Line data Source code
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 : *
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 : *
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 : *
19 : * PRIVATE futexes by Eric Dumazet
20 : * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com>
21 : *
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 : *
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>
57 : #include <linux/signal.h>
58 : #include <linux/export.h>
59 : #include <linux/magic.h>
60 : #include <linux/pid.h>
61 : #include <linux/nsproxy.h>
62 : #include <linux/ptrace.h>
63 : #include <linux/sched/rt.h>
64 : #include <linux/hugetlb.h>
65 : #include <linux/freezer.h>
66 : #include <linux/bootmem.h>
67 :
68 : #include <asm/futex.h>
69 :
70 : #include "locking/rtmutex_common.h"
71 :
72 : /*
73 : * READ this before attempting to hack on futexes!
74 : *
75 : * Basic futex operation and ordering guarantees
76 : * =============================================
77 : *
78 : * The waiter reads the futex value in user space and calls
79 : * futex_wait(). This function computes the hash bucket and acquires
80 : * the hash bucket lock. After that it reads the futex user space value
81 : * again and verifies that the data has not changed. If it has not changed
82 : * it enqueues itself into the hash bucket, releases the hash bucket lock
83 : * and schedules.
84 : *
85 : * The waker side modifies the user space value of the futex and calls
86 : * futex_wake(). This function computes the hash bucket and acquires the
87 : * hash bucket lock. Then it looks for waiters on that futex in the hash
88 : * bucket and wakes them.
89 : *
90 : * In futex wake up scenarios where no tasks are blocked on a futex, taking
91 : * the hb spinlock can be avoided and simply return. In order for this
92 : * optimization to work, ordering guarantees must exist so that the waiter
93 : * being added to the list is acknowledged when the list is concurrently being
94 : * checked by the waker, avoiding scenarios like the following:
95 : *
96 : * CPU 0 CPU 1
97 : * val = *futex;
98 : * sys_futex(WAIT, futex, val);
99 : * futex_wait(futex, val);
100 : * uval = *futex;
101 : * *futex = newval;
102 : * sys_futex(WAKE, futex);
103 : * futex_wake(futex);
104 : * if (queue_empty())
105 : * return;
106 : * if (uval == val)
107 : * lock(hash_bucket(futex));
108 : * queue();
109 : * unlock(hash_bucket(futex));
110 : * schedule();
111 : *
112 : * This would cause the waiter on CPU 0 to wait forever because it
113 : * missed the transition of the user space value from val to newval
114 : * and the waker did not find the waiter in the hash bucket queue.
115 : *
116 : * The correct serialization ensures that a waiter either observes
117 : * the changed user space value before blocking or is woken by a
118 : * concurrent waker:
119 : *
120 : * CPU 0 CPU 1
121 : * val = *futex;
122 : * sys_futex(WAIT, futex, val);
123 : * futex_wait(futex, val);
124 : *
125 : * waiters++; (a)
126 : * mb(); (A) <-- paired with -.
127 : * |
128 : * lock(hash_bucket(futex)); |
129 : * |
130 : * uval = *futex; |
131 : * | *futex = newval;
132 : * | sys_futex(WAKE, futex);
133 : * | futex_wake(futex);
134 : * |
135 : * `-------> mb(); (B)
136 : * if (uval == val)
137 : * queue();
138 : * unlock(hash_bucket(futex));
139 : * schedule(); if (waiters)
140 : * lock(hash_bucket(futex));
141 : * else wake_waiters(futex);
142 : * waiters--; (b) unlock(hash_bucket(futex));
143 : *
144 : * Where (A) orders the waiters increment and the futex value read through
145 : * atomic operations (see hb_waiters_inc) and where (B) orders the write
146 : * to futex and the waiters read -- this is done by the barriers for both
147 : * shared and private futexes in get_futex_key_refs().
148 : *
149 : * This yields the following case (where X:=waiters, Y:=futex):
150 : *
151 : * X = Y = 0
152 : *
153 : * w[X]=1 w[Y]=1
154 : * MB MB
155 : * r[Y]=y r[X]=x
156 : *
157 : * Which guarantees that x==0 && y==0 is impossible; which translates back into
158 : * the guarantee that we cannot both miss the futex variable change and the
159 : * enqueue.
160 : *
161 : * Note that a new waiter is accounted for in (a) even when it is possible that
162 : * the wait call can return error, in which case we backtrack from it in (b).
163 : * Refer to the comment in queue_lock().
164 : *
165 : * Similarly, in order to account for waiters being requeued on another
166 : * address we always increment the waiters for the destination bucket before
167 : * acquiring the lock. It then decrements them again after releasing it -
168 : * the code that actually moves the futex(es) between hash buckets (requeue_futex)
169 : * will do the additional required waiter count housekeeping. This is done for
170 : * double_lock_hb() and double_unlock_hb(), respectively.
171 : */
172 :
173 : #ifndef CONFIG_HAVE_FUTEX_CMPXCHG
174 : int __read_mostly futex_cmpxchg_enabled;
175 : #endif
176 :
177 : /*
178 : * Futex flags used to encode options to functions and preserve them across
179 : * restarts.
180 : */
181 : #define FLAGS_SHARED 0x01
182 : #define FLAGS_CLOCKRT 0x02
183 : #define FLAGS_HAS_TIMEOUT 0x04
184 :
185 : /*
186 : * Priority Inheritance state:
187 : */
188 : struct futex_pi_state {
189 : /*
190 : * list of 'owned' pi_state instances - these have to be
191 : * cleaned up in do_exit() if the task exits prematurely:
192 : */
193 : struct list_head list;
194 :
195 : /*
196 : * The PI object:
197 : */
198 : struct rt_mutex pi_mutex;
199 :
200 : struct task_struct *owner;
201 : atomic_t refcount;
202 :
203 : union futex_key key;
204 : };
205 :
206 : /**
207 : * struct futex_q - The hashed futex queue entry, one per waiting task
208 : * @list: priority-sorted list of tasks waiting on this futex
209 : * @task: the task waiting on the futex
210 : * @lock_ptr: the hash bucket lock
211 : * @key: the key the futex is hashed on
212 : * @pi_state: optional priority inheritance state
213 : * @rt_waiter: rt_waiter storage for use with requeue_pi
214 : * @requeue_pi_key: the requeue_pi target futex key
215 : * @bitset: bitset for the optional bitmasked wakeup
216 : *
217 : * We use this hashed waitqueue, instead of a normal wait_queue_t, so
218 : * we can wake only the relevant ones (hashed queues may be shared).
219 : *
220 : * A futex_q has a woken state, just like tasks have TASK_RUNNING.
221 : * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
222 : * The order of wakeup is always to make the first condition true, then
223 : * the second.
224 : *
225 : * PI futexes are typically woken before they are removed from the hash list via
226 : * the rt_mutex code. See unqueue_me_pi().
227 : */
228 : struct futex_q {
229 : struct plist_node list;
230 :
231 : struct task_struct *task;
232 : spinlock_t *lock_ptr;
233 : union futex_key key;
234 : struct futex_pi_state *pi_state;
235 : struct rt_mutex_waiter *rt_waiter;
236 : union futex_key *requeue_pi_key;
237 : u32 bitset;
238 : };
239 :
240 : static const struct futex_q futex_q_init = {
241 : /* list gets initialized in queue_me()*/
242 : .key = FUTEX_KEY_INIT,
243 : .bitset = FUTEX_BITSET_MATCH_ANY
244 : };
245 :
246 : /*
247 : * Hash buckets are shared by all the futex_keys that hash to the same
248 : * location. Each key may have multiple futex_q structures, one for each task
249 : * waiting on a futex.
250 : */
251 : struct futex_hash_bucket {
252 : atomic_t waiters;
253 : spinlock_t lock;
254 : struct plist_head chain;
255 : } ____cacheline_aligned_in_smp;
256 :
257 : static unsigned long __read_mostly futex_hashsize;
258 :
259 : static struct futex_hash_bucket *futex_queues;
260 :
261 : static inline void futex_get_mm(union futex_key *key)
262 : {
263 0 : atomic_inc(&key->private.mm->mm_count);
264 : /*
265 : * Ensure futex_get_mm() implies a full barrier such that
266 : * get_futex_key() implies a full barrier. This is relied upon
267 : * as full barrier (B), see the ordering comment above.
268 : */
269 0 : smp_mb__after_atomic();
270 : }
271 :
272 : /*
273 : * Reflects a new waiter being added to the waitqueue.
274 : */
275 : static inline void hb_waiters_inc(struct futex_hash_bucket *hb)
276 : {
277 : #ifdef CONFIG_SMP
278 : atomic_inc(&hb->waiters);
279 : /*
280 : * Full barrier (A), see the ordering comment above.
281 : */
282 : smp_mb__after_atomic();
283 : #endif
284 : }
285 :
286 : /*
287 : * Reflects a waiter being removed from the waitqueue by wakeup
288 : * paths.
289 : */
290 : static inline void hb_waiters_dec(struct futex_hash_bucket *hb)
291 : {
292 : #ifdef CONFIG_SMP
293 : atomic_dec(&hb->waiters);
294 : #endif
295 : }
296 :
297 : static inline int hb_waiters_pending(struct futex_hash_bucket *hb)
298 : {
299 : #ifdef CONFIG_SMP
300 : return atomic_read(&hb->waiters);
301 : #else
302 : return 1;
303 : #endif
304 : }
305 :
306 : /*
307 : * We hash on the keys returned from get_futex_key (see below).
308 : */
309 1037619 : static struct futex_hash_bucket *hash_futex(union futex_key *key)
310 : {
311 2075238 : u32 hash = jhash2((u32*)&key->both.word,
312 : (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
313 1037619 : key->both.offset);
314 1037619 : return &futex_queues[hash & (futex_hashsize - 1)];
315 : }
316 :
317 : /*
318 : * Return 1 if two futex_keys are equal, 0 otherwise.
319 : */
320 : static inline int match_futex(union futex_key *key1, union futex_key *key2)
321 : {
322 0 : return (key1 && key2
323 345260 : && key1->both.word == key2->both.word
324 345257 : && key1->both.ptr == key2->both.ptr
325 690517 : && key1->both.offset == key2->both.offset);
326 : }
327 :
328 : /*
329 : * Take a reference to the resource addressed by a key.
330 : * Can be called while holding spinlocks.
331 : *
332 : */
333 1037619 : static void get_futex_key_refs(union futex_key *key)
334 : {
335 1037619 : if (!key->both.ptr)
336 1037619 : return;
337 :
338 1037619 : switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
339 : case FUT_OFF_INODE:
340 0 : ihold(key->shared.inode); /* implies MB (B) */
341 : break;
342 : case FUT_OFF_MMSHARED:
343 : futex_get_mm(key); /* implies MB (B) */
344 : break;
345 : default:
346 : /*
347 : * Private futexes do not hold reference on an inode or
348 : * mm, therefore the only purpose of calling get_futex_key_refs
349 : * is because we need the barrier for the lockless waiter check.
350 : */
351 1037619 : smp_mb(); /* explicit MB (B) */
352 : }
353 : }
354 :
355 : /*
356 : * Drop a reference to the resource addressed by a key.
357 : * The hash bucket spinlock must not be held. This is
358 : * a no-op for private futexes, see comment in the get
359 : * counterpart.
360 : */
361 1037616 : static void drop_futex_key_refs(union futex_key *key)
362 : {
363 1037616 : if (!key->both.ptr) {
364 : /* If we're here then we tried to put a key we failed to get */
365 : WARN_ON_ONCE(1);
366 1037616 : return;
367 : }
368 :
369 1037616 : switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
370 : case FUT_OFF_INODE:
371 0 : iput(key->shared.inode);
372 : break;
373 : case FUT_OFF_MMSHARED:
374 : mmdrop(key->private.mm);
375 : break;
376 : }
377 : }
378 :
379 : /**
380 : * get_futex_key() - Get parameters which are the keys for a futex
381 : * @uaddr: virtual address of the futex
382 : * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
383 : * @key: address where result is stored.
384 : * @rw: mapping needs to be read/write (values: VERIFY_READ,
385 : * VERIFY_WRITE)
386 : *
387 : * Return: a negative error code or 0
388 : *
389 : * The key words are stored in *key on success.
390 : *
391 : * For shared mappings, it's (page->index, file_inode(vma->vm_file),
392 : * offset_within_page). For private mappings, it's (uaddr, current->mm).
393 : * We can usually work out the index without swapping in the page.
394 : *
395 : * lock_page() might sleep, the caller should not hold a spinlock.
396 : */
397 : static int
398 1037619 : get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw)
399 : {
400 1037619 : unsigned long address = (unsigned long)uaddr;
401 1037619 : struct mm_struct *mm = current->mm;
402 : struct page *page, *page_head;
403 : int err, ro = 0;
404 :
405 : /*
406 : * The futex address must be "naturally" aligned.
407 : */
408 1037619 : key->both.offset = address % PAGE_SIZE;
409 1037619 : if (unlikely((address % sizeof(u32)) != 0))
410 : return -EINVAL;
411 1037619 : address -= key->both.offset;
412 :
413 1037619 : if (unlikely(!access_ok(rw, uaddr, sizeof(u32))))
414 : return -EFAULT;
415 :
416 : /*
417 : * PROCESS_PRIVATE futexes are fast.
418 : * As the mm cannot disappear under us and the 'key' only needs
419 : * virtual address, we dont even have to find the underlying vma.
420 : * Note : We do have to check 'uaddr' is a valid user address,
421 : * but access_ok() should be faster than find_vma()
422 : */
423 1037619 : if (!fshared) {
424 1037619 : key->private.mm = mm;
425 1037619 : key->private.address = address;
426 1037619 : get_futex_key_refs(key); /* implies MB (B) */
427 1037619 : return 0;
428 : }
429 :
430 : again:
431 0 : err = get_user_pages_fast(address, 1, 1, &page);
432 : /*
433 : * If write access is not required (eg. FUTEX_WAIT), try
434 : * and get read-only access.
435 : */
436 0 : if (err == -EFAULT && rw == VERIFY_READ) {
437 0 : err = get_user_pages_fast(address, 1, 0, &page);
438 : ro = 1;
439 : }
440 0 : if (err < 0)
441 : return err;
442 : else
443 : err = 0;
444 :
445 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
446 : page_head = page;
447 : if (unlikely(PageTail(page))) {
448 : put_page(page);
449 : /* serialize against __split_huge_page_splitting() */
450 : local_irq_disable();
451 : if (likely(__get_user_pages_fast(address, 1, !ro, &page) == 1)) {
452 : page_head = compound_head(page);
453 : /*
454 : * page_head is valid pointer but we must pin
455 : * it before taking the PG_lock and/or
456 : * PG_compound_lock. The moment we re-enable
457 : * irqs __split_huge_page_splitting() can
458 : * return and the head page can be freed from
459 : * under us. We can't take the PG_lock and/or
460 : * PG_compound_lock on a page that could be
461 : * freed from under us.
462 : */
463 : if (page != page_head) {
464 : get_page(page_head);
465 : put_page(page);
466 : }
467 : local_irq_enable();
468 : } else {
469 : local_irq_enable();
470 : goto again;
471 : }
472 : }
473 : #else
474 0 : page_head = compound_head(page);
475 0 : if (page != page_head) {
476 : get_page(page_head);
477 0 : put_page(page);
478 : }
479 : #endif
480 :
481 : lock_page(page_head);
482 :
483 : /*
484 : * If page_head->mapping is NULL, then it cannot be a PageAnon
485 : * page; but it might be the ZERO_PAGE or in the gate area or
486 : * in a special mapping (all cases which we are happy to fail);
487 : * or it may have been a good file page when get_user_pages_fast
488 : * found it, but truncated or holepunched or subjected to
489 : * invalidate_complete_page2 before we got the page lock (also
490 : * cases which we are happy to fail). And we hold a reference,
491 : * so refcount care in invalidate_complete_page's remove_mapping
492 : * prevents drop_caches from setting mapping to NULL beneath us.
493 : *
494 : * The case we do have to guard against is when memory pressure made
495 : * shmem_writepage move it from filecache to swapcache beneath us:
496 : * an unlikely race, but we do need to retry for page_head->mapping.
497 : */
498 0 : if (!page_head->mapping) {
499 : int shmem_swizzled = PageSwapCache(page_head);
500 0 : unlock_page(page_head);
501 0 : put_page(page_head);
502 0 : if (shmem_swizzled)
503 : goto again;
504 : return -EFAULT;
505 : }
506 :
507 : /*
508 : * Private mappings are handled in a simple way.
509 : *
510 : * NOTE: When userspace waits on a MAP_SHARED mapping, even if
511 : * it's a read-only handle, it's expected that futexes attach to
512 : * the object not the particular process.
513 : */
514 0 : if (PageAnon(page_head)) {
515 : /*
516 : * A RO anonymous page will never change and thus doesn't make
517 : * sense for futex operations.
518 : */
519 0 : if (ro) {
520 : err = -EFAULT;
521 : goto out;
522 : }
523 :
524 0 : key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
525 0 : key->private.mm = mm;
526 0 : key->private.address = address;
527 : } else {
528 0 : key->both.offset |= FUT_OFF_INODE; /* inode-based key */
529 0 : key->shared.inode = page_head->mapping->host;
530 0 : key->shared.pgoff = basepage_index(page);
531 : }
532 :
533 0 : get_futex_key_refs(key); /* implies MB (B) */
534 :
535 : out:
536 0 : unlock_page(page_head);
537 0 : put_page(page_head);
538 0 : return err;
539 : }
540 :
541 : static inline void put_futex_key(union futex_key *key)
542 : {
543 692359 : drop_futex_key_refs(key);
544 : }
545 :
546 : /**
547 : * fault_in_user_writeable() - Fault in user address and verify RW access
548 : * @uaddr: pointer to faulting user space address
549 : *
550 : * Slow path to fixup the fault we just took in the atomic write
551 : * access to @uaddr.
552 : *
553 : * We have no generic implementation of a non-destructive write to the
554 : * user address. We know that we faulted in the atomic pagefault
555 : * disabled section so we can as well avoid the #PF overhead by
556 : * calling get_user_pages() right away.
557 : */
558 0 : static int fault_in_user_writeable(u32 __user *uaddr)
559 : {
560 0 : struct mm_struct *mm = current->mm;
561 : int ret;
562 :
563 0 : down_read(&mm->mmap_sem);
564 0 : ret = fixup_user_fault(current, mm, (unsigned long)uaddr,
565 : FAULT_FLAG_WRITE);
566 0 : up_read(&mm->mmap_sem);
567 :
568 0 : return ret < 0 ? ret : 0;
569 : }
570 :
571 : /**
572 : * futex_top_waiter() - Return the highest priority waiter on a futex
573 : * @hb: the hash bucket the futex_q's reside in
574 : * @key: the futex key (to distinguish it from other futex futex_q's)
575 : *
576 : * Must be called with the hb lock held.
577 : */
578 0 : static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb,
579 : union futex_key *key)
580 : {
581 : struct futex_q *this;
582 :
583 0 : plist_for_each_entry(this, &hb->chain, list) {
584 0 : if (match_futex(&this->key, key))
585 : return this;
586 : }
587 : return NULL;
588 : }
589 :
590 1 : static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr,
591 : u32 uval, u32 newval)
592 : {
593 : int ret;
594 :
595 : pagefault_disable();
596 : ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval);
597 : pagefault_enable();
598 :
599 1 : return ret;
600 : }
601 :
602 345538 : static int get_futex_value_locked(u32 *dest, u32 __user *from)
603 : {
604 : int ret;
605 :
606 : pagefault_disable();
607 345538 : ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
608 : pagefault_enable();
609 :
610 345538 : return ret ? -EFAULT : 0;
611 : }
612 :
613 :
614 : /*
615 : * PI code:
616 : */
617 0 : static int refill_pi_state_cache(void)
618 : {
619 : struct futex_pi_state *pi_state;
620 :
621 0 : if (likely(current->pi_state_cache))
622 : return 0;
623 :
624 : pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
625 :
626 0 : if (!pi_state)
627 : return -ENOMEM;
628 :
629 0 : INIT_LIST_HEAD(&pi_state->list);
630 : /* pi_mutex gets initialized later */
631 0 : pi_state->owner = NULL;
632 0 : atomic_set(&pi_state->refcount, 1);
633 0 : pi_state->key = FUTEX_KEY_INIT;
634 :
635 0 : current->pi_state_cache = pi_state;
636 :
637 0 : return 0;
638 : }
639 :
640 : static struct futex_pi_state * alloc_pi_state(void)
641 : {
642 0 : struct futex_pi_state *pi_state = current->pi_state_cache;
643 :
644 : WARN_ON(!pi_state);
645 0 : current->pi_state_cache = NULL;
646 :
647 : return pi_state;
648 : }
649 :
650 : /*
651 : * Must be called with the hb lock held.
652 : */
653 0 : static void free_pi_state(struct futex_pi_state *pi_state)
654 : {
655 0 : if (!pi_state)
656 : return;
657 :
658 0 : if (!atomic_dec_and_test(&pi_state->refcount))
659 : return;
660 :
661 : /*
662 : * If pi_state->owner is NULL, the owner is most probably dying
663 : * and has cleaned up the pi_state already
664 : */
665 0 : if (pi_state->owner) {
666 0 : raw_spin_lock_irq(&pi_state->owner->pi_lock);
667 0 : list_del_init(&pi_state->list);
668 0 : raw_spin_unlock_irq(&pi_state->owner->pi_lock);
669 :
670 0 : rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
671 : }
672 :
673 0 : if (current->pi_state_cache)
674 0 : kfree(pi_state);
675 : else {
676 : /*
677 : * pi_state->list is already empty.
678 : * clear pi_state->owner.
679 : * refcount is at 0 - put it back to 1.
680 : */
681 0 : pi_state->owner = NULL;
682 0 : atomic_set(&pi_state->refcount, 1);
683 0 : current->pi_state_cache = pi_state;
684 : }
685 : }
686 :
687 : /*
688 : * Look up the task based on what TID userspace gave us.
689 : * We dont trust it.
690 : */
691 0 : static struct task_struct * futex_find_get_task(pid_t pid)
692 : {
693 : struct task_struct *p;
694 :
695 : rcu_read_lock();
696 0 : p = find_task_by_vpid(pid);
697 0 : if (p)
698 0 : get_task_struct(p);
699 :
700 : rcu_read_unlock();
701 :
702 0 : return p;
703 : }
704 :
705 : /*
706 : * This task is holding PI mutexes at exit time => bad.
707 : * Kernel cleans up PI-state, but userspace is likely hosed.
708 : * (Robust-futex cleanup is separate and might save the day for userspace.)
709 : */
710 0 : void exit_pi_state_list(struct task_struct *curr)
711 : {
712 0 : struct list_head *next, *head = &curr->pi_state_list;
713 : struct futex_pi_state *pi_state;
714 : struct futex_hash_bucket *hb;
715 0 : union futex_key key = FUTEX_KEY_INIT;
716 :
717 0 : if (!futex_cmpxchg_enabled)
718 0 : return;
719 : /*
720 : * We are a ZOMBIE and nobody can enqueue itself on
721 : * pi_state_list anymore, but we have to be careful
722 : * versus waiters unqueueing themselves:
723 : */
724 0 : raw_spin_lock_irq(&curr->pi_lock);
725 0 : while (!list_empty(head)) {
726 :
727 : next = head->next;
728 : pi_state = list_entry(next, struct futex_pi_state, list);
729 0 : key = pi_state->key;
730 : hb = hash_futex(&key);
731 0 : raw_spin_unlock_irq(&curr->pi_lock);
732 :
733 : spin_lock(&hb->lock);
734 :
735 0 : raw_spin_lock_irq(&curr->pi_lock);
736 : /*
737 : * We dropped the pi-lock, so re-check whether this
738 : * task still owns the PI-state:
739 : */
740 0 : if (head->next != next) {
741 : spin_unlock(&hb->lock);
742 0 : continue;
743 : }
744 :
745 : WARN_ON(pi_state->owner != curr);
746 : WARN_ON(list_empty(&pi_state->list));
747 0 : list_del_init(&pi_state->list);
748 0 : pi_state->owner = NULL;
749 0 : raw_spin_unlock_irq(&curr->pi_lock);
750 :
751 0 : rt_mutex_unlock(&pi_state->pi_mutex);
752 :
753 : spin_unlock(&hb->lock);
754 :
755 0 : raw_spin_lock_irq(&curr->pi_lock);
756 : }
757 0 : raw_spin_unlock_irq(&curr->pi_lock);
758 : }
759 :
760 : /*
761 : * We need to check the following states:
762 : *
763 : * Waiter | pi_state | pi->owner | uTID | uODIED | ?
764 : *
765 : * [1] NULL | --- | --- | 0 | 0/1 | Valid
766 : * [2] NULL | --- | --- | >0 | 0/1 | Valid
767 : *
768 : * [3] Found | NULL | -- | Any | 0/1 | Invalid
769 : *
770 : * [4] Found | Found | NULL | 0 | 1 | Valid
771 : * [5] Found | Found | NULL | >0 | 1 | Invalid
772 : *
773 : * [6] Found | Found | task | 0 | 1 | Valid
774 : *
775 : * [7] Found | Found | NULL | Any | 0 | Invalid
776 : *
777 : * [8] Found | Found | task | ==taskTID | 0/1 | Valid
778 : * [9] Found | Found | task | 0 | 0 | Invalid
779 : * [10] Found | Found | task | !=taskTID | 0/1 | Invalid
780 : *
781 : * [1] Indicates that the kernel can acquire the futex atomically. We
782 : * came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit.
783 : *
784 : * [2] Valid, if TID does not belong to a kernel thread. If no matching
785 : * thread is found then it indicates that the owner TID has died.
786 : *
787 : * [3] Invalid. The waiter is queued on a non PI futex
788 : *
789 : * [4] Valid state after exit_robust_list(), which sets the user space
790 : * value to FUTEX_WAITERS | FUTEX_OWNER_DIED.
791 : *
792 : * [5] The user space value got manipulated between exit_robust_list()
793 : * and exit_pi_state_list()
794 : *
795 : * [6] Valid state after exit_pi_state_list() which sets the new owner in
796 : * the pi_state but cannot access the user space value.
797 : *
798 : * [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set.
799 : *
800 : * [8] Owner and user space value match
801 : *
802 : * [9] There is no transient state which sets the user space TID to 0
803 : * except exit_robust_list(), but this is indicated by the
804 : * FUTEX_OWNER_DIED bit. See [4]
805 : *
806 : * [10] There is no transient state which leaves owner and user space
807 : * TID out of sync.
808 : */
809 :
810 : /*
811 : * Validate that the existing waiter has a pi_state and sanity check
812 : * the pi_state against the user space value. If correct, attach to
813 : * it.
814 : */
815 0 : static int attach_to_pi_state(u32 uval, struct futex_pi_state *pi_state,
816 : struct futex_pi_state **ps)
817 : {
818 0 : pid_t pid = uval & FUTEX_TID_MASK;
819 :
820 : /*
821 : * Userspace might have messed up non-PI and PI futexes [3]
822 : */
823 0 : if (unlikely(!pi_state))
824 : return -EINVAL;
825 :
826 0 : WARN_ON(!atomic_read(&pi_state->refcount));
827 :
828 : /*
829 : * Handle the owner died case:
830 : */
831 0 : if (uval & FUTEX_OWNER_DIED) {
832 : /*
833 : * exit_pi_state_list sets owner to NULL and wakes the
834 : * topmost waiter. The task which acquires the
835 : * pi_state->rt_mutex will fixup owner.
836 : */
837 0 : if (!pi_state->owner) {
838 : /*
839 : * No pi state owner, but the user space TID
840 : * is not 0. Inconsistent state. [5]
841 : */
842 0 : if (pid)
843 : return -EINVAL;
844 : /*
845 : * Take a ref on the state and return success. [4]
846 : */
847 : goto out_state;
848 : }
849 :
850 : /*
851 : * If TID is 0, then either the dying owner has not
852 : * yet executed exit_pi_state_list() or some waiter
853 : * acquired the rtmutex in the pi state, but did not
854 : * yet fixup the TID in user space.
855 : *
856 : * Take a ref on the state and return success. [6]
857 : */
858 0 : if (!pid)
859 : goto out_state;
860 : } else {
861 : /*
862 : * If the owner died bit is not set, then the pi_state
863 : * must have an owner. [7]
864 : */
865 0 : if (!pi_state->owner)
866 : return -EINVAL;
867 : }
868 :
869 : /*
870 : * Bail out if user space manipulated the futex value. If pi
871 : * state exists then the owner TID must be the same as the
872 : * user space TID. [9/10]
873 : */
874 0 : if (pid != task_pid_vnr(pi_state->owner))
875 : return -EINVAL;
876 : out_state:
877 0 : atomic_inc(&pi_state->refcount);
878 0 : *ps = pi_state;
879 0 : return 0;
880 : }
881 :
882 : /*
883 : * Lookup the task for the TID provided from user space and attach to
884 : * it after doing proper sanity checks.
885 : */
886 0 : static int attach_to_pi_owner(u32 uval, union futex_key *key,
887 : struct futex_pi_state **ps)
888 : {
889 0 : pid_t pid = uval & FUTEX_TID_MASK;
890 : struct futex_pi_state *pi_state;
891 : struct task_struct *p;
892 :
893 : /*
894 : * We are the first waiter - try to look up the real owner and attach
895 : * the new pi_state to it, but bail out when TID = 0 [1]
896 : */
897 0 : if (!pid)
898 : return -ESRCH;
899 0 : p = futex_find_get_task(pid);
900 0 : if (!p)
901 : return -ESRCH;
902 :
903 0 : if (!p->mm) {
904 : put_task_struct(p);
905 : return -EPERM;
906 : }
907 :
908 : /*
909 : * We need to look at the task state flags to figure out,
910 : * whether the task is exiting. To protect against the do_exit
911 : * change of the task flags, we do this protected by
912 : * p->pi_lock:
913 : */
914 0 : raw_spin_lock_irq(&p->pi_lock);
915 0 : if (unlikely(p->flags & PF_EXITING)) {
916 : /*
917 : * The task is on the way out. When PF_EXITPIDONE is
918 : * set, we know that the task has finished the
919 : * cleanup:
920 : */
921 0 : int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
922 :
923 0 : raw_spin_unlock_irq(&p->pi_lock);
924 : put_task_struct(p);
925 0 : return ret;
926 : }
927 :
928 : /*
929 : * No existing pi state. First waiter. [2]
930 : */
931 : pi_state = alloc_pi_state();
932 :
933 : /*
934 : * Initialize the pi_mutex in locked state and make @p
935 : * the owner of it:
936 : */
937 0 : rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
938 :
939 : /* Store the key for possible exit cleanups: */
940 0 : pi_state->key = *key;
941 :
942 : WARN_ON(!list_empty(&pi_state->list));
943 0 : list_add(&pi_state->list, &p->pi_state_list);
944 0 : pi_state->owner = p;
945 0 : raw_spin_unlock_irq(&p->pi_lock);
946 :
947 : put_task_struct(p);
948 :
949 0 : *ps = pi_state;
950 :
951 0 : return 0;
952 : }
953 :
954 0 : static int lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
955 : union futex_key *key, struct futex_pi_state **ps)
956 : {
957 0 : struct futex_q *match = futex_top_waiter(hb, key);
958 :
959 : /*
960 : * If there is a waiter on that futex, validate it and
961 : * attach to the pi_state when the validation succeeds.
962 : */
963 0 : if (match)
964 0 : return attach_to_pi_state(uval, match->pi_state, ps);
965 :
966 : /*
967 : * We are the first waiter - try to look up the owner based on
968 : * @uval and attach to it.
969 : */
970 0 : return attach_to_pi_owner(uval, key, ps);
971 : }
972 :
973 0 : static int lock_pi_update_atomic(u32 __user *uaddr, u32 uval, u32 newval)
974 : {
975 : u32 uninitialized_var(curval);
976 :
977 0 : if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)))
978 : return -EFAULT;
979 :
980 : /*If user space value changed, let the caller retry */
981 0 : return curval != uval ? -EAGAIN : 0;
982 : }
983 :
984 : /**
985 : * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
986 : * @uaddr: the pi futex user address
987 : * @hb: the pi futex hash bucket
988 : * @key: the futex key associated with uaddr and hb
989 : * @ps: the pi_state pointer where we store the result of the
990 : * lookup
991 : * @task: the task to perform the atomic lock work for. This will
992 : * be "current" except in the case of requeue pi.
993 : * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
994 : *
995 : * Return:
996 : * 0 - ready to wait;
997 : * 1 - acquired the lock;
998 : * <0 - error
999 : *
1000 : * The hb->lock and futex_key refs shall be held by the caller.
1001 : */
1002 0 : static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb,
1003 : union futex_key *key,
1004 : struct futex_pi_state **ps,
1005 : struct task_struct *task, int set_waiters)
1006 : {
1007 0 : u32 uval, newval, vpid = task_pid_vnr(task);
1008 : struct futex_q *match;
1009 : int ret;
1010 :
1011 : /*
1012 : * Read the user space value first so we can validate a few
1013 : * things before proceeding further.
1014 : */
1015 0 : if (get_futex_value_locked(&uval, uaddr))
1016 : return -EFAULT;
1017 :
1018 : /*
1019 : * Detect deadlocks.
1020 : */
1021 0 : if ((unlikely((uval & FUTEX_TID_MASK) == vpid)))
1022 : return -EDEADLK;
1023 :
1024 : /*
1025 : * Lookup existing state first. If it exists, try to attach to
1026 : * its pi_state.
1027 : */
1028 0 : match = futex_top_waiter(hb, key);
1029 0 : if (match)
1030 0 : return attach_to_pi_state(uval, match->pi_state, ps);
1031 :
1032 : /*
1033 : * No waiter and user TID is 0. We are here because the
1034 : * waiters or the owner died bit is set or called from
1035 : * requeue_cmp_pi or for whatever reason something took the
1036 : * syscall.
1037 : */
1038 0 : if (!(uval & FUTEX_TID_MASK)) {
1039 : /*
1040 : * We take over the futex. No other waiters and the user space
1041 : * TID is 0. We preserve the owner died bit.
1042 : */
1043 0 : newval = uval & FUTEX_OWNER_DIED;
1044 0 : newval |= vpid;
1045 :
1046 : /* The futex requeue_pi code can enforce the waiters bit */
1047 0 : if (set_waiters)
1048 0 : newval |= FUTEX_WAITERS;
1049 :
1050 0 : ret = lock_pi_update_atomic(uaddr, uval, newval);
1051 : /* If the take over worked, return 1 */
1052 0 : return ret < 0 ? ret : 1;
1053 : }
1054 :
1055 : /*
1056 : * First waiter. Set the waiters bit before attaching ourself to
1057 : * the owner. If owner tries to unlock, it will be forced into
1058 : * the kernel and blocked on hb->lock.
1059 : */
1060 0 : newval = uval | FUTEX_WAITERS;
1061 0 : ret = lock_pi_update_atomic(uaddr, uval, newval);
1062 0 : if (ret)
1063 : return ret;
1064 : /*
1065 : * If the update of the user space value succeeded, we try to
1066 : * attach to the owner. If that fails, no harm done, we only
1067 : * set the FUTEX_WAITERS bit in the user space variable.
1068 : */
1069 0 : return attach_to_pi_owner(uval, key, ps);
1070 : }
1071 :
1072 : /**
1073 : * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket
1074 : * @q: The futex_q to unqueue
1075 : *
1076 : * The q->lock_ptr must not be NULL and must be held by the caller.
1077 : */
1078 0 : static void __unqueue_futex(struct futex_q *q)
1079 : {
1080 : struct futex_hash_bucket *hb;
1081 :
1082 345257 : if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr))
1083 345257 : || WARN_ON(plist_node_empty(&q->list)))
1084 0 : return;
1085 :
1086 345257 : hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock);
1087 345257 : plist_del(&q->list, &hb->chain);
1088 : hb_waiters_dec(hb);
1089 : }
1090 :
1091 : /*
1092 : * The hash bucket lock must be held when this is called.
1093 : * Afterwards, the futex_q must not be accessed.
1094 : */
1095 345257 : static void wake_futex(struct futex_q *q)
1096 : {
1097 345257 : struct task_struct *p = q->task;
1098 :
1099 345257 : if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n"))
1100 345257 : return;
1101 :
1102 : /*
1103 : * We set q->lock_ptr = NULL _before_ we wake up the task. If
1104 : * a non-futex wake up happens on another CPU then the task
1105 : * might exit and p would dereference a non-existing task
1106 : * struct. Prevent this by holding a reference on p across the
1107 : * wake up.
1108 : */
1109 345257 : get_task_struct(p);
1110 :
1111 : __unqueue_futex(q);
1112 : /*
1113 : * The waiting task can free the futex_q as soon as
1114 : * q->lock_ptr = NULL is written, without taking any locks. A
1115 : * memory barrier is required here to prevent the following
1116 : * store to lock_ptr from getting ahead of the plist_del.
1117 : */
1118 345257 : smp_wmb();
1119 345257 : q->lock_ptr = NULL;
1120 :
1121 345257 : wake_up_state(p, TASK_NORMAL);
1122 : put_task_struct(p);
1123 : }
1124 :
1125 0 : static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
1126 : {
1127 : struct task_struct *new_owner;
1128 0 : struct futex_pi_state *pi_state = this->pi_state;
1129 : u32 uninitialized_var(curval), newval;
1130 : int ret = 0;
1131 :
1132 0 : if (!pi_state)
1133 : return -EINVAL;
1134 :
1135 : /*
1136 : * If current does not own the pi_state then the futex is
1137 : * inconsistent and user space fiddled with the futex value.
1138 : */
1139 0 : if (pi_state->owner != current)
1140 : return -EINVAL;
1141 :
1142 0 : raw_spin_lock(&pi_state->pi_mutex.wait_lock);
1143 0 : new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
1144 :
1145 : /*
1146 : * It is possible that the next waiter (the one that brought
1147 : * this owner to the kernel) timed out and is no longer
1148 : * waiting on the lock.
1149 : */
1150 0 : if (!new_owner)
1151 0 : new_owner = this->task;
1152 :
1153 : /*
1154 : * We pass it to the next owner. The WAITERS bit is always
1155 : * kept enabled while there is PI state around. We cleanup the
1156 : * owner died bit, because we are the owner.
1157 : */
1158 0 : newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
1159 :
1160 0 : if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
1161 : ret = -EFAULT;
1162 0 : else if (curval != uval)
1163 : ret = -EINVAL;
1164 0 : if (ret) {
1165 0 : raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
1166 : return ret;
1167 : }
1168 :
1169 0 : raw_spin_lock_irq(&pi_state->owner->pi_lock);
1170 : WARN_ON(list_empty(&pi_state->list));
1171 0 : list_del_init(&pi_state->list);
1172 0 : raw_spin_unlock_irq(&pi_state->owner->pi_lock);
1173 :
1174 0 : raw_spin_lock_irq(&new_owner->pi_lock);
1175 : WARN_ON(!list_empty(&pi_state->list));
1176 0 : list_add(&pi_state->list, &new_owner->pi_state_list);
1177 0 : pi_state->owner = new_owner;
1178 0 : raw_spin_unlock_irq(&new_owner->pi_lock);
1179 :
1180 0 : raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
1181 0 : rt_mutex_unlock(&pi_state->pi_mutex);
1182 :
1183 : return 0;
1184 : }
1185 :
1186 : /*
1187 : * Express the locking dependencies for lockdep:
1188 : */
1189 : static inline void
1190 : double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
1191 : {
1192 173137 : if (hb1 <= hb2) {
1193 : spin_lock(&hb1->lock);
1194 172824 : if (hb1 < hb2)
1195 172824 : spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
1196 : } else { /* hb1 > hb2 */
1197 : spin_lock(&hb2->lock);
1198 313 : spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
1199 : }
1200 : }
1201 :
1202 : static inline void
1203 : double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
1204 : {
1205 : spin_unlock(&hb1->lock);
1206 173137 : if (hb1 != hb2)
1207 : spin_unlock(&hb2->lock);
1208 : }
1209 :
1210 : /*
1211 : * Wake up waiters matching bitset queued on this futex (uaddr).
1212 : */
1213 : static int
1214 345807 : futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
1215 : {
1216 : struct futex_hash_bucket *hb;
1217 : struct futex_q *this, *next;
1218 345807 : union futex_key key = FUTEX_KEY_INIT;
1219 : int ret;
1220 :
1221 345807 : if (!bitset)
1222 : return -EINVAL;
1223 :
1224 345807 : ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ);
1225 345807 : if (unlikely(ret != 0))
1226 : goto out;
1227 :
1228 345807 : hb = hash_futex(&key);
1229 :
1230 : /* Make sure we really have tasks to wakeup */
1231 : if (!hb_waiters_pending(hb))
1232 : goto out_put_key;
1233 :
1234 : spin_lock(&hb->lock);
1235 :
1236 345810 : plist_for_each_entry_safe(this, next, &hb->chain, list) {
1237 344324 : if (match_futex (&this->key, &key)) {
1238 172159 : if (this->pi_state || this->rt_waiter) {
1239 : ret = -EINVAL;
1240 : break;
1241 : }
1242 :
1243 : /* Check if one of the bits is set in both bitsets */
1244 172159 : if (!(this->bitset & bitset))
1245 0 : continue;
1246 :
1247 172159 : wake_futex(this);
1248 172159 : if (++ret >= nr_wake)
1249 : break;
1250 : }
1251 : }
1252 :
1253 : spin_unlock(&hb->lock);
1254 : out_put_key:
1255 345807 : put_futex_key(&key);
1256 : out:
1257 345807 : return ret;
1258 : }
1259 :
1260 : /*
1261 : * Wake up all waiters hashed on the physical page that is mapped
1262 : * to this virtual address:
1263 : */
1264 : static int
1265 173137 : futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
1266 : int nr_wake, int nr_wake2, int op)
1267 : {
1268 173137 : union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
1269 : struct futex_hash_bucket *hb1, *hb2;
1270 : struct futex_q *this, *next;
1271 : int ret, op_ret;
1272 :
1273 : retry:
1274 173137 : ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
1275 173137 : if (unlikely(ret != 0))
1276 : goto out;
1277 173137 : ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
1278 173137 : if (unlikely(ret != 0))
1279 : goto out_put_key1;
1280 :
1281 173137 : hb1 = hash_futex(&key1);
1282 173137 : hb2 = hash_futex(&key2);
1283 :
1284 : retry_private:
1285 : double_lock_hb(hb1, hb2);
1286 : op_ret = futex_atomic_op_inuser(op, uaddr2);
1287 173137 : if (unlikely(op_ret < 0)) {
1288 :
1289 : double_unlock_hb(hb1, hb2);
1290 :
1291 : #ifndef CONFIG_MMU
1292 : /*
1293 : * we don't get EFAULT from MMU faults if we don't have an MMU,
1294 : * but we might get them from range checking
1295 : */
1296 : ret = op_ret;
1297 : goto out_put_keys;
1298 : #endif
1299 :
1300 0 : if (unlikely(op_ret != -EFAULT)) {
1301 : ret = op_ret;
1302 : goto out_put_keys;
1303 : }
1304 :
1305 0 : ret = fault_in_user_writeable(uaddr2);
1306 0 : if (ret)
1307 : goto out_put_keys;
1308 :
1309 0 : if (!(flags & FLAGS_SHARED))
1310 : goto retry_private;
1311 :
1312 0 : put_futex_key(&key2);
1313 0 : put_futex_key(&key1);
1314 : goto retry;
1315 : }
1316 :
1317 173137 : plist_for_each_entry_safe(this, next, &hb1->chain, list) {
1318 346196 : if (match_futex (&this->key, &key1)) {
1319 173098 : if (this->pi_state || this->rt_waiter) {
1320 : ret = -EINVAL;
1321 : goto out_unlock;
1322 : }
1323 173098 : wake_futex(this);
1324 173098 : if (++ret >= nr_wake)
1325 : break;
1326 : }
1327 : }
1328 :
1329 173137 : if (op_ret > 0) {
1330 : op_ret = 0;
1331 38 : plist_for_each_entry_safe(this, next, &hb2->chain, list) {
1332 0 : if (match_futex (&this->key, &key2)) {
1333 0 : if (this->pi_state || this->rt_waiter) {
1334 : ret = -EINVAL;
1335 : goto out_unlock;
1336 : }
1337 0 : wake_futex(this);
1338 0 : if (++op_ret >= nr_wake2)
1339 : break;
1340 : }
1341 : }
1342 38 : ret += op_ret;
1343 : }
1344 :
1345 : out_unlock:
1346 : double_unlock_hb(hb1, hb2);
1347 : out_put_keys:
1348 173137 : put_futex_key(&key2);
1349 : out_put_key1:
1350 173137 : put_futex_key(&key1);
1351 : out:
1352 173137 : return ret;
1353 : }
1354 :
1355 : /**
1356 : * requeue_futex() - Requeue a futex_q from one hb to another
1357 : * @q: the futex_q to requeue
1358 : * @hb1: the source hash_bucket
1359 : * @hb2: the target hash_bucket
1360 : * @key2: the new key for the requeued futex_q
1361 : */
1362 : static inline
1363 : void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1,
1364 : struct futex_hash_bucket *hb2, union futex_key *key2)
1365 : {
1366 :
1367 : /*
1368 : * If key1 and key2 hash to the same bucket, no need to
1369 : * requeue.
1370 : */
1371 0 : if (likely(&hb1->chain != &hb2->chain)) {
1372 0 : plist_del(&q->list, &hb1->chain);
1373 : hb_waiters_dec(hb1);
1374 0 : plist_add(&q->list, &hb2->chain);
1375 : hb_waiters_inc(hb2);
1376 0 : q->lock_ptr = &hb2->lock;
1377 : }
1378 0 : get_futex_key_refs(key2);
1379 0 : q->key = *key2;
1380 : }
1381 :
1382 : /**
1383 : * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
1384 : * @q: the futex_q
1385 : * @key: the key of the requeue target futex
1386 : * @hb: the hash_bucket of the requeue target futex
1387 : *
1388 : * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1389 : * target futex if it is uncontended or via a lock steal. Set the futex_q key
1390 : * to the requeue target futex so the waiter can detect the wakeup on the right
1391 : * futex, but remove it from the hb and NULL the rt_waiter so it can detect
1392 : * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1393 : * to protect access to the pi_state to fixup the owner later. Must be called
1394 : * with both q->lock_ptr and hb->lock held.
1395 : */
1396 : static inline
1397 : void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key,
1398 : struct futex_hash_bucket *hb)
1399 : {
1400 0 : get_futex_key_refs(key);
1401 0 : q->key = *key;
1402 :
1403 0 : __unqueue_futex(q);
1404 :
1405 : WARN_ON(!q->rt_waiter);
1406 0 : q->rt_waiter = NULL;
1407 :
1408 0 : q->lock_ptr = &hb->lock;
1409 :
1410 0 : wake_up_state(q->task, TASK_NORMAL);
1411 : }
1412 :
1413 : /**
1414 : * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
1415 : * @pifutex: the user address of the to futex
1416 : * @hb1: the from futex hash bucket, must be locked by the caller
1417 : * @hb2: the to futex hash bucket, must be locked by the caller
1418 : * @key1: the from futex key
1419 : * @key2: the to futex key
1420 : * @ps: address to store the pi_state pointer
1421 : * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
1422 : *
1423 : * Try and get the lock on behalf of the top waiter if we can do it atomically.
1424 : * Wake the top waiter if we succeed. If the caller specified set_waiters,
1425 : * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1426 : * hb1 and hb2 must be held by the caller.
1427 : *
1428 : * Return:
1429 : * 0 - failed to acquire the lock atomically;
1430 : * >0 - acquired the lock, return value is vpid of the top_waiter
1431 : * <0 - error
1432 : */
1433 0 : static int futex_proxy_trylock_atomic(u32 __user *pifutex,
1434 : struct futex_hash_bucket *hb1,
1435 : struct futex_hash_bucket *hb2,
1436 : union futex_key *key1, union futex_key *key2,
1437 : struct futex_pi_state **ps, int set_waiters)
1438 : {
1439 : struct futex_q *top_waiter = NULL;
1440 : u32 curval;
1441 : int ret, vpid;
1442 :
1443 0 : if (get_futex_value_locked(&curval, pifutex))
1444 : return -EFAULT;
1445 :
1446 : /*
1447 : * Find the top_waiter and determine if there are additional waiters.
1448 : * If the caller intends to requeue more than 1 waiter to pifutex,
1449 : * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1450 : * as we have means to handle the possible fault. If not, don't set
1451 : * the bit unecessarily as it will force the subsequent unlock to enter
1452 : * the kernel.
1453 : */
1454 0 : top_waiter = futex_top_waiter(hb1, key1);
1455 :
1456 : /* There are no waiters, nothing for us to do. */
1457 0 : if (!top_waiter)
1458 : return 0;
1459 :
1460 : /* Ensure we requeue to the expected futex. */
1461 0 : if (!match_futex(top_waiter->requeue_pi_key, key2))
1462 : return -EINVAL;
1463 :
1464 : /*
1465 : * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1466 : * the contended case or if set_waiters is 1. The pi_state is returned
1467 : * in ps in contended cases.
1468 : */
1469 0 : vpid = task_pid_vnr(top_waiter->task);
1470 0 : ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task,
1471 : set_waiters);
1472 0 : if (ret == 1) {
1473 : requeue_pi_wake_futex(top_waiter, key2, hb2);
1474 0 : return vpid;
1475 : }
1476 : return ret;
1477 : }
1478 :
1479 : /**
1480 : * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
1481 : * @uaddr1: source futex user address
1482 : * @flags: futex flags (FLAGS_SHARED, etc.)
1483 : * @uaddr2: target futex user address
1484 : * @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1485 : * @nr_requeue: number of waiters to requeue (0-INT_MAX)
1486 : * @cmpval: @uaddr1 expected value (or %NULL)
1487 : * @requeue_pi: if we are attempting to requeue from a non-pi futex to a
1488 : * pi futex (pi to pi requeue is not supported)
1489 : *
1490 : * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1491 : * uaddr2 atomically on behalf of the top waiter.
1492 : *
1493 : * Return:
1494 : * >=0 - on success, the number of tasks requeued or woken;
1495 : * <0 - on error
1496 : */
1497 0 : static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
1498 : u32 __user *uaddr2, int nr_wake, int nr_requeue,
1499 : u32 *cmpval, int requeue_pi)
1500 : {
1501 0 : union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
1502 : int drop_count = 0, task_count = 0, ret;
1503 0 : struct futex_pi_state *pi_state = NULL;
1504 : struct futex_hash_bucket *hb1, *hb2;
1505 : struct futex_q *this, *next;
1506 :
1507 0 : if (requeue_pi) {
1508 : /*
1509 : * Requeue PI only works on two distinct uaddrs. This
1510 : * check is only valid for private futexes. See below.
1511 : */
1512 0 : if (uaddr1 == uaddr2)
1513 : return -EINVAL;
1514 :
1515 : /*
1516 : * requeue_pi requires a pi_state, try to allocate it now
1517 : * without any locks in case it fails.
1518 : */
1519 0 : if (refill_pi_state_cache())
1520 : return -ENOMEM;
1521 : /*
1522 : * requeue_pi must wake as many tasks as it can, up to nr_wake
1523 : * + nr_requeue, since it acquires the rt_mutex prior to
1524 : * returning to userspace, so as to not leave the rt_mutex with
1525 : * waiters and no owner. However, second and third wake-ups
1526 : * cannot be predicted as they involve race conditions with the
1527 : * first wake and a fault while looking up the pi_state. Both
1528 : * pthread_cond_signal() and pthread_cond_broadcast() should
1529 : * use nr_wake=1.
1530 : */
1531 0 : if (nr_wake != 1)
1532 : return -EINVAL;
1533 : }
1534 :
1535 : retry:
1536 0 : ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
1537 0 : if (unlikely(ret != 0))
1538 : goto out;
1539 0 : ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2,
1540 : requeue_pi ? VERIFY_WRITE : VERIFY_READ);
1541 0 : if (unlikely(ret != 0))
1542 : goto out_put_key1;
1543 :
1544 : /*
1545 : * The check above which compares uaddrs is not sufficient for
1546 : * shared futexes. We need to compare the keys:
1547 : */
1548 0 : if (requeue_pi && match_futex(&key1, &key2)) {
1549 : ret = -EINVAL;
1550 : goto out_put_keys;
1551 : }
1552 :
1553 0 : hb1 = hash_futex(&key1);
1554 0 : hb2 = hash_futex(&key2);
1555 :
1556 : retry_private:
1557 : hb_waiters_inc(hb2);
1558 : double_lock_hb(hb1, hb2);
1559 :
1560 0 : if (likely(cmpval != NULL)) {
1561 : u32 curval;
1562 :
1563 0 : ret = get_futex_value_locked(&curval, uaddr1);
1564 :
1565 0 : if (unlikely(ret)) {
1566 : double_unlock_hb(hb1, hb2);
1567 : hb_waiters_dec(hb2);
1568 :
1569 0 : ret = get_user(curval, uaddr1);
1570 0 : if (ret)
1571 : goto out_put_keys;
1572 :
1573 0 : if (!(flags & FLAGS_SHARED))
1574 : goto retry_private;
1575 :
1576 : put_futex_key(&key2);
1577 : put_futex_key(&key1);
1578 0 : goto retry;
1579 : }
1580 0 : if (curval != *cmpval) {
1581 : ret = -EAGAIN;
1582 0 : goto out_unlock;
1583 : }
1584 : }
1585 :
1586 0 : if (requeue_pi && (task_count - nr_wake < nr_requeue)) {
1587 : /*
1588 : * Attempt to acquire uaddr2 and wake the top waiter. If we
1589 : * intend to requeue waiters, force setting the FUTEX_WAITERS
1590 : * bit. We force this here where we are able to easily handle
1591 : * faults rather in the requeue loop below.
1592 : */
1593 0 : ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1,
1594 : &key2, &pi_state, nr_requeue);
1595 :
1596 : /*
1597 : * At this point the top_waiter has either taken uaddr2 or is
1598 : * waiting on it. If the former, then the pi_state will not
1599 : * exist yet, look it up one more time to ensure we have a
1600 : * reference to it. If the lock was taken, ret contains the
1601 : * vpid of the top waiter task.
1602 : */
1603 0 : if (ret > 0) {
1604 : WARN_ON(pi_state);
1605 0 : drop_count++;
1606 0 : task_count++;
1607 : /*
1608 : * If we acquired the lock, then the user
1609 : * space value of uaddr2 should be vpid. It
1610 : * cannot be changed by the top waiter as it
1611 : * is blocked on hb2 lock if it tries to do
1612 : * so. If something fiddled with it behind our
1613 : * back the pi state lookup might unearth
1614 : * it. So we rather use the known value than
1615 : * rereading and handing potential crap to
1616 : * lookup_pi_state.
1617 : */
1618 0 : ret = lookup_pi_state(ret, hb2, &key2, &pi_state);
1619 : }
1620 :
1621 0 : switch (ret) {
1622 : case 0:
1623 : break;
1624 : case -EFAULT:
1625 0 : free_pi_state(pi_state);
1626 0 : pi_state = NULL;
1627 : double_unlock_hb(hb1, hb2);
1628 : hb_waiters_dec(hb2);
1629 : put_futex_key(&key2);
1630 : put_futex_key(&key1);
1631 0 : ret = fault_in_user_writeable(uaddr2);
1632 0 : if (!ret)
1633 : goto retry;
1634 : goto out;
1635 : case -EAGAIN:
1636 : /*
1637 : * Two reasons for this:
1638 : * - Owner is exiting and we just wait for the
1639 : * exit to complete.
1640 : * - The user space value changed.
1641 : */
1642 0 : free_pi_state(pi_state);
1643 0 : pi_state = NULL;
1644 : double_unlock_hb(hb1, hb2);
1645 : hb_waiters_dec(hb2);
1646 : put_futex_key(&key2);
1647 : put_futex_key(&key1);
1648 0 : cond_resched();
1649 0 : goto retry;
1650 : default:
1651 : goto out_unlock;
1652 : }
1653 : }
1654 :
1655 0 : plist_for_each_entry_safe(this, next, &hb1->chain, list) {
1656 0 : if (task_count - nr_wake >= nr_requeue)
1657 : break;
1658 :
1659 0 : if (!match_futex(&this->key, &key1))
1660 0 : continue;
1661 :
1662 : /*
1663 : * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
1664 : * be paired with each other and no other futex ops.
1665 : *
1666 : * We should never be requeueing a futex_q with a pi_state,
1667 : * which is awaiting a futex_unlock_pi().
1668 : */
1669 0 : if ((requeue_pi && !this->rt_waiter) ||
1670 0 : (!requeue_pi && this->rt_waiter) ||
1671 0 : this->pi_state) {
1672 : ret = -EINVAL;
1673 : break;
1674 : }
1675 :
1676 : /*
1677 : * Wake nr_wake waiters. For requeue_pi, if we acquired the
1678 : * lock, we already woke the top_waiter. If not, it will be
1679 : * woken by futex_unlock_pi().
1680 : */
1681 0 : if (++task_count <= nr_wake && !requeue_pi) {
1682 0 : wake_futex(this);
1683 0 : continue;
1684 : }
1685 :
1686 : /* Ensure we requeue to the expected futex for requeue_pi. */
1687 0 : if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) {
1688 : ret = -EINVAL;
1689 : break;
1690 : }
1691 :
1692 : /*
1693 : * Requeue nr_requeue waiters and possibly one more in the case
1694 : * of requeue_pi if we couldn't acquire the lock atomically.
1695 : */
1696 0 : if (requeue_pi) {
1697 : /* Prepare the waiter to take the rt_mutex. */
1698 0 : atomic_inc(&pi_state->refcount);
1699 0 : this->pi_state = pi_state;
1700 0 : ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex,
1701 : this->rt_waiter,
1702 : this->task);
1703 0 : if (ret == 1) {
1704 : /* We got the lock. */
1705 : requeue_pi_wake_futex(this, &key2, hb2);
1706 0 : drop_count++;
1707 0 : continue;
1708 0 : } else if (ret) {
1709 : /* -EDEADLK */
1710 0 : this->pi_state = NULL;
1711 0 : free_pi_state(pi_state);
1712 0 : goto out_unlock;
1713 : }
1714 : }
1715 : requeue_futex(this, hb1, hb2, &key2);
1716 0 : drop_count++;
1717 : }
1718 :
1719 : out_unlock:
1720 0 : free_pi_state(pi_state);
1721 : double_unlock_hb(hb1, hb2);
1722 : hb_waiters_dec(hb2);
1723 :
1724 : /*
1725 : * drop_futex_key_refs() must be called outside the spinlocks. During
1726 : * the requeue we moved futex_q's from the hash bucket at key1 to the
1727 : * one at key2 and updated their key pointer. We no longer need to
1728 : * hold the references to key1.
1729 : */
1730 0 : while (--drop_count >= 0)
1731 0 : drop_futex_key_refs(&key1);
1732 :
1733 : out_put_keys:
1734 : put_futex_key(&key2);
1735 : out_put_key1:
1736 : put_futex_key(&key1);
1737 : out:
1738 0 : return ret ? ret : task_count;
1739 : }
1740 :
1741 : /* The key must be already stored in q->key. */
1742 : static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
1743 : __acquires(&hb->lock)
1744 : {
1745 : struct futex_hash_bucket *hb;
1746 :
1747 345538 : hb = hash_futex(&q->key);
1748 :
1749 : /*
1750 : * Increment the counter before taking the lock so that
1751 : * a potential waker won't miss a to-be-slept task that is
1752 : * waiting for the spinlock. This is safe as all queue_lock()
1753 : * users end up calling queue_me(). Similarly, for housekeeping,
1754 : * decrement the counter at queue_unlock() when some error has
1755 : * occurred and we don't end up adding the task to the list.
1756 : */
1757 : hb_waiters_inc(hb);
1758 :
1759 345538 : q->lock_ptr = &hb->lock;
1760 :
1761 : spin_lock(&hb->lock); /* implies MB (A) */
1762 : return hb;
1763 : }
1764 :
1765 : static inline void
1766 : queue_unlock(struct futex_hash_bucket *hb)
1767 : __releases(&hb->lock)
1768 : {
1769 : spin_unlock(&hb->lock);
1770 : hb_waiters_dec(hb);
1771 : }
1772 :
1773 : /**
1774 : * queue_me() - Enqueue the futex_q on the futex_hash_bucket
1775 : * @q: The futex_q to enqueue
1776 : * @hb: The destination hash bucket
1777 : *
1778 : * The hb->lock must be held by the caller, and is released here. A call to
1779 : * queue_me() is typically paired with exactly one call to unqueue_me(). The
1780 : * exceptions involve the PI related operations, which may use unqueue_me_pi()
1781 : * or nothing if the unqueue is done as part of the wake process and the unqueue
1782 : * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
1783 : * an example).
1784 : */
1785 : static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
1786 : __releases(&hb->lock)
1787 : {
1788 : int prio;
1789 :
1790 : /*
1791 : * The priority used to register this element is
1792 : * - either the real thread-priority for the real-time threads
1793 : * (i.e. threads with a priority lower than MAX_RT_PRIO)
1794 : * - or MAX_RT_PRIO for non-RT threads.
1795 : * Thus, all RT-threads are woken first in priority order, and
1796 : * the others are woken last, in FIFO order.
1797 : */
1798 345260 : prio = min(current->normal_prio, MAX_RT_PRIO);
1799 :
1800 : plist_node_init(&q->list, prio);
1801 345260 : plist_add(&q->list, &hb->chain);
1802 345260 : q->task = current;
1803 : spin_unlock(&hb->lock);
1804 : }
1805 :
1806 : /**
1807 : * unqueue_me() - Remove the futex_q from its futex_hash_bucket
1808 : * @q: The futex_q to unqueue
1809 : *
1810 : * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must
1811 : * be paired with exactly one earlier call to queue_me().
1812 : *
1813 : * Return:
1814 : * 1 - if the futex_q was still queued (and we removed unqueued it);
1815 : * 0 - if the futex_q was already removed by the waking thread
1816 : */
1817 345257 : static int unqueue_me(struct futex_q *q)
1818 : {
1819 : spinlock_t *lock_ptr;
1820 : int ret = 0;
1821 :
1822 : /* In the common case we don't take the spinlock, which is nice. */
1823 : retry:
1824 345257 : lock_ptr = q->lock_ptr;
1825 345257 : barrier();
1826 345257 : if (lock_ptr != NULL) {
1827 : spin_lock(lock_ptr);
1828 : /*
1829 : * q->lock_ptr can change between reading it and
1830 : * spin_lock(), causing us to take the wrong lock. This
1831 : * corrects the race condition.
1832 : *
1833 : * Reasoning goes like this: if we have the wrong lock,
1834 : * q->lock_ptr must have changed (maybe several times)
1835 : * between reading it and the spin_lock(). It can
1836 : * change again after the spin_lock() but only if it was
1837 : * already changed before the spin_lock(). It cannot,
1838 : * however, change back to the original value. Therefore
1839 : * we can detect whether we acquired the correct lock.
1840 : */
1841 0 : if (unlikely(lock_ptr != q->lock_ptr)) {
1842 : spin_unlock(lock_ptr);
1843 : goto retry;
1844 : }
1845 : __unqueue_futex(q);
1846 :
1847 : BUG_ON(q->pi_state);
1848 :
1849 : spin_unlock(lock_ptr);
1850 : ret = 1;
1851 : }
1852 :
1853 345257 : drop_futex_key_refs(&q->key);
1854 345257 : return ret;
1855 : }
1856 :
1857 : /*
1858 : * PI futexes can not be requeued and must remove themself from the
1859 : * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1860 : * and dropped here.
1861 : */
1862 0 : static void unqueue_me_pi(struct futex_q *q)
1863 : __releases(q->lock_ptr)
1864 : {
1865 : __unqueue_futex(q);
1866 :
1867 : BUG_ON(!q->pi_state);
1868 0 : free_pi_state(q->pi_state);
1869 0 : q->pi_state = NULL;
1870 :
1871 : spin_unlock(q->lock_ptr);
1872 0 : }
1873 :
1874 : /*
1875 : * Fixup the pi_state owner with the new owner.
1876 : *
1877 : * Must be called with hash bucket lock held and mm->sem held for non
1878 : * private futexes.
1879 : */
1880 0 : static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
1881 : struct task_struct *newowner)
1882 : {
1883 0 : u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;
1884 0 : struct futex_pi_state *pi_state = q->pi_state;
1885 0 : struct task_struct *oldowner = pi_state->owner;
1886 : u32 uval, uninitialized_var(curval), newval;
1887 : int ret;
1888 :
1889 : /* Owner died? */
1890 0 : if (!pi_state->owner)
1891 0 : newtid |= FUTEX_OWNER_DIED;
1892 :
1893 : /*
1894 : * We are here either because we stole the rtmutex from the
1895 : * previous highest priority waiter or we are the highest priority
1896 : * waiter but failed to get the rtmutex the first time.
1897 : * We have to replace the newowner TID in the user space variable.
1898 : * This must be atomic as we have to preserve the owner died bit here.
1899 : *
1900 : * Note: We write the user space value _before_ changing the pi_state
1901 : * because we can fault here. Imagine swapped out pages or a fork
1902 : * that marked all the anonymous memory readonly for cow.
1903 : *
1904 : * Modifying pi_state _before_ the user space value would
1905 : * leave the pi_state in an inconsistent state when we fault
1906 : * here, because we need to drop the hash bucket lock to
1907 : * handle the fault. This might be observed in the PID check
1908 : * in lookup_pi_state.
1909 : */
1910 : retry:
1911 0 : if (get_futex_value_locked(&uval, uaddr))
1912 : goto handle_fault;
1913 :
1914 : while (1) {
1915 0 : newval = (uval & FUTEX_OWNER_DIED) | newtid;
1916 :
1917 0 : if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
1918 : goto handle_fault;
1919 0 : if (curval == uval)
1920 : break;
1921 0 : uval = curval;
1922 : }
1923 :
1924 : /*
1925 : * We fixed up user space. Now we need to fix the pi_state
1926 : * itself.
1927 : */
1928 0 : if (pi_state->owner != NULL) {
1929 0 : raw_spin_lock_irq(&pi_state->owner->pi_lock);
1930 : WARN_ON(list_empty(&pi_state->list));
1931 0 : list_del_init(&pi_state->list);
1932 0 : raw_spin_unlock_irq(&pi_state->owner->pi_lock);
1933 : }
1934 :
1935 0 : pi_state->owner = newowner;
1936 :
1937 0 : raw_spin_lock_irq(&newowner->pi_lock);
1938 : WARN_ON(!list_empty(&pi_state->list));
1939 0 : list_add(&pi_state->list, &newowner->pi_state_list);
1940 0 : raw_spin_unlock_irq(&newowner->pi_lock);
1941 : return 0;
1942 :
1943 : /*
1944 : * To handle the page fault we need to drop the hash bucket
1945 : * lock here. That gives the other task (either the highest priority
1946 : * waiter itself or the task which stole the rtmutex) the
1947 : * chance to try the fixup of the pi_state. So once we are
1948 : * back from handling the fault we need to check the pi_state
1949 : * after reacquiring the hash bucket lock and before trying to
1950 : * do another fixup. When the fixup has been done already we
1951 : * simply return.
1952 : */
1953 : handle_fault:
1954 : spin_unlock(q->lock_ptr);
1955 :
1956 0 : ret = fault_in_user_writeable(uaddr);
1957 :
1958 : spin_lock(q->lock_ptr);
1959 :
1960 : /*
1961 : * Check if someone else fixed it for us:
1962 : */
1963 0 : if (pi_state->owner != oldowner)
1964 : return 0;
1965 :
1966 0 : if (ret)
1967 : return ret;
1968 :
1969 : goto retry;
1970 : }
1971 :
1972 : static long futex_wait_restart(struct restart_block *restart);
1973 :
1974 : /**
1975 : * fixup_owner() - Post lock pi_state and corner case management
1976 : * @uaddr: user address of the futex
1977 : * @q: futex_q (contains pi_state and access to the rt_mutex)
1978 : * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
1979 : *
1980 : * After attempting to lock an rt_mutex, this function is called to cleanup
1981 : * the pi_state owner as well as handle race conditions that may allow us to
1982 : * acquire the lock. Must be called with the hb lock held.
1983 : *
1984 : * Return:
1985 : * 1 - success, lock taken;
1986 : * 0 - success, lock not taken;
1987 : * <0 - on error (-EFAULT)
1988 : */
1989 0 : static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked)
1990 : {
1991 : struct task_struct *owner;
1992 : int ret = 0;
1993 :
1994 0 : if (locked) {
1995 : /*
1996 : * Got the lock. We might not be the anticipated owner if we
1997 : * did a lock-steal - fix up the PI-state in that case:
1998 : */
1999 0 : if (q->pi_state->owner != current)
2000 0 : ret = fixup_pi_state_owner(uaddr, q, current);
2001 : goto out;
2002 : }
2003 :
2004 : /*
2005 : * Catch the rare case, where the lock was released when we were on the
2006 : * way back before we locked the hash bucket.
2007 : */
2008 0 : if (q->pi_state->owner == current) {
2009 : /*
2010 : * Try to get the rt_mutex now. This might fail as some other
2011 : * task acquired the rt_mutex after we removed ourself from the
2012 : * rt_mutex waiters list.
2013 : */
2014 0 : if (rt_mutex_trylock(&q->pi_state->pi_mutex)) {
2015 : locked = 1;
2016 : goto out;
2017 : }
2018 :
2019 : /*
2020 : * pi_state is incorrect, some other task did a lock steal and
2021 : * we returned due to timeout or signal without taking the
2022 : * rt_mutex. Too late.
2023 : */
2024 0 : raw_spin_lock(&q->pi_state->pi_mutex.wait_lock);
2025 0 : owner = rt_mutex_owner(&q->pi_state->pi_mutex);
2026 0 : if (!owner)
2027 0 : owner = rt_mutex_next_owner(&q->pi_state->pi_mutex);
2028 0 : raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock);
2029 0 : ret = fixup_pi_state_owner(uaddr, q, owner);
2030 0 : goto out;
2031 : }
2032 :
2033 : /*
2034 : * Paranoia check. If we did not take the lock, then we should not be
2035 : * the owner of the rt_mutex.
2036 : */
2037 0 : if (rt_mutex_owner(&q->pi_state->pi_mutex) == current)
2038 0 : printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p "
2039 : "pi-state %p\n", ret,
2040 : q->pi_state->pi_mutex.owner,
2041 : q->pi_state->owner);
2042 :
2043 : out:
2044 0 : return ret ? ret : locked;
2045 : }
2046 :
2047 : /**
2048 : * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
2049 : * @hb: the futex hash bucket, must be locked by the caller
2050 : * @q: the futex_q to queue up on
2051 : * @timeout: the prepared hrtimer_sleeper, or null for no timeout
2052 : */
2053 345260 : static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
2054 : struct hrtimer_sleeper *timeout)
2055 : {
2056 : /*
2057 : * The task state is guaranteed to be set before another task can
2058 : * wake it. set_current_state() is implemented using set_mb() and
2059 : * queue_me() calls spin_unlock() upon completion, both serializing
2060 : * access to the hash list and forcing another memory barrier.
2061 : */
2062 345260 : set_current_state(TASK_INTERRUPTIBLE);
2063 : queue_me(q, hb);
2064 :
2065 : /* Arm the timer */
2066 345260 : if (timeout) {
2067 0 : hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
2068 0 : if (!hrtimer_active(&timeout->timer))
2069 0 : timeout->task = NULL;
2070 : }
2071 :
2072 : /*
2073 : * If we have been removed from the hash list, then another task
2074 : * has tried to wake us, and we can skip the call to schedule().
2075 : */
2076 345260 : if (likely(!plist_node_empty(&q->list))) {
2077 : /*
2078 : * If the timer has already expired, current will already be
2079 : * flagged for rescheduling. Only call schedule if there
2080 : * is no timeout, or if it has yet to expire.
2081 : */
2082 345260 : if (!timeout || timeout->task)
2083 : freezable_schedule();
2084 : }
2085 345257 : __set_current_state(TASK_RUNNING);
2086 345257 : }
2087 :
2088 : /**
2089 : * futex_wait_setup() - Prepare to wait on a futex
2090 : * @uaddr: the futex userspace address
2091 : * @val: the expected value
2092 : * @flags: futex flags (FLAGS_SHARED, etc.)
2093 : * @q: the associated futex_q
2094 : * @hb: storage for hash_bucket pointer to be returned to caller
2095 : *
2096 : * Setup the futex_q and locate the hash_bucket. Get the futex value and
2097 : * compare it with the expected value. Handle atomic faults internally.
2098 : * Return with the hb lock held and a q.key reference on success, and unlocked
2099 : * with no q.key reference on failure.
2100 : *
2101 : * Return:
2102 : * 0 - uaddr contains val and hb has been locked;
2103 : * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked
2104 : */
2105 345538 : static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
2106 : struct futex_q *q, struct futex_hash_bucket **hb)
2107 : {
2108 : u32 uval;
2109 : int ret;
2110 :
2111 : /*
2112 : * Access the page AFTER the hash-bucket is locked.
2113 : * Order is important:
2114 : *
2115 : * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
2116 : * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
2117 : *
2118 : * The basic logical guarantee of a futex is that it blocks ONLY
2119 : * if cond(var) is known to be true at the time of blocking, for
2120 : * any cond. If we locked the hash-bucket after testing *uaddr, that
2121 : * would open a race condition where we could block indefinitely with
2122 : * cond(var) false, which would violate the guarantee.
2123 : *
2124 : * On the other hand, we insert q and release the hash-bucket only
2125 : * after testing *uaddr. This guarantees that futex_wait() will NOT
2126 : * absorb a wakeup if *uaddr does not match the desired values
2127 : * while the syscall executes.
2128 : */
2129 : retry:
2130 345538 : ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ);
2131 345538 : if (unlikely(ret != 0))
2132 : return ret;
2133 :
2134 : retry_private:
2135 345538 : *hb = queue_lock(q);
2136 :
2137 345538 : ret = get_futex_value_locked(&uval, uaddr);
2138 :
2139 345538 : if (ret) {
2140 : queue_unlock(*hb);
2141 :
2142 0 : ret = get_user(uval, uaddr);
2143 0 : if (ret)
2144 : goto out;
2145 :
2146 0 : if (!(flags & FLAGS_SHARED))
2147 : goto retry_private;
2148 :
2149 : put_futex_key(&q->key);
2150 : goto retry;
2151 : }
2152 :
2153 345538 : if (uval != val) {
2154 : queue_unlock(*hb);
2155 : ret = -EWOULDBLOCK;
2156 : }
2157 :
2158 : out:
2159 345538 : if (ret)
2160 : put_futex_key(&q->key);
2161 345538 : return ret;
2162 : }
2163 :
2164 345538 : static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val,
2165 : ktime_t *abs_time, u32 bitset)
2166 : {
2167 : struct hrtimer_sleeper timeout, *to = NULL;
2168 : struct restart_block *restart;
2169 : struct futex_hash_bucket *hb;
2170 345538 : struct futex_q q = futex_q_init;
2171 : int ret;
2172 :
2173 345538 : if (!bitset)
2174 : return -EINVAL;
2175 345538 : q.bitset = bitset;
2176 :
2177 345538 : if (abs_time) {
2178 : to = &timeout;
2179 :
2180 0 : hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
2181 : CLOCK_REALTIME : CLOCK_MONOTONIC,
2182 : HRTIMER_MODE_ABS);
2183 0 : hrtimer_init_sleeper(to, current);
2184 0 : hrtimer_set_expires_range_ns(&to->timer, *abs_time,
2185 0 : current->timer_slack_ns);
2186 : }
2187 :
2188 : retry:
2189 : /*
2190 : * Prepare to wait on uaddr. On success, holds hb lock and increments
2191 : * q.key refs.
2192 : */
2193 345538 : ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
2194 345538 : if (ret)
2195 : goto out;
2196 :
2197 : /* queue_me and wait for wakeup, timeout, or a signal. */
2198 345260 : futex_wait_queue_me(hb, &q, to);
2199 :
2200 : /* If we were woken (and unqueued), we succeeded, whatever. */
2201 : ret = 0;
2202 : /* unqueue_me() drops q.key ref */
2203 345257 : if (!unqueue_me(&q))
2204 : goto out;
2205 : ret = -ETIMEDOUT;
2206 0 : if (to && !to->task)
2207 : goto out;
2208 :
2209 : /*
2210 : * We expect signal_pending(current), but we might be the
2211 : * victim of a spurious wakeup as well.
2212 : */
2213 0 : if (!signal_pending(current))
2214 : goto retry;
2215 :
2216 : ret = -ERESTARTSYS;
2217 0 : if (!abs_time)
2218 : goto out;
2219 :
2220 : restart = ¤t_thread_info()->restart_block;
2221 0 : restart->fn = futex_wait_restart;
2222 0 : restart->futex.uaddr = uaddr;
2223 0 : restart->futex.val = val;
2224 0 : restart->futex.time = abs_time->tv64;
2225 0 : restart->futex.bitset = bitset;
2226 0 : restart->futex.flags = flags | FLAGS_HAS_TIMEOUT;
2227 :
2228 : ret = -ERESTART_RESTARTBLOCK;
2229 :
2230 : out:
2231 345535 : if (to) {
2232 0 : hrtimer_cancel(&to->timer);
2233 : destroy_hrtimer_on_stack(&to->timer);
2234 : }
2235 345535 : return ret;
2236 : }
2237 :
2238 :
2239 0 : static long futex_wait_restart(struct restart_block *restart)
2240 : {
2241 0 : u32 __user *uaddr = restart->futex.uaddr;
2242 : ktime_t t, *tp = NULL;
2243 :
2244 0 : if (restart->futex.flags & FLAGS_HAS_TIMEOUT) {
2245 0 : t.tv64 = restart->futex.time;
2246 : tp = &t;
2247 : }
2248 0 : restart->fn = do_no_restart_syscall;
2249 :
2250 0 : return (long)futex_wait(uaddr, restart->futex.flags,
2251 : restart->futex.val, tp, restart->futex.bitset);
2252 : }
2253 :
2254 :
2255 : /*
2256 : * Userspace tried a 0 -> TID atomic transition of the futex value
2257 : * and failed. The kernel side here does the whole locking operation:
2258 : * if there are waiters then it will block, it does PI, etc. (Due to
2259 : * races the kernel might see a 0 value of the futex too.)
2260 : */
2261 0 : static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect,
2262 : ktime_t *time, int trylock)
2263 : {
2264 : struct hrtimer_sleeper timeout, *to = NULL;
2265 : struct futex_hash_bucket *hb;
2266 0 : struct futex_q q = futex_q_init;
2267 : int res, ret;
2268 :
2269 0 : if (refill_pi_state_cache())
2270 : return -ENOMEM;
2271 :
2272 0 : if (time) {
2273 : to = &timeout;
2274 : hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME,
2275 : HRTIMER_MODE_ABS);
2276 0 : hrtimer_init_sleeper(to, current);
2277 : hrtimer_set_expires(&to->timer, *time);
2278 : }
2279 :
2280 : retry:
2281 0 : ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE);
2282 0 : if (unlikely(ret != 0))
2283 : goto out;
2284 :
2285 : retry_private:
2286 : hb = queue_lock(&q);
2287 :
2288 0 : ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0);
2289 0 : if (unlikely(ret)) {
2290 0 : switch (ret) {
2291 : case 1:
2292 : /* We got the lock. */
2293 : ret = 0;
2294 : goto out_unlock_put_key;
2295 : case -EFAULT:
2296 : goto uaddr_faulted;
2297 : case -EAGAIN:
2298 : /*
2299 : * Two reasons for this:
2300 : * - Task is exiting and we just wait for the
2301 : * exit to complete.
2302 : * - The user space value changed.
2303 : */
2304 : queue_unlock(hb);
2305 : put_futex_key(&q.key);
2306 0 : cond_resched();
2307 : goto retry;
2308 : default:
2309 : goto out_unlock_put_key;
2310 : }
2311 : }
2312 :
2313 : /*
2314 : * Only actually queue now that the atomic ops are done:
2315 : */
2316 : queue_me(&q, hb);
2317 :
2318 : WARN_ON(!q.pi_state);
2319 : /*
2320 : * Block on the PI mutex:
2321 : */
2322 0 : if (!trylock) {
2323 0 : ret = rt_mutex_timed_futex_lock(&q.pi_state->pi_mutex, to);
2324 : } else {
2325 0 : ret = rt_mutex_trylock(&q.pi_state->pi_mutex);
2326 : /* Fixup the trylock return value: */
2327 0 : ret = ret ? 0 : -EWOULDBLOCK;
2328 : }
2329 :
2330 : spin_lock(q.lock_ptr);
2331 : /*
2332 : * Fixup the pi_state owner and possibly acquire the lock if we
2333 : * haven't already.
2334 : */
2335 0 : res = fixup_owner(uaddr, &q, !ret);
2336 : /*
2337 : * If fixup_owner() returned an error, proprogate that. If it acquired
2338 : * the lock, clear our -ETIMEDOUT or -EINTR.
2339 : */
2340 0 : if (res)
2341 0 : ret = (res < 0) ? res : 0;
2342 :
2343 : /*
2344 : * If fixup_owner() faulted and was unable to handle the fault, unlock
2345 : * it and return the fault to userspace.
2346 : */
2347 0 : if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current))
2348 0 : rt_mutex_unlock(&q.pi_state->pi_mutex);
2349 :
2350 : /* Unqueue and drop the lock */
2351 0 : unqueue_me_pi(&q);
2352 :
2353 : goto out_put_key;
2354 :
2355 : out_unlock_put_key:
2356 : queue_unlock(hb);
2357 :
2358 : out_put_key:
2359 : put_futex_key(&q.key);
2360 : out:
2361 : if (to)
2362 : destroy_hrtimer_on_stack(&to->timer);
2363 0 : return ret != -EINTR ? ret : -ERESTARTNOINTR;
2364 :
2365 : uaddr_faulted:
2366 : queue_unlock(hb);
2367 :
2368 0 : ret = fault_in_user_writeable(uaddr);
2369 0 : if (ret)
2370 : goto out_put_key;
2371 :
2372 0 : if (!(flags & FLAGS_SHARED))
2373 : goto retry_private;
2374 :
2375 : put_futex_key(&q.key);
2376 : goto retry;
2377 : }
2378 :
2379 : /*
2380 : * Userspace attempted a TID -> 0 atomic transition, and failed.
2381 : * This is the in-kernel slowpath: we look up the PI state (if any),
2382 : * and do the rt-mutex unlock.
2383 : */
2384 0 : static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags)
2385 : {
2386 0 : u32 uninitialized_var(curval), uval, vpid = task_pid_vnr(current);
2387 0 : union futex_key key = FUTEX_KEY_INIT;
2388 : struct futex_hash_bucket *hb;
2389 0 : struct futex_q *match;
2390 : int ret;
2391 :
2392 : retry:
2393 0 : if (get_user(uval, uaddr))
2394 : return -EFAULT;
2395 : /*
2396 : * We release only a lock we actually own:
2397 : */
2398 0 : if ((uval & FUTEX_TID_MASK) != vpid)
2399 : return -EPERM;
2400 :
2401 0 : ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE);
2402 0 : if (ret)
2403 : return ret;
2404 :
2405 0 : hb = hash_futex(&key);
2406 : spin_lock(&hb->lock);
2407 :
2408 : /*
2409 : * Check waiters first. We do not trust user space values at
2410 : * all and we at least want to know if user space fiddled
2411 : * with the futex value instead of blindly unlocking.
2412 : */
2413 0 : match = futex_top_waiter(hb, &key);
2414 0 : if (match) {
2415 0 : ret = wake_futex_pi(uaddr, uval, match);
2416 : /*
2417 : * The atomic access to the futex value generated a
2418 : * pagefault, so retry the user-access and the wakeup:
2419 : */
2420 0 : if (ret == -EFAULT)
2421 : goto pi_faulted;
2422 : goto out_unlock;
2423 : }
2424 :
2425 : /*
2426 : * We have no kernel internal state, i.e. no waiters in the
2427 : * kernel. Waiters which are about to queue themselves are stuck
2428 : * on hb->lock. So we can safely ignore them. We do neither
2429 : * preserve the WAITERS bit not the OWNER_DIED one. We are the
2430 : * owner.
2431 : */
2432 0 : if (cmpxchg_futex_value_locked(&curval, uaddr, uval, 0))
2433 : goto pi_faulted;
2434 :
2435 : /*
2436 : * If uval has changed, let user space handle it.
2437 : */
2438 0 : ret = (curval == uval) ? 0 : -EAGAIN;
2439 :
2440 : out_unlock:
2441 : spin_unlock(&hb->lock);
2442 : put_futex_key(&key);
2443 0 : return ret;
2444 :
2445 : pi_faulted:
2446 : spin_unlock(&hb->lock);
2447 : put_futex_key(&key);
2448 :
2449 0 : ret = fault_in_user_writeable(uaddr);
2450 0 : if (!ret)
2451 : goto retry;
2452 :
2453 : return ret;
2454 : }
2455 :
2456 : /**
2457 : * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2458 : * @hb: the hash_bucket futex_q was original enqueued on
2459 : * @q: the futex_q woken while waiting to be requeued
2460 : * @key2: the futex_key of the requeue target futex
2461 : * @timeout: the timeout associated with the wait (NULL if none)
2462 : *
2463 : * Detect if the task was woken on the initial futex as opposed to the requeue
2464 : * target futex. If so, determine if it was a timeout or a signal that caused
2465 : * the wakeup and return the appropriate error code to the caller. Must be
2466 : * called with the hb lock held.
2467 : *
2468 : * Return:
2469 : * 0 = no early wakeup detected;
2470 : * <0 = -ETIMEDOUT or -ERESTARTNOINTR
2471 : */
2472 : static inline
2473 : int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb,
2474 : struct futex_q *q, union futex_key *key2,
2475 : struct hrtimer_sleeper *timeout)
2476 : {
2477 : int ret = 0;
2478 :
2479 : /*
2480 : * With the hb lock held, we avoid races while we process the wakeup.
2481 : * We only need to hold hb (and not hb2) to ensure atomicity as the
2482 : * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2483 : * It can't be requeued from uaddr2 to something else since we don't
2484 : * support a PI aware source futex for requeue.
2485 : */
2486 0 : if (!match_futex(&q->key, key2)) {
2487 : WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr));
2488 : /*
2489 : * We were woken prior to requeue by a timeout or a signal.
2490 : * Unqueue the futex_q and determine which it was.
2491 : */
2492 0 : plist_del(&q->list, &hb->chain);
2493 : hb_waiters_dec(hb);
2494 :
2495 : /* Handle spurious wakeups gracefully */
2496 : ret = -EWOULDBLOCK;
2497 0 : if (timeout && !timeout->task)
2498 : ret = -ETIMEDOUT;
2499 0 : else if (signal_pending(current))
2500 : ret = -ERESTARTNOINTR;
2501 : }
2502 : return ret;
2503 : }
2504 :
2505 : /**
2506 : * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
2507 : * @uaddr: the futex we initially wait on (non-pi)
2508 : * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be
2509 : * the same type, no requeueing from private to shared, etc.
2510 : * @val: the expected value of uaddr
2511 : * @abs_time: absolute timeout
2512 : * @bitset: 32 bit wakeup bitset set by userspace, defaults to all
2513 : * @uaddr2: the pi futex we will take prior to returning to user-space
2514 : *
2515 : * The caller will wait on uaddr and will be requeued by futex_requeue() to
2516 : * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake
2517 : * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to
2518 : * userspace. This ensures the rt_mutex maintains an owner when it has waiters;
2519 : * without one, the pi logic would not know which task to boost/deboost, if
2520 : * there was a need to.
2521 : *
2522 : * We call schedule in futex_wait_queue_me() when we enqueue and return there
2523 : * via the following--
2524 : * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
2525 : * 2) wakeup on uaddr2 after a requeue
2526 : * 3) signal
2527 : * 4) timeout
2528 : *
2529 : * If 3, cleanup and return -ERESTARTNOINTR.
2530 : *
2531 : * If 2, we may then block on trying to take the rt_mutex and return via:
2532 : * 5) successful lock
2533 : * 6) signal
2534 : * 7) timeout
2535 : * 8) other lock acquisition failure
2536 : *
2537 : * If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
2538 : *
2539 : * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2540 : *
2541 : * Return:
2542 : * 0 - On success;
2543 : * <0 - On error
2544 : */
2545 0 : static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
2546 : u32 val, ktime_t *abs_time, u32 bitset,
2547 : u32 __user *uaddr2)
2548 : {
2549 : struct hrtimer_sleeper timeout, *to = NULL;
2550 : struct rt_mutex_waiter rt_waiter;
2551 0 : struct rt_mutex *pi_mutex = NULL;
2552 : struct futex_hash_bucket *hb;
2553 0 : union futex_key key2 = FUTEX_KEY_INIT;
2554 0 : struct futex_q q = futex_q_init;
2555 : int res, ret;
2556 :
2557 0 : if (uaddr == uaddr2)
2558 : return -EINVAL;
2559 :
2560 0 : if (!bitset)
2561 : return -EINVAL;
2562 :
2563 0 : if (abs_time) {
2564 : to = &timeout;
2565 0 : hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
2566 : CLOCK_REALTIME : CLOCK_MONOTONIC,
2567 : HRTIMER_MODE_ABS);
2568 0 : hrtimer_init_sleeper(to, current);
2569 0 : hrtimer_set_expires_range_ns(&to->timer, *abs_time,
2570 0 : current->timer_slack_ns);
2571 : }
2572 :
2573 : /*
2574 : * The waiter is allocated on our stack, manipulated by the requeue
2575 : * code while we sleep on uaddr.
2576 : */
2577 : debug_rt_mutex_init_waiter(&rt_waiter);
2578 0 : RB_CLEAR_NODE(&rt_waiter.pi_tree_entry);
2579 0 : RB_CLEAR_NODE(&rt_waiter.tree_entry);
2580 0 : rt_waiter.task = NULL;
2581 :
2582 0 : ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
2583 0 : if (unlikely(ret != 0))
2584 : goto out;
2585 :
2586 0 : q.bitset = bitset;
2587 0 : q.rt_waiter = &rt_waiter;
2588 0 : q.requeue_pi_key = &key2;
2589 :
2590 : /*
2591 : * Prepare to wait on uaddr. On success, increments q.key (key1) ref
2592 : * count.
2593 : */
2594 0 : ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
2595 0 : if (ret)
2596 : goto out_key2;
2597 :
2598 : /*
2599 : * The check above which compares uaddrs is not sufficient for
2600 : * shared futexes. We need to compare the keys:
2601 : */
2602 0 : if (match_futex(&q.key, &key2)) {
2603 : queue_unlock(hb);
2604 : ret = -EINVAL;
2605 : goto out_put_keys;
2606 : }
2607 :
2608 : /* Queue the futex_q, drop the hb lock, wait for wakeup. */
2609 0 : futex_wait_queue_me(hb, &q, to);
2610 :
2611 : spin_lock(&hb->lock);
2612 0 : ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
2613 : spin_unlock(&hb->lock);
2614 0 : if (ret)
2615 : goto out_put_keys;
2616 :
2617 : /*
2618 : * In order for us to be here, we know our q.key == key2, and since
2619 : * we took the hb->lock above, we also know that futex_requeue() has
2620 : * completed and we no longer have to concern ourselves with a wakeup
2621 : * race with the atomic proxy lock acquisition by the requeue code. The
2622 : * futex_requeue dropped our key1 reference and incremented our key2
2623 : * reference count.
2624 : */
2625 :
2626 : /* Check if the requeue code acquired the second futex for us. */
2627 0 : if (!q.rt_waiter) {
2628 : /*
2629 : * Got the lock. We might not be the anticipated owner if we
2630 : * did a lock-steal - fix up the PI-state in that case.
2631 : */
2632 0 : if (q.pi_state && (q.pi_state->owner != current)) {
2633 : spin_lock(q.lock_ptr);
2634 0 : ret = fixup_pi_state_owner(uaddr2, &q, current);
2635 : spin_unlock(q.lock_ptr);
2636 : }
2637 : } else {
2638 : /*
2639 : * We have been woken up by futex_unlock_pi(), a timeout, or a
2640 : * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2641 : * the pi_state.
2642 : */
2643 : WARN_ON(!q.pi_state);
2644 0 : pi_mutex = &q.pi_state->pi_mutex;
2645 0 : ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter);
2646 : debug_rt_mutex_free_waiter(&rt_waiter);
2647 :
2648 : spin_lock(q.lock_ptr);
2649 : /*
2650 : * Fixup the pi_state owner and possibly acquire the lock if we
2651 : * haven't already.
2652 : */
2653 0 : res = fixup_owner(uaddr2, &q, !ret);
2654 : /*
2655 : * If fixup_owner() returned an error, proprogate that. If it
2656 : * acquired the lock, clear -ETIMEDOUT or -EINTR.
2657 : */
2658 0 : if (res)
2659 0 : ret = (res < 0) ? res : 0;
2660 :
2661 : /* Unqueue and drop the lock. */
2662 0 : unqueue_me_pi(&q);
2663 : }
2664 :
2665 : /*
2666 : * If fixup_pi_state_owner() faulted and was unable to handle the
2667 : * fault, unlock the rt_mutex and return the fault to userspace.
2668 : */
2669 0 : if (ret == -EFAULT) {
2670 0 : if (pi_mutex && rt_mutex_owner(pi_mutex) == current)
2671 0 : rt_mutex_unlock(pi_mutex);
2672 0 : } else if (ret == -EINTR) {
2673 : /*
2674 : * We've already been requeued, but cannot restart by calling
2675 : * futex_lock_pi() directly. We could restart this syscall, but
2676 : * it would detect that the user space "val" changed and return
2677 : * -EWOULDBLOCK. Save the overhead of the restart and return
2678 : * -EWOULDBLOCK directly.
2679 : */
2680 : ret = -EWOULDBLOCK;
2681 : }
2682 :
2683 : out_put_keys:
2684 : put_futex_key(&q.key);
2685 : out_key2:
2686 : put_futex_key(&key2);
2687 :
2688 : out:
2689 0 : if (to) {
2690 0 : hrtimer_cancel(&to->timer);
2691 : destroy_hrtimer_on_stack(&to->timer);
2692 : }
2693 0 : return ret;
2694 : }
2695 :
2696 : /*
2697 : * Support for robust futexes: the kernel cleans up held futexes at
2698 : * thread exit time.
2699 : *
2700 : * Implementation: user-space maintains a per-thread list of locks it
2701 : * is holding. Upon do_exit(), the kernel carefully walks this list,
2702 : * and marks all locks that are owned by this thread with the
2703 : * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
2704 : * always manipulated with the lock held, so the list is private and
2705 : * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2706 : * field, to allow the kernel to clean up if the thread dies after
2707 : * acquiring the lock, but just before it could have added itself to
2708 : * the list. There can only be one such pending lock.
2709 : */
2710 :
2711 : /**
2712 : * sys_set_robust_list() - Set the robust-futex list head of a task
2713 : * @head: pointer to the list-head
2714 : * @len: length of the list-head, as userspace expects
2715 : */
2716 490 : SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head,
2717 : size_t, len)
2718 : {
2719 245 : if (!futex_cmpxchg_enabled)
2720 : return -ENOSYS;
2721 : /*
2722 : * The kernel knows only one size for now:
2723 : */
2724 245 : if (unlikely(len != sizeof(*head)))
2725 : return -EINVAL;
2726 :
2727 245 : current->robust_list = head;
2728 :
2729 : return 0;
2730 : }
2731 :
2732 : /**
2733 : * sys_get_robust_list() - Get the robust-futex list head of a task
2734 : * @pid: pid of the process [zero for current task]
2735 : * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2736 : * @len_ptr: pointer to a length field, the kernel fills in the header size
2737 : */
2738 0 : SYSCALL_DEFINE3(get_robust_list, int, pid,
2739 : struct robust_list_head __user * __user *, head_ptr,
2740 : size_t __user *, len_ptr)
2741 : {
2742 : struct robust_list_head __user *head;
2743 : unsigned long ret;
2744 : struct task_struct *p;
2745 :
2746 0 : if (!futex_cmpxchg_enabled)
2747 : return -ENOSYS;
2748 :
2749 : rcu_read_lock();
2750 :
2751 : ret = -ESRCH;
2752 0 : if (!pid)
2753 0 : p = current;
2754 : else {
2755 0 : p = find_task_by_vpid(pid);
2756 0 : if (!p)
2757 : goto err_unlock;
2758 : }
2759 :
2760 : ret = -EPERM;
2761 0 : if (!ptrace_may_access(p, PTRACE_MODE_READ))
2762 : goto err_unlock;
2763 :
2764 0 : head = p->robust_list;
2765 : rcu_read_unlock();
2766 :
2767 0 : if (put_user(sizeof(*head), len_ptr))
2768 : return -EFAULT;
2769 0 : return put_user(head, head_ptr);
2770 :
2771 : err_unlock:
2772 : rcu_read_unlock();
2773 :
2774 0 : return ret;
2775 : }
2776 :
2777 : /*
2778 : * Process a futex-list entry, check whether it's owned by the
2779 : * dying task, and do notification if so:
2780 : */
2781 0 : int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
2782 : {
2783 : u32 uval, uninitialized_var(nval), mval;
2784 :
2785 : retry:
2786 0 : if (get_user(uval, uaddr))
2787 : return -1;
2788 :
2789 0 : if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) {
2790 : /*
2791 : * Ok, this dying thread is truly holding a futex
2792 : * of interest. Set the OWNER_DIED bit atomically
2793 : * via cmpxchg, and if the value had FUTEX_WAITERS
2794 : * set, wake up a waiter (if any). (We have to do a
2795 : * futex_wake() even if OWNER_DIED is already set -
2796 : * to handle the rare but possible case of recursive
2797 : * thread-death.) The rest of the cleanup is done in
2798 : * userspace.
2799 : */
2800 0 : mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
2801 : /*
2802 : * We are not holding a lock here, but we want to have
2803 : * the pagefault_disable/enable() protection because
2804 : * we want to handle the fault gracefully. If the
2805 : * access fails we try to fault in the futex with R/W
2806 : * verification via get_user_pages. get_user() above
2807 : * does not guarantee R/W access. If that fails we
2808 : * give up and leave the futex locked.
2809 : */
2810 0 : if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) {
2811 0 : if (fault_in_user_writeable(uaddr))
2812 : return -1;
2813 : goto retry;
2814 : }
2815 0 : if (nval != uval)
2816 : goto retry;
2817 :
2818 : /*
2819 : * Wake robust non-PI futexes here. The wakeup of
2820 : * PI futexes happens in exit_pi_state():
2821 : */
2822 0 : if (!pi && (uval & FUTEX_WAITERS))
2823 0 : futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
2824 : }
2825 : return 0;
2826 : }
2827 :
2828 : /*
2829 : * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2830 : */
2831 : static inline int fetch_robust_entry(struct robust_list __user **entry,
2832 : struct robust_list __user * __user *head,
2833 : unsigned int *pi)
2834 : {
2835 : unsigned long uentry;
2836 :
2837 472 : if (get_user(uentry, (unsigned long __user *)head))
2838 : return -EFAULT;
2839 :
2840 472 : *entry = (void __user *)(uentry & ~1UL);
2841 472 : *pi = uentry & 1;
2842 :
2843 : return 0;
2844 : }
2845 :
2846 : /*
2847 : * Walk curr->robust_list (very carefully, it's a userspace list!)
2848 : * and mark any locks found there dead, and notify any waiters.
2849 : *
2850 : * We silently return on any sign of list-walking problem.
2851 : */
2852 236 : void exit_robust_list(struct task_struct *curr)
2853 : {
2854 236 : struct robust_list_head __user *head = curr->robust_list;
2855 : struct robust_list __user *entry, *next_entry, *pending;
2856 : unsigned int limit = ROBUST_LIST_LIMIT, pi, pip;
2857 : unsigned int uninitialized_var(next_pi);
2858 : unsigned long futex_offset;
2859 : int rc;
2860 :
2861 236 : if (!futex_cmpxchg_enabled)
2862 : return;
2863 :
2864 : /*
2865 : * Fetch the list head (which was registered earlier, via
2866 : * sys_set_robust_list()):
2867 : */
2868 472 : if (fetch_robust_entry(&entry, &head->list.next, &pi))
2869 : return;
2870 : /*
2871 : * Fetch the relative futex offset:
2872 : */
2873 236 : if (get_user(futex_offset, &head->futex_offset))
2874 : return;
2875 : /*
2876 : * Fetch any possibly pending lock-add first, and handle it
2877 : * if it exists:
2878 : */
2879 472 : if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
2880 : return;
2881 :
2882 : next_entry = NULL; /* avoid warning with gcc */
2883 236 : while (entry != &head->list) {
2884 : /*
2885 : * Fetch the next entry in the list before calling
2886 : * handle_futex_death:
2887 : */
2888 0 : rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
2889 : /*
2890 : * A pending lock might already be on the list, so
2891 : * don't process it twice:
2892 : */
2893 0 : if (entry != pending)
2894 0 : if (handle_futex_death((void __user *)entry + futex_offset,
2895 : curr, pi))
2896 : return;
2897 0 : if (rc)
2898 : return;
2899 : entry = next_entry;
2900 : pi = next_pi;
2901 : /*
2902 : * Avoid excessively long or circular lists:
2903 : */
2904 0 : if (!--limit)
2905 : break;
2906 :
2907 0 : cond_resched();
2908 : }
2909 :
2910 236 : if (pending)
2911 0 : handle_futex_death((void __user *)pending + futex_offset,
2912 : curr, pip);
2913 : }
2914 :
2915 864482 : long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
2916 : u32 __user *uaddr2, u32 val2, u32 val3)
2917 : {
2918 864482 : int cmd = op & FUTEX_CMD_MASK;
2919 : unsigned int flags = 0;
2920 :
2921 864482 : if (!(op & FUTEX_PRIVATE_FLAG))
2922 : flags |= FLAGS_SHARED;
2923 :
2924 864482 : if (op & FUTEX_CLOCK_REALTIME) {
2925 239 : flags |= FLAGS_CLOCKRT;
2926 239 : if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
2927 : return -ENOSYS;
2928 : }
2929 :
2930 : switch (cmd) {
2931 : case FUTEX_LOCK_PI:
2932 : case FUTEX_UNLOCK_PI:
2933 : case FUTEX_TRYLOCK_PI:
2934 : case FUTEX_WAIT_REQUEUE_PI:
2935 : case FUTEX_CMP_REQUEUE_PI:
2936 0 : if (!futex_cmpxchg_enabled)
2937 : return -ENOSYS;
2938 : }
2939 :
2940 864482 : switch (cmd) {
2941 : case FUTEX_WAIT:
2942 345299 : val3 = FUTEX_BITSET_MATCH_ANY;
2943 : case FUTEX_WAIT_BITSET:
2944 345538 : return futex_wait(uaddr, flags, val, timeout, val3);
2945 : case FUTEX_WAKE:
2946 345807 : val3 = FUTEX_BITSET_MATCH_ANY;
2947 : case FUTEX_WAKE_BITSET:
2948 345807 : return futex_wake(uaddr, flags, val, val3);
2949 : case FUTEX_REQUEUE:
2950 0 : return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0);
2951 : case FUTEX_CMP_REQUEUE:
2952 0 : return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0);
2953 : case FUTEX_WAKE_OP:
2954 173137 : return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3);
2955 : case FUTEX_LOCK_PI:
2956 0 : return futex_lock_pi(uaddr, flags, val, timeout, 0);
2957 : case FUTEX_UNLOCK_PI:
2958 0 : return futex_unlock_pi(uaddr, flags);
2959 : case FUTEX_TRYLOCK_PI:
2960 0 : return futex_lock_pi(uaddr, flags, 0, timeout, 1);
2961 : case FUTEX_WAIT_REQUEUE_PI:
2962 0 : val3 = FUTEX_BITSET_MATCH_ANY;
2963 0 : return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3,
2964 : uaddr2);
2965 : case FUTEX_CMP_REQUEUE_PI:
2966 0 : return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1);
2967 : }
2968 : return -ENOSYS;
2969 : }
2970 :
2971 :
2972 1728961 : SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
2973 : struct timespec __user *, utime, u32 __user *, uaddr2,
2974 : u32, val3)
2975 : {
2976 : struct timespec ts;
2977 : ktime_t t, *tp = NULL;
2978 : u32 val2 = 0;
2979 864482 : int cmd = op & FUTEX_CMD_MASK;
2980 :
2981 864482 : if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
2982 346274 : cmd == FUTEX_WAIT_BITSET ||
2983 173137 : cmd == FUTEX_WAIT_REQUEUE_PI)) {
2984 0 : if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
2985 : return -EFAULT;
2986 0 : if (!timespec_valid(&ts))
2987 : return -EINVAL;
2988 :
2989 0 : t = timespec_to_ktime(ts);
2990 0 : if (cmd == FUTEX_WAIT)
2991 0 : t = ktime_add_safe(ktime_get(), t);
2992 : tp = &t;
2993 : }
2994 : /*
2995 : * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
2996 : * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
2997 : */
2998 1728964 : if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
2999 1728964 : cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP)
3000 173137 : val2 = (u32) (unsigned long) utime;
3001 :
3002 864482 : return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
3003 : }
3004 :
3005 1 : static void __init futex_detect_cmpxchg(void)
3006 : {
3007 : #ifndef CONFIG_HAVE_FUTEX_CMPXCHG
3008 : u32 curval;
3009 :
3010 : /*
3011 : * This will fail and we want it. Some arch implementations do
3012 : * runtime detection of the futex_atomic_cmpxchg_inatomic()
3013 : * functionality. We want to know that before we call in any
3014 : * of the complex code paths. Also we want to prevent
3015 : * registration of robust lists in that case. NULL is
3016 : * guaranteed to fault and we get -EFAULT on functional
3017 : * implementation, the non-functional ones will return
3018 : * -ENOSYS.
3019 : */
3020 1 : if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT)
3021 1 : futex_cmpxchg_enabled = 1;
3022 : #endif
3023 1 : }
3024 :
3025 1 : static int __init futex_init(void)
3026 : {
3027 : unsigned int futex_shift;
3028 : unsigned long i;
3029 :
3030 : #if CONFIG_BASE_SMALL
3031 : futex_hashsize = 16;
3032 : #else
3033 1 : futex_hashsize = roundup_pow_of_two(256 * num_possible_cpus());
3034 : #endif
3035 :
3036 1 : futex_queues = alloc_large_system_hash("futex", sizeof(*futex_queues),
3037 : futex_hashsize, 0,
3038 : futex_hashsize < 256 ? HASH_SMALL : 0,
3039 : &futex_shift, NULL,
3040 : futex_hashsize, futex_hashsize);
3041 1 : futex_hashsize = 1UL << futex_shift;
3042 :
3043 1 : futex_detect_cmpxchg();
3044 :
3045 257 : for (i = 0; i < futex_hashsize; i++) {
3046 256 : atomic_set(&futex_queues[i].waiters, 0);
3047 : plist_head_init(&futex_queues[i].chain);
3048 : spin_lock_init(&futex_queues[i].lock);
3049 : }
3050 :
3051 1 : return 0;
3052 : }
3053 : __initcall(futex_init);
|
