blob: 06e8240d2abe63baeab90d17e028e273fc525767 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Ingo Molnar0771dfe2006-03-27 01:16:22 -080011 * 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 *
Ingo Molnarc87e2832006-06-27 02:54:58 -070015 * 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 *
Eric Dumazet34f01cc2007-05-09 02:35:04 -070019 * PRIVATE futexes by Eric Dumazet
20 * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com>
21 *
Darren Hart52400ba2009-04-03 13:40:49 -070022 * 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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070026 * 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>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070057#include <linux/signal.h>
Rusty Russell9adef582007-05-08 00:26:42 -070058#include <linux/module.h>
Andrey Mirkinfd5eea42007-10-16 23:30:13 -070059#include <linux/magic.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070060#include <linux/pid.h>
61#include <linux/nsproxy.h>
62
Jakub Jelinek4732efb2005-09-06 15:16:25 -070063#include <asm/futex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Ingo Molnarc87e2832006-06-27 02:54:58 -070065#include "rtmutex_common.h"
66
Thomas Gleixnera0c1e902008-02-23 15:23:57 -080067int __read_mostly futex_cmpxchg_enabled;
68
Linus Torvalds1da177e2005-04-16 15:20:36 -070069#define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
70
71/*
Ingo Molnarc87e2832006-06-27 02:54:58 -070072 * Priority Inheritance state:
73 */
74struct futex_pi_state {
75 /*
76 * list of 'owned' pi_state instances - these have to be
77 * cleaned up in do_exit() if the task exits prematurely:
78 */
79 struct list_head list;
80
81 /*
82 * The PI object:
83 */
84 struct rt_mutex pi_mutex;
85
86 struct task_struct *owner;
87 atomic_t refcount;
88
89 union futex_key key;
90};
91
Darren Hartd8d88fb2009-09-21 22:30:30 -070092/**
93 * struct futex_q - The hashed futex queue entry, one per waiting task
94 * @task: the task waiting on the futex
95 * @lock_ptr: the hash bucket lock
96 * @key: the key the futex is hashed on
97 * @pi_state: optional priority inheritance state
98 * @rt_waiter: rt_waiter storage for use with requeue_pi
99 * @requeue_pi_key: the requeue_pi target futex key
100 * @bitset: bitset for the optional bitmasked wakeup
101 *
102 * We use this hashed waitqueue, instead of a normal wait_queue_t, so
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 * we can wake only the relevant ones (hashed queues may be shared).
104 *
105 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
Pierre Peifferec92d082007-05-09 02:35:00 -0700106 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * The order of wakup is always to make the first condition true, then
Darren Hartd8d88fb2009-09-21 22:30:30 -0700108 * the second.
109 *
110 * PI futexes are typically woken before they are removed from the hash list via
111 * the rt_mutex code. See unqueue_me_pi().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 */
113struct futex_q {
Pierre Peifferec92d082007-05-09 02:35:00 -0700114 struct plist_node list;
Darren Hartd8d88fb2009-09-21 22:30:30 -0700115
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200116 struct task_struct *task;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 spinlock_t *lock_ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 union futex_key key;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700119 struct futex_pi_state *pi_state;
Darren Hart52400ba2009-04-03 13:40:49 -0700120 struct rt_mutex_waiter *rt_waiter;
Darren Hart84bc4af2009-08-13 17:36:53 -0700121 union futex_key *requeue_pi_key;
Thomas Gleixnercd689982008-02-01 17:45:14 +0100122 u32 bitset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123};
124
125/*
Darren Hartb2d09942009-03-12 00:55:37 -0700126 * Hash buckets are shared by all the futex_keys that hash to the same
127 * location. Each key may have multiple futex_q structures, one for each task
128 * waiting on a futex.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 */
130struct futex_hash_bucket {
Pierre Peifferec92d082007-05-09 02:35:00 -0700131 spinlock_t lock;
132 struct plist_head chain;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133};
134
135static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS];
136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137/*
138 * We hash on the keys returned from get_futex_key (see below).
139 */
140static struct futex_hash_bucket *hash_futex(union futex_key *key)
141{
142 u32 hash = jhash2((u32*)&key->both.word,
143 (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
144 key->both.offset);
145 return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)];
146}
147
148/*
149 * Return 1 if two futex_keys are equal, 0 otherwise.
150 */
151static inline int match_futex(union futex_key *key1, union futex_key *key2)
152{
Darren Hart2bc87202009-10-14 10:12:39 -0700153 return (key1 && key2
154 && key1->both.word == key2->both.word
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 && key1->both.ptr == key2->both.ptr
156 && key1->both.offset == key2->both.offset);
157}
158
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200159/*
160 * Take a reference to the resource addressed by a key.
161 * Can be called while holding spinlocks.
162 *
163 */
164static void get_futex_key_refs(union futex_key *key)
165{
166 if (!key->both.ptr)
167 return;
168
169 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
170 case FUT_OFF_INODE:
171 atomic_inc(&key->shared.inode->i_count);
172 break;
173 case FUT_OFF_MMSHARED:
174 atomic_inc(&key->private.mm->mm_count);
175 break;
176 }
177}
178
179/*
180 * Drop a reference to the resource addressed by a key.
181 * The hash bucket spinlock must not be held.
182 */
183static void drop_futex_key_refs(union futex_key *key)
184{
Darren Hart90621c42008-12-29 19:43:21 -0800185 if (!key->both.ptr) {
186 /* If we're here then we tried to put a key we failed to get */
187 WARN_ON_ONCE(1);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200188 return;
Darren Hart90621c42008-12-29 19:43:21 -0800189 }
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200190
191 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
192 case FUT_OFF_INODE:
193 iput(key->shared.inode);
194 break;
195 case FUT_OFF_MMSHARED:
196 mmdrop(key->private.mm);
197 break;
198 }
199}
200
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700201/**
Darren Hartd96ee562009-09-21 22:30:22 -0700202 * get_futex_key() - Get parameters which are the keys for a futex
203 * @uaddr: virtual address of the futex
204 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
205 * @key: address where result is stored.
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700206 *
207 * Returns a negative error code or 0
208 * The key words are stored in *key on success.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 *
Josef "Jeff" Sipekf3a43f32006-12-08 02:36:43 -0800210 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 * offset_within_page). For private mappings, it's (uaddr, current->mm).
212 * We can usually work out the index without swapping in the page.
213 *
Darren Hartb2d09942009-03-12 00:55:37 -0700214 * lock_page() might sleep, the caller should not hold a spinlock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 */
Thomas Gleixner64d13042009-05-18 21:20:10 +0200216static int
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +0900217get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218{
Ingo Molnare2970f22006-06-27 02:54:47 -0700219 unsigned long address = (unsigned long)uaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 struct page *page;
222 int err;
223
224 /*
225 * The futex address must be "naturally" aligned.
226 */
Ingo Molnare2970f22006-06-27 02:54:47 -0700227 key->both.offset = address % PAGE_SIZE;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700228 if (unlikely((address % sizeof(u32)) != 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 return -EINVAL;
Ingo Molnare2970f22006-06-27 02:54:47 -0700230 address -= key->both.offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
232 /*
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700233 * PROCESS_PRIVATE futexes are fast.
234 * As the mm cannot disappear under us and the 'key' only needs
235 * virtual address, we dont even have to find the underlying vma.
236 * Note : We do have to check 'uaddr' is a valid user address,
237 * but access_ok() should be faster than find_vma()
238 */
239 if (!fshared) {
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +0900240 if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))))
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700241 return -EFAULT;
242 key->private.mm = mm;
243 key->private.address = address;
Peter Zijlstra42569c32008-09-30 12:33:07 +0200244 get_futex_key_refs(key);
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700245 return 0;
246 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200248again:
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +0900249 err = get_user_pages_fast(address, 1, 1, &page);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200250 if (err < 0)
251 return err;
252
Sonny Raoce2ae532009-07-10 18:13:13 -0500253 page = compound_head(page);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200254 lock_page(page);
255 if (!page->mapping) {
256 unlock_page(page);
257 put_page(page);
258 goto again;
259 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261 /*
262 * Private mappings are handled in a simple way.
263 *
264 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
265 * it's a read-only handle, it's expected that futexes attach to
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200266 * the object not the particular process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 */
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200268 if (PageAnon(page)) {
269 key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 key->private.mm = mm;
Ingo Molnare2970f22006-06-27 02:54:47 -0700271 key->private.address = address;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200272 } else {
273 key->both.offset |= FUT_OFF_INODE; /* inode-based key */
274 key->shared.inode = page->mapping->host;
275 key->shared.pgoff = page->index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 }
277
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200278 get_futex_key_refs(key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200280 unlock_page(page);
281 put_page(page);
282 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283}
284
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200285static inline
Peter Zijlstrac2f9f202008-09-26 19:32:23 +0200286void put_futex_key(int fshared, union futex_key *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287{
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200288 drop_futex_key_refs(key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289}
290
Darren Hartd96ee562009-09-21 22:30:22 -0700291/**
292 * fault_in_user_writeable() - Fault in user address and verify RW access
Thomas Gleixnerd0725992009-06-11 23:15:43 +0200293 * @uaddr: pointer to faulting user space address
294 *
295 * Slow path to fixup the fault we just took in the atomic write
296 * access to @uaddr.
297 *
298 * We have no generic implementation of a non destructive write to the
299 * user address. We know that we faulted in the atomic pagefault
300 * disabled section so we can as well avoid the #PF overhead by
301 * calling get_user_pages() right away.
302 */
303static int fault_in_user_writeable(u32 __user *uaddr)
304{
Andi Kleen722d0172009-12-08 13:19:42 +0100305 struct mm_struct *mm = current->mm;
306 int ret;
307
308 down_read(&mm->mmap_sem);
309 ret = get_user_pages(current, mm, (unsigned long)uaddr,
310 1, 1, 0, NULL, NULL);
311 up_read(&mm->mmap_sem);
312
Thomas Gleixnerd0725992009-06-11 23:15:43 +0200313 return ret < 0 ? ret : 0;
314}
315
Darren Hart4b1c4862009-04-03 13:39:42 -0700316/**
317 * futex_top_waiter() - Return the highest priority waiter on a futex
Darren Hartd96ee562009-09-21 22:30:22 -0700318 * @hb: the hash bucket the futex_q's reside in
319 * @key: the futex key (to distinguish it from other futex futex_q's)
Darren Hart4b1c4862009-04-03 13:39:42 -0700320 *
321 * Must be called with the hb lock held.
322 */
323static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb,
324 union futex_key *key)
325{
326 struct futex_q *this;
327
328 plist_for_each_entry(this, &hb->chain, list) {
329 if (match_futex(&this->key, key))
330 return this;
331 }
332 return NULL;
333}
334
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700335static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval)
336{
337 u32 curval;
338
339 pagefault_disable();
340 curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval);
341 pagefault_enable();
342
343 return curval;
344}
345
346static int get_futex_value_locked(u32 *dest, u32 __user *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347{
348 int ret;
349
Peter Zijlstraa8663742006-12-06 20:32:20 -0800350 pagefault_disable();
Ingo Molnare2970f22006-06-27 02:54:47 -0700351 ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
Peter Zijlstraa8663742006-12-06 20:32:20 -0800352 pagefault_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
354 return ret ? -EFAULT : 0;
355}
356
Ingo Molnarc87e2832006-06-27 02:54:58 -0700357
358/*
359 * PI code:
360 */
361static int refill_pi_state_cache(void)
362{
363 struct futex_pi_state *pi_state;
364
365 if (likely(current->pi_state_cache))
366 return 0;
367
Burman Yan4668edc2006-12-06 20:38:51 -0800368 pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700369
370 if (!pi_state)
371 return -ENOMEM;
372
Ingo Molnarc87e2832006-06-27 02:54:58 -0700373 INIT_LIST_HEAD(&pi_state->list);
374 /* pi_mutex gets initialized later */
375 pi_state->owner = NULL;
376 atomic_set(&pi_state->refcount, 1);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200377 pi_state->key = FUTEX_KEY_INIT;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700378
379 current->pi_state_cache = pi_state;
380
381 return 0;
382}
383
384static struct futex_pi_state * alloc_pi_state(void)
385{
386 struct futex_pi_state *pi_state = current->pi_state_cache;
387
388 WARN_ON(!pi_state);
389 current->pi_state_cache = NULL;
390
391 return pi_state;
392}
393
394static void free_pi_state(struct futex_pi_state *pi_state)
395{
396 if (!atomic_dec_and_test(&pi_state->refcount))
397 return;
398
399 /*
400 * If pi_state->owner is NULL, the owner is most probably dying
401 * and has cleaned up the pi_state already
402 */
403 if (pi_state->owner) {
Thomas Gleixner1d615482009-11-17 14:54:03 +0100404 raw_spin_lock_irq(&pi_state->owner->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700405 list_del_init(&pi_state->list);
Thomas Gleixner1d615482009-11-17 14:54:03 +0100406 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700407
408 rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
409 }
410
411 if (current->pi_state_cache)
412 kfree(pi_state);
413 else {
414 /*
415 * pi_state->list is already empty.
416 * clear pi_state->owner.
417 * refcount is at 0 - put it back to 1.
418 */
419 pi_state->owner = NULL;
420 atomic_set(&pi_state->refcount, 1);
421 current->pi_state_cache = pi_state;
422 }
423}
424
425/*
426 * Look up the task based on what TID userspace gave us.
427 * We dont trust it.
428 */
429static struct task_struct * futex_find_get_task(pid_t pid)
430{
431 struct task_struct *p;
David Howellsc69e8d92008-11-14 10:39:19 +1100432 const struct cred *cred = current_cred(), *pcred;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700433
Oleg Nesterovd359b542006-09-29 02:00:55 -0700434 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700435 p = find_task_by_vpid(pid);
David Howellsc69e8d92008-11-14 10:39:19 +1100436 if (!p) {
Thomas Gleixnera06381f2007-06-23 11:48:40 +0200437 p = ERR_PTR(-ESRCH);
David Howellsc69e8d92008-11-14 10:39:19 +1100438 } else {
439 pcred = __task_cred(p);
440 if (cred->euid != pcred->euid &&
441 cred->euid != pcred->uid)
442 p = ERR_PTR(-ESRCH);
443 else
444 get_task_struct(p);
445 }
Thomas Gleixnera06381f2007-06-23 11:48:40 +0200446
Oleg Nesterovd359b542006-09-29 02:00:55 -0700447 rcu_read_unlock();
Ingo Molnarc87e2832006-06-27 02:54:58 -0700448
449 return p;
450}
451
452/*
453 * This task is holding PI mutexes at exit time => bad.
454 * Kernel cleans up PI-state, but userspace is likely hosed.
455 * (Robust-futex cleanup is separate and might save the day for userspace.)
456 */
457void exit_pi_state_list(struct task_struct *curr)
458{
Ingo Molnarc87e2832006-06-27 02:54:58 -0700459 struct list_head *next, *head = &curr->pi_state_list;
460 struct futex_pi_state *pi_state;
Ingo Molnar627371d2006-07-29 05:16:20 +0200461 struct futex_hash_bucket *hb;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200462 union futex_key key = FUTEX_KEY_INIT;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700463
Thomas Gleixnera0c1e902008-02-23 15:23:57 -0800464 if (!futex_cmpxchg_enabled)
465 return;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700466 /*
467 * We are a ZOMBIE and nobody can enqueue itself on
468 * pi_state_list anymore, but we have to be careful
Ingo Molnar627371d2006-07-29 05:16:20 +0200469 * versus waiters unqueueing themselves:
Ingo Molnarc87e2832006-06-27 02:54:58 -0700470 */
Thomas Gleixner1d615482009-11-17 14:54:03 +0100471 raw_spin_lock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700472 while (!list_empty(head)) {
473
474 next = head->next;
475 pi_state = list_entry(next, struct futex_pi_state, list);
476 key = pi_state->key;
Ingo Molnar627371d2006-07-29 05:16:20 +0200477 hb = hash_futex(&key);
Thomas Gleixner1d615482009-11-17 14:54:03 +0100478 raw_spin_unlock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700479
Ingo Molnarc87e2832006-06-27 02:54:58 -0700480 spin_lock(&hb->lock);
481
Thomas Gleixner1d615482009-11-17 14:54:03 +0100482 raw_spin_lock_irq(&curr->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200483 /*
484 * We dropped the pi-lock, so re-check whether this
485 * task still owns the PI-state:
486 */
Ingo Molnarc87e2832006-06-27 02:54:58 -0700487 if (head->next != next) {
488 spin_unlock(&hb->lock);
489 continue;
490 }
491
Ingo Molnarc87e2832006-06-27 02:54:58 -0700492 WARN_ON(pi_state->owner != curr);
Ingo Molnar627371d2006-07-29 05:16:20 +0200493 WARN_ON(list_empty(&pi_state->list));
494 list_del_init(&pi_state->list);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700495 pi_state->owner = NULL;
Thomas Gleixner1d615482009-11-17 14:54:03 +0100496 raw_spin_unlock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700497
498 rt_mutex_unlock(&pi_state->pi_mutex);
499
500 spin_unlock(&hb->lock);
501
Thomas Gleixner1d615482009-11-17 14:54:03 +0100502 raw_spin_lock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700503 }
Thomas Gleixner1d615482009-11-17 14:54:03 +0100504 raw_spin_unlock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700505}
506
507static int
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700508lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
509 union futex_key *key, struct futex_pi_state **ps)
Ingo Molnarc87e2832006-06-27 02:54:58 -0700510{
511 struct futex_pi_state *pi_state = NULL;
512 struct futex_q *this, *next;
Pierre Peifferec92d082007-05-09 02:35:00 -0700513 struct plist_head *head;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700514 struct task_struct *p;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700515 pid_t pid = uval & FUTEX_TID_MASK;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700516
517 head = &hb->chain;
518
Pierre Peifferec92d082007-05-09 02:35:00 -0700519 plist_for_each_entry_safe(this, next, head, list) {
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700520 if (match_futex(&this->key, key)) {
Ingo Molnarc87e2832006-06-27 02:54:58 -0700521 /*
522 * Another waiter already exists - bump up
523 * the refcount and return its pi_state:
524 */
525 pi_state = this->pi_state;
Thomas Gleixner06a9ec22006-07-10 04:44:30 -0700526 /*
527 * Userspace might have messed up non PI and PI futexes
528 */
529 if (unlikely(!pi_state))
530 return -EINVAL;
531
Ingo Molnar627371d2006-07-29 05:16:20 +0200532 WARN_ON(!atomic_read(&pi_state->refcount));
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700533 WARN_ON(pid && pi_state->owner &&
534 pi_state->owner->pid != pid);
Ingo Molnar627371d2006-07-29 05:16:20 +0200535
Ingo Molnarc87e2832006-06-27 02:54:58 -0700536 atomic_inc(&pi_state->refcount);
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700537 *ps = pi_state;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700538
539 return 0;
540 }
541 }
542
543 /*
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200544 * We are the first waiter - try to look up the real owner and attach
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700545 * the new pi_state to it, but bail out when TID = 0
Ingo Molnarc87e2832006-06-27 02:54:58 -0700546 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700547 if (!pid)
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200548 return -ESRCH;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700549 p = futex_find_get_task(pid);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700550 if (IS_ERR(p))
551 return PTR_ERR(p);
552
553 /*
554 * We need to look at the task state flags to figure out,
555 * whether the task is exiting. To protect against the do_exit
556 * change of the task flags, we do this protected by
557 * p->pi_lock:
558 */
Thomas Gleixner1d615482009-11-17 14:54:03 +0100559 raw_spin_lock_irq(&p->pi_lock);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700560 if (unlikely(p->flags & PF_EXITING)) {
561 /*
562 * The task is on the way out. When PF_EXITPIDONE is
563 * set, we know that the task has finished the
564 * cleanup:
565 */
566 int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
567
Thomas Gleixner1d615482009-11-17 14:54:03 +0100568 raw_spin_unlock_irq(&p->pi_lock);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700569 put_task_struct(p);
570 return ret;
571 }
Ingo Molnarc87e2832006-06-27 02:54:58 -0700572
573 pi_state = alloc_pi_state();
574
575 /*
576 * Initialize the pi_mutex in locked state and make 'p'
577 * the owner of it:
578 */
579 rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
580
581 /* Store the key for possible exit cleanups: */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700582 pi_state->key = *key;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700583
Ingo Molnar627371d2006-07-29 05:16:20 +0200584 WARN_ON(!list_empty(&pi_state->list));
Ingo Molnarc87e2832006-06-27 02:54:58 -0700585 list_add(&pi_state->list, &p->pi_state_list);
586 pi_state->owner = p;
Thomas Gleixner1d615482009-11-17 14:54:03 +0100587 raw_spin_unlock_irq(&p->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700588
589 put_task_struct(p);
590
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700591 *ps = pi_state;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700592
593 return 0;
594}
595
Darren Hart1a520842009-04-03 13:39:52 -0700596/**
Darren Hartd96ee562009-09-21 22:30:22 -0700597 * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
Darren Hartbab5bc92009-04-07 23:23:50 -0700598 * @uaddr: the pi futex user address
599 * @hb: the pi futex hash bucket
600 * @key: the futex key associated with uaddr and hb
601 * @ps: the pi_state pointer where we store the result of the
602 * lookup
603 * @task: the task to perform the atomic lock work for. This will
604 * be "current" except in the case of requeue pi.
605 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
Darren Hart1a520842009-04-03 13:39:52 -0700606 *
607 * Returns:
608 * 0 - ready to wait
609 * 1 - acquired the lock
610 * <0 - error
611 *
612 * The hb->lock and futex_key refs shall be held by the caller.
613 */
614static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb,
615 union futex_key *key,
616 struct futex_pi_state **ps,
Darren Hartbab5bc92009-04-07 23:23:50 -0700617 struct task_struct *task, int set_waiters)
Darren Hart1a520842009-04-03 13:39:52 -0700618{
619 int lock_taken, ret, ownerdied = 0;
620 u32 uval, newval, curval;
621
622retry:
623 ret = lock_taken = 0;
624
625 /*
626 * To avoid races, we attempt to take the lock here again
627 * (by doing a 0 -> TID atomic cmpxchg), while holding all
628 * the locks. It will most likely not succeed.
629 */
630 newval = task_pid_vnr(task);
Darren Hartbab5bc92009-04-07 23:23:50 -0700631 if (set_waiters)
632 newval |= FUTEX_WAITERS;
Darren Hart1a520842009-04-03 13:39:52 -0700633
634 curval = cmpxchg_futex_value_locked(uaddr, 0, newval);
635
636 if (unlikely(curval == -EFAULT))
637 return -EFAULT;
638
639 /*
640 * Detect deadlocks.
641 */
642 if ((unlikely((curval & FUTEX_TID_MASK) == task_pid_vnr(task))))
643 return -EDEADLK;
644
645 /*
646 * Surprise - we got the lock. Just return to userspace:
647 */
648 if (unlikely(!curval))
649 return 1;
650
651 uval = curval;
652
653 /*
654 * Set the FUTEX_WAITERS flag, so the owner will know it has someone
655 * to wake at the next unlock.
656 */
657 newval = curval | FUTEX_WAITERS;
658
659 /*
660 * There are two cases, where a futex might have no owner (the
661 * owner TID is 0): OWNER_DIED. We take over the futex in this
662 * case. We also do an unconditional take over, when the owner
663 * of the futex died.
664 *
665 * This is safe as we are protected by the hash bucket lock !
666 */
667 if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) {
668 /* Keep the OWNER_DIED bit */
669 newval = (curval & ~FUTEX_TID_MASK) | task_pid_vnr(task);
670 ownerdied = 0;
671 lock_taken = 1;
672 }
673
674 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
675
676 if (unlikely(curval == -EFAULT))
677 return -EFAULT;
678 if (unlikely(curval != uval))
679 goto retry;
680
681 /*
682 * We took the lock due to owner died take over.
683 */
684 if (unlikely(lock_taken))
685 return 1;
686
687 /*
688 * We dont have the lock. Look up the PI state (or create it if
689 * we are the first waiter):
690 */
691 ret = lookup_pi_state(uval, hb, key, ps);
692
693 if (unlikely(ret)) {
694 switch (ret) {
695 case -ESRCH:
696 /*
697 * No owner found for this futex. Check if the
698 * OWNER_DIED bit is set to figure out whether
699 * this is a robust futex or not.
700 */
701 if (get_futex_value_locked(&curval, uaddr))
702 return -EFAULT;
703
704 /*
705 * We simply start over in case of a robust
706 * futex. The code above will take the futex
707 * and return happy.
708 */
709 if (curval & FUTEX_OWNER_DIED) {
710 ownerdied = 1;
711 goto retry;
712 }
713 default:
714 break;
715 }
716 }
717
718 return ret;
719}
720
Ingo Molnarc87e2832006-06-27 02:54:58 -0700721/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 * The hash bucket lock must be held when this is called.
723 * Afterwards, the futex_q must not be accessed.
724 */
725static void wake_futex(struct futex_q *q)
726{
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200727 struct task_struct *p = q->task;
728
729 /*
730 * We set q->lock_ptr = NULL _before_ we wake up the task. If
731 * a non futex wake up happens on another CPU then the task
732 * might exit and p would dereference a non existing task
733 * struct. Prevent this by holding a reference on p across the
734 * wake up.
735 */
736 get_task_struct(p);
737
Pierre Peifferec92d082007-05-09 02:35:00 -0700738 plist_del(&q->list, &q->list.plist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 /*
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200740 * The waiting task can free the futex_q as soon as
741 * q->lock_ptr = NULL is written, without taking any locks. A
742 * memory barrier is required here to prevent the following
743 * store to lock_ptr from getting ahead of the plist_del.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 */
Ralf Baechleccdea2f2006-12-06 20:40:26 -0800745 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 q->lock_ptr = NULL;
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200747
748 wake_up_state(p, TASK_NORMAL);
749 put_task_struct(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750}
751
Ingo Molnarc87e2832006-06-27 02:54:58 -0700752static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
753{
754 struct task_struct *new_owner;
755 struct futex_pi_state *pi_state = this->pi_state;
756 u32 curval, newval;
757
758 if (!pi_state)
759 return -EINVAL;
760
Thomas Gleixner51246bf2010-02-02 11:40:27 +0100761 /*
762 * If current does not own the pi_state then the futex is
763 * inconsistent and user space fiddled with the futex value.
764 */
765 if (pi_state->owner != current)
766 return -EINVAL;
767
Thomas Gleixnerd209d742009-11-17 18:22:11 +0100768 raw_spin_lock(&pi_state->pi_mutex.wait_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700769 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
770
771 /*
772 * This happens when we have stolen the lock and the original
773 * pending owner did not enqueue itself back on the rt_mutex.
774 * Thats not a tragedy. We know that way, that a lock waiter
775 * is on the fly. We make the futex_q waiter the pending owner.
776 */
777 if (!new_owner)
778 new_owner = this->task;
779
780 /*
781 * We pass it to the next owner. (The WAITERS bit is always
782 * kept enabled while there is PI state around. We must also
783 * preserve the owner died bit.)
784 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200785 if (!(uval & FUTEX_OWNER_DIED)) {
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700786 int ret = 0;
787
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700788 newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700789
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700790 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700791
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200792 if (curval == -EFAULT)
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700793 ret = -EFAULT;
Thomas Gleixnercde898f2007-12-05 15:46:09 +0100794 else if (curval != uval)
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700795 ret = -EINVAL;
796 if (ret) {
Thomas Gleixnerd209d742009-11-17 18:22:11 +0100797 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700798 return ret;
799 }
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200800 }
Ingo Molnarc87e2832006-06-27 02:54:58 -0700801
Thomas Gleixner1d615482009-11-17 14:54:03 +0100802 raw_spin_lock_irq(&pi_state->owner->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200803 WARN_ON(list_empty(&pi_state->list));
804 list_del_init(&pi_state->list);
Thomas Gleixner1d615482009-11-17 14:54:03 +0100805 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200806
Thomas Gleixner1d615482009-11-17 14:54:03 +0100807 raw_spin_lock_irq(&new_owner->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200808 WARN_ON(!list_empty(&pi_state->list));
Ingo Molnarc87e2832006-06-27 02:54:58 -0700809 list_add(&pi_state->list, &new_owner->pi_state_list);
810 pi_state->owner = new_owner;
Thomas Gleixner1d615482009-11-17 14:54:03 +0100811 raw_spin_unlock_irq(&new_owner->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200812
Thomas Gleixnerd209d742009-11-17 18:22:11 +0100813 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700814 rt_mutex_unlock(&pi_state->pi_mutex);
815
816 return 0;
817}
818
819static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
820{
821 u32 oldval;
822
823 /*
824 * There is no waiter, so we unlock the futex. The owner died
825 * bit has not to be preserved here. We are the owner:
826 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700827 oldval = cmpxchg_futex_value_locked(uaddr, uval, 0);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700828
829 if (oldval == -EFAULT)
830 return oldval;
831 if (oldval != uval)
832 return -EAGAIN;
833
834 return 0;
835}
836
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837/*
Ingo Molnar8b8f3192006-07-03 00:25:05 -0700838 * Express the locking dependencies for lockdep:
839 */
840static inline void
841double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
842{
843 if (hb1 <= hb2) {
844 spin_lock(&hb1->lock);
845 if (hb1 < hb2)
846 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
847 } else { /* hb1 > hb2 */
848 spin_lock(&hb2->lock);
849 spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
850 }
851}
852
Darren Hart5eb3dc62009-03-12 00:55:52 -0700853static inline void
854double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
855{
Darren Hartf061d352009-03-12 15:11:18 -0700856 spin_unlock(&hb1->lock);
Ingo Molnar88f502f2009-03-13 10:32:07 +0100857 if (hb1 != hb2)
858 spin_unlock(&hb2->lock);
Darren Hart5eb3dc62009-03-12 00:55:52 -0700859}
860
Ingo Molnar8b8f3192006-07-03 00:25:05 -0700861/*
Darren Hartb2d09942009-03-12 00:55:37 -0700862 * Wake up waiters matching bitset queued on this futex (uaddr).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 */
Peter Zijlstrac2f9f202008-09-26 19:32:23 +0200864static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865{
Ingo Molnare2970f22006-06-27 02:54:47 -0700866 struct futex_hash_bucket *hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 struct futex_q *this, *next;
Pierre Peifferec92d082007-05-09 02:35:00 -0700868 struct plist_head *head;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200869 union futex_key key = FUTEX_KEY_INIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 int ret;
871
Thomas Gleixnercd689982008-02-01 17:45:14 +0100872 if (!bitset)
873 return -EINVAL;
874
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +0900875 ret = get_futex_key(uaddr, fshared, &key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 if (unlikely(ret != 0))
877 goto out;
878
Ingo Molnare2970f22006-06-27 02:54:47 -0700879 hb = hash_futex(&key);
880 spin_lock(&hb->lock);
881 head = &hb->chain;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
Pierre Peifferec92d082007-05-09 02:35:00 -0700883 plist_for_each_entry_safe(this, next, head, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 if (match_futex (&this->key, &key)) {
Darren Hart52400ba2009-04-03 13:40:49 -0700885 if (this->pi_state || this->rt_waiter) {
Ingo Molnared6f7b12006-07-01 04:35:46 -0700886 ret = -EINVAL;
887 break;
888 }
Thomas Gleixnercd689982008-02-01 17:45:14 +0100889
890 /* Check if one of the bits is set in both bitsets */
891 if (!(this->bitset & bitset))
892 continue;
893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 wake_futex(this);
895 if (++ret >= nr_wake)
896 break;
897 }
898 }
899
Ingo Molnare2970f22006-06-27 02:54:47 -0700900 spin_unlock(&hb->lock);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200901 put_futex_key(fshared, &key);
Darren Hart42d35d42008-12-29 15:49:53 -0800902out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 return ret;
904}
905
906/*
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700907 * Wake up all waiters hashed on the physical page that is mapped
908 * to this virtual address:
909 */
Ingo Molnare2970f22006-06-27 02:54:47 -0700910static int
Peter Zijlstrac2f9f202008-09-26 19:32:23 +0200911futex_wake_op(u32 __user *uaddr1, int fshared, u32 __user *uaddr2,
Ingo Molnare2970f22006-06-27 02:54:47 -0700912 int nr_wake, int nr_wake2, int op)
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700913{
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200914 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
Ingo Molnare2970f22006-06-27 02:54:47 -0700915 struct futex_hash_bucket *hb1, *hb2;
Pierre Peifferec92d082007-05-09 02:35:00 -0700916 struct plist_head *head;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700917 struct futex_q *this, *next;
Darren Harte4dc5b72009-03-12 00:56:13 -0700918 int ret, op_ret;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700919
Darren Harte4dc5b72009-03-12 00:56:13 -0700920retry:
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +0900921 ret = get_futex_key(uaddr1, fshared, &key1);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700922 if (unlikely(ret != 0))
923 goto out;
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +0900924 ret = get_futex_key(uaddr2, fshared, &key2);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700925 if (unlikely(ret != 0))
Darren Hart42d35d42008-12-29 15:49:53 -0800926 goto out_put_key1;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700927
Ingo Molnare2970f22006-06-27 02:54:47 -0700928 hb1 = hash_futex(&key1);
929 hb2 = hash_futex(&key2);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700930
Darren Harte4dc5b72009-03-12 00:56:13 -0700931retry_private:
Thomas Gleixnereaaea802009-10-04 09:34:17 +0200932 double_lock_hb(hb1, hb2);
Ingo Molnare2970f22006-06-27 02:54:47 -0700933 op_ret = futex_atomic_op_inuser(op, uaddr2);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700934 if (unlikely(op_ret < 0)) {
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700935
Darren Hart5eb3dc62009-03-12 00:55:52 -0700936 double_unlock_hb(hb1, hb2);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700937
David Howells7ee1dd32006-01-06 00:11:44 -0800938#ifndef CONFIG_MMU
Ingo Molnare2970f22006-06-27 02:54:47 -0700939 /*
940 * we don't get EFAULT from MMU faults if we don't have an MMU,
941 * but we might get them from range checking
942 */
David Howells7ee1dd32006-01-06 00:11:44 -0800943 ret = op_ret;
Darren Hart42d35d42008-12-29 15:49:53 -0800944 goto out_put_keys;
David Howells7ee1dd32006-01-06 00:11:44 -0800945#endif
946
David Gibson796f8d92005-11-07 00:59:33 -0800947 if (unlikely(op_ret != -EFAULT)) {
948 ret = op_ret;
Darren Hart42d35d42008-12-29 15:49:53 -0800949 goto out_put_keys;
David Gibson796f8d92005-11-07 00:59:33 -0800950 }
951
Thomas Gleixnerd0725992009-06-11 23:15:43 +0200952 ret = fault_in_user_writeable(uaddr2);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700953 if (ret)
Darren Hartde87fcc2009-03-12 00:55:46 -0700954 goto out_put_keys;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700955
Darren Harte4dc5b72009-03-12 00:56:13 -0700956 if (!fshared)
957 goto retry_private;
958
Darren Hartde87fcc2009-03-12 00:55:46 -0700959 put_futex_key(fshared, &key2);
960 put_futex_key(fshared, &key1);
Darren Harte4dc5b72009-03-12 00:56:13 -0700961 goto retry;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700962 }
963
Ingo Molnare2970f22006-06-27 02:54:47 -0700964 head = &hb1->chain;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700965
Pierre Peifferec92d082007-05-09 02:35:00 -0700966 plist_for_each_entry_safe(this, next, head, list) {
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700967 if (match_futex (&this->key, &key1)) {
968 wake_futex(this);
969 if (++ret >= nr_wake)
970 break;
971 }
972 }
973
974 if (op_ret > 0) {
Ingo Molnare2970f22006-06-27 02:54:47 -0700975 head = &hb2->chain;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700976
977 op_ret = 0;
Pierre Peifferec92d082007-05-09 02:35:00 -0700978 plist_for_each_entry_safe(this, next, head, list) {
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700979 if (match_futex (&this->key, &key2)) {
980 wake_futex(this);
981 if (++op_ret >= nr_wake2)
982 break;
983 }
984 }
985 ret += op_ret;
986 }
987
Darren Hart5eb3dc62009-03-12 00:55:52 -0700988 double_unlock_hb(hb1, hb2);
Darren Hart42d35d42008-12-29 15:49:53 -0800989out_put_keys:
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200990 put_futex_key(fshared, &key2);
Darren Hart42d35d42008-12-29 15:49:53 -0800991out_put_key1:
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200992 put_futex_key(fshared, &key1);
Darren Hart42d35d42008-12-29 15:49:53 -0800993out:
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700994 return ret;
995}
996
Darren Hart9121e472009-04-03 13:40:31 -0700997/**
998 * requeue_futex() - Requeue a futex_q from one hb to another
999 * @q: the futex_q to requeue
1000 * @hb1: the source hash_bucket
1001 * @hb2: the target hash_bucket
1002 * @key2: the new key for the requeued futex_q
1003 */
1004static inline
1005void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1,
1006 struct futex_hash_bucket *hb2, union futex_key *key2)
1007{
1008
1009 /*
1010 * If key1 and key2 hash to the same bucket, no need to
1011 * requeue.
1012 */
1013 if (likely(&hb1->chain != &hb2->chain)) {
1014 plist_del(&q->list, &hb1->chain);
1015 plist_add(&q->list, &hb2->chain);
1016 q->lock_ptr = &hb2->lock;
1017#ifdef CONFIG_DEBUG_PI_LIST
Thomas Gleixnera2672452009-11-17 14:46:14 +01001018 q->list.plist.spinlock = &hb2->lock;
Darren Hart9121e472009-04-03 13:40:31 -07001019#endif
1020 }
1021 get_futex_key_refs(key2);
1022 q->key = *key2;
1023}
1024
Darren Hart52400ba2009-04-03 13:40:49 -07001025/**
1026 * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
Darren Hartd96ee562009-09-21 22:30:22 -07001027 * @q: the futex_q
1028 * @key: the key of the requeue target futex
1029 * @hb: the hash_bucket of the requeue target futex
Darren Hart52400ba2009-04-03 13:40:49 -07001030 *
1031 * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1032 * target futex if it is uncontended or via a lock steal. Set the futex_q key
1033 * to the requeue target futex so the waiter can detect the wakeup on the right
1034 * futex, but remove it from the hb and NULL the rt_waiter so it can detect
Darren Hartbeda2c72009-08-09 15:34:39 -07001035 * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1036 * to protect access to the pi_state to fixup the owner later. Must be called
1037 * with both q->lock_ptr and hb->lock held.
Darren Hart52400ba2009-04-03 13:40:49 -07001038 */
1039static inline
Darren Hartbeda2c72009-08-09 15:34:39 -07001040void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key,
1041 struct futex_hash_bucket *hb)
Darren Hart52400ba2009-04-03 13:40:49 -07001042{
Darren Hart52400ba2009-04-03 13:40:49 -07001043 get_futex_key_refs(key);
1044 q->key = *key;
1045
1046 WARN_ON(plist_node_empty(&q->list));
1047 plist_del(&q->list, &q->list.plist);
1048
1049 WARN_ON(!q->rt_waiter);
1050 q->rt_waiter = NULL;
1051
Darren Hartbeda2c72009-08-09 15:34:39 -07001052 q->lock_ptr = &hb->lock;
1053#ifdef CONFIG_DEBUG_PI_LIST
Thomas Gleixnera2672452009-11-17 14:46:14 +01001054 q->list.plist.spinlock = &hb->lock;
Darren Hartbeda2c72009-08-09 15:34:39 -07001055#endif
1056
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02001057 wake_up_state(q->task, TASK_NORMAL);
Darren Hart52400ba2009-04-03 13:40:49 -07001058}
1059
1060/**
1061 * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
Darren Hartbab5bc92009-04-07 23:23:50 -07001062 * @pifutex: the user address of the to futex
1063 * @hb1: the from futex hash bucket, must be locked by the caller
1064 * @hb2: the to futex hash bucket, must be locked by the caller
1065 * @key1: the from futex key
1066 * @key2: the to futex key
1067 * @ps: address to store the pi_state pointer
1068 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
Darren Hart52400ba2009-04-03 13:40:49 -07001069 *
1070 * Try and get the lock on behalf of the top waiter if we can do it atomically.
Darren Hartbab5bc92009-04-07 23:23:50 -07001071 * Wake the top waiter if we succeed. If the caller specified set_waiters,
1072 * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1073 * hb1 and hb2 must be held by the caller.
Darren Hart52400ba2009-04-03 13:40:49 -07001074 *
1075 * Returns:
1076 * 0 - failed to acquire the lock atomicly
1077 * 1 - acquired the lock
1078 * <0 - error
1079 */
1080static int futex_proxy_trylock_atomic(u32 __user *pifutex,
1081 struct futex_hash_bucket *hb1,
1082 struct futex_hash_bucket *hb2,
1083 union futex_key *key1, union futex_key *key2,
Darren Hartbab5bc92009-04-07 23:23:50 -07001084 struct futex_pi_state **ps, int set_waiters)
Darren Hart52400ba2009-04-03 13:40:49 -07001085{
Darren Hartbab5bc92009-04-07 23:23:50 -07001086 struct futex_q *top_waiter = NULL;
Darren Hart52400ba2009-04-03 13:40:49 -07001087 u32 curval;
1088 int ret;
1089
1090 if (get_futex_value_locked(&curval, pifutex))
1091 return -EFAULT;
1092
Darren Hartbab5bc92009-04-07 23:23:50 -07001093 /*
1094 * Find the top_waiter and determine if there are additional waiters.
1095 * If the caller intends to requeue more than 1 waiter to pifutex,
1096 * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1097 * as we have means to handle the possible fault. If not, don't set
1098 * the bit unecessarily as it will force the subsequent unlock to enter
1099 * the kernel.
1100 */
Darren Hart52400ba2009-04-03 13:40:49 -07001101 top_waiter = futex_top_waiter(hb1, key1);
1102
1103 /* There are no waiters, nothing for us to do. */
1104 if (!top_waiter)
1105 return 0;
1106
Darren Hart84bc4af2009-08-13 17:36:53 -07001107 /* Ensure we requeue to the expected futex. */
1108 if (!match_futex(top_waiter->requeue_pi_key, key2))
1109 return -EINVAL;
1110
Darren Hart52400ba2009-04-03 13:40:49 -07001111 /*
Darren Hartbab5bc92009-04-07 23:23:50 -07001112 * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1113 * the contended case or if set_waiters is 1. The pi_state is returned
1114 * in ps in contended cases.
Darren Hart52400ba2009-04-03 13:40:49 -07001115 */
Darren Hartbab5bc92009-04-07 23:23:50 -07001116 ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task,
1117 set_waiters);
Darren Hart52400ba2009-04-03 13:40:49 -07001118 if (ret == 1)
Darren Hartbeda2c72009-08-09 15:34:39 -07001119 requeue_pi_wake_futex(top_waiter, key2, hb2);
Darren Hart52400ba2009-04-03 13:40:49 -07001120
1121 return ret;
1122}
1123
1124/**
1125 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
1126 * uaddr1: source futex user address
1127 * uaddr2: target futex user address
1128 * nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1129 * nr_requeue: number of waiters to requeue (0-INT_MAX)
1130 * requeue_pi: if we are attempting to requeue from a non-pi futex to a
1131 * pi futex (pi to pi requeue is not supported)
1132 *
1133 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1134 * uaddr2 atomically on behalf of the top waiter.
1135 *
1136 * Returns:
1137 * >=0 - on success, the number of tasks requeued or woken
1138 * <0 - on error
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 */
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001140static int futex_requeue(u32 __user *uaddr1, int fshared, u32 __user *uaddr2,
Darren Hart52400ba2009-04-03 13:40:49 -07001141 int nr_wake, int nr_requeue, u32 *cmpval,
1142 int requeue_pi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143{
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001144 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
Darren Hart52400ba2009-04-03 13:40:49 -07001145 int drop_count = 0, task_count = 0, ret;
1146 struct futex_pi_state *pi_state = NULL;
Ingo Molnare2970f22006-06-27 02:54:47 -07001147 struct futex_hash_bucket *hb1, *hb2;
Pierre Peifferec92d082007-05-09 02:35:00 -07001148 struct plist_head *head1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 struct futex_q *this, *next;
Darren Hart52400ba2009-04-03 13:40:49 -07001150 u32 curval2;
1151
1152 if (requeue_pi) {
1153 /*
1154 * requeue_pi requires a pi_state, try to allocate it now
1155 * without any locks in case it fails.
1156 */
1157 if (refill_pi_state_cache())
1158 return -ENOMEM;
1159 /*
1160 * requeue_pi must wake as many tasks as it can, up to nr_wake
1161 * + nr_requeue, since it acquires the rt_mutex prior to
1162 * returning to userspace, so as to not leave the rt_mutex with
1163 * waiters and no owner. However, second and third wake-ups
1164 * cannot be predicted as they involve race conditions with the
1165 * first wake and a fault while looking up the pi_state. Both
1166 * pthread_cond_signal() and pthread_cond_broadcast() should
1167 * use nr_wake=1.
1168 */
1169 if (nr_wake != 1)
1170 return -EINVAL;
1171 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
Darren Hart42d35d42008-12-29 15:49:53 -08001173retry:
Darren Hart52400ba2009-04-03 13:40:49 -07001174 if (pi_state != NULL) {
1175 /*
1176 * We will have to lookup the pi_state again, so free this one
1177 * to keep the accounting correct.
1178 */
1179 free_pi_state(pi_state);
1180 pi_state = NULL;
1181 }
1182
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +09001183 ret = get_futex_key(uaddr1, fshared, &key1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 if (unlikely(ret != 0))
1185 goto out;
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +09001186 ret = get_futex_key(uaddr2, fshared, &key2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 if (unlikely(ret != 0))
Darren Hart42d35d42008-12-29 15:49:53 -08001188 goto out_put_key1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
Ingo Molnare2970f22006-06-27 02:54:47 -07001190 hb1 = hash_futex(&key1);
1191 hb2 = hash_futex(&key2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
Darren Harte4dc5b72009-03-12 00:56:13 -07001193retry_private:
Ingo Molnar8b8f3192006-07-03 00:25:05 -07001194 double_lock_hb(hb1, hb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
Ingo Molnare2970f22006-06-27 02:54:47 -07001196 if (likely(cmpval != NULL)) {
1197 u32 curval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
Ingo Molnare2970f22006-06-27 02:54:47 -07001199 ret = get_futex_value_locked(&curval, uaddr1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200
1201 if (unlikely(ret)) {
Darren Hart5eb3dc62009-03-12 00:55:52 -07001202 double_unlock_hb(hb1, hb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
Darren Harte4dc5b72009-03-12 00:56:13 -07001204 ret = get_user(curval, uaddr1);
1205 if (ret)
1206 goto out_put_keys;
1207
1208 if (!fshared)
1209 goto retry_private;
1210
Darren Hartde87fcc2009-03-12 00:55:46 -07001211 put_futex_key(fshared, &key2);
1212 put_futex_key(fshared, &key1);
Darren Harte4dc5b72009-03-12 00:56:13 -07001213 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 }
Ingo Molnare2970f22006-06-27 02:54:47 -07001215 if (curval != *cmpval) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 ret = -EAGAIN;
1217 goto out_unlock;
1218 }
1219 }
1220
Darren Hart52400ba2009-04-03 13:40:49 -07001221 if (requeue_pi && (task_count - nr_wake < nr_requeue)) {
Darren Hartbab5bc92009-04-07 23:23:50 -07001222 /*
1223 * Attempt to acquire uaddr2 and wake the top waiter. If we
1224 * intend to requeue waiters, force setting the FUTEX_WAITERS
1225 * bit. We force this here where we are able to easily handle
1226 * faults rather in the requeue loop below.
1227 */
Darren Hart52400ba2009-04-03 13:40:49 -07001228 ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1,
Darren Hartbab5bc92009-04-07 23:23:50 -07001229 &key2, &pi_state, nr_requeue);
Darren Hart52400ba2009-04-03 13:40:49 -07001230
1231 /*
1232 * At this point the top_waiter has either taken uaddr2 or is
1233 * waiting on it. If the former, then the pi_state will not
1234 * exist yet, look it up one more time to ensure we have a
1235 * reference to it.
1236 */
1237 if (ret == 1) {
1238 WARN_ON(pi_state);
Darren Hart89061d32009-10-15 15:30:48 -07001239 drop_count++;
Darren Hart52400ba2009-04-03 13:40:49 -07001240 task_count++;
1241 ret = get_futex_value_locked(&curval2, uaddr2);
1242 if (!ret)
1243 ret = lookup_pi_state(curval2, hb2, &key2,
1244 &pi_state);
1245 }
1246
1247 switch (ret) {
1248 case 0:
1249 break;
1250 case -EFAULT:
1251 double_unlock_hb(hb1, hb2);
1252 put_futex_key(fshared, &key2);
1253 put_futex_key(fshared, &key1);
Thomas Gleixnerd0725992009-06-11 23:15:43 +02001254 ret = fault_in_user_writeable(uaddr2);
Darren Hart52400ba2009-04-03 13:40:49 -07001255 if (!ret)
1256 goto retry;
1257 goto out;
1258 case -EAGAIN:
1259 /* The owner was exiting, try again. */
1260 double_unlock_hb(hb1, hb2);
1261 put_futex_key(fshared, &key2);
1262 put_futex_key(fshared, &key1);
1263 cond_resched();
1264 goto retry;
1265 default:
1266 goto out_unlock;
1267 }
1268 }
1269
Ingo Molnare2970f22006-06-27 02:54:47 -07001270 head1 = &hb1->chain;
Pierre Peifferec92d082007-05-09 02:35:00 -07001271 plist_for_each_entry_safe(this, next, head1, list) {
Darren Hart52400ba2009-04-03 13:40:49 -07001272 if (task_count - nr_wake >= nr_requeue)
1273 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Darren Hart52400ba2009-04-03 13:40:49 -07001275 if (!match_futex(&this->key, &key1))
1276 continue;
1277
Darren Hart392741e2009-08-07 15:20:48 -07001278 /*
1279 * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
1280 * be paired with each other and no other futex ops.
1281 */
1282 if ((requeue_pi && !this->rt_waiter) ||
1283 (!requeue_pi && this->rt_waiter)) {
1284 ret = -EINVAL;
1285 break;
1286 }
Darren Hart52400ba2009-04-03 13:40:49 -07001287
1288 /*
1289 * Wake nr_wake waiters. For requeue_pi, if we acquired the
1290 * lock, we already woke the top_waiter. If not, it will be
1291 * woken by futex_unlock_pi().
1292 */
1293 if (++task_count <= nr_wake && !requeue_pi) {
1294 wake_futex(this);
1295 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 }
Darren Hart52400ba2009-04-03 13:40:49 -07001297
Darren Hart84bc4af2009-08-13 17:36:53 -07001298 /* Ensure we requeue to the expected futex for requeue_pi. */
1299 if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) {
1300 ret = -EINVAL;
1301 break;
1302 }
1303
Darren Hart52400ba2009-04-03 13:40:49 -07001304 /*
1305 * Requeue nr_requeue waiters and possibly one more in the case
1306 * of requeue_pi if we couldn't acquire the lock atomically.
1307 */
1308 if (requeue_pi) {
1309 /* Prepare the waiter to take the rt_mutex. */
1310 atomic_inc(&pi_state->refcount);
1311 this->pi_state = pi_state;
1312 ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex,
1313 this->rt_waiter,
1314 this->task, 1);
1315 if (ret == 1) {
1316 /* We got the lock. */
Darren Hartbeda2c72009-08-09 15:34:39 -07001317 requeue_pi_wake_futex(this, &key2, hb2);
Darren Hart89061d32009-10-15 15:30:48 -07001318 drop_count++;
Darren Hart52400ba2009-04-03 13:40:49 -07001319 continue;
1320 } else if (ret) {
1321 /* -EDEADLK */
1322 this->pi_state = NULL;
1323 free_pi_state(pi_state);
1324 goto out_unlock;
1325 }
1326 }
1327 requeue_futex(this, hb1, hb2, &key2);
1328 drop_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 }
1330
1331out_unlock:
Darren Hart5eb3dc62009-03-12 00:55:52 -07001332 double_unlock_hb(hb1, hb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Darren Hartcd84a422009-04-02 14:19:38 -07001334 /*
1335 * drop_futex_key_refs() must be called outside the spinlocks. During
1336 * the requeue we moved futex_q's from the hash bucket at key1 to the
1337 * one at key2 and updated their key pointer. We no longer need to
1338 * hold the references to key1.
1339 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 while (--drop_count >= 0)
Rusty Russell9adef582007-05-08 00:26:42 -07001341 drop_futex_key_refs(&key1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
Darren Hart42d35d42008-12-29 15:49:53 -08001343out_put_keys:
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001344 put_futex_key(fshared, &key2);
Darren Hart42d35d42008-12-29 15:49:53 -08001345out_put_key1:
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001346 put_futex_key(fshared, &key1);
Darren Hart42d35d42008-12-29 15:49:53 -08001347out:
Darren Hart52400ba2009-04-03 13:40:49 -07001348 if (pi_state != NULL)
1349 free_pi_state(pi_state);
1350 return ret ? ret : task_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351}
1352
1353/* The key must be already stored in q->key. */
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01001354static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355{
Ingo Molnare2970f22006-06-27 02:54:47 -07001356 struct futex_hash_bucket *hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
Rusty Russell9adef582007-05-08 00:26:42 -07001358 get_futex_key_refs(&q->key);
Ingo Molnare2970f22006-06-27 02:54:47 -07001359 hb = hash_futex(&q->key);
1360 q->lock_ptr = &hb->lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
Ingo Molnare2970f22006-06-27 02:54:47 -07001362 spin_lock(&hb->lock);
1363 return hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364}
1365
Darren Hartd40d65c2009-09-21 22:30:15 -07001366static inline void
1367queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
1368{
1369 spin_unlock(&hb->lock);
1370 drop_futex_key_refs(&q->key);
1371}
1372
1373/**
1374 * queue_me() - Enqueue the futex_q on the futex_hash_bucket
1375 * @q: The futex_q to enqueue
1376 * @hb: The destination hash bucket
1377 *
1378 * The hb->lock must be held by the caller, and is released here. A call to
1379 * queue_me() is typically paired with exactly one call to unqueue_me(). The
1380 * exceptions involve the PI related operations, which may use unqueue_me_pi()
1381 * or nothing if the unqueue is done as part of the wake process and the unqueue
1382 * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
1383 * an example).
1384 */
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01001385static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386{
Pierre Peifferec92d082007-05-09 02:35:00 -07001387 int prio;
1388
1389 /*
1390 * The priority used to register this element is
1391 * - either the real thread-priority for the real-time threads
1392 * (i.e. threads with a priority lower than MAX_RT_PRIO)
1393 * - or MAX_RT_PRIO for non-RT threads.
1394 * Thus, all RT-threads are woken first in priority order, and
1395 * the others are woken last, in FIFO order.
1396 */
1397 prio = min(current->normal_prio, MAX_RT_PRIO);
1398
1399 plist_node_init(&q->list, prio);
1400#ifdef CONFIG_DEBUG_PI_LIST
Thomas Gleixnera2672452009-11-17 14:46:14 +01001401 q->list.plist.spinlock = &hb->lock;
Pierre Peifferec92d082007-05-09 02:35:00 -07001402#endif
1403 plist_add(&q->list, &hb->chain);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001404 q->task = current;
Ingo Molnare2970f22006-06-27 02:54:47 -07001405 spin_unlock(&hb->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406}
1407
Darren Hartd40d65c2009-09-21 22:30:15 -07001408/**
1409 * unqueue_me() - Remove the futex_q from its futex_hash_bucket
1410 * @q: The futex_q to unqueue
1411 *
1412 * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must
1413 * be paired with exactly one earlier call to queue_me().
1414 *
1415 * Returns:
1416 * 1 - if the futex_q was still queued (and we removed unqueued it)
1417 * 0 - if the futex_q was already removed by the waking thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419static int unqueue_me(struct futex_q *q)
1420{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 spinlock_t *lock_ptr;
Ingo Molnare2970f22006-06-27 02:54:47 -07001422 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
1424 /* In the common case we don't take the spinlock, which is nice. */
Darren Hart42d35d42008-12-29 15:49:53 -08001425retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 lock_ptr = q->lock_ptr;
Christian Borntraegere91467e2006-08-05 12:13:52 -07001427 barrier();
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001428 if (lock_ptr != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 spin_lock(lock_ptr);
1430 /*
1431 * q->lock_ptr can change between reading it and
1432 * spin_lock(), causing us to take the wrong lock. This
1433 * corrects the race condition.
1434 *
1435 * Reasoning goes like this: if we have the wrong lock,
1436 * q->lock_ptr must have changed (maybe several times)
1437 * between reading it and the spin_lock(). It can
1438 * change again after the spin_lock() but only if it was
1439 * already changed before the spin_lock(). It cannot,
1440 * however, change back to the original value. Therefore
1441 * we can detect whether we acquired the correct lock.
1442 */
1443 if (unlikely(lock_ptr != q->lock_ptr)) {
1444 spin_unlock(lock_ptr);
1445 goto retry;
1446 }
Pierre Peifferec92d082007-05-09 02:35:00 -07001447 WARN_ON(plist_node_empty(&q->list));
1448 plist_del(&q->list, &q->list.plist);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001449
1450 BUG_ON(q->pi_state);
1451
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 spin_unlock(lock_ptr);
1453 ret = 1;
1454 }
1455
Rusty Russell9adef582007-05-08 00:26:42 -07001456 drop_futex_key_refs(&q->key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 return ret;
1458}
1459
Ingo Molnarc87e2832006-06-27 02:54:58 -07001460/*
1461 * PI futexes can not be requeued and must remove themself from the
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001462 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1463 * and dropped here.
Ingo Molnarc87e2832006-06-27 02:54:58 -07001464 */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001465static void unqueue_me_pi(struct futex_q *q)
Ingo Molnarc87e2832006-06-27 02:54:58 -07001466{
Pierre Peifferec92d082007-05-09 02:35:00 -07001467 WARN_ON(plist_node_empty(&q->list));
1468 plist_del(&q->list, &q->list.plist);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001469
1470 BUG_ON(!q->pi_state);
1471 free_pi_state(q->pi_state);
1472 q->pi_state = NULL;
1473
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001474 spin_unlock(q->lock_ptr);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001475
Rusty Russell9adef582007-05-08 00:26:42 -07001476 drop_futex_key_refs(&q->key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001477}
1478
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001479/*
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001480 * Fixup the pi_state owner with the new owner.
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001481 *
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001482 * Must be called with hash bucket lock held and mm->sem held for non
1483 * private futexes.
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001484 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001485static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001486 struct task_struct *newowner, int fshared)
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001487{
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001488 u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001489 struct futex_pi_state *pi_state = q->pi_state;
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001490 struct task_struct *oldowner = pi_state->owner;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001491 u32 uval, curval, newval;
Darren Harte4dc5b72009-03-12 00:56:13 -07001492 int ret;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001493
1494 /* Owner died? */
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001495 if (!pi_state->owner)
1496 newtid |= FUTEX_OWNER_DIED;
1497
1498 /*
1499 * We are here either because we stole the rtmutex from the
1500 * pending owner or we are the pending owner which failed to
1501 * get the rtmutex. We have to replace the pending owner TID
1502 * in the user space variable. This must be atomic as we have
1503 * to preserve the owner died bit here.
1504 *
Darren Hartb2d09942009-03-12 00:55:37 -07001505 * Note: We write the user space value _before_ changing the pi_state
1506 * because we can fault here. Imagine swapped out pages or a fork
1507 * that marked all the anonymous memory readonly for cow.
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001508 *
1509 * Modifying pi_state _before_ the user space value would
1510 * leave the pi_state in an inconsistent state when we fault
1511 * here, because we need to drop the hash bucket lock to
1512 * handle the fault. This might be observed in the PID check
1513 * in lookup_pi_state.
1514 */
1515retry:
1516 if (get_futex_value_locked(&uval, uaddr))
1517 goto handle_fault;
1518
1519 while (1) {
1520 newval = (uval & FUTEX_OWNER_DIED) | newtid;
1521
1522 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
1523
1524 if (curval == -EFAULT)
1525 goto handle_fault;
1526 if (curval == uval)
1527 break;
1528 uval = curval;
1529 }
1530
1531 /*
1532 * We fixed up user space. Now we need to fix the pi_state
1533 * itself.
1534 */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001535 if (pi_state->owner != NULL) {
Thomas Gleixner1d615482009-11-17 14:54:03 +01001536 raw_spin_lock_irq(&pi_state->owner->pi_lock);
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001537 WARN_ON(list_empty(&pi_state->list));
1538 list_del_init(&pi_state->list);
Thomas Gleixner1d615482009-11-17 14:54:03 +01001539 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001540 }
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001541
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001542 pi_state->owner = newowner;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001543
Thomas Gleixner1d615482009-11-17 14:54:03 +01001544 raw_spin_lock_irq(&newowner->pi_lock);
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001545 WARN_ON(!list_empty(&pi_state->list));
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001546 list_add(&pi_state->list, &newowner->pi_state_list);
Thomas Gleixner1d615482009-11-17 14:54:03 +01001547 raw_spin_unlock_irq(&newowner->pi_lock);
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001548 return 0;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001549
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001550 /*
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001551 * To handle the page fault we need to drop the hash bucket
1552 * lock here. That gives the other task (either the pending
1553 * owner itself or the task which stole the rtmutex) the
1554 * chance to try the fixup of the pi_state. So once we are
1555 * back from handling the fault we need to check the pi_state
1556 * after reacquiring the hash bucket lock and before trying to
1557 * do another fixup. When the fixup has been done already we
1558 * simply return.
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001559 */
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001560handle_fault:
1561 spin_unlock(q->lock_ptr);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001562
Thomas Gleixnerd0725992009-06-11 23:15:43 +02001563 ret = fault_in_user_writeable(uaddr);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001564
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001565 spin_lock(q->lock_ptr);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001566
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001567 /*
1568 * Check if someone else fixed it for us:
1569 */
1570 if (pi_state->owner != oldowner)
1571 return 0;
1572
1573 if (ret)
1574 return ret;
1575
1576 goto retry;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001577}
1578
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001579/*
1580 * In case we must use restart_block to restart a futex_wait,
Steven Rostedtce6bd422007-12-05 15:46:09 +01001581 * we encode in the 'flags' shared capability
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001582 */
Thomas Gleixner1acdac12008-11-20 10:02:53 -08001583#define FLAGS_SHARED 0x01
1584#define FLAGS_CLOCKRT 0x02
Darren Harta72188d2009-04-03 13:40:22 -07001585#define FLAGS_HAS_TIMEOUT 0x04
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001586
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001587static long futex_wait_restart(struct restart_block *restart);
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001588
Darren Hartca5f9522009-04-03 13:39:33 -07001589/**
Darren Hartdd973992009-04-03 13:40:02 -07001590 * fixup_owner() - Post lock pi_state and corner case management
1591 * @uaddr: user address of the futex
1592 * @fshared: whether the futex is shared (1) or not (0)
1593 * @q: futex_q (contains pi_state and access to the rt_mutex)
1594 * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
1595 *
1596 * After attempting to lock an rt_mutex, this function is called to cleanup
1597 * the pi_state owner as well as handle race conditions that may allow us to
1598 * acquire the lock. Must be called with the hb lock held.
1599 *
1600 * Returns:
1601 * 1 - success, lock taken
1602 * 0 - success, lock not taken
1603 * <0 - on error (-EFAULT)
1604 */
1605static int fixup_owner(u32 __user *uaddr, int fshared, struct futex_q *q,
1606 int locked)
1607{
1608 struct task_struct *owner;
1609 int ret = 0;
1610
1611 if (locked) {
1612 /*
1613 * Got the lock. We might not be the anticipated owner if we
1614 * did a lock-steal - fix up the PI-state in that case:
1615 */
1616 if (q->pi_state->owner != current)
1617 ret = fixup_pi_state_owner(uaddr, q, current, fshared);
1618 goto out;
1619 }
1620
1621 /*
1622 * Catch the rare case, where the lock was released when we were on the
1623 * way back before we locked the hash bucket.
1624 */
1625 if (q->pi_state->owner == current) {
1626 /*
1627 * Try to get the rt_mutex now. This might fail as some other
1628 * task acquired the rt_mutex after we removed ourself from the
1629 * rt_mutex waiters list.
1630 */
1631 if (rt_mutex_trylock(&q->pi_state->pi_mutex)) {
1632 locked = 1;
1633 goto out;
1634 }
1635
1636 /*
1637 * pi_state is incorrect, some other task did a lock steal and
1638 * we returned due to timeout or signal without taking the
1639 * rt_mutex. Too late. We can access the rt_mutex_owner without
1640 * locking, as the other task is now blocked on the hash bucket
1641 * lock. Fix the state up.
1642 */
1643 owner = rt_mutex_owner(&q->pi_state->pi_mutex);
1644 ret = fixup_pi_state_owner(uaddr, q, owner, fshared);
1645 goto out;
1646 }
1647
1648 /*
1649 * Paranoia check. If we did not take the lock, then we should not be
1650 * the owner, nor the pending owner, of the rt_mutex.
1651 */
1652 if (rt_mutex_owner(&q->pi_state->pi_mutex) == current)
1653 printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p "
1654 "pi-state %p\n", ret,
1655 q->pi_state->pi_mutex.owner,
1656 q->pi_state->owner);
1657
1658out:
1659 return ret ? ret : locked;
1660}
1661
1662/**
Darren Hartca5f9522009-04-03 13:39:33 -07001663 * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
1664 * @hb: the futex hash bucket, must be locked by the caller
1665 * @q: the futex_q to queue up on
1666 * @timeout: the prepared hrtimer_sleeper, or null for no timeout
Darren Hartca5f9522009-04-03 13:39:33 -07001667 */
1668static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02001669 struct hrtimer_sleeper *timeout)
Darren Hartca5f9522009-04-03 13:39:33 -07001670{
Darren Hart9beba3c2009-09-24 11:54:47 -07001671 /*
1672 * The task state is guaranteed to be set before another task can
1673 * wake it. set_current_state() is implemented using set_mb() and
1674 * queue_me() calls spin_unlock() upon completion, both serializing
1675 * access to the hash list and forcing another memory barrier.
1676 */
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02001677 set_current_state(TASK_INTERRUPTIBLE);
Darren Hart0729e192009-09-21 22:30:38 -07001678 queue_me(q, hb);
Darren Hartca5f9522009-04-03 13:39:33 -07001679
1680 /* Arm the timer */
1681 if (timeout) {
1682 hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
1683 if (!hrtimer_active(&timeout->timer))
1684 timeout->task = NULL;
1685 }
1686
1687 /*
Darren Hart0729e192009-09-21 22:30:38 -07001688 * If we have been removed from the hash list, then another task
1689 * has tried to wake us, and we can skip the call to schedule().
Darren Hartca5f9522009-04-03 13:39:33 -07001690 */
1691 if (likely(!plist_node_empty(&q->list))) {
1692 /*
1693 * If the timer has already expired, current will already be
1694 * flagged for rescheduling. Only call schedule if there
1695 * is no timeout, or if it has yet to expire.
1696 */
1697 if (!timeout || timeout->task)
1698 schedule();
1699 }
1700 __set_current_state(TASK_RUNNING);
1701}
1702
Darren Hartf8010732009-04-03 13:40:40 -07001703/**
1704 * futex_wait_setup() - Prepare to wait on a futex
1705 * @uaddr: the futex userspace address
1706 * @val: the expected value
1707 * @fshared: whether the futex is shared (1) or not (0)
1708 * @q: the associated futex_q
1709 * @hb: storage for hash_bucket pointer to be returned to caller
1710 *
1711 * Setup the futex_q and locate the hash_bucket. Get the futex value and
1712 * compare it with the expected value. Handle atomic faults internally.
1713 * Return with the hb lock held and a q.key reference on success, and unlocked
1714 * with no q.key reference on failure.
1715 *
1716 * Returns:
1717 * 0 - uaddr contains val and hb has been locked
1718 * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlcoked
1719 */
1720static int futex_wait_setup(u32 __user *uaddr, u32 val, int fshared,
1721 struct futex_q *q, struct futex_hash_bucket **hb)
1722{
1723 u32 uval;
1724 int ret;
1725
1726 /*
1727 * Access the page AFTER the hash-bucket is locked.
1728 * Order is important:
1729 *
1730 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1731 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1732 *
1733 * The basic logical guarantee of a futex is that it blocks ONLY
1734 * if cond(var) is known to be true at the time of blocking, for
1735 * any cond. If we queued after testing *uaddr, that would open
1736 * a race condition where we could block indefinitely with
1737 * cond(var) false, which would violate the guarantee.
1738 *
1739 * A consequence is that futex_wait() can return zero and absorb
1740 * a wakeup when *uaddr != val on entry to the syscall. This is
1741 * rare, but normal.
1742 */
1743retry:
1744 q->key = FUTEX_KEY_INIT;
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +09001745 ret = get_futex_key(uaddr, fshared, &q->key);
Darren Hartf8010732009-04-03 13:40:40 -07001746 if (unlikely(ret != 0))
Darren Harta5a2a0c2009-04-10 09:50:05 -07001747 return ret;
Darren Hartf8010732009-04-03 13:40:40 -07001748
1749retry_private:
1750 *hb = queue_lock(q);
1751
1752 ret = get_futex_value_locked(&uval, uaddr);
1753
1754 if (ret) {
1755 queue_unlock(q, *hb);
1756
1757 ret = get_user(uval, uaddr);
1758 if (ret)
1759 goto out;
1760
1761 if (!fshared)
1762 goto retry_private;
1763
1764 put_futex_key(fshared, &q->key);
1765 goto retry;
1766 }
1767
1768 if (uval != val) {
1769 queue_unlock(q, *hb);
1770 ret = -EWOULDBLOCK;
1771 }
1772
1773out:
1774 if (ret)
1775 put_futex_key(fshared, &q->key);
1776 return ret;
1777}
1778
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001779static int futex_wait(u32 __user *uaddr, int fshared,
Thomas Gleixner1acdac12008-11-20 10:02:53 -08001780 u32 val, ktime_t *abs_time, u32 bitset, int clockrt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781{
Darren Hartca5f9522009-04-03 13:39:33 -07001782 struct hrtimer_sleeper timeout, *to = NULL;
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001783 struct restart_block *restart;
Ingo Molnare2970f22006-06-27 02:54:47 -07001784 struct futex_hash_bucket *hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 struct futex_q q;
Ingo Molnare2970f22006-06-27 02:54:47 -07001786 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787
Thomas Gleixnercd689982008-02-01 17:45:14 +01001788 if (!bitset)
1789 return -EINVAL;
1790
Ingo Molnarc87e2832006-06-27 02:54:58 -07001791 q.pi_state = NULL;
Thomas Gleixnercd689982008-02-01 17:45:14 +01001792 q.bitset = bitset;
Darren Hart52400ba2009-04-03 13:40:49 -07001793 q.rt_waiter = NULL;
Darren Hart84bc4af2009-08-13 17:36:53 -07001794 q.requeue_pi_key = NULL;
Darren Hartca5f9522009-04-03 13:39:33 -07001795
1796 if (abs_time) {
1797 to = &timeout;
1798
1799 hrtimer_init_on_stack(&to->timer, clockrt ? CLOCK_REALTIME :
1800 CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1801 hrtimer_init_sleeper(to, current);
1802 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
1803 current->timer_slack_ns);
1804 }
1805
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02001806retry:
Darren Hartf8010732009-04-03 13:40:40 -07001807 /* Prepare to wait on uaddr. */
1808 ret = futex_wait_setup(uaddr, val, fshared, &q, &hb);
1809 if (ret)
Darren Hart42d35d42008-12-29 15:49:53 -08001810 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
Darren Hartca5f9522009-04-03 13:39:33 -07001812 /* queue_me and wait for wakeup, timeout, or a signal. */
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02001813 futex_wait_queue_me(hb, &q, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814
1815 /* If we were woken (and unqueued), we succeeded, whatever. */
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001816 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 if (!unqueue_me(&q))
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001818 goto out_put_key;
1819 ret = -ETIMEDOUT;
Darren Hartca5f9522009-04-03 13:39:33 -07001820 if (to && !to->task)
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001821 goto out_put_key;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001822
Ingo Molnare2970f22006-06-27 02:54:47 -07001823 /*
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02001824 * We expect signal_pending(current), but we might be the
1825 * victim of a spurious wakeup as well.
Ingo Molnare2970f22006-06-27 02:54:47 -07001826 */
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02001827 if (!signal_pending(current)) {
1828 put_futex_key(fshared, &q.key);
1829 goto retry;
1830 }
1831
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001832 ret = -ERESTARTSYS;
Pierre Peifferc19384b2007-05-09 02:35:02 -07001833 if (!abs_time)
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001834 goto out_put_key;
Steven Rostedtce6bd422007-12-05 15:46:09 +01001835
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001836 restart = &current_thread_info()->restart_block;
1837 restart->fn = futex_wait_restart;
1838 restart->futex.uaddr = (u32 *)uaddr;
1839 restart->futex.val = val;
1840 restart->futex.time = abs_time->tv64;
1841 restart->futex.bitset = bitset;
Darren Harta72188d2009-04-03 13:40:22 -07001842 restart->futex.flags = FLAGS_HAS_TIMEOUT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001844 if (fshared)
1845 restart->futex.flags |= FLAGS_SHARED;
1846 if (clockrt)
1847 restart->futex.flags |= FLAGS_CLOCKRT;
1848
1849 ret = -ERESTART_RESTARTBLOCK;
1850
1851out_put_key:
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001852 put_futex_key(fshared, &q.key);
Darren Hart42d35d42008-12-29 15:49:53 -08001853out:
Darren Hartca5f9522009-04-03 13:39:33 -07001854 if (to) {
1855 hrtimer_cancel(&to->timer);
1856 destroy_hrtimer_on_stack(&to->timer);
1857 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001858 return ret;
1859}
1860
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001861
1862static long futex_wait_restart(struct restart_block *restart)
1863{
Steven Rostedtce6bd422007-12-05 15:46:09 +01001864 u32 __user *uaddr = (u32 __user *)restart->futex.uaddr;
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001865 int fshared = 0;
Darren Harta72188d2009-04-03 13:40:22 -07001866 ktime_t t, *tp = NULL;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001867
Darren Harta72188d2009-04-03 13:40:22 -07001868 if (restart->futex.flags & FLAGS_HAS_TIMEOUT) {
1869 t.tv64 = restart->futex.time;
1870 tp = &t;
1871 }
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001872 restart->fn = do_no_restart_syscall;
Steven Rostedtce6bd422007-12-05 15:46:09 +01001873 if (restart->futex.flags & FLAGS_SHARED)
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001874 fshared = 1;
Darren Harta72188d2009-04-03 13:40:22 -07001875 return (long)futex_wait(uaddr, fshared, restart->futex.val, tp,
Thomas Gleixner1acdac12008-11-20 10:02:53 -08001876 restart->futex.bitset,
1877 restart->futex.flags & FLAGS_CLOCKRT);
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001878}
1879
1880
Ingo Molnarc87e2832006-06-27 02:54:58 -07001881/*
1882 * Userspace tried a 0 -> TID atomic transition of the futex value
1883 * and failed. The kernel side here does the whole locking operation:
1884 * if there are waiters then it will block, it does PI, etc. (Due to
1885 * races the kernel might see a 0 value of the futex too.)
1886 */
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001887static int futex_lock_pi(u32 __user *uaddr, int fshared,
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001888 int detect, ktime_t *time, int trylock)
Ingo Molnarc87e2832006-06-27 02:54:58 -07001889{
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001890 struct hrtimer_sleeper timeout, *to = NULL;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001891 struct futex_hash_bucket *hb;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001892 struct futex_q q;
Darren Hartdd973992009-04-03 13:40:02 -07001893 int res, ret;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001894
1895 if (refill_pi_state_cache())
1896 return -ENOMEM;
1897
Pierre Peifferc19384b2007-05-09 02:35:02 -07001898 if (time) {
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001899 to = &timeout;
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001900 hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME,
1901 HRTIMER_MODE_ABS);
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001902 hrtimer_init_sleeper(to, current);
Arjan van de Vencc584b22008-09-01 15:02:30 -07001903 hrtimer_set_expires(&to->timer, *time);
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001904 }
1905
Ingo Molnarc87e2832006-06-27 02:54:58 -07001906 q.pi_state = NULL;
Darren Hart52400ba2009-04-03 13:40:49 -07001907 q.rt_waiter = NULL;
Darren Hart84bc4af2009-08-13 17:36:53 -07001908 q.requeue_pi_key = NULL;
Darren Hart42d35d42008-12-29 15:49:53 -08001909retry:
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001910 q.key = FUTEX_KEY_INIT;
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +09001911 ret = get_futex_key(uaddr, fshared, &q.key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001912 if (unlikely(ret != 0))
Darren Hart42d35d42008-12-29 15:49:53 -08001913 goto out;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001914
Darren Harte4dc5b72009-03-12 00:56:13 -07001915retry_private:
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01001916 hb = queue_lock(&q);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001917
Darren Hartbab5bc92009-04-07 23:23:50 -07001918 ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001919 if (unlikely(ret)) {
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001920 switch (ret) {
Darren Hart1a520842009-04-03 13:39:52 -07001921 case 1:
1922 /* We got the lock. */
1923 ret = 0;
1924 goto out_unlock_put_key;
1925 case -EFAULT:
1926 goto uaddr_faulted;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001927 case -EAGAIN:
1928 /*
1929 * Task is exiting and we just wait for the
1930 * exit to complete.
1931 */
1932 queue_unlock(&q, hb);
Darren Hartde87fcc2009-03-12 00:55:46 -07001933 put_futex_key(fshared, &q.key);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001934 cond_resched();
1935 goto retry;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001936 default:
Darren Hart42d35d42008-12-29 15:49:53 -08001937 goto out_unlock_put_key;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001938 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001939 }
1940
1941 /*
1942 * Only actually queue now that the atomic ops are done:
1943 */
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01001944 queue_me(&q, hb);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001945
Ingo Molnarc87e2832006-06-27 02:54:58 -07001946 WARN_ON(!q.pi_state);
1947 /*
1948 * Block on the PI mutex:
1949 */
1950 if (!trylock)
1951 ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1);
1952 else {
1953 ret = rt_mutex_trylock(&q.pi_state->pi_mutex);
1954 /* Fixup the trylock return value: */
1955 ret = ret ? 0 : -EWOULDBLOCK;
1956 }
1957
Vernon Mauerya99e4e42006-07-01 04:35:42 -07001958 spin_lock(q.lock_ptr);
Darren Hartdd973992009-04-03 13:40:02 -07001959 /*
1960 * Fixup the pi_state owner and possibly acquire the lock if we
1961 * haven't already.
1962 */
1963 res = fixup_owner(uaddr, fshared, &q, !ret);
1964 /*
1965 * If fixup_owner() returned an error, proprogate that. If it acquired
1966 * the lock, clear our -ETIMEDOUT or -EINTR.
1967 */
1968 if (res)
1969 ret = (res < 0) ? res : 0;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001970
Darren Harte8f63862009-03-12 00:56:06 -07001971 /*
Darren Hartdd973992009-04-03 13:40:02 -07001972 * If fixup_owner() faulted and was unable to handle the fault, unlock
1973 * it and return the fault to userspace.
Darren Harte8f63862009-03-12 00:56:06 -07001974 */
1975 if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current))
1976 rt_mutex_unlock(&q.pi_state->pi_mutex);
1977
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001978 /* Unqueue and drop the lock */
1979 unqueue_me_pi(&q);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001980
Mikael Pettersson5ecb01c2010-01-23 22:36:29 +01001981 goto out_put_key;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001982
Darren Hart42d35d42008-12-29 15:49:53 -08001983out_unlock_put_key:
Ingo Molnarc87e2832006-06-27 02:54:58 -07001984 queue_unlock(&q, hb);
1985
Darren Hart42d35d42008-12-29 15:49:53 -08001986out_put_key:
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001987 put_futex_key(fshared, &q.key);
Darren Hart42d35d42008-12-29 15:49:53 -08001988out:
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001989 if (to)
1990 destroy_hrtimer_on_stack(&to->timer);
Darren Hartdd973992009-04-03 13:40:02 -07001991 return ret != -EINTR ? ret : -ERESTARTNOINTR;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001992
Darren Hart42d35d42008-12-29 15:49:53 -08001993uaddr_faulted:
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001994 queue_unlock(&q, hb);
1995
Thomas Gleixnerd0725992009-06-11 23:15:43 +02001996 ret = fault_in_user_writeable(uaddr);
Darren Harte4dc5b72009-03-12 00:56:13 -07001997 if (ret)
1998 goto out_put_key;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001999
Darren Harte4dc5b72009-03-12 00:56:13 -07002000 if (!fshared)
2001 goto retry_private;
2002
2003 put_futex_key(fshared, &q.key);
2004 goto retry;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002005}
2006
2007/*
Ingo Molnarc87e2832006-06-27 02:54:58 -07002008 * Userspace attempted a TID -> 0 atomic transition, and failed.
2009 * This is the in-kernel slowpath: we look up the PI state (if any),
2010 * and do the rt-mutex unlock.
2011 */
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02002012static int futex_unlock_pi(u32 __user *uaddr, int fshared)
Ingo Molnarc87e2832006-06-27 02:54:58 -07002013{
2014 struct futex_hash_bucket *hb;
2015 struct futex_q *this, *next;
2016 u32 uval;
Pierre Peifferec92d082007-05-09 02:35:00 -07002017 struct plist_head *head;
Peter Zijlstra38d47c12008-09-26 19:32:20 +02002018 union futex_key key = FUTEX_KEY_INIT;
Darren Harte4dc5b72009-03-12 00:56:13 -07002019 int ret;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002020
2021retry:
2022 if (get_user(uval, uaddr))
2023 return -EFAULT;
2024 /*
2025 * We release only a lock we actually own:
2026 */
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002027 if ((uval & FUTEX_TID_MASK) != task_pid_vnr(current))
Ingo Molnarc87e2832006-06-27 02:54:58 -07002028 return -EPERM;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002029
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +09002030 ret = get_futex_key(uaddr, fshared, &key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002031 if (unlikely(ret != 0))
2032 goto out;
2033
2034 hb = hash_futex(&key);
2035 spin_lock(&hb->lock);
2036
Ingo Molnarc87e2832006-06-27 02:54:58 -07002037 /*
2038 * To avoid races, try to do the TID -> 0 atomic transition
2039 * again. If it succeeds then we can return without waking
2040 * anyone else up:
2041 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07002042 if (!(uval & FUTEX_OWNER_DIED))
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002043 uval = cmpxchg_futex_value_locked(uaddr, task_pid_vnr(current), 0);
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07002044
Ingo Molnarc87e2832006-06-27 02:54:58 -07002045
2046 if (unlikely(uval == -EFAULT))
2047 goto pi_faulted;
2048 /*
2049 * Rare case: we managed to release the lock atomically,
2050 * no need to wake anyone else up:
2051 */
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002052 if (unlikely(uval == task_pid_vnr(current)))
Ingo Molnarc87e2832006-06-27 02:54:58 -07002053 goto out_unlock;
2054
2055 /*
2056 * Ok, other tasks may need to be woken up - check waiters
2057 * and do the wakeup if necessary:
2058 */
2059 head = &hb->chain;
2060
Pierre Peifferec92d082007-05-09 02:35:00 -07002061 plist_for_each_entry_safe(this, next, head, list) {
Ingo Molnarc87e2832006-06-27 02:54:58 -07002062 if (!match_futex (&this->key, &key))
2063 continue;
2064 ret = wake_futex_pi(uaddr, uval, this);
2065 /*
2066 * The atomic access to the futex value
2067 * generated a pagefault, so retry the
2068 * user-access and the wakeup:
2069 */
2070 if (ret == -EFAULT)
2071 goto pi_faulted;
2072 goto out_unlock;
2073 }
2074 /*
2075 * No waiters - kernel unlocks the futex:
2076 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002077 if (!(uval & FUTEX_OWNER_DIED)) {
2078 ret = unlock_futex_pi(uaddr, uval);
2079 if (ret == -EFAULT)
2080 goto pi_faulted;
2081 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07002082
2083out_unlock:
2084 spin_unlock(&hb->lock);
Peter Zijlstra38d47c12008-09-26 19:32:20 +02002085 put_futex_key(fshared, &key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002086
Darren Hart42d35d42008-12-29 15:49:53 -08002087out:
Ingo Molnarc87e2832006-06-27 02:54:58 -07002088 return ret;
2089
2090pi_faulted:
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002091 spin_unlock(&hb->lock);
Darren Harte4dc5b72009-03-12 00:56:13 -07002092 put_futex_key(fshared, &key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002093
Thomas Gleixnerd0725992009-06-11 23:15:43 +02002094 ret = fault_in_user_writeable(uaddr);
Darren Hartb5686362008-12-18 15:06:34 -08002095 if (!ret)
Ingo Molnarc87e2832006-06-27 02:54:58 -07002096 goto retry;
2097
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 return ret;
2099}
2100
Darren Hart52400ba2009-04-03 13:40:49 -07002101/**
2102 * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2103 * @hb: the hash_bucket futex_q was original enqueued on
2104 * @q: the futex_q woken while waiting to be requeued
2105 * @key2: the futex_key of the requeue target futex
2106 * @timeout: the timeout associated with the wait (NULL if none)
2107 *
2108 * Detect if the task was woken on the initial futex as opposed to the requeue
2109 * target futex. If so, determine if it was a timeout or a signal that caused
2110 * the wakeup and return the appropriate error code to the caller. Must be
2111 * called with the hb lock held.
2112 *
2113 * Returns
2114 * 0 - no early wakeup detected
Thomas Gleixner1c840c12009-05-20 09:22:40 +02002115 * <0 - -ETIMEDOUT or -ERESTARTNOINTR
Darren Hart52400ba2009-04-03 13:40:49 -07002116 */
2117static inline
2118int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb,
2119 struct futex_q *q, union futex_key *key2,
2120 struct hrtimer_sleeper *timeout)
2121{
2122 int ret = 0;
2123
2124 /*
2125 * With the hb lock held, we avoid races while we process the wakeup.
2126 * We only need to hold hb (and not hb2) to ensure atomicity as the
2127 * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2128 * It can't be requeued from uaddr2 to something else since we don't
2129 * support a PI aware source futex for requeue.
2130 */
2131 if (!match_futex(&q->key, key2)) {
2132 WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr));
2133 /*
2134 * We were woken prior to requeue by a timeout or a signal.
2135 * Unqueue the futex_q and determine which it was.
2136 */
2137 plist_del(&q->list, &q->list.plist);
Darren Hart52400ba2009-04-03 13:40:49 -07002138
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02002139 /* Handle spurious wakeups gracefully */
Thomas Gleixner11df6dd2009-10-28 20:26:48 +01002140 ret = -EWOULDBLOCK;
Darren Hart52400ba2009-04-03 13:40:49 -07002141 if (timeout && !timeout->task)
2142 ret = -ETIMEDOUT;
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02002143 else if (signal_pending(current))
Thomas Gleixner1c840c12009-05-20 09:22:40 +02002144 ret = -ERESTARTNOINTR;
Darren Hart52400ba2009-04-03 13:40:49 -07002145 }
2146 return ret;
2147}
2148
2149/**
2150 * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
Darren Hart56ec1602009-09-21 22:29:59 -07002151 * @uaddr: the futex we initially wait on (non-pi)
Darren Hart52400ba2009-04-03 13:40:49 -07002152 * @fshared: whether the futexes are shared (1) or not (0). They must be
2153 * the same type, no requeueing from private to shared, etc.
2154 * @val: the expected value of uaddr
2155 * @abs_time: absolute timeout
Darren Hart56ec1602009-09-21 22:29:59 -07002156 * @bitset: 32 bit wakeup bitset set by userspace, defaults to all
Darren Hart52400ba2009-04-03 13:40:49 -07002157 * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0)
2158 * @uaddr2: the pi futex we will take prior to returning to user-space
2159 *
2160 * The caller will wait on uaddr and will be requeued by futex_requeue() to
2161 * uaddr2 which must be PI aware. Normal wakeup will wake on uaddr2 and
2162 * complete the acquisition of the rt_mutex prior to returning to userspace.
2163 * This ensures the rt_mutex maintains an owner when it has waiters; without
2164 * one, the pi logic wouldn't know which task to boost/deboost, if there was a
2165 * need to.
2166 *
2167 * We call schedule in futex_wait_queue_me() when we enqueue and return there
2168 * via the following:
2169 * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
Darren Hartcc6db4e2009-07-31 16:20:10 -07002170 * 2) wakeup on uaddr2 after a requeue
2171 * 3) signal
2172 * 4) timeout
Darren Hart52400ba2009-04-03 13:40:49 -07002173 *
Darren Hartcc6db4e2009-07-31 16:20:10 -07002174 * If 3, cleanup and return -ERESTARTNOINTR.
Darren Hart52400ba2009-04-03 13:40:49 -07002175 *
2176 * If 2, we may then block on trying to take the rt_mutex and return via:
2177 * 5) successful lock
2178 * 6) signal
2179 * 7) timeout
2180 * 8) other lock acquisition failure
2181 *
Darren Hartcc6db4e2009-07-31 16:20:10 -07002182 * If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
Darren Hart52400ba2009-04-03 13:40:49 -07002183 *
2184 * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2185 *
2186 * Returns:
2187 * 0 - On success
2188 * <0 - On error
2189 */
2190static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared,
2191 u32 val, ktime_t *abs_time, u32 bitset,
2192 int clockrt, u32 __user *uaddr2)
2193{
2194 struct hrtimer_sleeper timeout, *to = NULL;
2195 struct rt_mutex_waiter rt_waiter;
2196 struct rt_mutex *pi_mutex = NULL;
Darren Hart52400ba2009-04-03 13:40:49 -07002197 struct futex_hash_bucket *hb;
2198 union futex_key key2;
2199 struct futex_q q;
2200 int res, ret;
Darren Hart52400ba2009-04-03 13:40:49 -07002201
2202 if (!bitset)
2203 return -EINVAL;
2204
2205 if (abs_time) {
2206 to = &timeout;
2207 hrtimer_init_on_stack(&to->timer, clockrt ? CLOCK_REALTIME :
2208 CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
2209 hrtimer_init_sleeper(to, current);
2210 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
2211 current->timer_slack_ns);
2212 }
2213
2214 /*
2215 * The waiter is allocated on our stack, manipulated by the requeue
2216 * code while we sleep on uaddr.
2217 */
2218 debug_rt_mutex_init_waiter(&rt_waiter);
2219 rt_waiter.task = NULL;
2220
Darren Hart52400ba2009-04-03 13:40:49 -07002221 key2 = FUTEX_KEY_INIT;
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +09002222 ret = get_futex_key(uaddr2, fshared, &key2);
Darren Hart52400ba2009-04-03 13:40:49 -07002223 if (unlikely(ret != 0))
2224 goto out;
2225
Darren Hart84bc4af2009-08-13 17:36:53 -07002226 q.pi_state = NULL;
2227 q.bitset = bitset;
2228 q.rt_waiter = &rt_waiter;
2229 q.requeue_pi_key = &key2;
2230
Darren Hart52400ba2009-04-03 13:40:49 -07002231 /* Prepare to wait on uaddr. */
2232 ret = futex_wait_setup(uaddr, val, fshared, &q, &hb);
Thomas Gleixnerc8b15a72009-05-20 09:18:50 +02002233 if (ret)
2234 goto out_key2;
Darren Hart52400ba2009-04-03 13:40:49 -07002235
2236 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02002237 futex_wait_queue_me(hb, &q, to);
Darren Hart52400ba2009-04-03 13:40:49 -07002238
2239 spin_lock(&hb->lock);
2240 ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
2241 spin_unlock(&hb->lock);
2242 if (ret)
2243 goto out_put_keys;
2244
2245 /*
2246 * In order for us to be here, we know our q.key == key2, and since
2247 * we took the hb->lock above, we also know that futex_requeue() has
2248 * completed and we no longer have to concern ourselves with a wakeup
2249 * race with the atomic proxy lock acquition by the requeue code.
2250 */
2251
2252 /* Check if the requeue code acquired the second futex for us. */
2253 if (!q.rt_waiter) {
2254 /*
2255 * Got the lock. We might not be the anticipated owner if we
2256 * did a lock-steal - fix up the PI-state in that case.
2257 */
2258 if (q.pi_state && (q.pi_state->owner != current)) {
2259 spin_lock(q.lock_ptr);
2260 ret = fixup_pi_state_owner(uaddr2, &q, current,
2261 fshared);
2262 spin_unlock(q.lock_ptr);
2263 }
2264 } else {
2265 /*
2266 * We have been woken up by futex_unlock_pi(), a timeout, or a
2267 * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2268 * the pi_state.
2269 */
2270 WARN_ON(!&q.pi_state);
2271 pi_mutex = &q.pi_state->pi_mutex;
2272 ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1);
2273 debug_rt_mutex_free_waiter(&rt_waiter);
2274
2275 spin_lock(q.lock_ptr);
2276 /*
2277 * Fixup the pi_state owner and possibly acquire the lock if we
2278 * haven't already.
2279 */
2280 res = fixup_owner(uaddr2, fshared, &q, !ret);
2281 /*
2282 * If fixup_owner() returned an error, proprogate that. If it
Darren Hart56ec1602009-09-21 22:29:59 -07002283 * acquired the lock, clear -ETIMEDOUT or -EINTR.
Darren Hart52400ba2009-04-03 13:40:49 -07002284 */
2285 if (res)
2286 ret = (res < 0) ? res : 0;
2287
2288 /* Unqueue and drop the lock. */
2289 unqueue_me_pi(&q);
2290 }
2291
2292 /*
2293 * If fixup_pi_state_owner() faulted and was unable to handle the
2294 * fault, unlock the rt_mutex and return the fault to userspace.
2295 */
2296 if (ret == -EFAULT) {
2297 if (rt_mutex_owner(pi_mutex) == current)
2298 rt_mutex_unlock(pi_mutex);
2299 } else if (ret == -EINTR) {
Darren Hart52400ba2009-04-03 13:40:49 -07002300 /*
Darren Hartcc6db4e2009-07-31 16:20:10 -07002301 * We've already been requeued, but cannot restart by calling
2302 * futex_lock_pi() directly. We could restart this syscall, but
2303 * it would detect that the user space "val" changed and return
2304 * -EWOULDBLOCK. Save the overhead of the restart and return
2305 * -EWOULDBLOCK directly.
Darren Hart52400ba2009-04-03 13:40:49 -07002306 */
Thomas Gleixner20708872009-05-19 23:04:59 +02002307 ret = -EWOULDBLOCK;
Darren Hart52400ba2009-04-03 13:40:49 -07002308 }
2309
2310out_put_keys:
2311 put_futex_key(fshared, &q.key);
Thomas Gleixnerc8b15a72009-05-20 09:18:50 +02002312out_key2:
Darren Hart52400ba2009-04-03 13:40:49 -07002313 put_futex_key(fshared, &key2);
2314
2315out:
2316 if (to) {
2317 hrtimer_cancel(&to->timer);
2318 destroy_hrtimer_on_stack(&to->timer);
2319 }
2320 return ret;
2321}
2322
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002323/*
2324 * Support for robust futexes: the kernel cleans up held futexes at
2325 * thread exit time.
2326 *
2327 * Implementation: user-space maintains a per-thread list of locks it
2328 * is holding. Upon do_exit(), the kernel carefully walks this list,
2329 * and marks all locks that are owned by this thread with the
Ingo Molnarc87e2832006-06-27 02:54:58 -07002330 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002331 * always manipulated with the lock held, so the list is private and
2332 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2333 * field, to allow the kernel to clean up if the thread dies after
2334 * acquiring the lock, but just before it could have added itself to
2335 * the list. There can only be one such pending lock.
2336 */
2337
2338/**
Darren Hartd96ee562009-09-21 22:30:22 -07002339 * sys_set_robust_list() - Set the robust-futex list head of a task
2340 * @head: pointer to the list-head
2341 * @len: length of the list-head, as userspace expects
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002342 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01002343SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head,
2344 size_t, len)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002345{
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002346 if (!futex_cmpxchg_enabled)
2347 return -ENOSYS;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002348 /*
2349 * The kernel knows only one size for now:
2350 */
2351 if (unlikely(len != sizeof(*head)))
2352 return -EINVAL;
2353
2354 current->robust_list = head;
2355
2356 return 0;
2357}
2358
2359/**
Darren Hartd96ee562009-09-21 22:30:22 -07002360 * sys_get_robust_list() - Get the robust-futex list head of a task
2361 * @pid: pid of the process [zero for current task]
2362 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2363 * @len_ptr: pointer to a length field, the kernel fills in the header size
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002364 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01002365SYSCALL_DEFINE3(get_robust_list, int, pid,
2366 struct robust_list_head __user * __user *, head_ptr,
2367 size_t __user *, len_ptr)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002368{
Al Viroba46df92006-10-10 22:46:07 +01002369 struct robust_list_head __user *head;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002370 unsigned long ret;
David Howellsc69e8d92008-11-14 10:39:19 +11002371 const struct cred *cred = current_cred(), *pcred;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002372
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002373 if (!futex_cmpxchg_enabled)
2374 return -ENOSYS;
2375
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002376 if (!pid)
2377 head = current->robust_list;
2378 else {
2379 struct task_struct *p;
2380
2381 ret = -ESRCH;
Oleg Nesterovaaa2a972006-09-29 02:00:55 -07002382 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07002383 p = find_task_by_vpid(pid);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002384 if (!p)
2385 goto err_unlock;
2386 ret = -EPERM;
David Howellsc69e8d92008-11-14 10:39:19 +11002387 pcred = __task_cred(p);
2388 if (cred->euid != pcred->euid &&
2389 cred->euid != pcred->uid &&
David Howells76aac0e2008-11-14 10:39:12 +11002390 !capable(CAP_SYS_PTRACE))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002391 goto err_unlock;
2392 head = p->robust_list;
Oleg Nesterovaaa2a972006-09-29 02:00:55 -07002393 rcu_read_unlock();
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002394 }
2395
2396 if (put_user(sizeof(*head), len_ptr))
2397 return -EFAULT;
2398 return put_user(head, head_ptr);
2399
2400err_unlock:
Oleg Nesterovaaa2a972006-09-29 02:00:55 -07002401 rcu_read_unlock();
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002402
2403 return ret;
2404}
2405
2406/*
2407 * Process a futex-list entry, check whether it's owned by the
2408 * dying task, and do notification if so:
2409 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002410int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002411{
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002412 u32 uval, nval, mval;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002413
Ingo Molnar8f17d3a2006-03-27 01:16:27 -08002414retry:
2415 if (get_user(uval, uaddr))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002416 return -1;
2417
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002418 if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) {
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002419 /*
2420 * Ok, this dying thread is truly holding a futex
2421 * of interest. Set the OWNER_DIED bit atomically
2422 * via cmpxchg, and if the value had FUTEX_WAITERS
2423 * set, wake up a waiter (if any). (We have to do a
2424 * futex_wake() even if OWNER_DIED is already set -
2425 * to handle the rare but possible case of recursive
2426 * thread-death.) The rest of the cleanup is done in
2427 * userspace.
2428 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002429 mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
2430 nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval);
2431
Ingo Molnarc87e2832006-06-27 02:54:58 -07002432 if (nval == -EFAULT)
2433 return -1;
2434
2435 if (nval != uval)
Ingo Molnar8f17d3a2006-03-27 01:16:27 -08002436 goto retry;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002437
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002438 /*
2439 * Wake robust non-PI futexes here. The wakeup of
2440 * PI futexes happens in exit_pi_state():
2441 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07002442 if (!pi && (uval & FUTEX_WAITERS))
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02002443 futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002444 }
2445 return 0;
2446}
2447
2448/*
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002449 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2450 */
2451static inline int fetch_robust_entry(struct robust_list __user **entry,
Al Viroba46df92006-10-10 22:46:07 +01002452 struct robust_list __user * __user *head,
2453 int *pi)
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002454{
2455 unsigned long uentry;
2456
Al Viroba46df92006-10-10 22:46:07 +01002457 if (get_user(uentry, (unsigned long __user *)head))
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002458 return -EFAULT;
2459
Al Viroba46df92006-10-10 22:46:07 +01002460 *entry = (void __user *)(uentry & ~1UL);
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002461 *pi = uentry & 1;
2462
2463 return 0;
2464}
2465
2466/*
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002467 * Walk curr->robust_list (very carefully, it's a userspace list!)
2468 * and mark any locks found there dead, and notify any waiters.
2469 *
2470 * We silently return on any sign of list-walking problem.
2471 */
2472void exit_robust_list(struct task_struct *curr)
2473{
2474 struct robust_list_head __user *head = curr->robust_list;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002475 struct robust_list __user *entry, *next_entry, *pending;
2476 unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002477 unsigned long futex_offset;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002478 int rc;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002479
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002480 if (!futex_cmpxchg_enabled)
2481 return;
2482
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002483 /*
2484 * Fetch the list head (which was registered earlier, via
2485 * sys_set_robust_list()):
2486 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002487 if (fetch_robust_entry(&entry, &head->list.next, &pi))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002488 return;
2489 /*
2490 * Fetch the relative futex offset:
2491 */
2492 if (get_user(futex_offset, &head->futex_offset))
2493 return;
2494 /*
2495 * Fetch any possibly pending lock-add first, and handle it
2496 * if it exists:
2497 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002498 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002499 return;
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002500
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002501 next_entry = NULL; /* avoid warning with gcc */
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002502 while (entry != &head->list) {
2503 /*
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002504 * Fetch the next entry in the list before calling
2505 * handle_futex_death:
2506 */
2507 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
2508 /*
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002509 * A pending lock might already be on the list, so
Ingo Molnarc87e2832006-06-27 02:54:58 -07002510 * don't process it twice:
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002511 */
2512 if (entry != pending)
Al Viroba46df92006-10-10 22:46:07 +01002513 if (handle_futex_death((void __user *)entry + futex_offset,
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002514 curr, pi))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002515 return;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002516 if (rc)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002517 return;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002518 entry = next_entry;
2519 pi = next_pi;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002520 /*
2521 * Avoid excessively long or circular lists:
2522 */
2523 if (!--limit)
2524 break;
2525
2526 cond_resched();
2527 }
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002528
2529 if (pending)
2530 handle_futex_death((void __user *)pending + futex_offset,
2531 curr, pip);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002532}
2533
Pierre Peifferc19384b2007-05-09 02:35:02 -07002534long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
Ingo Molnare2970f22006-06-27 02:54:47 -07002535 u32 __user *uaddr2, u32 val2, u32 val3)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536{
Thomas Gleixner1acdac12008-11-20 10:02:53 -08002537 int clockrt, ret = -ENOSYS;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002538 int cmd = op & FUTEX_CMD_MASK;
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02002539 int fshared = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002541 if (!(op & FUTEX_PRIVATE_FLAG))
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02002542 fshared = 1;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002543
Thomas Gleixner1acdac12008-11-20 10:02:53 -08002544 clockrt = op & FUTEX_CLOCK_REALTIME;
Darren Hart52400ba2009-04-03 13:40:49 -07002545 if (clockrt && cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
Thomas Gleixner1acdac12008-11-20 10:02:53 -08002546 return -ENOSYS;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002547
2548 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 case FUTEX_WAIT:
Thomas Gleixnercd689982008-02-01 17:45:14 +01002550 val3 = FUTEX_BITSET_MATCH_ANY;
2551 case FUTEX_WAIT_BITSET:
Thomas Gleixner1acdac12008-11-20 10:02:53 -08002552 ret = futex_wait(uaddr, fshared, val, timeout, val3, clockrt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 break;
2554 case FUTEX_WAKE:
Thomas Gleixnercd689982008-02-01 17:45:14 +01002555 val3 = FUTEX_BITSET_MATCH_ANY;
2556 case FUTEX_WAKE_BITSET:
2557 ret = futex_wake(uaddr, fshared, val, val3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 case FUTEX_REQUEUE:
Darren Hart52400ba2009-04-03 13:40:49 -07002560 ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 break;
2562 case FUTEX_CMP_REQUEUE:
Darren Hart52400ba2009-04-03 13:40:49 -07002563 ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3,
2564 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 break;
Jakub Jelinek4732efb2005-09-06 15:16:25 -07002566 case FUTEX_WAKE_OP:
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002567 ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3);
Jakub Jelinek4732efb2005-09-06 15:16:25 -07002568 break;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002569 case FUTEX_LOCK_PI:
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002570 if (futex_cmpxchg_enabled)
2571 ret = futex_lock_pi(uaddr, fshared, val, timeout, 0);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002572 break;
2573 case FUTEX_UNLOCK_PI:
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002574 if (futex_cmpxchg_enabled)
2575 ret = futex_unlock_pi(uaddr, fshared);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002576 break;
2577 case FUTEX_TRYLOCK_PI:
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002578 if (futex_cmpxchg_enabled)
2579 ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002580 break;
Darren Hart52400ba2009-04-03 13:40:49 -07002581 case FUTEX_WAIT_REQUEUE_PI:
2582 val3 = FUTEX_BITSET_MATCH_ANY;
2583 ret = futex_wait_requeue_pi(uaddr, fshared, val, timeout, val3,
2584 clockrt, uaddr2);
2585 break;
Darren Hart52400ba2009-04-03 13:40:49 -07002586 case FUTEX_CMP_REQUEUE_PI:
2587 ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3,
2588 1);
2589 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 default:
2591 ret = -ENOSYS;
2592 }
2593 return ret;
2594}
2595
2596
Heiko Carstens17da2bd2009-01-14 14:14:10 +01002597SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
2598 struct timespec __user *, utime, u32 __user *, uaddr2,
2599 u32, val3)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600{
Pierre Peifferc19384b2007-05-09 02:35:02 -07002601 struct timespec ts;
2602 ktime_t t, *tp = NULL;
Ingo Molnare2970f22006-06-27 02:54:47 -07002603 u32 val2 = 0;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002604 int cmd = op & FUTEX_CMD_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
Thomas Gleixnercd689982008-02-01 17:45:14 +01002606 if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
Darren Hart52400ba2009-04-03 13:40:49 -07002607 cmd == FUTEX_WAIT_BITSET ||
2608 cmd == FUTEX_WAIT_REQUEUE_PI)) {
Pierre Peifferc19384b2007-05-09 02:35:02 -07002609 if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 return -EFAULT;
Pierre Peifferc19384b2007-05-09 02:35:02 -07002611 if (!timespec_valid(&ts))
Thomas Gleixner9741ef962006-03-31 02:31:32 -08002612 return -EINVAL;
Pierre Peifferc19384b2007-05-09 02:35:02 -07002613
2614 t = timespec_to_ktime(ts);
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002615 if (cmd == FUTEX_WAIT)
Thomas Gleixner5a7780e2008-02-13 09:20:43 +01002616 t = ktime_add_safe(ktime_get(), t);
Pierre Peifferc19384b2007-05-09 02:35:02 -07002617 tp = &t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 }
2619 /*
Darren Hart52400ba2009-04-03 13:40:49 -07002620 * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
Andreas Schwabf54f0982007-07-31 00:38:51 -07002621 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 */
Andreas Schwabf54f0982007-07-31 00:38:51 -07002623 if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
Darren Hartba9c22f2009-04-20 22:22:22 -07002624 cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP)
Ingo Molnare2970f22006-06-27 02:54:47 -07002625 val2 = (u32) (unsigned long) utime;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626
Pierre Peifferc19384b2007-05-09 02:35:02 -07002627 return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628}
2629
Benjamin Herrenschmidtf6d107f2008-03-27 14:52:15 +11002630static int __init futex_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631{
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002632 u32 curval;
Thomas Gleixner3e4ab742008-02-23 15:23:55 -08002633 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002635 /*
2636 * This will fail and we want it. Some arch implementations do
2637 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2638 * functionality. We want to know that before we call in any
2639 * of the complex code paths. Also we want to prevent
2640 * registration of robust lists in that case. NULL is
2641 * guaranteed to fault and we get -EFAULT on functional
2642 * implementation, the non functional ones will return
2643 * -ENOSYS.
2644 */
2645 curval = cmpxchg_futex_value_locked(NULL, 0, 0);
2646 if (curval == -EFAULT)
2647 futex_cmpxchg_enabled = 1;
2648
Thomas Gleixner3e4ab742008-02-23 15:23:55 -08002649 for (i = 0; i < ARRAY_SIZE(futex_queues); i++) {
2650 plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock);
2651 spin_lock_init(&futex_queues[i].lock);
2652 }
2653
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 return 0;
2655}
Benjamin Herrenschmidtf6d107f2008-03-27 14:52:15 +11002656__initcall(futex_init);