345 | retry: |
52400ba9 DH |
1346 | if (pi_state != NULL) { |
1347 | /* | |
1348 | * We will have to lookup the pi_state again, so free this one | |
1349 | * to keep the accounting correct. | |
1350 | */ | |
1351 | free_pi_state(pi_state); | |
1352 | pi_state = NULL; | |
1353 | } | |
1354 | ||
9ea71503 | 1355 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ); |
1da177e4 LT |
1356 | if (unlikely(ret != 0)) |
1357 | goto out; | |
9ea71503 SB |
1358 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, |
1359 | requeue_pi ? VERIFY_WRITE : VERIFY_READ); | |
1da177e4 | 1360 | if (unlikely(ret != 0)) |
42d35d48 | 1361 | goto out_put_key1; |
1da177e4 | 1362 | |
e2970f2f IM |
1363 | hb1 = hash_futex(&key1); |
1364 | hb2 = hash_futex(&key2); | |
1da177e4 | 1365 | |
e4dc5b7a | 1366 | retry_private: |
8b8f319f | 1367 | double_lock_hb(hb1, hb2); |
1da177e4 | 1368 | |
e2970f2f IM |
1369 | if (likely(cmpval != NULL)) { |
1370 | u32 curval; | |
1da177e4 | 1371 | |
e2970f2f | 1372 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1373 | |
1374 | if (unlikely(ret)) { | |
5eb3dc62 | 1375 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1376 | |
e2970f2f | 1377 | ret = get_user(curval, uaddr1); |
e4dc5b7a DH |
1378 | if (ret) |
1379 | goto out_put_keys; | |
1da177e4 | 1380 | |
b41277dc | 1381 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 1382 | goto retry_private; |
1da177e4 | 1383 | |
ae791a2d TG |
1384 | put_futex_key(&key2); |
1385 | put_futex_key(&key1); | |
e4dc5b7a | 1386 | goto retry; |
1da177e4 | 1387 | } |
e2970f2f | 1388 | if (curval != *cmpval) { |
1da177e4 LT |
1389 | ret = -EAGAIN; |
1390 | goto out_unlock; | |
1391 | } | |
1392 | } | |
1393 | ||
52400ba9 | 1394 | if (requeue_pi && (task_count - nr_wake < nr_requeue)) { |
bab5bc9e DH |
1395 | /* |
1396 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
1397 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
1398 | * bit. We force this here where we are able to easily handle | |
1399 | * faults rather in the requeue loop below. | |
1400 | */ | |
52400ba9 | 1401 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
bab5bc9e | 1402 | &key2, &pi_state, nr_requeue); |
52400ba9 DH |
1403 | |
1404 | /* | |
1405 | * At this point the top_waiter has either taken uaddr2 or is | |
1406 | * waiting on it. If the former, then the pi_state will not | |
1407 | * exist yet, look it up one more time to ensure we have a | |
1408 | * reference to it. | |
1409 | */ | |
1410 | if (ret == 1) { | |
1411 | WARN_ON(pi_state); | |
89061d3d | 1412 | drop_count++; |
52400ba9 DH |
1413 | task_count++; |
1414 | ret = get_futex_value_locked(&curval2, uaddr2); | |
1415 | if (!ret) | |
1416 | ret = lookup_pi_state(curval2, hb2, &key2, | |
1417 | &pi_state); | |
1418 | } | |
1419 | ||
1420 | switch (ret) { | |
1421 | case 0: | |
1422 | break; | |
1423 | case -EFAULT: | |
1424 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1425 | put_futex_key(&key2); |
1426 | put_futex_key(&key1); | |
d0725992 | 1427 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
1428 | if (!ret) |
1429 | goto retry; | |
1430 | goto out; | |
1431 | case -EAGAIN: | |
1432 | /* The owner was exiting, try again. */ | |
1433 | double_unlock_hb(hb1, hb2); | |
ae791a2d TG |
1434 | put_futex_key(&key2); |
1435 | put_futex_key(&key1); | |
52400ba9 DH |
1436 | cond_resched(); |
1437 | goto retry; | |
1438 | default: | |
1439 | goto out_unlock; | |
1440 | } | |
1441 | } | |
1442 | ||
0d00c7b2 | 1443 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
52400ba9 DH |
1444 | if (task_count - nr_wake >= nr_requeue) |
1445 | break; | |
1446 | ||
1447 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 1448 | continue; |
52400ba9 | 1449 | |
392741e0 DH |
1450 | /* |
1451 | * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always | |
1452 | * be paired with each other and no other futex ops. | |
aa10990e DH |
1453 | * |
1454 | * We should never be requeueing a futex_q with a pi_state, | |
1455 | * which is awaiting a futex_unlock_pi(). | |
392741e0 DH |
1456 | */ |
1457 | if ((requeue_pi && !this->rt_waiter) || | |
aa10990e DH |
1458 | (!requeue_pi && this->rt_waiter) || |
1459 | this->pi_state) { | |
392741e0 DH |
1460 | ret = -EINVAL; |
1461 | break; | |
1462 | } | |
52400ba9 DH |
1463 | |
1464 | /* | |
1465 | * Wake nr_wake waiters. For requeue_pi, if we acquired the | |
1466 | * lock, we already woke the top_waiter. If not, it will be | |
1467 | * woken by futex_unlock_pi(). | |
1468 | */ | |
1469 | if (++task_count <= nr_wake && !requeue_pi) { | |
1da177e4 | 1470 | wake_futex(this); |
52400ba9 DH |
1471 | continue; |
1472 | } | |
1da177e4 | 1473 | |
84bc4af5 DH |
1474 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
1475 | if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) { | |
1476 | ret = -EINVAL; | |
1477 | break; | |
1478 | } | |
1479 | ||
52400ba9 DH |
1480 | /* |
1481 | * Requeue nr_requeue waiters and possibly one more in the case | |
1482 | * of requeue_pi if we couldn't acquire the lock atomically. | |
1483 | */ | |
1484 | if (requeue_pi) { | |
1485 | /* Prepare the waiter to take the rt_mutex. */ | |
1486 | atomic_inc(&pi_state->refcount); | |
1487 | this->pi_state = pi_state; | |
1488 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
1489 | this->rt_waiter, | |
1490 | this->task, 1); | |
1491 | if (ret == 1) { | |
1492 | /* We got the lock. */ | |
beda2c7e | 1493 | requeue_pi_wake_futex(this, &key2, hb2); |
89061d3d | 1494 | drop_count++; |
52400ba9 DH |
1495 | continue; |
1496 | } else if (ret) { | |
1497 | /* -EDEADLK */ | |
1498 | this->pi_state = NULL; | |
1499 | free_pi_state(pi_state); | |
1500 | goto out_unlock; | |
1501 | } | |
1da177e4 | 1502 | } |
52400ba9 DH |
1503 | requeue_futex(this, hb1, hb2, &key2); |
1504 | drop_count++; | |
1da177e4 LT |
1505 | } |
1506 | ||
1507 | out_unlock: | |
5eb3dc62 | 1508 | double_unlock_hb(hb1, hb2); |
1da177e4 | 1509 | |
cd84a42f DH |
1510 | /* |
1511 | * drop_futex_key_refs() must be called outside the spinlocks. During | |
1512 | * the requeue we moved futex_q's from the hash bucket at key1 to the | |
1513 | * one at key2 and updated their key pointer. We no longer need to | |
1514 | * hold the references to key1. | |
1515 | */ | |
1da177e4 | 1516 | while (--drop_count >= 0) |
9adef58b | 1517 | drop_futex_key_refs(&key1); |
1da177e4 | 1518 | |
42d35d48 | 1519 | out_put_keys: |
ae791a2d | 1520 | put_futex_key(&key2); |
42d35d48 | 1521 | out_put_key1: |
ae791a2d | 1522 | put_futex_key(&key1); |
42d35d48 | 1523 | out: |
52400ba9 DH |
1524 | if (pi_state != NULL) |
1525 | free_pi_state(pi_state); | |
1526 | return ret ? ret : task_count; | |
1da177e4 LT |
1527 | } |
1528 | ||
1529 | /* The key must be already stored in q->key. */ | |
82af7aca | 1530 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 1531 | __acquires(&hb->lock) |
1da177e4 | 1532 | { |
e2970f2f | 1533 | struct futex_hash_bucket *hb; |
1da177e4 | 1534 | |
e2970f2f IM |
1535 | hb = hash_futex(&q->key); |
1536 | q->lock_ptr = &hb->lock; | |
1da177e4 | 1537 | |
e2970f2f IM |
1538 | spin_lock(&hb->lock); |
1539 | return hb; | |
1da177e4 LT |
1540 | } |
1541 | ||
d40d65c8 | 1542 | static inline void |
0d00c7b2 | 1543 | queue_unlock(struct futex_hash_bucket *hb) |
15e408cd | 1544 | __releases(&hb->lock) |
d40d65c8 DH |
1545 | { |
1546 | spin_unlock(&hb->lock); | |
d40d65c8 DH |
1547 | } |
1548 | ||
1549 | /** | |
1550 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
1551 | * @q: The futex_q to enqueue | |
1552 | * @hb: The destination hash bucket | |
1553 | * | |
1554 | * The hb->lock must be held by the caller, and is released here. A call to | |
1555 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
1556 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
1557 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
1558 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
1559 | * an example). | |
1560 | */ | |
82af7aca | 1561 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
15e408cd | 1562 | __releases(&hb->lock) |
1da177e4 | 1563 | { |
ec92d082 PP |
1564 | int prio; |
1565 | ||
1566 | /* | |
1567 | * The priority used to register this element is | |
1568 | * - either the real thread-priority for the real-time threads | |
1569 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
1570 | * - or MAX_RT_PRIO for non-RT threads. | |
1571 | * Thus, all RT-threads are woken first in priority order, and | |
1572 | * the others are woken last, in FIFO order. | |
1573 | */ | |
1574 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
1575 | ||
1576 | plist_node_init(&q->list, prio); | |
ec92d082 | 1577 | plist_add(&q->list, &hb->chain); |
c87e2837 | 1578 | q->task = current; |
e2970f2f | 1579 | spin_unlock(&hb->lock); |
1da177e4 LT |
1580 | } |
1581 | ||
d40d65c8 DH |
1582 | /** |
1583 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
1584 | * @q: The futex_q to unqueue | |
1585 | * | |
1586 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
1587 | * be paired with exactly one earlier call to queue_me(). | |
1588 | * | |
6c23cbbd RD |
1589 | * Return: |
1590 | * 1 - if the futex_q was still queued (and we removed unqueued it); | |
d40d65c8 | 1591 | * 0 - if the futex_q was already removed by the waking thread |
1da177e4 | 1592 | */ |
1da177e4 LT |
1593 | static int unqueue_me(struct futex_q *q) |
1594 | { | |
1da177e4 | 1595 | spinlock_t *lock_ptr; |
e2970f2f | 1596 | int ret = 0; |
1da177e4 LT |
1597 | |
1598 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 1599 | retry: |
1da177e4 | 1600 | lock_ptr = q->lock_ptr; |
e91467ec | 1601 | barrier(); |
c80544dc | 1602 | if (lock_ptr != NULL) { |
1da177e4 LT |
1603 | spin_lock(lock_ptr); |
1604 | /* | |
1605 | * q->lock_ptr can change between reading it and | |
1606 | * spin_lock(), causing us to take the wrong lock. This | |
1607 | * corrects the race condition. | |
1608 | * | |
1609 | * Reasoning goes like this: if we have the wrong lock, | |
1610 | * q->lock_ptr must have changed (maybe several times) | |
1611 | * between reading it and the spin_lock(). It can | |
1612 | * change again after the spin_lock() but only if it was | |
1613 | * already changed before the spin_lock(). It cannot, | |
1614 | * however, change back to the original value. Therefore | |
1615 | * we can detect whether we acquired the correct lock. | |
1616 | */ | |
1617 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
1618 | spin_unlock(lock_ptr); | |
1619 | goto retry; | |
1620 | } | |
2e12978a | 1621 | __unqueue_futex(q); |
c87e2837 IM |
1622 | |
1623 | BUG_ON(q->pi_state); | |
1624 | ||
1da177e4 LT |
1625 | spin_unlock(lock_ptr); |
1626 | ret = 1; | |
1627 | } | |
1628 | ||
9adef58b | 1629 | drop_futex_key_refs(&q->key); |
1da177e4 LT |
1630 | return ret; |
1631 | } | |
1632 | ||
c87e2837 IM |
1633 | /* |
1634 | * PI futexes can not be requeued and must remove themself from the | |
d0aa7a70 PP |
1635 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
1636 | * and dropped here. | |
c87e2837 | 1637 | */ |
d0aa7a70 | 1638 | static void unqueue_me_pi(struct futex_q *q) |
15e408cd | 1639 | __releases(q->lock_ptr) |
c87e2837 | 1640 | { |
2e12978a | 1641 | __unqueue_futex(q); |
c87e2837 IM |
1642 | |
1643 | BUG_ON(!q->pi_state); | |
1644 | free_pi_state(q->pi_state); | |
1645 | q->pi_state = NULL; | |
1646 | ||
d0aa7a70 | 1647 | spin_unlock(q->lock_ptr); |
c87e2837 IM |
1648 | } |
1649 | ||
d0aa7a70 | 1650 | /* |
cdf71a10 | 1651 | * Fixup the pi_state owner with the new owner. |
d0aa7a70 | 1652 | * |
778e9a9c AK |
1653 | * Must be called with hash bucket lock held and mm->sem held for non |
1654 | * private futexes. | |
d0aa7a70 | 1655 | */ |
778e9a9c | 1656 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
ae791a2d | 1657 | struct task_struct *newowner) |
d0aa7a70 | 1658 | { |
cdf71a10 | 1659 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
d0aa7a70 | 1660 | struct futex_pi_state *pi_state = q->pi_state; |
1b7558e4 | 1661 | struct task_struct *oldowner = pi_state->owner; |
7cfdaf38 | 1662 | u32 uval, uninitialized_var(curval), newval; |
e4dc5b7a | 1663 | int ret; |
d0aa7a70 PP |
1664 | |
1665 | /* Owner died? */ | |
1b7558e4 TG |
1666 | if (!pi_state->owner) |
1667 | newtid |= FUTEX_OWNER_DIED; | |
1668 | ||
1669 | /* | |
1670 | * We are here either because we stole the rtmutex from the | |
8161239a LJ |
1671 | * previous highest priority waiter or we are the highest priority |
1672 | * waiter but failed to get the rtmutex the first time. | |
1673 | * We have to replace the newowner TID in the user space variable. | |
1674 | * This must be atomic as we have to preserve the owner died bit here. | |
1b7558e4 | 1675 | * |
b2d0994b DH |
1676 | * Note: We write the user space value _before_ changing the pi_state |
1677 | * because we can fault here. Imagine swapped out pages or a fork | |
1678 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 TG |
1679 | * |
1680 | * Modifying pi_state _before_ the user space value would | |
1681 | * leave the pi_state in an inconsistent state when we fault | |
1682 | * here, because we need to drop the hash bucket lock to | |
1683 | * handle the fault. This might be observed in the PID check | |
1684 | * in lookup_pi_state. | |
1685 | */ | |
1686 | retry: | |
1687 | if (get_futex_value_locked(&uval, uaddr)) | |
1688 | goto handle_fault; | |
1689 | ||
1690 | while (1) { | |
1691 | newval = (uval & FUTEX_OWNER_DIED) | newtid; | |
1692 | ||
37a9d912 | 1693 | if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) |
1b7558e4 TG |
1694 | goto handle_fault; |
1695 | if (curval == uval) | |
1696 | break; | |
1697 | uval = curval; | |
1698 | } | |
1699 | ||
1700 | /* | |
1701 | * We fixed up user space. Now we need to fix the pi_state | |
1702 | * itself. | |
1703 | */ | |
d0aa7a70 | 1704 | if (pi_state->owner != NULL) { |
1d615482 | 1705 | raw_spin_lock_irq(&pi_state->owner->pi_lock); |
d0aa7a70 PP |
1706 | WARN_ON(list_empty(&pi_state->list)); |
1707 | list_del_init(&pi_state->list); | |
1d615482 | 1708 | raw_spin_unlock_irq(&pi_state->owner->pi_lock); |
1b7558e4 | 1709 | } |
d0aa7a70 | 1710 | |
cdf71a10 | 1711 | pi_state->owner = newowner; |
d0aa7a70 | 1712 | |
1d615482 | 1713 | raw_spin_lock_irq(&newowner->pi_lock); |
d0aa7a70 | 1714 | WARN_ON(!list_empty(&pi_state->list)); |
cdf71a10 | 1715 | list_add(&pi_state->list, &newowner->pi_state_list); |
1d615482 | 1716 | raw_spin_unlock_irq(&newowner->pi_lock); |
1b7558e4 | 1717 | return 0; |
d0aa7a70 | 1718 | |
d0aa7a70 | 1719 | /* |
1b7558e4 | 1720 | * To handle the page fault we need to drop the hash bucket |
8161239a LJ |
1721 | * lock here. That gives the other task (either the highest priority |
1722 | * waiter itself or the task which stole the rtmutex) the | |
1b7558e4 TG |
1723 | * chance to try the fixup of the pi_state. So once we are |
1724 | * back from handling the fault we need to check the pi_state | |
1725 | * after reacquiring the hash bucket lock and before trying to | |
1726 | * do another fixup. When the fixup has been done already we | |
1727 | * simply return. | |
d0aa7a70 | 1728 | */ |
1b7558e4 TG |
1729 | handle_fault: |
1730 | spin_unlock(q->lock_ptr); | |
778e9a9c | 1731 | |
d0725992 | 1732 | ret = fault_in_user_writeable(uaddr); |
778e9a9c | 1733 | |
1b7558e4 | 1734 | spin_lock(q->lock_ptr); |
778e9a9c | 1735 | |
1b7558e4 TG |
1736 | /* |
1737 | * Check if someone else fixed it for us: | |
1738 | */ | |
1739 | if (pi_state->owner != oldowner) | |
1740 | return 0; | |
1741 | ||
1742 | if (ret) | |
1743 | return ret; | |
1744 | ||
1745 | goto retry; | |
d0aa7a70 PP |
1746 | } |
1747 | ||
72c1bbf3 | 1748 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 1749 | |
dd973998 DH |
1750 | /** |
1751 | * fixup_owner() - Post lock pi_state and corner case management | |
1752 | * @uaddr: user address of the futex | |
dd973998 DH |
1753 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
1754 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
1755 | * | |
1756 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
1757 | * the pi_state owner as well as handle race conditions that may allow us to | |
1758 | * acquire the lock. Must be called with the hb lock held. | |
1759 | * | |
6c23cbbd RD |
1760 | * Return: |
1761 | * 1 - success, lock taken; | |
1762 | * 0 - success, lock not taken; | |
dd973998 DH |
1763 | * <0 - on error (-EFAULT) |
1764 | */ | |
ae791a2d | 1765 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 DH |
1766 | { |
1767 | struct task_struct *owner; | |
1768 | int ret = 0; | |
1769 | ||
1770 | if (locked) { | |
1771 | /* | |
1772 | * Got the lock. We might not be the anticipated owner if we | |
1773 | * did a lock-steal - fix up the PI-state in that case: | |
1774 | */ | |
1775 | if (q->pi_state->owner != current) | |
ae791a2d | 1776 | ret = fixup_pi_state_owner(uaddr, q, current); |
dd973998 DH |
1777 | goto out; |
1778 | } | |
1779 | ||
1780 | /* | |
1781 | * Catch the rare case, where the lock was released when we were on the | |
1782 | * way back before we locked the hash bucket. | |
1783 | */ | |
1784 | if (q->pi_state->owner == current) { | |
1785 | /* | |
1786 | * Try to get the rt_mutex now. This might fail as some other | |
1787 | * task acquired the rt_mutex after we removed ourself from the | |
1788 | * rt_mutex waiters list. | |
1789 | */ | |
1790 | if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { | |
1791 | locked = 1; | |
1792 | goto out; | |
1793 | } | |
1794 | ||
1795 | /* | |
1796 | * pi_state is incorrect, some other task did a lock steal and | |
1797 | * we returned due to timeout or signal without taking the | |
8161239a | 1798 | * rt_mutex. Too late. |
dd973998 | 1799 | */ |
8161239a | 1800 | raw_spin_lock(&q->pi_state->pi_mutex.wait_lock); |
dd973998 | 1801 | owner = rt_mutex_owner(&q->pi_state->pi_mutex); |
8161239a LJ |
1802 | if (!owner) |
1803 | owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); | |
1804 | raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock); | |
ae791a2d | 1805 | ret = fixup_pi_state_owner(uaddr, q, owner); |
dd973998 DH |
1806 | goto out; |
1807 | } | |
1808 | ||
1809 | /* | |
1810 | * Paranoia check. If we did not take the lock, then we should not be | |
8161239a | 1811 | * the owner of the rt_mutex. |
dd973998 DH |
1812 | */ |
1813 | if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) | |
1814 | printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " | |
1815 | "pi-state %p\n", ret, | |
1816 | q->pi_state->pi_mutex.owner, | |
1817 | q->pi_state->owner); | |
1818 | ||
1819 | out: | |
1820 | return ret ? ret : locked; | |
1821 | } | |
1822 | ||
ca5f9524 DH |
1823 | /** |
1824 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
1825 | * @hb: the futex hash bucket, must be locked by the caller | |
1826 | * @q: the futex_q to queue up on | |
1827 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
1828 | */ |
1829 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 1830 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 1831 | { |
9beba3c5 DH |
1832 | /* |
1833 | * The task state is guaranteed to be set before another task can | |
1834 | * wake it. set_current_state() is implemented using set_mb() and | |
1835 | * queue_me() calls spin_unlock() upon completion, both serializing | |
1836 | * access to the hash list and forcing another memory barrier. | |
1837 | */ | |
f1a11e05 | 1838 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 1839 | queue_me(q, hb); |
ca5f9524 DH |
1840 | |
1841 | /* Arm the timer */ | |
1842 | if (timeout) { | |
1843 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); | |
1844 | if (!hrtimer_active(&timeout->timer)) | |
1845 | timeout->task = NULL; | |
1846 | } | |
1847 | ||
1848 | /* | |
0729e196 DH |
1849 | * If we have been removed from the hash list, then another task |
1850 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
1851 | */ |
1852 | if (likely(!plist_node_empty(&q->list))) { | |
1853 | /* | |
1854 | * If the timer has already expired, current will already be | |
1855 | * flagged for rescheduling. Only call schedule if there | |
1856 | * is no timeout, or if it has yet to expire. | |
1857 | */ | |
1858 | if (!timeout || timeout->task) | |
88c8004f | 1859 | freezable_schedule(); |
ca5f9524 DH |
1860 | } |
1861 | __set_current_state(TASK_RUNNING); | |
1862 | } | |
1863 | ||
f801073f DH |
1864 | /** |
1865 | * futex_wait_setup() - Prepare to wait on a futex | |
1866 | * @uaddr: the futex userspace address | |
1867 | * @val: the expected value | |
b41277dc | 1868 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
1869 | * @q: the associated futex_q |
1870 | * @hb: storage for hash_bucket pointer to be returned to caller | |
1871 | * | |
1872 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
1873 | * compare it with the expected value. Handle atomic faults internally. | |
1874 | * Return with the hb lock held and a q.key reference on success, and unlocked | |
1875 | * with no q.key reference on failure. | |
1876 | * | |
6c23cbbd RD |
1877 | * Return: |
1878 | * 0 - uaddr contains val and hb has been locked; | |
ca4a04cf | 1879 | * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked |
f801073f | 1880 | */ |
b41277dc | 1881 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 1882 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 1883 | { |
e2970f2f IM |
1884 | u32 uval; |
1885 | int ret; | |
1da177e4 | 1886 | |
1da177e4 | 1887 | /* |
b2d0994b | 1888 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
1889 | * Order is important: |
1890 | * | |
1891 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
1892 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
1893 | * | |
1894 | * The basic logical guarantee of a futex is that it blocks ONLY | |
1895 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
1896 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
1897 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
1898 | * cond(var) false, which would violate the guarantee. |
1899 | * | |
8fe8f545 ML |
1900 | * On the other hand, we insert q and release the hash-bucket only |
1901 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
1902 | * absorb a wakeup if *uaddr does not match the desired values | |
1903 | * while the syscall executes. | |
1da177e4 | 1904 | */ |
f801073f | 1905 | retry: |
9ea71503 | 1906 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ); |
f801073f | 1907 | if (unlikely(ret != 0)) |
a5a2a0c7 | 1908 | return ret; |
f801073f DH |
1909 | |
1910 | retry_private: | |
1911 | *hb = queue_lock(q); | |
1912 | ||
e2970f2f | 1913 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 1914 | |
f801073f | 1915 | if (ret) { |
0d00c7b2 | 1916 | queue_unlock(*hb); |
1da177e4 | 1917 | |
e2970f2f | 1918 | ret = get_user(uval, uaddr); |
e4dc5b7a | 1919 | if (ret) |
f801073f | 1920 | goto out; |
1da177e4 | 1921 | |
b41277dc | 1922 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
1923 | goto retry_private; |
1924 | ||
ae791a2d | 1925 | put_futex_key(&q->key); |
e4dc5b7a | 1926 | goto retry; |
1da177e4 | 1927 | } |
ca5f9524 | 1928 | |
f801073f | 1929 | if (uval != val) { |
0d00c7b2 | 1930 | queue_unlock(*hb); |
f801073f | 1931 | ret = -EWOULDBLOCK; |
2fff78c7 | 1932 | } |
1da177e4 | 1933 | |
f801073f DH |
1934 | out: |
1935 | if (ret) | |
ae791a2d | 1936 | put_futex_key(&q->key); |
f801073f DH |
1937 | return ret; |
1938 | } | |
1939 | ||
b41277dc DH |
1940 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
1941 | ktime_t *abs_time, u32 bitset) | |
f801073f DH |
1942 | { |
1943 | struct hrtimer_sleeper timeout, *to = NULL; | |
f801073f DH |
1944 | struct restart_block *restart; |
1945 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 1946 | struct futex_q q = futex_q_init; |
f801073f DH |
1947 | int ret; |
1948 | ||
1949 | if (!bitset) | |
1950 | return -EINVAL; | |
f801073f DH |
1951 | q.bitset = bitset; |
1952 | ||
1953 | if (abs_time) { | |
1954 | to = &timeout; | |
1955 | ||
b41277dc DH |
1956 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
1957 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
1958 | HRTIMER_MODE_ABS); | |
f801073f DH |
1959 | hrtimer_init_sleeper(to, current); |
1960 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
1961 | current->timer_slack_ns); | |
1962 | } | |
1963 | ||
d58e6576 | 1964 | retry: |
7ada876a DH |
1965 | /* |
1966 | * Prepare to wait on uaddr. On success, holds hb lock and increments | |
1967 | * q.key refs. | |
1968 | */ | |
b41277dc | 1969 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
1970 | if (ret) |
1971 | goto out; | |
1972 | ||
ca5f9524 | 1973 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 1974 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
1975 | |
1976 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 1977 | ret = 0; |
7ada876a | 1978 | /* unqueue_me() drops q.key ref */ |
1da177e4 | 1979 | if (!unqueue_me(&q)) |
7ada876a | 1980 | goto out; |
2fff78c7 | 1981 | ret = -ETIMEDOUT; |
ca5f9524 | 1982 | if (to && !to->task) |
7ada876a | 1983 | goto out; |
72c1bbf3 | 1984 | |
e2970f2f | 1985 | /* |
d58e6576 TG |
1986 | * We expect signal_pending(current), but we might be the |
1987 | * victim of a spurious wakeup as well. | |
e2970f2f | 1988 | */ |
7ada876a | 1989 | if (!signal_pending(current)) |
d58e6576 | 1990 | goto retry; |
d58e6576 | 1991 | |
2fff78c7 | 1992 | ret = -ERESTARTSYS; |
c19384b5 | 1993 | if (!abs_time) |
7ada876a | 1994 | goto out; |
1da177e4 | 1995 | |
2fff78c7 PZ |
1996 | restart = ¤t_thread_info()->restart_block; |
1997 | restart->fn = futex_wait_restart; | |
a3c74c52 | 1998 | restart->futex.uaddr = uaddr; |
2fff78c7 PZ |
1999 | restart->futex.val = val; |
2000 | restart->futex.time = abs_time->tv64; | |
2001 | restart->futex.bitset = bitset; | |
0cd9c649 | 2002 | restart->futex.flags = flags | FLAGS_HAS_TIMEOUT; |
42d35d48 | 2003 | |
2fff78c7 PZ |
2004 | ret = -ERESTART_RESTARTBLOCK; |
2005 | ||
42d35d48 | 2006 | out: |
ca5f9524 DH |
2007 | if (to) { |
2008 | hrtimer_cancel(&to->timer); | |
2009 | destroy_hrtimer_on_stack(&to->timer); | |
2010 | } | |
c87e2837 IM |
2011 | return ret; |
2012 | } | |
2013 | ||
72c1bbf3 NP |
2014 | |
2015 | static long futex_wait_restart(struct restart_block *restart) | |
2016 | { | |
a3c74c52 | 2017 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 2018 | ktime_t t, *tp = NULL; |
72c1bbf3 | 2019 | |
a72188d8 DH |
2020 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
2021 | t.tv64 = restart->futex.time; | |
2022 | tp = &t; | |
2023 | } | |
72c1bbf3 | 2024 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
2025 | |
2026 | return (long)futex_wait(uaddr, restart->futex.flags, | |
2027 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
2028 | } |
2029 | ||
2030 | ||
c87e2837 IM |
2031 | /* |
2032 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
2033 | * and failed. The kernel side here does the whole locking operation: | |
2034 | * if there are waiters then it will block, it does PI, etc. (Due to | |
2035 | * races the kernel might see a 0 value of the futex too.) | |
2036 | */ | |
b41277dc DH |
2037 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect, |
2038 | ktime_t *time, int trylock) | |
c87e2837 | 2039 | { |
c5780e97 | 2040 | struct hrtimer_sleeper timeout, *to = NULL; |
c87e2837 | 2041 | struct futex_hash_bucket *hb; |
5bdb05f9 | 2042 | struct futex_q q = futex_q_init; |
dd973998 | 2043 | int res, ret; |
c87e2837 IM |
2044 | |
2045 | if (refill_pi_state_cache()) | |
2046 | return -ENOMEM; | |
2047 | ||
c19384b5 | 2048 | if (time) { |
c5780e97 | 2049 | to = &timeout; |
237fc6e7 TG |
2050 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
2051 | HRTIMER_MODE_ABS); | |
c5780e97 | 2052 | hrtimer_init_sleeper(to, current); |
cc584b21 | 2053 | hrtimer_set_expires(&to->timer, *time); |
c5780e97 TG |
2054 | } |
2055 | ||
42d35d48 | 2056 | retry: |
9ea71503 | 2057 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE); |
c87e2837 | 2058 | if (unlikely(ret != 0)) |
42d35d48 | 2059 | goto out; |
c87e2837 | 2060 | |
e4dc5b7a | 2061 | retry_private: |
82af7aca | 2062 | hb = queue_lock(&q); |
c87e2837 | 2063 | |
bab5bc9e | 2064 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); |
c87e2837 | 2065 | if (unlikely(ret)) { |
778e9a9c | 2066 | switch (ret) { |
1a52084d DH |
2067 | case 1: |
2068 | /* We got the lock. */ | |
2069 | ret = 0; | |
2070 | goto out_unlock_put_key; | |
2071 | case -EFAULT: | |
2072 | goto uaddr_faulted; | |
778e9a9c AK |
2073 | case -EAGAIN: |
2074 | /* | |
2075 | * Task is exiting and we just wait for the | |
2076 | * exit to complete. | |
2077 | */ | |
0d00c7b2 | 2078 | queue_unlock(hb); |
ae791a2d | 2079 | put_futex_key(&q.key); |
778e9a9c AK |
2080 | cond_resched(); |
2081 | goto retry; | |
778e9a9c | 2082 | default: |
42d35d48 | 2083 | goto out_unlock_put_key; |
c87e2837 | 2084 | } |
c87e2837 IM |
2085 | } |
2086 | ||
2087 | /* | |
2088 | * Only actually queue now that the atomic ops are done: | |
2089 | */ | |
82af7aca | 2090 | queue_me(&q, hb); |
c87e2837 | 2091 | |
c87e2837 IM |
2092 | WARN_ON(!q.pi_state); |
2093 | /* | |
2094 | * Block on the PI mutex: | |
2095 | */ | |
2096 | if (!trylock) | |
2097 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); | |
2098 | else { | |
2099 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); | |
2100 | /* Fixup the trylock return value: */ | |
2101 | ret = ret ? 0 : -EWOULDBLOCK; | |
2102 | } | |
2103 | ||
a99e4e41 | 2104 | spin_lock(q.lock_ptr); |
dd973998 DH |
2105 | /* |
2106 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2107 | * haven't already. | |
2108 | */ | |
ae791a2d | 2109 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 DH |
2110 | /* |
2111 | * If fixup_owner() returned an error, proprogate that. If it acquired | |
2112 | * the lock, clear our -ETIMEDOUT or -EINTR. | |
2113 | */ | |
2114 | if (res) | |
2115 | ret = (res < 0) ? res : 0; | |
c87e2837 | 2116 | |
e8f6386c | 2117 | /* |
dd973998 DH |
2118 | * If fixup_owner() faulted and was unable to handle the fault, unlock |
2119 | * it and return the fault to userspace. | |
e8f6386c DH |
2120 | */ |
2121 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) | |
2122 | rt_mutex_unlock(&q.pi_state->pi_mutex); | |
2123 | ||
778e9a9c AK |
2124 | /* Unqueue and drop the lock */ |
2125 | unqueue_me_pi(&q); | |
c87e2837 | 2126 | |
5ecb01cf | 2127 | goto out_put_key; |
c87e2837 | 2128 | |
42d35d48 | 2129 | out_unlock_put_key: |
0d00c7b2 | 2130 | queue_unlock(hb); |
c87e2837 | 2131 | |
42d35d48 | 2132 | out_put_key: |
ae791a2d | 2133 | put_futex_key(&q.key); |
42d35d48 | 2134 | out: |
237fc6e7 TG |
2135 | if (to) |
2136 | destroy_hrtimer_on_stack(&to->timer); | |
dd973998 | 2137 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 2138 | |
42d35d48 | 2139 | uaddr_faulted: |
0d00c7b2 | 2140 | queue_unlock(hb); |
778e9a9c | 2141 | |
d0725992 | 2142 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a DH |
2143 | if (ret) |
2144 | goto out_put_key; | |
c87e2837 | 2145 | |
b41277dc | 2146 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2147 | goto retry_private; |
2148 | ||
ae791a2d | 2149 | put_futex_key(&q.key); |
e4dc5b7a | 2150 | goto retry; |
c87e2837 IM |
2151 | } |
2152 | ||
c87e2837 IM |
2153 | /* |
2154 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
2155 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
2156 | * and do the rt-mutex unlock. | |
2157 | */ | |
b41277dc | 2158 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 IM |
2159 | { |
2160 | struct futex_hash_bucket *hb; | |
2161 | struct futex_q *this, *next; | |
38d47c1b | 2162 | union futex_key key = FUTEX_KEY_INIT; |
c0c9ed15 | 2163 | u32 uval, vpid = task_pid_vnr(current); |
e4dc5b7a | 2164 | int ret; |
c87e2837 IM |
2165 | |
2166 | retry: | |
2167 | if (get_user(uval, uaddr)) | |
2168 | return -EFAULT; | |
2169 | /* | |
2170 | * We release only a lock we actually own: | |
2171 | */ | |
c0c9ed15 | 2172 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 2173 | return -EPERM; |
c87e2837 | 2174 | |
9ea71503 | 2175 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE); |
c87e2837 IM |
2176 | if (unlikely(ret != 0)) |
2177 | goto out; | |
2178 | ||
2179 | hb = hash_futex(&key); | |
2180 | spin_lock(&hb->lock); | |
2181 | ||
c87e2837 IM |
2182 | /* |
2183 | * To avoid races, try to do the TID -> 0 atomic transition | |
2184 | * again. If it succeeds then we can return without waking | |
2185 | * anyone else up: | |
2186 | */ | |
37a9d912 ML |
2187 | if (!(uval & FUTEX_OWNER_DIED) && |
2188 | cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0)) | |
c87e2837 IM |
2189 | goto pi_faulted; |
2190 | /* | |
2191 | * Rare case: we managed to release the lock atomically, | |
2192 | * no need to wake anyone else up: | |
2193 | */ | |
c0c9ed15 | 2194 | if (unlikely(uval == vpid)) |
c87e2837 IM |
2195 | goto out_unlock; |
2196 | ||
2197 | /* | |
2198 | * Ok, other tasks may need to be woken up - check waiters | |
2199 | * and do the wakeup if necessary: | |
2200 | */ | |
0d00c7b2 | 2201 | plist_for_each_entry_safe(this, next, &hb->chain, list) { |
c87e2837 IM |
2202 | if (!match_futex (&this->key, &key)) |
2203 | continue; | |
2204 | ret = wake_futex_pi(uaddr, uval, this); | |
2205 | /* | |
2206 | * The atomic access to the futex value | |
2207 | * generated a pagefault, so retry the | |
2208 | * user-access and the wakeup: | |
2209 | */ | |
2210 | if (ret == -EFAULT) | |
2211 | goto pi_faulted; | |
2212 | goto out_unlock; | |
2213 | } | |
2214 | /* | |
2215 | * No waiters - kernel unlocks the futex: | |
2216 | */ | |
e3f2ddea IM |
2217 | if (!(uval & FUTEX_OWNER_DIED)) { |
2218 | ret = unlock_futex_pi(uaddr, uval); | |
2219 | if (ret == -EFAULT) | |
2220 | goto pi_faulted; | |
2221 | } | |
c87e2837 IM |
2222 | |
2223 | out_unlock: | |
2224 | spin_unlock(&hb->lock); | |
ae791a2d | 2225 | put_futex_key(&key); |
c87e2837 | 2226 | |
42d35d48 | 2227 | out: |
c87e2837 IM |
2228 | return ret; |
2229 | ||
2230 | pi_faulted: | |
778e9a9c | 2231 | spin_unlock(&hb->lock); |
ae791a2d | 2232 | put_futex_key(&key); |
c87e2837 | 2233 | |
d0725992 | 2234 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 2235 | if (!ret) |
c87e2837 IM |
2236 | goto retry; |
2237 | ||
1da177e4 LT |
2238 | return ret; |
2239 | } | |
2240 | ||
52400ba9 DH |
2241 | /** |
2242 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
2243 | * @hb: the hash_bucket futex_q was original enqueued on | |
2244 | * @q: the futex_q woken while waiting to be requeued | |
2245 | * @key2: the futex_key of the requeue target futex | |
2246 | * @timeout: the timeout associated with the wait (NULL if none) | |
2247 | * | |
2248 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
2249 | * target futex. If so, determine if it was a timeout or a signal that caused | |
2250 | * the wakeup and return the appropriate error code to the caller. Must be | |
2251 | * called with the hb lock held. | |
2252 | * | |
6c23cbbd RD |
2253 | * Return: |
2254 | * 0 = no early wakeup detected; | |
2255 | * <0 = -ETIMEDOUT or -ERESTARTNOINTR | |
52400ba9 DH |
2256 | */ |
2257 | static inline | |
2258 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
2259 | struct futex_q *q, union futex_key *key2, | |
2260 | struct hrtimer_sleeper *timeout) | |
2261 | { | |
2262 | int ret = 0; | |
2263 | ||
2264 | /* | |
2265 | * With the hb lock held, we avoid races while we process the wakeup. | |
2266 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
2267 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
2268 | * It can't be requeued from uaddr2 to something else since we don't | |
2269 | * support a PI aware source futex for requeue. | |
2270 | */ | |
2271 | if (!match_futex(&q->key, key2)) { | |
2272 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
2273 | /* | |
2274 | * We were woken prior to requeue by a timeout or a signal. | |
2275 | * Unqueue the futex_q and determine which it was. | |
2276 | */ | |
2e12978a | 2277 | plist_del(&q->list, &hb->chain); |
52400ba9 | 2278 | |
d58e6576 | 2279 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 2280 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2281 | if (timeout && !timeout->task) |
2282 | ret = -ETIMEDOUT; | |
d58e6576 | 2283 | else if (signal_pending(current)) |
1c840c14 | 2284 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
2285 | } |
2286 | return ret; | |
2287 | } | |
2288 | ||
2289 | /** | |
2290 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 2291 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 2292 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
52400ba9 DH |
2293 | * the same type, no requeueing from private to shared, etc. |
2294 | * @val: the expected value of uaddr | |
2295 | * @abs_time: absolute timeout | |
56ec1607 | 2296 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
2297 | * @uaddr2: the pi futex we will take prior to returning to user-space |
2298 | * | |
2299 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
6f7b0a2a DH |
2300 | * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake |
2301 | * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to | |
2302 | * userspace. This ensures the rt_mutex maintains an owner when it has waiters; | |
2303 | * without one, the pi logic would not know which task to boost/deboost, if | |
2304 | * there was a need to. | |
52400ba9 DH |
2305 | * |
2306 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
6c23cbbd | 2307 | * via the following-- |
52400ba9 | 2308 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() |
cc6db4e6 DH |
2309 | * 2) wakeup on uaddr2 after a requeue |
2310 | * 3) signal | |
2311 | * 4) timeout | |
52400ba9 | 2312 | * |
cc6db4e6 | 2313 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
2314 | * |
2315 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
2316 | * 5) successful lock | |
2317 | * 6) signal | |
2318 | * 7) timeout | |
2319 | * 8) other lock acquisition failure | |
2320 | * | |
cc6db4e6 | 2321 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
2322 | * |
2323 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
2324 | * | |
6c23cbbd RD |
2325 | * Return: |
2326 | * 0 - On success; | |
52400ba9 DH |
2327 | * <0 - On error |
2328 | */ | |
b41277dc | 2329 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 2330 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 2331 | u32 __user *uaddr2) |
52400ba9 DH |
2332 | { |
2333 | struct hrtimer_sleeper timeout, *to = NULL; | |
2334 | struct rt_mutex_waiter rt_waiter; | |
2335 | struct rt_mutex *pi_mutex = NULL; | |
52400ba9 | 2336 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
2337 | union futex_key key2 = FUTEX_KEY_INIT; |
2338 | struct futex_q q = futex_q_init; | |
52400ba9 | 2339 | int res, ret; |
52400ba9 | 2340 | |
6f7b0a2a DH |
2341 | if (uaddr == uaddr2) |
2342 | return -EINVAL; | |
2343 | ||
52400ba9 DH |
2344 | if (!bitset) |
2345 | return -EINVAL; | |
2346 | ||
2347 | if (abs_time) { | |
2348 | to = &timeout; | |
b41277dc DH |
2349 | hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ? |
2350 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
2351 | HRTIMER_MODE_ABS); | |
52400ba9 DH |
2352 | hrtimer_init_sleeper(to, current); |
2353 | hrtimer_set_expires_range_ns(&to->timer, *abs_time, | |
2354 | current->timer_slack_ns); | |
2355 | } | |
2356 | ||
2357 | /* | |
2358 | * The waiter is allocated on our stack, manipulated by the requeue | |
2359 | * code while we sleep on uaddr. | |
2360 | */ | |
2361 | debug_rt_mutex_init_waiter(&rt_waiter); | |
2362 | rt_waiter.task = NULL; | |
2363 | ||
9ea71503 | 2364 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE); |
52400ba9 DH |
2365 | if (unlikely(ret != 0)) |
2366 | goto out; | |
2367 | ||
84bc4af5 DH |
2368 | q.bitset = bitset; |
2369 | q.rt_waiter = &rt_waiter; | |
2370 | q.requeue_pi_key = &key2; | |
2371 | ||
7ada876a DH |
2372 | /* |
2373 | * Prepare to wait on uaddr. On success, increments q.key (key1) ref | |
2374 | * count. | |
2375 | */ | |
b41277dc | 2376 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 TG |
2377 | if (ret) |
2378 | goto out_key2; | |
52400ba9 DH |
2379 | |
2380 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ | |
f1a11e05 | 2381 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
2382 | |
2383 | spin_lock(&hb->lock); | |
2384 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
2385 | spin_unlock(&hb->lock); | |
2386 | if (ret) | |
2387 | goto out_put_keys; | |
2388 | ||
2389 | /* | |
2390 | * In order for us to be here, we know our q.key == key2, and since | |
2391 | * we took the hb->lock above, we also know that futex_requeue() has | |
2392 | * completed and we no longer have to concern ourselves with a wakeup | |
7ada876a DH |
2393 | * race with the atomic proxy lock acquisition by the requeue code. The |
2394 | * futex_requeue dropped our key1 reference and incremented our key2 | |
2395 | * reference count. | |
52400ba9 DH |
2396 | */ |
2397 | ||
2398 | /* Check if the requeue code acquired the second futex for us. */ | |
2399 | if (!q.rt_waiter) { | |
2400 | /* | |
2401 | * Got the lock. We might not be the anticipated owner if we | |
2402 | * did a lock-steal - fix up the PI-state in that case. | |
2403 | */ | |
2404 | if (q.pi_state && (q.pi_state->owner != current)) { | |
2405 | spin_lock(q.lock_ptr); | |
ae791a2d | 2406 | ret = fixup_pi_state_owner(uaddr2, &q, current); |
52400ba9 DH |
2407 | spin_unlock(q.lock_ptr); |
2408 | } | |
2409 | } else { | |
2410 | /* | |
2411 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
2412 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
2413 | * the pi_state. | |
2414 | */ | |
f27071cb | 2415 | WARN_ON(!q.pi_state); |
52400ba9 DH |
2416 | pi_mutex = &q.pi_state->pi_mutex; |
2417 | ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1); | |
2418 | debug_rt_mutex_free_waiter(&rt_waiter); | |
2419 | ||
2420 | spin_lock(q.lock_ptr); | |
2421 | /* | |
2422 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2423 | * haven't already. | |
2424 | */ | |
ae791a2d | 2425 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 DH |
2426 | /* |
2427 | * If fixup_owner() returned an error, proprogate that. If it | |
56ec1607 | 2428 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
2429 | */ |
2430 | if (res) | |
2431 | ret = (res < 0) ? res : 0; | |
2432 | ||
2433 | /* Unqueue and drop the lock. */ | |
2434 | unqueue_me_pi(&q); | |
2435 | } | |
2436 | ||
2437 | /* | |
2438 | * If fixup_pi_state_owner() faulted and was unable to handle the | |
2439 | * fault, unlock the rt_mutex and return the fault to userspace. | |
2440 | */ | |
2441 | if (ret == -EFAULT) { | |
b6070a8d | 2442 | if (pi_mutex && rt_mutex_owner(pi_mutex) == current) |
52400ba9 DH |
2443 | rt_mutex_unlock(pi_mutex); |
2444 | } else if (ret == -EINTR) { | |
52400ba9 | 2445 | /* |
cc6db4e6 DH |
2446 | * We've already been requeued, but cannot restart by calling |
2447 | * futex_lock_pi() directly. We could restart this syscall, but | |
2448 | * it would detect that the user space "val" changed and return | |
2449 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
2450 | * -EWOULDBLOCK directly. | |
52400ba9 | 2451 | */ |
2070887f | 2452 | ret = -EWOULDBLOCK; |
52400ba9 DH |
2453 | } |
2454 | ||
2455 | out_put_keys: | |
ae791a2d | 2456 | put_futex_key(&q.key); |
c8b15a70 | 2457 | out_key2: |
ae791a2d | 2458 | put_futex_key(&key2); |
52400ba9 DH |
2459 | |
2460 | out: | |
2461 | if (to) { | |
2462 | hrtimer_cancel(&to->timer); | |
2463 | destroy_hrtimer_on_stack(&to->timer); | |
2464 | } | |
2465 | return ret; | |
2466 | } | |
2467 | ||
0771dfef IM |
2468 | /* |
2469 | * Support for robust futexes: the kernel cleans up held futexes at | |
2470 | * thread exit time. | |
2471 | * | |
2472 | * Implementation: user-space maintains a per-thread list of locks it | |
2473 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
2474 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 2475 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
2476 | * always manipulated with the lock held, so the list is private and |
2477 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
2478 | * field, to allow the kernel to clean up if the thread dies after | |
2479 | * acquiring the lock, but just before it could have added itself to | |
2480 | * the list. There can only be one such pending lock. | |
2481 | */ | |
2482 | ||
2483 | /** | |
d96ee56c DH |
2484 | * sys_set_robust_list() - Set the robust-futex list head of a task |
2485 | * @head: pointer to the list-head | |
2486 | * @len: length of the list-head, as userspace expects | |
0771dfef | 2487 | */ |
836f92ad HC |
2488 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
2489 | size_t, len) | |
0771dfef | 2490 | { |
a0c1e907 TG |
2491 | if (!futex_cmpxchg_enabled) |
2492 | return -ENOSYS; | |
0771dfef IM |
2493 | /* |
2494 | * The kernel knows only one size for now: | |
2495 | */ | |
2496 | if (unlikely(len != sizeof(*head))) | |
2497 | return -EINVAL; | |
2498 | ||
2499 | current->robust_list = head; | |
2500 | ||
2501 | return 0; | |
2502 | } | |
2503 | ||
2504 | /** | |
d96ee56c DH |
2505 | * sys_get_robust_list() - Get the robust-futex list head of a task |
2506 | * @pid: pid of the process [zero for current task] | |
2507 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
2508 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 2509 | */ |
836f92ad HC |
2510 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
2511 | struct robust_list_head __user * __user *, head_ptr, | |
2512 | size_t __user *, len_ptr) | |
0771dfef | 2513 | { |
ba46df98 | 2514 | struct robust_list_head __user *head; |
0771dfef | 2515 | unsigned long ret; |
bdbb776f | 2516 | struct task_struct *p; |
0771dfef | 2517 | |
a0c1e907 TG |
2518 | if (!futex_cmpxchg_enabled) |
2519 | return -ENOSYS; | |
2520 | ||
bdbb776f KC |
2521 | rcu_read_lock(); |
2522 | ||
2523 | ret = -ESRCH; | |
0771dfef | 2524 | if (!pid) |
bdbb776f | 2525 | p = current; |
0771dfef | 2526 | else { |
228ebcbe | 2527 | p = find_task_by_vpid(pid); |
0771dfef IM |
2528 | if (!p) |
2529 | goto err_unlock; | |
0771dfef IM |
2530 | } |
2531 | ||
bdbb776f KC |
2532 | ret = -EPERM; |
2533 | if (!ptrace_may_access(p, PTRACE_MODE_READ)) | |
2534 | goto err_unlock; | |
2535 | ||
2536 | head = p->robust_list; | |
2537 | rcu_read_unlock(); | |
2538 | ||
0771dfef IM |
2539 | if (put_user(sizeof(*head), len_ptr)) |
2540 | return -EFAULT; | |
2541 | return put_user(head, head_ptr); | |
2542 | ||
2543 | err_unlock: | |
aaa2a97e | 2544 | rcu_read_unlock(); |
0771dfef IM |
2545 | |
2546 | return ret; | |
2547 | } | |
2548 | ||
2549 | /* | |
2550 | * Process a futex-list entry, check whether it's owned by the | |
2551 | * dying task, and do notification if so: | |
2552 | */ | |
e3f2ddea | 2553 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
0771dfef | 2554 | { |
7cfdaf38 | 2555 | u32 uval, uninitialized_var(nval), mval; |
0771dfef | 2556 | |
8f17d3a5 IM |
2557 | retry: |
2558 | if (get_user(uval, uaddr)) | |
0771dfef IM |
2559 | return -1; |
2560 | ||
b488893a | 2561 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
0771dfef IM |
2562 | /* |
2563 | * Ok, this dying thread is truly holding a futex | |
2564 | * of interest. Set the OWNER_DIED bit atomically | |
2565 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
2566 | * set, wake up a waiter (if any). (We have to do a | |
2567 | * futex_wake() even if OWNER_DIED is already set - | |
2568 | * to handle the rare but possible case of recursive | |
2569 | * thread-death.) The rest of the cleanup is done in | |
2570 | * userspace. | |
2571 | */ | |
e3f2ddea | 2572 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
6e0aa9f8 TG |
2573 | /* |
2574 | * We are not holding a lock here, but we want to have | |
2575 | * the pagefault_disable/enable() protection because | |
2576 | * we want to handle the fault gracefully. If the | |
2577 | * access fails we try to fault in the futex with R/W | |
2578 | * verification via get_user_pages. get_user() above | |
2579 | * does not guarantee R/W access. If that fails we | |
2580 | * give up and leave the futex locked. | |
2581 | */ | |
2582 | if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) { | |
2583 | if (fault_in_user_writeable(uaddr)) | |
2584 | return -1; | |
2585 | goto retry; | |
2586 | } | |
c87e2837 | 2587 | if (nval != uval) |
8f17d3a5 | 2588 | goto retry; |
0771dfef | 2589 | |
e3f2ddea IM |
2590 | /* |
2591 | * Wake robust non-PI futexes here. The wakeup of | |
2592 | * PI futexes happens in exit_pi_state(): | |
2593 | */ | |
36cf3b5c | 2594 | if (!pi && (uval & FUTEX_WAITERS)) |
c2f9f201 | 2595 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
0771dfef IM |
2596 | } |
2597 | return 0; | |
2598 | } | |
2599 | ||
e3f2ddea IM |
2600 | /* |
2601 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
2602 | */ | |
2603 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 2604 | struct robust_list __user * __user *head, |
1dcc41bb | 2605 | unsigned int *pi) |
e3f2ddea IM |
2606 | { |
2607 | unsigned long uentry; | |
2608 | ||
ba46df98 | 2609 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
2610 | return -EFAULT; |
2611 | ||
ba46df98 | 2612 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
2613 | *pi = uentry & 1; |
2614 | ||
2615 | return 0; | |
2616 | } | |
2617 | ||
0771dfef IM |
2618 | /* |
2619 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
2620 | * and mark any locks found there dead, and notify any waiters. | |
2621 | * | |
2622 | * We silently return on any sign of list-walking problem. | |
2623 | */ | |
2624 | void exit_robust_list(struct task_struct *curr) | |
2625 | { | |
2626 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 2627 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 DH |
2628 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
2629 | unsigned int uninitialized_var(next_pi); | |
0771dfef | 2630 | unsigned long futex_offset; |
9f96cb1e | 2631 | int rc; |
0771dfef | 2632 | |
a0c1e907 TG |
2633 | if (!futex_cmpxchg_enabled) |
2634 | return; | |
2635 | ||
0771dfef IM |
2636 | /* |
2637 | * Fetch the list head (which was registered earlier, via | |
2638 | * sys_set_robust_list()): | |
2639 | */ | |
e3f2ddea | 2640 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
2641 | return; |
2642 | /* | |
2643 | * Fetch the relative futex offset: | |
2644 | */ | |
2645 | if (get_user(futex_offset, &head->futex_offset)) | |
2646 | return; | |
2647 | /* | |
2648 | * Fetch any possibly pending lock-add first, and handle it | |
2649 | * if it exists: | |
2650 | */ | |
e3f2ddea | 2651 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 2652 | return; |
e3f2ddea | 2653 | |
9f96cb1e | 2654 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 2655 | while (entry != &head->list) { |
9f96cb1e MS |
2656 | /* |
2657 | * Fetch the next entry in the list before calling | |
2658 | * handle_futex_death: | |
2659 | */ | |
2660 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
2661 | /* |
2662 | * A pending lock might already be on the list, so | |
c87e2837 | 2663 | * don't process it twice: |
0771dfef IM |
2664 | */ |
2665 | if (entry != pending) | |
ba46df98 | 2666 | if (handle_futex_death((void __user *)entry + futex_offset, |
e3f2ddea | 2667 | curr, pi)) |
0771dfef | 2668 | return; |
9f96cb1e | 2669 | if (rc) |
0771dfef | 2670 | return; |
9f96cb1e MS |
2671 | entry = next_entry; |
2672 | pi = next_pi; | |
0771dfef IM |
2673 | /* |
2674 | * Avoid excessively long or circular lists: | |
2675 | */ | |
2676 | if (!--limit) | |
2677 | break; | |
2678 | ||
2679 | cond_resched(); | |
2680 | } | |
9f96cb1e MS |
2681 | |
2682 | if (pending) | |
2683 | handle_futex_death((void __user *)pending + futex_offset, | |
2684 | curr, pip); | |
0771dfef IM |
2685 | } |
2686 | ||
c19384b5 | 2687 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 2688 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 2689 | { |
81b40539 | 2690 | int cmd = op & FUTEX_CMD_MASK; |
b41277dc | 2691 | unsigned int flags = 0; |
34f01cc1 ED |
2692 | |
2693 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 2694 | flags |= FLAGS_SHARED; |
1da177e4 | 2695 | |
b41277dc DH |
2696 | if (op & FUTEX_CLOCK_REALTIME) { |
2697 | flags |= FLAGS_CLOCKRT; | |
2698 | if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI) | |
2699 | return -ENOSYS; | |
2700 | } | |
1da177e4 | 2701 | |
59263b51 TG |
2702 | switch (cmd) { |
2703 | case FUTEX_LOCK_PI: | |
2704 | case FUTEX_UNLOCK_PI: | |
2705 | case FUTEX_TRYLOCK_PI: | |
2706 | case FUTEX_WAIT_REQUEUE_PI: | |
2707 | case FUTEX_CMP_REQUEUE_PI: | |
2708 | if (!futex_cmpxchg_enabled) | |
2709 | return -ENOSYS; | |
2710 | } | |
2711 | ||
34f01cc1 | 2712 | switch (cmd) { |
1da177e4 | 2713 | case FUTEX_WAIT: |
cd689985 TG |
2714 | val3 = FUTEX_BITSET_MATCH_ANY; |
2715 | case FUTEX_WAIT_BITSET: | |
81b40539 | 2716 | return futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 | 2717 | case FUTEX_WAKE: |
cd689985 TG |
2718 | val3 = FUTEX_BITSET_MATCH_ANY; |
2719 | case FUTEX_WAKE_BITSET: | |
81b40539 | 2720 | return futex_wake(uaddr, flags, val, val3); |
1da177e4 | 2721 | case FUTEX_REQUEUE: |
81b40539 | 2722 | return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 | 2723 | case FUTEX_CMP_REQUEUE: |
81b40539 | 2724 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
4732efbe | 2725 | case FUTEX_WAKE_OP: |
81b40539 | 2726 | return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
c87e2837 | 2727 | case FUTEX_LOCK_PI: |
81b40539 | 2728 | return futex_lock_pi(uaddr, flags, val, timeout, 0); |
c87e2837 | 2729 | case FUTEX_UNLOCK_PI: |
81b40539 | 2730 | return futex_unlock_pi(uaddr, flags); |
c87e2837 | 2731 | case FUTEX_TRYLOCK_PI: |
81b40539 | 2732 | return futex_lock_pi(uaddr, flags, 0, timeout, 1); |
52400ba9 DH |
2733 | case FUTEX_WAIT_REQUEUE_PI: |
2734 | val3 = FUTEX_BITSET_MATCH_ANY; | |
81b40539 TG |
2735 | return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
2736 | uaddr2); | |
52400ba9 | 2737 | case FUTEX_CMP_REQUEUE_PI: |
81b40539 | 2738 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
1da177e4 | 2739 | } |
81b40539 | 2740 | return -ENOSYS; |
1da177e4 LT |
2741 | } |
2742 | ||
2743 | ||
17da2bd9 HC |
2744 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
2745 | struct timespec __user *, utime, u32 __user *, uaddr2, | |
2746 | u32, val3) | |
1da177e4 | 2747 | { |
c19384b5 PP |
2748 | struct timespec ts; |
2749 | ktime_t t, *tp = NULL; | |
e2970f2f | 2750 | u32 val2 = 0; |
34f01cc1 | 2751 | int cmd = op & FUTEX_CMD_MASK; |
1da177e4 | 2752 | |
cd689985 | 2753 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
52400ba9 DH |
2754 | cmd == FUTEX_WAIT_BITSET || |
2755 | cmd == FUTEX_WAIT_REQUEUE_PI)) { | |
c19384b5 | 2756 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
1da177e4 | 2757 | return -EFAULT; |
c19384b5 | 2758 | if (!timespec_valid(&ts)) |
9741ef96 | 2759 | return -EINVAL; |
c19384b5 PP |
2760 | |
2761 | t = timespec_to_ktime(ts); | |
34f01cc1 | 2762 | if (cmd == FUTEX_WAIT) |
5a7780e7 | 2763 | t = ktime_add_safe(ktime_get(), t); |
c19384b5 | 2764 | tp = &t; |
1da177e4 LT |
2765 | } |
2766 | /* | |
52400ba9 | 2767 | * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. |
f54f0986 | 2768 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
1da177e4 | 2769 | */ |
f54f0986 | 2770 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
ba9c22f2 | 2771 | cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) |
e2970f2f | 2772 | val2 = (u32) (unsigned long) utime; |
1da177e4 | 2773 | |
c19384b5 | 2774 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
1da177e4 LT |
2775 | } |
2776 | ||
f6d107fb | 2777 | static int __init futex_init(void) |
1da177e4 | 2778 | { |
a0c1e907 | 2779 | u32 curval; |
a52b89eb DB |
2780 | unsigned long i; |
2781 | ||
2782 | #if CONFIG_BASE_SMALL | |
2783 | futex_hashsize = 16; | |
2784 | #else | |
2785 | futex_hashsize = roundup_pow_of_two(256 * num_possible_cpus()); | |
2786 | #endif | |
2787 | ||
2788 | futex_queues = alloc_large_system_hash("futex", sizeof(*futex_queues), | |
2789 | futex_hashsize, 0, | |
2790 | futex_hashsize < 256 ? HASH_SMALL : 0, | |
2791 | NULL, NULL, futex_hashsize, futex_hashsize); | |
95362fa9 | 2792 | |
a0c1e907 TG |
2793 | /* |
2794 | * This will fail and we want it. Some arch implementations do | |
2795 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
2796 | * functionality. We want to know that before we call in any | |
2797 | * of the complex code paths. Also we want to prevent | |
2798 | * registration of robust lists in that case. NULL is | |
2799 | * guaranteed to fault and we get -EFAULT on functional | |
fb62db2b | 2800 | * implementation, the non-functional ones will return |
a0c1e907 TG |
2801 | * -ENOSYS. |
2802 | */ | |
37a9d912 | 2803 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) |
a0c1e907 TG |
2804 | futex_cmpxchg_enabled = 1; |
2805 | ||
a52b89eb | 2806 | for (i = 0; i < futex_hashsize; i++) { |
732375c6 | 2807 | plist_head_init(&futex_queues[i].chain); |
3e4ab747 TG |
2808 | spin_lock_init(&futex_queues[i].lock); |
2809 | } | |
2810 | ||
1da177e4 LT |
2811 | return 0; |
2812 | } | |
f6d107fb | 2813 | __initcall(futex_init); |