blob: 4000454e4d8373f6128eb0735d0afd02ee583a6e [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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070022 * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly
23 * enough at me, Linus for the original (flawed) idea, Matthew
24 * Kirkwood for proof-of-concept implementation.
25 *
26 * "The futexes are also cursed."
27 * "But they come in a choice of three flavours!"
28 *
29 * This program is free software; you can redistribute it and/or modify
30 * it under the terms of the GNU General Public License as published by
31 * the Free Software Foundation; either version 2 of the License, or
32 * (at your option) any later version.
33 *
34 * This program is distributed in the hope that it will be useful,
35 * but WITHOUT ANY WARRANTY; without even the implied warranty of
36 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
37 * GNU General Public License for more details.
38 *
39 * You should have received a copy of the GNU General Public License
40 * along with this program; if not, write to the Free Software
41 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
42 */
43#include <linux/slab.h>
44#include <linux/poll.h>
45#include <linux/fs.h>
46#include <linux/file.h>
47#include <linux/jhash.h>
48#include <linux/init.h>
49#include <linux/futex.h>
50#include <linux/mount.h>
51#include <linux/pagemap.h>
52#include <linux/syscalls.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070053#include <linux/signal.h>
Rusty Russell9adef582007-05-08 00:26:42 -070054#include <linux/module.h>
Andrey Mirkinfd5eea42007-10-16 23:30:13 -070055#include <linux/magic.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070056#include <linux/pid.h>
57#include <linux/nsproxy.h>
58
Jakub Jelinek4732efb2005-09-06 15:16:25 -070059#include <asm/futex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Ingo Molnarc87e2832006-06-27 02:54:58 -070061#include "rtmutex_common.h"
62
Thomas Gleixnera0c1e902008-02-23 15:23:57 -080063int __read_mostly futex_cmpxchg_enabled;
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065#define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
66
67/*
Ingo Molnarc87e2832006-06-27 02:54:58 -070068 * Priority Inheritance state:
69 */
70struct futex_pi_state {
71 /*
72 * list of 'owned' pi_state instances - these have to be
73 * cleaned up in do_exit() if the task exits prematurely:
74 */
75 struct list_head list;
76
77 /*
78 * The PI object:
79 */
80 struct rt_mutex pi_mutex;
81
82 struct task_struct *owner;
83 atomic_t refcount;
84
85 union futex_key key;
86};
87
88/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 * We use this hashed waitqueue instead of a normal wait_queue_t, so
90 * we can wake only the relevant ones (hashed queues may be shared).
91 *
92 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
Pierre Peifferec92d082007-05-09 02:35:00 -070093 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 * The order of wakup is always to make the first condition true, then
Darren Hart73500ac2008-12-17 17:29:56 -080095 * wake up q->waiter, then make the second condition true.
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 */
97struct futex_q {
Pierre Peifferec92d082007-05-09 02:35:00 -070098 struct plist_node list;
Darren Hart73500ac2008-12-17 17:29:56 -080099 /* There can only be a single waiter */
100 wait_queue_head_t waiter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Ingo Molnare2970f22006-06-27 02:54:47 -0700102 /* Which hash list lock to use: */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 spinlock_t *lock_ptr;
104
Ingo Molnare2970f22006-06-27 02:54:47 -0700105 /* Key which the futex is hashed on: */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 union futex_key key;
107
Ingo Molnarc87e2832006-06-27 02:54:58 -0700108 /* Optional priority inheritance state: */
109 struct futex_pi_state *pi_state;
110 struct task_struct *task;
Thomas Gleixnercd689982008-02-01 17:45:14 +0100111
112 /* Bitset for the optional bitmasked wakeup */
113 u32 bitset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114};
115
116/*
Darren Hartb2d09942009-03-12 00:55:37 -0700117 * Hash buckets are shared by all the futex_keys that hash to the same
118 * location. Each key may have multiple futex_q structures, one for each task
119 * waiting on a futex.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 */
121struct futex_hash_bucket {
Pierre Peifferec92d082007-05-09 02:35:00 -0700122 spinlock_t lock;
123 struct plist_head chain;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124};
125
126static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS];
127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128/*
129 * We hash on the keys returned from get_futex_key (see below).
130 */
131static struct futex_hash_bucket *hash_futex(union futex_key *key)
132{
133 u32 hash = jhash2((u32*)&key->both.word,
134 (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
135 key->both.offset);
136 return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)];
137}
138
139/*
140 * Return 1 if two futex_keys are equal, 0 otherwise.
141 */
142static inline int match_futex(union futex_key *key1, union futex_key *key2)
143{
144 return (key1->both.word == key2->both.word
145 && key1->both.ptr == key2->both.ptr
146 && key1->both.offset == key2->both.offset);
147}
148
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200149/*
150 * Take a reference to the resource addressed by a key.
151 * Can be called while holding spinlocks.
152 *
153 */
154static void get_futex_key_refs(union futex_key *key)
155{
156 if (!key->both.ptr)
157 return;
158
159 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
160 case FUT_OFF_INODE:
161 atomic_inc(&key->shared.inode->i_count);
162 break;
163 case FUT_OFF_MMSHARED:
164 atomic_inc(&key->private.mm->mm_count);
165 break;
166 }
167}
168
169/*
170 * Drop a reference to the resource addressed by a key.
171 * The hash bucket spinlock must not be held.
172 */
173static void drop_futex_key_refs(union futex_key *key)
174{
Darren Hart90621c42008-12-29 19:43:21 -0800175 if (!key->both.ptr) {
176 /* If we're here then we tried to put a key we failed to get */
177 WARN_ON_ONCE(1);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200178 return;
Darren Hart90621c42008-12-29 19:43:21 -0800179 }
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200180
181 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
182 case FUT_OFF_INODE:
183 iput(key->shared.inode);
184 break;
185 case FUT_OFF_MMSHARED:
186 mmdrop(key->private.mm);
187 break;
188 }
189}
190
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700191/**
192 * get_futex_key - Get parameters which are the keys for a futex.
193 * @uaddr: virtual address of the futex
Darren Hartb2d09942009-03-12 00:55:37 -0700194 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700195 * @key: address where result is stored.
196 *
197 * Returns a negative error code or 0
198 * The key words are stored in *key on success.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 *
Josef "Jeff" Sipekf3a43f32006-12-08 02:36:43 -0800200 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 * offset_within_page). For private mappings, it's (uaddr, current->mm).
202 * We can usually work out the index without swapping in the page.
203 *
Darren Hartb2d09942009-03-12 00:55:37 -0700204 * lock_page() might sleep, the caller should not hold a spinlock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 */
Peter Zijlstrac2f9f202008-09-26 19:32:23 +0200206static int get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207{
Ingo Molnare2970f22006-06-27 02:54:47 -0700208 unsigned long address = (unsigned long)uaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 struct page *page;
211 int err;
212
213 /*
214 * The futex address must be "naturally" aligned.
215 */
Ingo Molnare2970f22006-06-27 02:54:47 -0700216 key->both.offset = address % PAGE_SIZE;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700217 if (unlikely((address % sizeof(u32)) != 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 return -EINVAL;
Ingo Molnare2970f22006-06-27 02:54:47 -0700219 address -= key->both.offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
221 /*
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700222 * PROCESS_PRIVATE futexes are fast.
223 * As the mm cannot disappear under us and the 'key' only needs
224 * virtual address, we dont even have to find the underlying vma.
225 * Note : We do have to check 'uaddr' is a valid user address,
226 * but access_ok() should be faster than find_vma()
227 */
228 if (!fshared) {
229 if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))))
230 return -EFAULT;
231 key->private.mm = mm;
232 key->private.address = address;
Peter Zijlstra42569c32008-09-30 12:33:07 +0200233 get_futex_key_refs(key);
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700234 return 0;
235 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200237again:
Peter Zijlstra734b05b2008-09-26 19:32:22 +0200238 err = get_user_pages_fast(address, 1, 0, &page);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200239 if (err < 0)
240 return err;
241
242 lock_page(page);
243 if (!page->mapping) {
244 unlock_page(page);
245 put_page(page);
246 goto again;
247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
249 /*
250 * Private mappings are handled in a simple way.
251 *
252 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
253 * it's a read-only handle, it's expected that futexes attach to
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200254 * the object not the particular process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 */
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200256 if (PageAnon(page)) {
257 key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 key->private.mm = mm;
Ingo Molnare2970f22006-06-27 02:54:47 -0700259 key->private.address = address;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200260 } else {
261 key->both.offset |= FUT_OFF_INODE; /* inode-based key */
262 key->shared.inode = page->mapping->host;
263 key->shared.pgoff = page->index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 }
265
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200266 get_futex_key_refs(key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200268 unlock_page(page);
269 put_page(page);
270 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271}
272
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200273static inline
Peter Zijlstrac2f9f202008-09-26 19:32:23 +0200274void put_futex_key(int fshared, union futex_key *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200276 drop_futex_key_refs(key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277}
278
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700279static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval)
280{
281 u32 curval;
282
283 pagefault_disable();
284 curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval);
285 pagefault_enable();
286
287 return curval;
288}
289
290static int get_futex_value_locked(u32 *dest, u32 __user *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291{
292 int ret;
293
Peter Zijlstraa8663742006-12-06 20:32:20 -0800294 pagefault_disable();
Ingo Molnare2970f22006-06-27 02:54:47 -0700295 ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
Peter Zijlstraa8663742006-12-06 20:32:20 -0800296 pagefault_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
298 return ret ? -EFAULT : 0;
299}
300
301/*
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700302 * Fault handling.
Ingo Molnarc87e2832006-06-27 02:54:58 -0700303 */
Peter Zijlstrac2f9f202008-09-26 19:32:23 +0200304static int futex_handle_fault(unsigned long address, int attempt)
Ingo Molnarc87e2832006-06-27 02:54:58 -0700305{
306 struct vm_area_struct * vma;
307 struct mm_struct *mm = current->mm;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700308 int ret = -EFAULT;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700309
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700310 if (attempt > 2)
311 return ret;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700312
Peter Zijlstra61270702008-09-26 19:32:21 +0200313 down_read(&mm->mmap_sem);
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700314 vma = find_vma(mm, address);
315 if (vma && address >= vma->vm_start &&
316 (vma->vm_flags & VM_WRITE)) {
Nick Piggin83c54072007-07-19 01:47:05 -0700317 int fault;
318 fault = handle_mm_fault(mm, vma, address, 1);
319 if (unlikely((fault & VM_FAULT_ERROR))) {
320#if 0
321 /* XXX: let's do this when we verify it is OK */
322 if (ret & VM_FAULT_OOM)
323 ret = -ENOMEM;
324#endif
325 } else {
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700326 ret = 0;
Nick Piggin83c54072007-07-19 01:47:05 -0700327 if (fault & VM_FAULT_MAJOR)
328 current->maj_flt++;
329 else
330 current->min_flt++;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700331 }
Ingo Molnarc87e2832006-06-27 02:54:58 -0700332 }
Peter Zijlstra61270702008-09-26 19:32:21 +0200333 up_read(&mm->mmap_sem);
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700334 return ret;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700335}
336
337/*
338 * PI code:
339 */
340static int refill_pi_state_cache(void)
341{
342 struct futex_pi_state *pi_state;
343
344 if (likely(current->pi_state_cache))
345 return 0;
346
Burman Yan4668edc2006-12-06 20:38:51 -0800347 pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700348
349 if (!pi_state)
350 return -ENOMEM;
351
Ingo Molnarc87e2832006-06-27 02:54:58 -0700352 INIT_LIST_HEAD(&pi_state->list);
353 /* pi_mutex gets initialized later */
354 pi_state->owner = NULL;
355 atomic_set(&pi_state->refcount, 1);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200356 pi_state->key = FUTEX_KEY_INIT;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700357
358 current->pi_state_cache = pi_state;
359
360 return 0;
361}
362
363static struct futex_pi_state * alloc_pi_state(void)
364{
365 struct futex_pi_state *pi_state = current->pi_state_cache;
366
367 WARN_ON(!pi_state);
368 current->pi_state_cache = NULL;
369
370 return pi_state;
371}
372
373static void free_pi_state(struct futex_pi_state *pi_state)
374{
375 if (!atomic_dec_and_test(&pi_state->refcount))
376 return;
377
378 /*
379 * If pi_state->owner is NULL, the owner is most probably dying
380 * and has cleaned up the pi_state already
381 */
382 if (pi_state->owner) {
383 spin_lock_irq(&pi_state->owner->pi_lock);
384 list_del_init(&pi_state->list);
385 spin_unlock_irq(&pi_state->owner->pi_lock);
386
387 rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
388 }
389
390 if (current->pi_state_cache)
391 kfree(pi_state);
392 else {
393 /*
394 * pi_state->list is already empty.
395 * clear pi_state->owner.
396 * refcount is at 0 - put it back to 1.
397 */
398 pi_state->owner = NULL;
399 atomic_set(&pi_state->refcount, 1);
400 current->pi_state_cache = pi_state;
401 }
402}
403
404/*
405 * Look up the task based on what TID userspace gave us.
406 * We dont trust it.
407 */
408static struct task_struct * futex_find_get_task(pid_t pid)
409{
410 struct task_struct *p;
David Howellsc69e8d92008-11-14 10:39:19 +1100411 const struct cred *cred = current_cred(), *pcred;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700412
Oleg Nesterovd359b542006-09-29 02:00:55 -0700413 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700414 p = find_task_by_vpid(pid);
David Howellsc69e8d92008-11-14 10:39:19 +1100415 if (!p) {
Thomas Gleixnera06381f2007-06-23 11:48:40 +0200416 p = ERR_PTR(-ESRCH);
David Howellsc69e8d92008-11-14 10:39:19 +1100417 } else {
418 pcred = __task_cred(p);
419 if (cred->euid != pcred->euid &&
420 cred->euid != pcred->uid)
421 p = ERR_PTR(-ESRCH);
422 else
423 get_task_struct(p);
424 }
Thomas Gleixnera06381f2007-06-23 11:48:40 +0200425
Oleg Nesterovd359b542006-09-29 02:00:55 -0700426 rcu_read_unlock();
Ingo Molnarc87e2832006-06-27 02:54:58 -0700427
428 return p;
429}
430
431/*
432 * This task is holding PI mutexes at exit time => bad.
433 * Kernel cleans up PI-state, but userspace is likely hosed.
434 * (Robust-futex cleanup is separate and might save the day for userspace.)
435 */
436void exit_pi_state_list(struct task_struct *curr)
437{
Ingo Molnarc87e2832006-06-27 02:54:58 -0700438 struct list_head *next, *head = &curr->pi_state_list;
439 struct futex_pi_state *pi_state;
Ingo Molnar627371d2006-07-29 05:16:20 +0200440 struct futex_hash_bucket *hb;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200441 union futex_key key = FUTEX_KEY_INIT;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700442
Thomas Gleixnera0c1e902008-02-23 15:23:57 -0800443 if (!futex_cmpxchg_enabled)
444 return;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700445 /*
446 * We are a ZOMBIE and nobody can enqueue itself on
447 * pi_state_list anymore, but we have to be careful
Ingo Molnar627371d2006-07-29 05:16:20 +0200448 * versus waiters unqueueing themselves:
Ingo Molnarc87e2832006-06-27 02:54:58 -0700449 */
450 spin_lock_irq(&curr->pi_lock);
451 while (!list_empty(head)) {
452
453 next = head->next;
454 pi_state = list_entry(next, struct futex_pi_state, list);
455 key = pi_state->key;
Ingo Molnar627371d2006-07-29 05:16:20 +0200456 hb = hash_futex(&key);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700457 spin_unlock_irq(&curr->pi_lock);
458
Ingo Molnarc87e2832006-06-27 02:54:58 -0700459 spin_lock(&hb->lock);
460
461 spin_lock_irq(&curr->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200462 /*
463 * We dropped the pi-lock, so re-check whether this
464 * task still owns the PI-state:
465 */
Ingo Molnarc87e2832006-06-27 02:54:58 -0700466 if (head->next != next) {
467 spin_unlock(&hb->lock);
468 continue;
469 }
470
Ingo Molnarc87e2832006-06-27 02:54:58 -0700471 WARN_ON(pi_state->owner != curr);
Ingo Molnar627371d2006-07-29 05:16:20 +0200472 WARN_ON(list_empty(&pi_state->list));
473 list_del_init(&pi_state->list);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700474 pi_state->owner = NULL;
475 spin_unlock_irq(&curr->pi_lock);
476
477 rt_mutex_unlock(&pi_state->pi_mutex);
478
479 spin_unlock(&hb->lock);
480
481 spin_lock_irq(&curr->pi_lock);
482 }
483 spin_unlock_irq(&curr->pi_lock);
484}
485
486static int
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700487lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
488 union futex_key *key, struct futex_pi_state **ps)
Ingo Molnarc87e2832006-06-27 02:54:58 -0700489{
490 struct futex_pi_state *pi_state = NULL;
491 struct futex_q *this, *next;
Pierre Peifferec92d082007-05-09 02:35:00 -0700492 struct plist_head *head;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700493 struct task_struct *p;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700494 pid_t pid = uval & FUTEX_TID_MASK;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700495
496 head = &hb->chain;
497
Pierre Peifferec92d082007-05-09 02:35:00 -0700498 plist_for_each_entry_safe(this, next, head, list) {
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700499 if (match_futex(&this->key, key)) {
Ingo Molnarc87e2832006-06-27 02:54:58 -0700500 /*
501 * Another waiter already exists - bump up
502 * the refcount and return its pi_state:
503 */
504 pi_state = this->pi_state;
Thomas Gleixner06a9ec22006-07-10 04:44:30 -0700505 /*
506 * Userspace might have messed up non PI and PI futexes
507 */
508 if (unlikely(!pi_state))
509 return -EINVAL;
510
Ingo Molnar627371d2006-07-29 05:16:20 +0200511 WARN_ON(!atomic_read(&pi_state->refcount));
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700512 WARN_ON(pid && pi_state->owner &&
513 pi_state->owner->pid != pid);
Ingo Molnar627371d2006-07-29 05:16:20 +0200514
Ingo Molnarc87e2832006-06-27 02:54:58 -0700515 atomic_inc(&pi_state->refcount);
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700516 *ps = pi_state;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700517
518 return 0;
519 }
520 }
521
522 /*
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200523 * We are the first waiter - try to look up the real owner and attach
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700524 * the new pi_state to it, but bail out when TID = 0
Ingo Molnarc87e2832006-06-27 02:54:58 -0700525 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700526 if (!pid)
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200527 return -ESRCH;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700528 p = futex_find_get_task(pid);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700529 if (IS_ERR(p))
530 return PTR_ERR(p);
531
532 /*
533 * We need to look at the task state flags to figure out,
534 * whether the task is exiting. To protect against the do_exit
535 * change of the task flags, we do this protected by
536 * p->pi_lock:
537 */
538 spin_lock_irq(&p->pi_lock);
539 if (unlikely(p->flags & PF_EXITING)) {
540 /*
541 * The task is on the way out. When PF_EXITPIDONE is
542 * set, we know that the task has finished the
543 * cleanup:
544 */
545 int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
546
547 spin_unlock_irq(&p->pi_lock);
548 put_task_struct(p);
549 return ret;
550 }
Ingo Molnarc87e2832006-06-27 02:54:58 -0700551
552 pi_state = alloc_pi_state();
553
554 /*
555 * Initialize the pi_mutex in locked state and make 'p'
556 * the owner of it:
557 */
558 rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
559
560 /* Store the key for possible exit cleanups: */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700561 pi_state->key = *key;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700562
Ingo Molnar627371d2006-07-29 05:16:20 +0200563 WARN_ON(!list_empty(&pi_state->list));
Ingo Molnarc87e2832006-06-27 02:54:58 -0700564 list_add(&pi_state->list, &p->pi_state_list);
565 pi_state->owner = p;
566 spin_unlock_irq(&p->pi_lock);
567
568 put_task_struct(p);
569
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700570 *ps = pi_state;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700571
572 return 0;
573}
574
575/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 * The hash bucket lock must be held when this is called.
577 * Afterwards, the futex_q must not be accessed.
578 */
579static void wake_futex(struct futex_q *q)
580{
Pierre Peifferec92d082007-05-09 02:35:00 -0700581 plist_del(&q->list, &q->list.plist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 /*
583 * The lock in wake_up_all() is a crucial memory barrier after the
Pierre Peifferec92d082007-05-09 02:35:00 -0700584 * plist_del() and also before assigning to q->lock_ptr.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 */
Darren Hart73500ac2008-12-17 17:29:56 -0800586 wake_up(&q->waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 /*
588 * The waiting task can free the futex_q as soon as this is written,
589 * without taking any locks. This must come last.
Andrew Morton8e311082005-12-23 19:54:46 -0800590 *
Darren Hartb2d09942009-03-12 00:55:37 -0700591 * A memory barrier is required here to prevent the following store to
592 * lock_ptr from getting ahead of the wakeup. Clearing the lock at the
593 * end of wake_up() does not prevent this store from moving.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 */
Ralf Baechleccdea2f2006-12-06 20:40:26 -0800595 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 q->lock_ptr = NULL;
597}
598
Ingo Molnarc87e2832006-06-27 02:54:58 -0700599static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
600{
601 struct task_struct *new_owner;
602 struct futex_pi_state *pi_state = this->pi_state;
603 u32 curval, newval;
604
605 if (!pi_state)
606 return -EINVAL;
607
Ingo Molnar217788672007-03-16 13:38:31 -0800608 spin_lock(&pi_state->pi_mutex.wait_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700609 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
610
611 /*
612 * This happens when we have stolen the lock and the original
613 * pending owner did not enqueue itself back on the rt_mutex.
614 * Thats not a tragedy. We know that way, that a lock waiter
615 * is on the fly. We make the futex_q waiter the pending owner.
616 */
617 if (!new_owner)
618 new_owner = this->task;
619
620 /*
621 * We pass it to the next owner. (The WAITERS bit is always
622 * kept enabled while there is PI state around. We must also
623 * preserve the owner died bit.)
624 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200625 if (!(uval & FUTEX_OWNER_DIED)) {
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700626 int ret = 0;
627
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700628 newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700629
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700630 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700631
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200632 if (curval == -EFAULT)
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700633 ret = -EFAULT;
Thomas Gleixnercde898f2007-12-05 15:46:09 +0100634 else if (curval != uval)
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700635 ret = -EINVAL;
636 if (ret) {
637 spin_unlock(&pi_state->pi_mutex.wait_lock);
638 return ret;
639 }
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200640 }
Ingo Molnarc87e2832006-06-27 02:54:58 -0700641
Ingo Molnar627371d2006-07-29 05:16:20 +0200642 spin_lock_irq(&pi_state->owner->pi_lock);
643 WARN_ON(list_empty(&pi_state->list));
644 list_del_init(&pi_state->list);
645 spin_unlock_irq(&pi_state->owner->pi_lock);
646
647 spin_lock_irq(&new_owner->pi_lock);
648 WARN_ON(!list_empty(&pi_state->list));
Ingo Molnarc87e2832006-06-27 02:54:58 -0700649 list_add(&pi_state->list, &new_owner->pi_state_list);
650 pi_state->owner = new_owner;
Ingo Molnar627371d2006-07-29 05:16:20 +0200651 spin_unlock_irq(&new_owner->pi_lock);
652
Ingo Molnar217788672007-03-16 13:38:31 -0800653 spin_unlock(&pi_state->pi_mutex.wait_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700654 rt_mutex_unlock(&pi_state->pi_mutex);
655
656 return 0;
657}
658
659static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
660{
661 u32 oldval;
662
663 /*
664 * There is no waiter, so we unlock the futex. The owner died
665 * bit has not to be preserved here. We are the owner:
666 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700667 oldval = cmpxchg_futex_value_locked(uaddr, uval, 0);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700668
669 if (oldval == -EFAULT)
670 return oldval;
671 if (oldval != uval)
672 return -EAGAIN;
673
674 return 0;
675}
676
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677/*
Ingo Molnar8b8f3192006-07-03 00:25:05 -0700678 * Express the locking dependencies for lockdep:
679 */
680static inline void
681double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
682{
683 if (hb1 <= hb2) {
684 spin_lock(&hb1->lock);
685 if (hb1 < hb2)
686 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
687 } else { /* hb1 > hb2 */
688 spin_lock(&hb2->lock);
689 spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
690 }
691}
692
693/*
Darren Hartb2d09942009-03-12 00:55:37 -0700694 * Wake up waiters matching bitset queued on this futex (uaddr).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 */
Peter Zijlstrac2f9f202008-09-26 19:32:23 +0200696static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697{
Ingo Molnare2970f22006-06-27 02:54:47 -0700698 struct futex_hash_bucket *hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 struct futex_q *this, *next;
Pierre Peifferec92d082007-05-09 02:35:00 -0700700 struct plist_head *head;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200701 union futex_key key = FUTEX_KEY_INIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 int ret;
703
Thomas Gleixnercd689982008-02-01 17:45:14 +0100704 if (!bitset)
705 return -EINVAL;
706
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700707 ret = get_futex_key(uaddr, fshared, &key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 if (unlikely(ret != 0))
709 goto out;
710
Ingo Molnare2970f22006-06-27 02:54:47 -0700711 hb = hash_futex(&key);
712 spin_lock(&hb->lock);
713 head = &hb->chain;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
Pierre Peifferec92d082007-05-09 02:35:00 -0700715 plist_for_each_entry_safe(this, next, head, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 if (match_futex (&this->key, &key)) {
Ingo Molnared6f7b12006-07-01 04:35:46 -0700717 if (this->pi_state) {
718 ret = -EINVAL;
719 break;
720 }
Thomas Gleixnercd689982008-02-01 17:45:14 +0100721
722 /* Check if one of the bits is set in both bitsets */
723 if (!(this->bitset & bitset))
724 continue;
725
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 wake_futex(this);
727 if (++ret >= nr_wake)
728 break;
729 }
730 }
731
Ingo Molnare2970f22006-06-27 02:54:47 -0700732 spin_unlock(&hb->lock);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200733 put_futex_key(fshared, &key);
Darren Hart42d35d42008-12-29 15:49:53 -0800734out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 return ret;
736}
737
738/*
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700739 * Wake up all waiters hashed on the physical page that is mapped
740 * to this virtual address:
741 */
Ingo Molnare2970f22006-06-27 02:54:47 -0700742static int
Peter Zijlstrac2f9f202008-09-26 19:32:23 +0200743futex_wake_op(u32 __user *uaddr1, int fshared, u32 __user *uaddr2,
Ingo Molnare2970f22006-06-27 02:54:47 -0700744 int nr_wake, int nr_wake2, int op)
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700745{
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200746 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
Ingo Molnare2970f22006-06-27 02:54:47 -0700747 struct futex_hash_bucket *hb1, *hb2;
Pierre Peifferec92d082007-05-09 02:35:00 -0700748 struct plist_head *head;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700749 struct futex_q *this, *next;
750 int ret, op_ret, attempt = 0;
751
752retryfull:
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700753 ret = get_futex_key(uaddr1, fshared, &key1);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700754 if (unlikely(ret != 0))
755 goto out;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700756 ret = get_futex_key(uaddr2, fshared, &key2);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700757 if (unlikely(ret != 0))
Darren Hart42d35d42008-12-29 15:49:53 -0800758 goto out_put_key1;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700759
Ingo Molnare2970f22006-06-27 02:54:47 -0700760 hb1 = hash_futex(&key1);
761 hb2 = hash_futex(&key2);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700762
763retry:
Ingo Molnar8b8f3192006-07-03 00:25:05 -0700764 double_lock_hb(hb1, hb2);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700765
Ingo Molnare2970f22006-06-27 02:54:47 -0700766 op_ret = futex_atomic_op_inuser(op, uaddr2);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700767 if (unlikely(op_ret < 0)) {
Ingo Molnare2970f22006-06-27 02:54:47 -0700768 u32 dummy;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700769
Ingo Molnare2970f22006-06-27 02:54:47 -0700770 spin_unlock(&hb1->lock);
771 if (hb1 != hb2)
772 spin_unlock(&hb2->lock);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700773
David Howells7ee1dd32006-01-06 00:11:44 -0800774#ifndef CONFIG_MMU
Ingo Molnare2970f22006-06-27 02:54:47 -0700775 /*
776 * we don't get EFAULT from MMU faults if we don't have an MMU,
777 * but we might get them from range checking
778 */
David Howells7ee1dd32006-01-06 00:11:44 -0800779 ret = op_ret;
Darren Hart42d35d42008-12-29 15:49:53 -0800780 goto out_put_keys;
David Howells7ee1dd32006-01-06 00:11:44 -0800781#endif
782
David Gibson796f8d92005-11-07 00:59:33 -0800783 if (unlikely(op_ret != -EFAULT)) {
784 ret = op_ret;
Darren Hart42d35d42008-12-29 15:49:53 -0800785 goto out_put_keys;
David Gibson796f8d92005-11-07 00:59:33 -0800786 }
787
Ingo Molnare2970f22006-06-27 02:54:47 -0700788 /*
789 * futex_atomic_op_inuser needs to both read and write
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700790 * *(int __user *)uaddr2, but we can't modify it
791 * non-atomically. Therefore, if get_user below is not
792 * enough, we need to handle the fault ourselves, while
Ingo Molnare2970f22006-06-27 02:54:47 -0700793 * still holding the mmap_sem.
794 */
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700795 if (attempt++) {
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700796 ret = futex_handle_fault((unsigned long)uaddr2,
Peter Zijlstrac2f9f202008-09-26 19:32:23 +0200797 attempt);
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700798 if (ret)
Darren Hart42d35d42008-12-29 15:49:53 -0800799 goto out_put_keys;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700800 goto retry;
801 }
802
Ingo Molnare2970f22006-06-27 02:54:47 -0700803 ret = get_user(dummy, uaddr2);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700804 if (ret)
Darren Hartde87fcc2009-03-12 00:55:46 -0700805 goto out_put_keys;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700806
Darren Hartde87fcc2009-03-12 00:55:46 -0700807 put_futex_key(fshared, &key2);
808 put_futex_key(fshared, &key1);
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700809 goto retryfull;
810 }
811
Ingo Molnare2970f22006-06-27 02:54:47 -0700812 head = &hb1->chain;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700813
Pierre Peifferec92d082007-05-09 02:35:00 -0700814 plist_for_each_entry_safe(this, next, head, list) {
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700815 if (match_futex (&this->key, &key1)) {
816 wake_futex(this);
817 if (++ret >= nr_wake)
818 break;
819 }
820 }
821
822 if (op_ret > 0) {
Ingo Molnare2970f22006-06-27 02:54:47 -0700823 head = &hb2->chain;
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700824
825 op_ret = 0;
Pierre Peifferec92d082007-05-09 02:35:00 -0700826 plist_for_each_entry_safe(this, next, head, list) {
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700827 if (match_futex (&this->key, &key2)) {
828 wake_futex(this);
829 if (++op_ret >= nr_wake2)
830 break;
831 }
832 }
833 ret += op_ret;
834 }
835
Ingo Molnare2970f22006-06-27 02:54:47 -0700836 spin_unlock(&hb1->lock);
837 if (hb1 != hb2)
838 spin_unlock(&hb2->lock);
Darren Hart42d35d42008-12-29 15:49:53 -0800839out_put_keys:
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200840 put_futex_key(fshared, &key2);
Darren Hart42d35d42008-12-29 15:49:53 -0800841out_put_key1:
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200842 put_futex_key(fshared, &key1);
Darren Hart42d35d42008-12-29 15:49:53 -0800843out:
Jakub Jelinek4732efb2005-09-06 15:16:25 -0700844 return ret;
845}
846
847/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 * Requeue all waiters hashed on one physical page to another
849 * physical page.
850 */
Peter Zijlstrac2f9f202008-09-26 19:32:23 +0200851static int futex_requeue(u32 __user *uaddr1, int fshared, u32 __user *uaddr2,
Ingo Molnare2970f22006-06-27 02:54:47 -0700852 int nr_wake, int nr_requeue, u32 *cmpval)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853{
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200854 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
Ingo Molnare2970f22006-06-27 02:54:47 -0700855 struct futex_hash_bucket *hb1, *hb2;
Pierre Peifferec92d082007-05-09 02:35:00 -0700856 struct plist_head *head1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 struct futex_q *this, *next;
858 int ret, drop_count = 0;
859
Darren Hart42d35d42008-12-29 15:49:53 -0800860retry:
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700861 ret = get_futex_key(uaddr1, fshared, &key1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 if (unlikely(ret != 0))
863 goto out;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700864 ret = get_futex_key(uaddr2, fshared, &key2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 if (unlikely(ret != 0))
Darren Hart42d35d42008-12-29 15:49:53 -0800866 goto out_put_key1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
Ingo Molnare2970f22006-06-27 02:54:47 -0700868 hb1 = hash_futex(&key1);
869 hb2 = hash_futex(&key2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
Ingo Molnar8b8f3192006-07-03 00:25:05 -0700871 double_lock_hb(hb1, hb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
Ingo Molnare2970f22006-06-27 02:54:47 -0700873 if (likely(cmpval != NULL)) {
874 u32 curval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
Ingo Molnare2970f22006-06-27 02:54:47 -0700876 ret = get_futex_value_locked(&curval, uaddr1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
878 if (unlikely(ret)) {
Ingo Molnare2970f22006-06-27 02:54:47 -0700879 spin_unlock(&hb1->lock);
880 if (hb1 != hb2)
881 spin_unlock(&hb2->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
Darren Hartde87fcc2009-03-12 00:55:46 -0700883 put_futex_key(fshared, &key2);
884 put_futex_key(fshared, &key1);
885
Ingo Molnare2970f22006-06-27 02:54:47 -0700886 ret = get_user(curval, uaddr1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
888 if (!ret)
889 goto retry;
890
Darren Hart42d35d42008-12-29 15:49:53 -0800891 goto out_put_keys;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 }
Ingo Molnare2970f22006-06-27 02:54:47 -0700893 if (curval != *cmpval) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 ret = -EAGAIN;
895 goto out_unlock;
896 }
897 }
898
Ingo Molnare2970f22006-06-27 02:54:47 -0700899 head1 = &hb1->chain;
Pierre Peifferec92d082007-05-09 02:35:00 -0700900 plist_for_each_entry_safe(this, next, head1, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 if (!match_futex (&this->key, &key1))
902 continue;
903 if (++ret <= nr_wake) {
904 wake_futex(this);
905 } else {
Sebastien Dugue59e0e0a2006-06-27 02:55:03 -0700906 /*
907 * If key1 and key2 hash to the same bucket, no need to
908 * requeue.
909 */
910 if (likely(head1 != &hb2->chain)) {
Pierre Peifferec92d082007-05-09 02:35:00 -0700911 plist_del(&this->list, &hb1->chain);
912 plist_add(&this->list, &hb2->chain);
Sebastien Dugue59e0e0a2006-06-27 02:55:03 -0700913 this->lock_ptr = &hb2->lock;
Pierre Peifferec92d082007-05-09 02:35:00 -0700914#ifdef CONFIG_DEBUG_PI_LIST
915 this->list.plist.lock = &hb2->lock;
916#endif
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700917 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 this->key = key2;
Rusty Russell9adef582007-05-08 00:26:42 -0700919 get_futex_key_refs(&key2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 drop_count++;
921
922 if (ret - nr_wake >= nr_requeue)
923 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 }
925 }
926
927out_unlock:
Ingo Molnare2970f22006-06-27 02:54:47 -0700928 spin_unlock(&hb1->lock);
929 if (hb1 != hb2)
930 spin_unlock(&hb2->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
Rusty Russell9adef582007-05-08 00:26:42 -0700932 /* drop_futex_key_refs() must be called outside the spinlocks. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 while (--drop_count >= 0)
Rusty Russell9adef582007-05-08 00:26:42 -0700934 drop_futex_key_refs(&key1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
Darren Hart42d35d42008-12-29 15:49:53 -0800936out_put_keys:
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200937 put_futex_key(fshared, &key2);
Darren Hart42d35d42008-12-29 15:49:53 -0800938out_put_key1:
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200939 put_futex_key(fshared, &key1);
Darren Hart42d35d42008-12-29 15:49:53 -0800940out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 return ret;
942}
943
944/* The key must be already stored in q->key. */
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +0100945static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946{
Ingo Molnare2970f22006-06-27 02:54:47 -0700947 struct futex_hash_bucket *hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
Darren Hart73500ac2008-12-17 17:29:56 -0800949 init_waitqueue_head(&q->waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
Rusty Russell9adef582007-05-08 00:26:42 -0700951 get_futex_key_refs(&q->key);
Ingo Molnare2970f22006-06-27 02:54:47 -0700952 hb = hash_futex(&q->key);
953 q->lock_ptr = &hb->lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
Ingo Molnare2970f22006-06-27 02:54:47 -0700955 spin_lock(&hb->lock);
956 return hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957}
958
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +0100959static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960{
Pierre Peifferec92d082007-05-09 02:35:00 -0700961 int prio;
962
963 /*
964 * The priority used to register this element is
965 * - either the real thread-priority for the real-time threads
966 * (i.e. threads with a priority lower than MAX_RT_PRIO)
967 * - or MAX_RT_PRIO for non-RT threads.
968 * Thus, all RT-threads are woken first in priority order, and
969 * the others are woken last, in FIFO order.
970 */
971 prio = min(current->normal_prio, MAX_RT_PRIO);
972
973 plist_node_init(&q->list, prio);
974#ifdef CONFIG_DEBUG_PI_LIST
975 q->list.plist.lock = &hb->lock;
976#endif
977 plist_add(&q->list, &hb->chain);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700978 q->task = current;
Ingo Molnare2970f22006-06-27 02:54:47 -0700979 spin_unlock(&hb->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980}
981
982static inline void
Ingo Molnare2970f22006-06-27 02:54:47 -0700983queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984{
Ingo Molnare2970f22006-06-27 02:54:47 -0700985 spin_unlock(&hb->lock);
Rusty Russell9adef582007-05-08 00:26:42 -0700986 drop_futex_key_refs(&q->key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987}
988
989/*
990 * queue_me and unqueue_me must be called as a pair, each
991 * exactly once. They are called with the hashed spinlock held.
992 */
993
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994/* Return 1 if we were still queued (ie. 0 means we were woken) */
995static int unqueue_me(struct futex_q *q)
996{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 spinlock_t *lock_ptr;
Ingo Molnare2970f22006-06-27 02:54:47 -0700998 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999
1000 /* In the common case we don't take the spinlock, which is nice. */
Darren Hart42d35d42008-12-29 15:49:53 -08001001retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 lock_ptr = q->lock_ptr;
Christian Borntraegere91467e2006-08-05 12:13:52 -07001003 barrier();
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001004 if (lock_ptr != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 spin_lock(lock_ptr);
1006 /*
1007 * q->lock_ptr can change between reading it and
1008 * spin_lock(), causing us to take the wrong lock. This
1009 * corrects the race condition.
1010 *
1011 * Reasoning goes like this: if we have the wrong lock,
1012 * q->lock_ptr must have changed (maybe several times)
1013 * between reading it and the spin_lock(). It can
1014 * change again after the spin_lock() but only if it was
1015 * already changed before the spin_lock(). It cannot,
1016 * however, change back to the original value. Therefore
1017 * we can detect whether we acquired the correct lock.
1018 */
1019 if (unlikely(lock_ptr != q->lock_ptr)) {
1020 spin_unlock(lock_ptr);
1021 goto retry;
1022 }
Pierre Peifferec92d082007-05-09 02:35:00 -07001023 WARN_ON(plist_node_empty(&q->list));
1024 plist_del(&q->list, &q->list.plist);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001025
1026 BUG_ON(q->pi_state);
1027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 spin_unlock(lock_ptr);
1029 ret = 1;
1030 }
1031
Rusty Russell9adef582007-05-08 00:26:42 -07001032 drop_futex_key_refs(&q->key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 return ret;
1034}
1035
Ingo Molnarc87e2832006-06-27 02:54:58 -07001036/*
1037 * PI futexes can not be requeued and must remove themself from the
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001038 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1039 * and dropped here.
Ingo Molnarc87e2832006-06-27 02:54:58 -07001040 */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001041static void unqueue_me_pi(struct futex_q *q)
Ingo Molnarc87e2832006-06-27 02:54:58 -07001042{
Pierre Peifferec92d082007-05-09 02:35:00 -07001043 WARN_ON(plist_node_empty(&q->list));
1044 plist_del(&q->list, &q->list.plist);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001045
1046 BUG_ON(!q->pi_state);
1047 free_pi_state(q->pi_state);
1048 q->pi_state = NULL;
1049
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001050 spin_unlock(q->lock_ptr);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001051
Rusty Russell9adef582007-05-08 00:26:42 -07001052 drop_futex_key_refs(&q->key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001053}
1054
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001055/*
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001056 * Fixup the pi_state owner with the new owner.
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001057 *
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001058 * Must be called with hash bucket lock held and mm->sem held for non
1059 * private futexes.
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001060 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001061static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001062 struct task_struct *newowner, int fshared)
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001063{
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001064 u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001065 struct futex_pi_state *pi_state = q->pi_state;
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001066 struct task_struct *oldowner = pi_state->owner;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001067 u32 uval, curval, newval;
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001068 int ret, attempt = 0;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001069
1070 /* Owner died? */
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001071 if (!pi_state->owner)
1072 newtid |= FUTEX_OWNER_DIED;
1073
1074 /*
1075 * We are here either because we stole the rtmutex from the
1076 * pending owner or we are the pending owner which failed to
1077 * get the rtmutex. We have to replace the pending owner TID
1078 * in the user space variable. This must be atomic as we have
1079 * to preserve the owner died bit here.
1080 *
Darren Hartb2d09942009-03-12 00:55:37 -07001081 * Note: We write the user space value _before_ changing the pi_state
1082 * because we can fault here. Imagine swapped out pages or a fork
1083 * that marked all the anonymous memory readonly for cow.
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001084 *
1085 * Modifying pi_state _before_ the user space value would
1086 * leave the pi_state in an inconsistent state when we fault
1087 * here, because we need to drop the hash bucket lock to
1088 * handle the fault. This might be observed in the PID check
1089 * in lookup_pi_state.
1090 */
1091retry:
1092 if (get_futex_value_locked(&uval, uaddr))
1093 goto handle_fault;
1094
1095 while (1) {
1096 newval = (uval & FUTEX_OWNER_DIED) | newtid;
1097
1098 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
1099
1100 if (curval == -EFAULT)
1101 goto handle_fault;
1102 if (curval == uval)
1103 break;
1104 uval = curval;
1105 }
1106
1107 /*
1108 * We fixed up user space. Now we need to fix the pi_state
1109 * itself.
1110 */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001111 if (pi_state->owner != NULL) {
1112 spin_lock_irq(&pi_state->owner->pi_lock);
1113 WARN_ON(list_empty(&pi_state->list));
1114 list_del_init(&pi_state->list);
1115 spin_unlock_irq(&pi_state->owner->pi_lock);
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001116 }
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001117
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001118 pi_state->owner = newowner;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001119
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001120 spin_lock_irq(&newowner->pi_lock);
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001121 WARN_ON(!list_empty(&pi_state->list));
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001122 list_add(&pi_state->list, &newowner->pi_state_list);
1123 spin_unlock_irq(&newowner->pi_lock);
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001124 return 0;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001125
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001126 /*
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001127 * To handle the page fault we need to drop the hash bucket
1128 * lock here. That gives the other task (either the pending
1129 * owner itself or the task which stole the rtmutex) the
1130 * chance to try the fixup of the pi_state. So once we are
1131 * back from handling the fault we need to check the pi_state
1132 * after reacquiring the hash bucket lock and before trying to
1133 * do another fixup. When the fixup has been done already we
1134 * simply return.
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001135 */
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001136handle_fault:
1137 spin_unlock(q->lock_ptr);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001138
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001139 ret = futex_handle_fault((unsigned long)uaddr, attempt++);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001140
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001141 spin_lock(q->lock_ptr);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001142
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001143 /*
1144 * Check if someone else fixed it for us:
1145 */
1146 if (pi_state->owner != oldowner)
1147 return 0;
1148
1149 if (ret)
1150 return ret;
1151
1152 goto retry;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001153}
1154
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001155/*
1156 * In case we must use restart_block to restart a futex_wait,
Steven Rostedtce6bd422007-12-05 15:46:09 +01001157 * we encode in the 'flags' shared capability
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001158 */
Thomas Gleixner1acdac12008-11-20 10:02:53 -08001159#define FLAGS_SHARED 0x01
1160#define FLAGS_CLOCKRT 0x02
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001161
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001162static long futex_wait_restart(struct restart_block *restart);
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001163
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001164static int futex_wait(u32 __user *uaddr, int fshared,
Thomas Gleixner1acdac12008-11-20 10:02:53 -08001165 u32 val, ktime_t *abs_time, u32 bitset, int clockrt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166{
Ingo Molnarc87e2832006-06-27 02:54:58 -07001167 struct task_struct *curr = current;
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001168 struct restart_block *restart;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001169 DECLARE_WAITQUEUE(wait, curr);
Ingo Molnare2970f22006-06-27 02:54:47 -07001170 struct futex_hash_bucket *hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 struct futex_q q;
Ingo Molnare2970f22006-06-27 02:54:47 -07001172 u32 uval;
1173 int ret;
Thomas Gleixnerbd197232007-06-17 21:11:10 +02001174 struct hrtimer_sleeper t;
Pierre Peifferc19384b2007-05-09 02:35:02 -07001175 int rem = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Thomas Gleixnercd689982008-02-01 17:45:14 +01001177 if (!bitset)
1178 return -EINVAL;
1179
Ingo Molnarc87e2832006-06-27 02:54:58 -07001180 q.pi_state = NULL;
Thomas Gleixnercd689982008-02-01 17:45:14 +01001181 q.bitset = bitset;
Darren Hart42d35d42008-12-29 15:49:53 -08001182retry:
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001183 q.key = FUTEX_KEY_INIT;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001184 ret = get_futex_key(uaddr, fshared, &q.key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 if (unlikely(ret != 0))
Darren Hart42d35d42008-12-29 15:49:53 -08001186 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01001188 hb = queue_lock(&q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
1190 /*
Darren Hartb2d09942009-03-12 00:55:37 -07001191 * Access the page AFTER the hash-bucket is locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 * Order is important:
1193 *
1194 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1195 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1196 *
1197 * The basic logical guarantee of a futex is that it blocks ONLY
1198 * if cond(var) is known to be true at the time of blocking, for
1199 * any cond. If we queued after testing *uaddr, that would open
1200 * a race condition where we could block indefinitely with
1201 * cond(var) false, which would violate the guarantee.
1202 *
1203 * A consequence is that futex_wait() can return zero and absorb
1204 * a wakeup when *uaddr != val on entry to the syscall. This is
1205 * rare, but normal.
1206 *
Darren Hartb2d09942009-03-12 00:55:37 -07001207 * For shared futexes, we hold the mmap semaphore, so the mapping
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001208 * cannot have changed since we looked it up in get_futex_key.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 */
Ingo Molnare2970f22006-06-27 02:54:47 -07001210 ret = get_futex_value_locked(&uval, uaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211
1212 if (unlikely(ret)) {
Ingo Molnare2970f22006-06-27 02:54:47 -07001213 queue_unlock(&q, hb);
Darren Hart42d35d42008-12-29 15:49:53 -08001214 put_futex_key(fshared, &q.key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215
Ingo Molnare2970f22006-06-27 02:54:47 -07001216 ret = get_user(uval, uaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217
1218 if (!ret)
1219 goto retry;
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001220 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001222 ret = -EWOULDBLOCK;
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001223 if (unlikely(uval != val)) {
1224 queue_unlock(&q, hb);
1225 goto out_put_key;
1226 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227
1228 /* Only actually queue if *uaddr contained val. */
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01001229 queue_me(&q, hb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
1231 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 * There might have been scheduling since the queue_me(), as we
1233 * cannot hold a spinlock across the get_user() in case it
1234 * faults, and we cannot just set TASK_INTERRUPTIBLE state when
1235 * queueing ourselves into the futex hash. This code thus has to
1236 * rely on the futex_wake() code removing us from hash when it
1237 * wakes us up.
1238 */
1239
1240 /* add_wait_queue is the barrier after __set_current_state. */
1241 __set_current_state(TASK_INTERRUPTIBLE);
Darren Hart73500ac2008-12-17 17:29:56 -08001242 add_wait_queue(&q.waiter, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 /*
Pierre Peifferec92d082007-05-09 02:35:00 -07001244 * !plist_node_empty() is safe here without any lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 * q.lock_ptr != 0 is not safe, because of ordering against wakeup.
1246 */
Pierre Peifferec92d082007-05-09 02:35:00 -07001247 if (likely(!plist_node_empty(&q.list))) {
Pierre Peifferc19384b2007-05-09 02:35:02 -07001248 if (!abs_time)
1249 schedule();
1250 else {
Arjan van de Venae4b7482008-09-08 09:03:57 -07001251 unsigned long slack;
1252 slack = current->timer_slack_ns;
1253 if (rt_task(current))
1254 slack = 0;
Thomas Gleixner1acdac12008-11-20 10:02:53 -08001255 hrtimer_init_on_stack(&t.timer,
1256 clockrt ? CLOCK_REALTIME :
1257 CLOCK_MONOTONIC,
1258 HRTIMER_MODE_ABS);
Pierre Peifferc19384b2007-05-09 02:35:02 -07001259 hrtimer_init_sleeper(&t, current);
Arjan van de Venae4b7482008-09-08 09:03:57 -07001260 hrtimer_set_expires_range_ns(&t.timer, *abs_time, slack);
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001261
Arjan van de Vencc584b22008-09-01 15:02:30 -07001262 hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS);
Peter Zijlstra3588a082008-02-01 17:45:13 +01001263 if (!hrtimer_active(&t.timer))
1264 t.task = NULL;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001265
Pierre Peifferc19384b2007-05-09 02:35:02 -07001266 /*
1267 * the timer could have already expired, in which
1268 * case current would be flagged for rescheduling.
1269 * Don't bother calling schedule.
1270 */
1271 if (likely(t.task))
1272 schedule();
1273
1274 hrtimer_cancel(&t.timer);
1275
1276 /* Flag if a timeout occured */
1277 rem = (t.task == NULL);
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001278
1279 destroy_hrtimer_on_stack(&t.timer);
Pierre Peifferc19384b2007-05-09 02:35:02 -07001280 }
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001281 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 __set_current_state(TASK_RUNNING);
1283
1284 /*
1285 * NOTE: we don't remove ourselves from the waitqueue because
1286 * we are the only user of it.
1287 */
1288
1289 /* If we were woken (and unqueued), we succeeded, whatever. */
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001290 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 if (!unqueue_me(&q))
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001292 goto out_put_key;
1293 ret = -ETIMEDOUT;
Pierre Peifferc19384b2007-05-09 02:35:02 -07001294 if (rem)
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001295 goto out_put_key;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001296
Ingo Molnare2970f22006-06-27 02:54:47 -07001297 /*
1298 * We expect signal_pending(current), but another thread may
1299 * have handled it for us already.
1300 */
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001301 ret = -ERESTARTSYS;
Pierre Peifferc19384b2007-05-09 02:35:02 -07001302 if (!abs_time)
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001303 goto out_put_key;
Steven Rostedtce6bd422007-12-05 15:46:09 +01001304
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001305 restart = &current_thread_info()->restart_block;
1306 restart->fn = futex_wait_restart;
1307 restart->futex.uaddr = (u32 *)uaddr;
1308 restart->futex.val = val;
1309 restart->futex.time = abs_time->tv64;
1310 restart->futex.bitset = bitset;
1311 restart->futex.flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001313 if (fshared)
1314 restart->futex.flags |= FLAGS_SHARED;
1315 if (clockrt)
1316 restart->futex.flags |= FLAGS_CLOCKRT;
1317
1318 ret = -ERESTART_RESTARTBLOCK;
1319
1320out_put_key:
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001321 put_futex_key(fshared, &q.key);
Darren Hart42d35d42008-12-29 15:49:53 -08001322out:
Ingo Molnarc87e2832006-06-27 02:54:58 -07001323 return ret;
1324}
1325
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001326
1327static long futex_wait_restart(struct restart_block *restart)
1328{
Steven Rostedtce6bd422007-12-05 15:46:09 +01001329 u32 __user *uaddr = (u32 __user *)restart->futex.uaddr;
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001330 int fshared = 0;
Steven Rostedtce6bd422007-12-05 15:46:09 +01001331 ktime_t t;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001332
Steven Rostedtce6bd422007-12-05 15:46:09 +01001333 t.tv64 = restart->futex.time;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001334 restart->fn = do_no_restart_syscall;
Steven Rostedtce6bd422007-12-05 15:46:09 +01001335 if (restart->futex.flags & FLAGS_SHARED)
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001336 fshared = 1;
Thomas Gleixnercd689982008-02-01 17:45:14 +01001337 return (long)futex_wait(uaddr, fshared, restart->futex.val, &t,
Thomas Gleixner1acdac12008-11-20 10:02:53 -08001338 restart->futex.bitset,
1339 restart->futex.flags & FLAGS_CLOCKRT);
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001340}
1341
1342
Ingo Molnarc87e2832006-06-27 02:54:58 -07001343/*
1344 * Userspace tried a 0 -> TID atomic transition of the futex value
1345 * and failed. The kernel side here does the whole locking operation:
1346 * if there are waiters then it will block, it does PI, etc. (Due to
1347 * races the kernel might see a 0 value of the futex too.)
1348 */
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001349static int futex_lock_pi(u32 __user *uaddr, int fshared,
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001350 int detect, ktime_t *time, int trylock)
Ingo Molnarc87e2832006-06-27 02:54:58 -07001351{
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001352 struct hrtimer_sleeper timeout, *to = NULL;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001353 struct task_struct *curr = current;
1354 struct futex_hash_bucket *hb;
1355 u32 uval, newval, curval;
1356 struct futex_q q;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001357 int ret, lock_taken, ownerdied = 0, attempt = 0;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001358
1359 if (refill_pi_state_cache())
1360 return -ENOMEM;
1361
Pierre Peifferc19384b2007-05-09 02:35:02 -07001362 if (time) {
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001363 to = &timeout;
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001364 hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME,
1365 HRTIMER_MODE_ABS);
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001366 hrtimer_init_sleeper(to, current);
Arjan van de Vencc584b22008-09-01 15:02:30 -07001367 hrtimer_set_expires(&to->timer, *time);
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001368 }
1369
Ingo Molnarc87e2832006-06-27 02:54:58 -07001370 q.pi_state = NULL;
Darren Hart42d35d42008-12-29 15:49:53 -08001371retry:
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001372 q.key = FUTEX_KEY_INIT;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001373 ret = get_futex_key(uaddr, fshared, &q.key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001374 if (unlikely(ret != 0))
Darren Hart42d35d42008-12-29 15:49:53 -08001375 goto out;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001376
Darren Hart42d35d42008-12-29 15:49:53 -08001377retry_unlocked:
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01001378 hb = queue_lock(&q);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001379
Darren Hart42d35d42008-12-29 15:49:53 -08001380retry_locked:
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001381 ret = lock_taken = 0;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001382
Ingo Molnarc87e2832006-06-27 02:54:58 -07001383 /*
1384 * To avoid races, we attempt to take the lock here again
1385 * (by doing a 0 -> TID atomic cmpxchg), while holding all
1386 * the locks. It will most likely not succeed.
1387 */
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001388 newval = task_pid_vnr(current);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001389
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001390 curval = cmpxchg_futex_value_locked(uaddr, 0, newval);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001391
1392 if (unlikely(curval == -EFAULT))
1393 goto uaddr_faulted;
1394
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001395 /*
1396 * Detect deadlocks. In case of REQUEUE_PI this is a valid
1397 * situation and we return success to user space.
1398 */
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001399 if (unlikely((curval & FUTEX_TID_MASK) == task_pid_vnr(current))) {
Thomas Gleixnerbd197232007-06-17 21:11:10 +02001400 ret = -EDEADLK;
Darren Hart42d35d42008-12-29 15:49:53 -08001401 goto out_unlock_put_key;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001402 }
1403
1404 /*
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001405 * Surprise - we got the lock. Just return to userspace:
Ingo Molnarc87e2832006-06-27 02:54:58 -07001406 */
1407 if (unlikely(!curval))
Darren Hart42d35d42008-12-29 15:49:53 -08001408 goto out_unlock_put_key;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001409
1410 uval = curval;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001411
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001412 /*
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001413 * Set the WAITERS flag, so the owner will know it has someone
1414 * to wake at next unlock
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001415 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001416 newval = curval | FUTEX_WAITERS;
1417
1418 /*
1419 * There are two cases, where a futex might have no owner (the
Thomas Gleixnerbd197232007-06-17 21:11:10 +02001420 * owner TID is 0): OWNER_DIED. We take over the futex in this
1421 * case. We also do an unconditional take over, when the owner
1422 * of the futex died.
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001423 *
1424 * This is safe as we are protected by the hash bucket lock !
1425 */
1426 if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) {
Thomas Gleixnerbd197232007-06-17 21:11:10 +02001427 /* Keep the OWNER_DIED bit */
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001428 newval = (curval & ~FUTEX_TID_MASK) | task_pid_vnr(current);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001429 ownerdied = 0;
1430 lock_taken = 1;
1431 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001432
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001433 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001434
1435 if (unlikely(curval == -EFAULT))
1436 goto uaddr_faulted;
1437 if (unlikely(curval != uval))
1438 goto retry_locked;
1439
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001440 /*
Thomas Gleixnerbd197232007-06-17 21:11:10 +02001441 * We took the lock due to owner died take over.
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001442 */
Thomas Gleixnerbd197232007-06-17 21:11:10 +02001443 if (unlikely(lock_taken))
Darren Hart42d35d42008-12-29 15:49:53 -08001444 goto out_unlock_put_key;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001445
Ingo Molnarc87e2832006-06-27 02:54:58 -07001446 /*
1447 * We dont have the lock. Look up the PI state (or create it if
1448 * we are the first waiter):
1449 */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001450 ret = lookup_pi_state(uval, hb, &q.key, &q.pi_state);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001451
1452 if (unlikely(ret)) {
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001453 switch (ret) {
Ingo Molnarc87e2832006-06-27 02:54:58 -07001454
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001455 case -EAGAIN:
1456 /*
1457 * Task is exiting and we just wait for the
1458 * exit to complete.
1459 */
1460 queue_unlock(&q, hb);
Darren Hartde87fcc2009-03-12 00:55:46 -07001461 put_futex_key(fshared, &q.key);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001462 cond_resched();
1463 goto retry;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001464
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001465 case -ESRCH:
1466 /*
1467 * No owner found for this futex. Check if the
1468 * OWNER_DIED bit is set to figure out whether
1469 * this is a robust futex or not.
1470 */
1471 if (get_futex_value_locked(&curval, uaddr))
Ingo Molnarc87e2832006-06-27 02:54:58 -07001472 goto uaddr_faulted;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001473
1474 /*
1475 * We simply start over in case of a robust
1476 * futex. The code above will take the futex
1477 * and return happy.
1478 */
1479 if (curval & FUTEX_OWNER_DIED) {
1480 ownerdied = 1;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001481 goto retry_locked;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001482 }
1483 default:
Darren Hart42d35d42008-12-29 15:49:53 -08001484 goto out_unlock_put_key;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001485 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001486 }
1487
1488 /*
1489 * Only actually queue now that the atomic ops are done:
1490 */
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01001491 queue_me(&q, hb);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001492
Ingo Molnarc87e2832006-06-27 02:54:58 -07001493 WARN_ON(!q.pi_state);
1494 /*
1495 * Block on the PI mutex:
1496 */
1497 if (!trylock)
1498 ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1);
1499 else {
1500 ret = rt_mutex_trylock(&q.pi_state->pi_mutex);
1501 /* Fixup the trylock return value: */
1502 ret = ret ? 0 : -EWOULDBLOCK;
1503 }
1504
Vernon Mauerya99e4e42006-07-01 04:35:42 -07001505 spin_lock(q.lock_ptr);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001506
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001507 if (!ret) {
1508 /*
1509 * Got the lock. We might not be the anticipated owner
1510 * if we did a lock-steal - fix up the PI-state in
1511 * that case:
1512 */
1513 if (q.pi_state->owner != curr)
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001514 ret = fixup_pi_state_owner(uaddr, &q, curr, fshared);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001515 } else {
Ingo Molnarc87e2832006-06-27 02:54:58 -07001516 /*
1517 * Catch the rare case, where the lock was released
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001518 * when we were on the way back before we locked the
1519 * hash bucket.
Ingo Molnarc87e2832006-06-27 02:54:58 -07001520 */
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001521 if (q.pi_state->owner == curr) {
1522 /*
1523 * Try to get the rt_mutex now. This might
1524 * fail as some other task acquired the
1525 * rt_mutex after we removed ourself from the
1526 * rt_mutex waiters list.
1527 */
1528 if (rt_mutex_trylock(&q.pi_state->pi_mutex))
1529 ret = 0;
1530 else {
1531 /*
1532 * pi_state is incorrect, some other
1533 * task did a lock steal and we
1534 * returned due to timeout or signal
1535 * without taking the rt_mutex. Too
1536 * late. We can access the
1537 * rt_mutex_owner without locking, as
1538 * the other task is now blocked on
1539 * the hash bucket lock. Fix the state
1540 * up.
1541 */
1542 struct task_struct *owner;
1543 int res;
1544
1545 owner = rt_mutex_owner(&q.pi_state->pi_mutex);
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001546 res = fixup_pi_state_owner(uaddr, &q, owner,
1547 fshared);
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001548
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001549 /* propagate -EFAULT, if the fixup failed */
1550 if (res)
1551 ret = res;
1552 }
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001553 } else {
1554 /*
1555 * Paranoia check. If we did not take the lock
1556 * in the trylock above, then we should not be
1557 * the owner of the rtmutex, neither the real
1558 * nor the pending one:
1559 */
1560 if (rt_mutex_owner(&q.pi_state->pi_mutex) == curr)
1561 printk(KERN_ERR "futex_lock_pi: ret = %d "
1562 "pi-mutex: %p pi-state %p\n", ret,
1563 q.pi_state->pi_mutex.owner,
1564 q.pi_state->owner);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001565 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001566 }
1567
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001568 /* Unqueue and drop the lock */
1569 unqueue_me_pi(&q);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001570
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001571 if (to)
1572 destroy_hrtimer_on_stack(&to->timer);
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001573 return ret != -EINTR ? ret : -ERESTARTNOINTR;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001574
Darren Hart42d35d42008-12-29 15:49:53 -08001575out_unlock_put_key:
Ingo Molnarc87e2832006-06-27 02:54:58 -07001576 queue_unlock(&q, hb);
1577
Darren Hart42d35d42008-12-29 15:49:53 -08001578out_put_key:
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001579 put_futex_key(fshared, &q.key);
Darren Hart42d35d42008-12-29 15:49:53 -08001580out:
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001581 if (to)
1582 destroy_hrtimer_on_stack(&to->timer);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001583 return ret;
1584
Darren Hart42d35d42008-12-29 15:49:53 -08001585uaddr_faulted:
Ingo Molnarc87e2832006-06-27 02:54:58 -07001586 /*
Darren Hartb5686362008-12-18 15:06:34 -08001587 * We have to r/w *(int __user *)uaddr, and we have to modify it
1588 * atomically. Therefore, if we continue to fault after get_user()
1589 * below, we need to handle the fault ourselves, while still holding
1590 * the mmap_sem. This can occur if the uaddr is under contention as
1591 * we have to drop the mmap_sem in order to call get_user().
Ingo Molnarc87e2832006-06-27 02:54:58 -07001592 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001593 queue_unlock(&q, hb);
1594
Ingo Molnarc87e2832006-06-27 02:54:58 -07001595 if (attempt++) {
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001596 ret = futex_handle_fault((unsigned long)uaddr, attempt);
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001597 if (ret)
Darren Hart42d35d42008-12-29 15:49:53 -08001598 goto out_put_key;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001599 goto retry_unlocked;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001600 }
1601
Ingo Molnarc87e2832006-06-27 02:54:58 -07001602 ret = get_user(uval, uaddr);
Darren Hartb5686362008-12-18 15:06:34 -08001603 if (!ret)
Darren Hartde87fcc2009-03-12 00:55:46 -07001604 goto retry_unlocked;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001605
Darren Hartde87fcc2009-03-12 00:55:46 -07001606 goto out_put_key;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001607}
1608
Darren Hartde87fcc2009-03-12 00:55:46 -07001609
Ingo Molnarc87e2832006-06-27 02:54:58 -07001610/*
Ingo Molnarc87e2832006-06-27 02:54:58 -07001611 * Userspace attempted a TID -> 0 atomic transition, and failed.
1612 * This is the in-kernel slowpath: we look up the PI state (if any),
1613 * and do the rt-mutex unlock.
1614 */
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001615static int futex_unlock_pi(u32 __user *uaddr, int fshared)
Ingo Molnarc87e2832006-06-27 02:54:58 -07001616{
1617 struct futex_hash_bucket *hb;
1618 struct futex_q *this, *next;
1619 u32 uval;
Pierre Peifferec92d082007-05-09 02:35:00 -07001620 struct plist_head *head;
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001621 union futex_key key = FUTEX_KEY_INIT;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001622 int ret, attempt = 0;
1623
1624retry:
1625 if (get_user(uval, uaddr))
1626 return -EFAULT;
1627 /*
1628 * We release only a lock we actually own:
1629 */
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001630 if ((uval & FUTEX_TID_MASK) != task_pid_vnr(current))
Ingo Molnarc87e2832006-06-27 02:54:58 -07001631 return -EPERM;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001632
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001633 ret = get_futex_key(uaddr, fshared, &key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001634 if (unlikely(ret != 0))
1635 goto out;
1636
1637 hb = hash_futex(&key);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001638retry_unlocked:
Ingo Molnarc87e2832006-06-27 02:54:58 -07001639 spin_lock(&hb->lock);
1640
Ingo Molnarc87e2832006-06-27 02:54:58 -07001641 /*
1642 * To avoid races, try to do the TID -> 0 atomic transition
1643 * again. If it succeeds then we can return without waking
1644 * anyone else up:
1645 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001646 if (!(uval & FUTEX_OWNER_DIED))
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001647 uval = cmpxchg_futex_value_locked(uaddr, task_pid_vnr(current), 0);
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001648
Ingo Molnarc87e2832006-06-27 02:54:58 -07001649
1650 if (unlikely(uval == -EFAULT))
1651 goto pi_faulted;
1652 /*
1653 * Rare case: we managed to release the lock atomically,
1654 * no need to wake anyone else up:
1655 */
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001656 if (unlikely(uval == task_pid_vnr(current)))
Ingo Molnarc87e2832006-06-27 02:54:58 -07001657 goto out_unlock;
1658
1659 /*
1660 * Ok, other tasks may need to be woken up - check waiters
1661 * and do the wakeup if necessary:
1662 */
1663 head = &hb->chain;
1664
Pierre Peifferec92d082007-05-09 02:35:00 -07001665 plist_for_each_entry_safe(this, next, head, list) {
Ingo Molnarc87e2832006-06-27 02:54:58 -07001666 if (!match_futex (&this->key, &key))
1667 continue;
1668 ret = wake_futex_pi(uaddr, uval, this);
1669 /*
1670 * The atomic access to the futex value
1671 * generated a pagefault, so retry the
1672 * user-access and the wakeup:
1673 */
1674 if (ret == -EFAULT)
1675 goto pi_faulted;
1676 goto out_unlock;
1677 }
1678 /*
1679 * No waiters - kernel unlocks the futex:
1680 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001681 if (!(uval & FUTEX_OWNER_DIED)) {
1682 ret = unlock_futex_pi(uaddr, uval);
1683 if (ret == -EFAULT)
1684 goto pi_faulted;
1685 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001686
1687out_unlock:
1688 spin_unlock(&hb->lock);
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001689 put_futex_key(fshared, &key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001690
Darren Hart42d35d42008-12-29 15:49:53 -08001691out:
Ingo Molnarc87e2832006-06-27 02:54:58 -07001692 return ret;
1693
1694pi_faulted:
1695 /*
Darren Hartb5686362008-12-18 15:06:34 -08001696 * We have to r/w *(int __user *)uaddr, and we have to modify it
1697 * atomically. Therefore, if we continue to fault after get_user()
1698 * below, we need to handle the fault ourselves, while still holding
1699 * the mmap_sem. This can occur if the uaddr is under contention as
1700 * we have to drop the mmap_sem in order to call get_user().
Ingo Molnarc87e2832006-06-27 02:54:58 -07001701 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001702 spin_unlock(&hb->lock);
1703
Ingo Molnarc87e2832006-06-27 02:54:58 -07001704 if (attempt++) {
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001705 ret = futex_handle_fault((unsigned long)uaddr, attempt);
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001706 if (ret)
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001707 goto out;
john stultz187226f2007-08-22 14:01:10 -07001708 uval = 0;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001709 goto retry_unlocked;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001710 }
1711
Ingo Molnarc87e2832006-06-27 02:54:58 -07001712 ret = get_user(uval, uaddr);
Darren Hartde87fcc2009-03-12 00:55:46 -07001713 put_futex_key(fshared, &key);
Darren Hartb5686362008-12-18 15:06:34 -08001714 if (!ret)
Ingo Molnarc87e2832006-06-27 02:54:58 -07001715 goto retry;
1716
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 return ret;
1718}
1719
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001720/*
1721 * Support for robust futexes: the kernel cleans up held futexes at
1722 * thread exit time.
1723 *
1724 * Implementation: user-space maintains a per-thread list of locks it
1725 * is holding. Upon do_exit(), the kernel carefully walks this list,
1726 * and marks all locks that are owned by this thread with the
Ingo Molnarc87e2832006-06-27 02:54:58 -07001727 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001728 * always manipulated with the lock held, so the list is private and
1729 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
1730 * field, to allow the kernel to clean up if the thread dies after
1731 * acquiring the lock, but just before it could have added itself to
1732 * the list. There can only be one such pending lock.
1733 */
1734
1735/**
1736 * sys_set_robust_list - set the robust-futex list head of a task
1737 * @head: pointer to the list-head
1738 * @len: length of the list-head, as userspace expects
1739 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001740SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head,
1741 size_t, len)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001742{
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08001743 if (!futex_cmpxchg_enabled)
1744 return -ENOSYS;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001745 /*
1746 * The kernel knows only one size for now:
1747 */
1748 if (unlikely(len != sizeof(*head)))
1749 return -EINVAL;
1750
1751 current->robust_list = head;
1752
1753 return 0;
1754}
1755
1756/**
1757 * sys_get_robust_list - get the robust-futex list head of a task
1758 * @pid: pid of the process [zero for current task]
1759 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
1760 * @len_ptr: pointer to a length field, the kernel fills in the header size
1761 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001762SYSCALL_DEFINE3(get_robust_list, int, pid,
1763 struct robust_list_head __user * __user *, head_ptr,
1764 size_t __user *, len_ptr)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001765{
Al Viroba46df92006-10-10 22:46:07 +01001766 struct robust_list_head __user *head;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001767 unsigned long ret;
David Howellsc69e8d92008-11-14 10:39:19 +11001768 const struct cred *cred = current_cred(), *pcred;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001769
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08001770 if (!futex_cmpxchg_enabled)
1771 return -ENOSYS;
1772
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001773 if (!pid)
1774 head = current->robust_list;
1775 else {
1776 struct task_struct *p;
1777
1778 ret = -ESRCH;
Oleg Nesterovaaa2a972006-09-29 02:00:55 -07001779 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001780 p = find_task_by_vpid(pid);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001781 if (!p)
1782 goto err_unlock;
1783 ret = -EPERM;
David Howellsc69e8d92008-11-14 10:39:19 +11001784 pcred = __task_cred(p);
1785 if (cred->euid != pcred->euid &&
1786 cred->euid != pcred->uid &&
David Howells76aac0e2008-11-14 10:39:12 +11001787 !capable(CAP_SYS_PTRACE))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001788 goto err_unlock;
1789 head = p->robust_list;
Oleg Nesterovaaa2a972006-09-29 02:00:55 -07001790 rcu_read_unlock();
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001791 }
1792
1793 if (put_user(sizeof(*head), len_ptr))
1794 return -EFAULT;
1795 return put_user(head, head_ptr);
1796
1797err_unlock:
Oleg Nesterovaaa2a972006-09-29 02:00:55 -07001798 rcu_read_unlock();
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001799
1800 return ret;
1801}
1802
1803/*
1804 * Process a futex-list entry, check whether it's owned by the
1805 * dying task, and do notification if so:
1806 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001807int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001808{
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001809 u32 uval, nval, mval;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001810
Ingo Molnar8f17d3a2006-03-27 01:16:27 -08001811retry:
1812 if (get_user(uval, uaddr))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001813 return -1;
1814
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001815 if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) {
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001816 /*
1817 * Ok, this dying thread is truly holding a futex
1818 * of interest. Set the OWNER_DIED bit atomically
1819 * via cmpxchg, and if the value had FUTEX_WAITERS
1820 * set, wake up a waiter (if any). (We have to do a
1821 * futex_wake() even if OWNER_DIED is already set -
1822 * to handle the rare but possible case of recursive
1823 * thread-death.) The rest of the cleanup is done in
1824 * userspace.
1825 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001826 mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
1827 nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval);
1828
Ingo Molnarc87e2832006-06-27 02:54:58 -07001829 if (nval == -EFAULT)
1830 return -1;
1831
1832 if (nval != uval)
Ingo Molnar8f17d3a2006-03-27 01:16:27 -08001833 goto retry;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001834
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001835 /*
1836 * Wake robust non-PI futexes here. The wakeup of
1837 * PI futexes happens in exit_pi_state():
1838 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001839 if (!pi && (uval & FUTEX_WAITERS))
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001840 futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001841 }
1842 return 0;
1843}
1844
1845/*
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001846 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
1847 */
1848static inline int fetch_robust_entry(struct robust_list __user **entry,
Al Viroba46df92006-10-10 22:46:07 +01001849 struct robust_list __user * __user *head,
1850 int *pi)
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001851{
1852 unsigned long uentry;
1853
Al Viroba46df92006-10-10 22:46:07 +01001854 if (get_user(uentry, (unsigned long __user *)head))
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001855 return -EFAULT;
1856
Al Viroba46df92006-10-10 22:46:07 +01001857 *entry = (void __user *)(uentry & ~1UL);
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001858 *pi = uentry & 1;
1859
1860 return 0;
1861}
1862
1863/*
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001864 * Walk curr->robust_list (very carefully, it's a userspace list!)
1865 * and mark any locks found there dead, and notify any waiters.
1866 *
1867 * We silently return on any sign of list-walking problem.
1868 */
1869void exit_robust_list(struct task_struct *curr)
1870{
1871 struct robust_list_head __user *head = curr->robust_list;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001872 struct robust_list __user *entry, *next_entry, *pending;
1873 unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001874 unsigned long futex_offset;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001875 int rc;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001876
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08001877 if (!futex_cmpxchg_enabled)
1878 return;
1879
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001880 /*
1881 * Fetch the list head (which was registered earlier, via
1882 * sys_set_robust_list()):
1883 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001884 if (fetch_robust_entry(&entry, &head->list.next, &pi))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001885 return;
1886 /*
1887 * Fetch the relative futex offset:
1888 */
1889 if (get_user(futex_offset, &head->futex_offset))
1890 return;
1891 /*
1892 * Fetch any possibly pending lock-add first, and handle it
1893 * if it exists:
1894 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001895 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001896 return;
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001897
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001898 next_entry = NULL; /* avoid warning with gcc */
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001899 while (entry != &head->list) {
1900 /*
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001901 * Fetch the next entry in the list before calling
1902 * handle_futex_death:
1903 */
1904 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
1905 /*
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001906 * A pending lock might already be on the list, so
Ingo Molnarc87e2832006-06-27 02:54:58 -07001907 * don't process it twice:
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001908 */
1909 if (entry != pending)
Al Viroba46df92006-10-10 22:46:07 +01001910 if (handle_futex_death((void __user *)entry + futex_offset,
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001911 curr, pi))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001912 return;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001913 if (rc)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001914 return;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001915 entry = next_entry;
1916 pi = next_pi;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001917 /*
1918 * Avoid excessively long or circular lists:
1919 */
1920 if (!--limit)
1921 break;
1922
1923 cond_resched();
1924 }
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001925
1926 if (pending)
1927 handle_futex_death((void __user *)pending + futex_offset,
1928 curr, pip);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001929}
1930
Pierre Peifferc19384b2007-05-09 02:35:02 -07001931long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
Ingo Molnare2970f22006-06-27 02:54:47 -07001932 u32 __user *uaddr2, u32 val2, u32 val3)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933{
Thomas Gleixner1acdac12008-11-20 10:02:53 -08001934 int clockrt, ret = -ENOSYS;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001935 int cmd = op & FUTEX_CMD_MASK;
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001936 int fshared = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001938 if (!(op & FUTEX_PRIVATE_FLAG))
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02001939 fshared = 1;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001940
Thomas Gleixner1acdac12008-11-20 10:02:53 -08001941 clockrt = op & FUTEX_CLOCK_REALTIME;
1942 if (clockrt && cmd != FUTEX_WAIT_BITSET)
1943 return -ENOSYS;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001944
1945 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 case FUTEX_WAIT:
Thomas Gleixnercd689982008-02-01 17:45:14 +01001947 val3 = FUTEX_BITSET_MATCH_ANY;
1948 case FUTEX_WAIT_BITSET:
Thomas Gleixner1acdac12008-11-20 10:02:53 -08001949 ret = futex_wait(uaddr, fshared, val, timeout, val3, clockrt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 break;
1951 case FUTEX_WAKE:
Thomas Gleixnercd689982008-02-01 17:45:14 +01001952 val3 = FUTEX_BITSET_MATCH_ANY;
1953 case FUTEX_WAKE_BITSET:
1954 ret = futex_wake(uaddr, fshared, val, val3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 case FUTEX_REQUEUE:
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001957 ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 break;
1959 case FUTEX_CMP_REQUEUE:
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001960 ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 break;
Jakub Jelinek4732efb2005-09-06 15:16:25 -07001962 case FUTEX_WAKE_OP:
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001963 ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3);
Jakub Jelinek4732efb2005-09-06 15:16:25 -07001964 break;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001965 case FUTEX_LOCK_PI:
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08001966 if (futex_cmpxchg_enabled)
1967 ret = futex_lock_pi(uaddr, fshared, val, timeout, 0);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001968 break;
1969 case FUTEX_UNLOCK_PI:
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08001970 if (futex_cmpxchg_enabled)
1971 ret = futex_unlock_pi(uaddr, fshared);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001972 break;
1973 case FUTEX_TRYLOCK_PI:
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08001974 if (futex_cmpxchg_enabled)
1975 ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001976 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 default:
1978 ret = -ENOSYS;
1979 }
1980 return ret;
1981}
1982
1983
Heiko Carstens17da2bd2009-01-14 14:14:10 +01001984SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
1985 struct timespec __user *, utime, u32 __user *, uaddr2,
1986 u32, val3)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987{
Pierre Peifferc19384b2007-05-09 02:35:02 -07001988 struct timespec ts;
1989 ktime_t t, *tp = NULL;
Ingo Molnare2970f22006-06-27 02:54:47 -07001990 u32 val2 = 0;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001991 int cmd = op & FUTEX_CMD_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992
Thomas Gleixnercd689982008-02-01 17:45:14 +01001993 if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
1994 cmd == FUTEX_WAIT_BITSET)) {
Pierre Peifferc19384b2007-05-09 02:35:02 -07001995 if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 return -EFAULT;
Pierre Peifferc19384b2007-05-09 02:35:02 -07001997 if (!timespec_valid(&ts))
Thomas Gleixner9741ef92006-03-31 02:31:32 -08001998 return -EINVAL;
Pierre Peifferc19384b2007-05-09 02:35:02 -07001999
2000 t = timespec_to_ktime(ts);
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002001 if (cmd == FUTEX_WAIT)
Thomas Gleixner5a7780e2008-02-13 09:20:43 +01002002 t = ktime_add_safe(ktime_get(), t);
Pierre Peifferc19384b2007-05-09 02:35:02 -07002003 tp = &t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 }
2005 /*
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002006 * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE.
Andreas Schwabf54f0982007-07-31 00:38:51 -07002007 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 */
Andreas Schwabf54f0982007-07-31 00:38:51 -07002009 if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
2010 cmd == FUTEX_WAKE_OP)
Ingo Molnare2970f22006-06-27 02:54:47 -07002011 val2 = (u32) (unsigned long) utime;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012
Pierre Peifferc19384b2007-05-09 02:35:02 -07002013 return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014}
2015
Benjamin Herrenschmidtf6d107f2008-03-27 14:52:15 +11002016static int __init futex_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017{
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002018 u32 curval;
Thomas Gleixner3e4ab742008-02-23 15:23:55 -08002019 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002021 /*
2022 * This will fail and we want it. Some arch implementations do
2023 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2024 * functionality. We want to know that before we call in any
2025 * of the complex code paths. Also we want to prevent
2026 * registration of robust lists in that case. NULL is
2027 * guaranteed to fault and we get -EFAULT on functional
2028 * implementation, the non functional ones will return
2029 * -ENOSYS.
2030 */
2031 curval = cmpxchg_futex_value_locked(NULL, 0, 0);
2032 if (curval == -EFAULT)
2033 futex_cmpxchg_enabled = 1;
2034
Thomas Gleixner3e4ab742008-02-23 15:23:55 -08002035 for (i = 0; i < ARRAY_SIZE(futex_queues); i++) {
2036 plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock);
2037 spin_lock_init(&futex_queues[i].lock);
2038 }
2039
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 return 0;
2041}
Benjamin Herrenschmidtf6d107f2008-03-27 14:52:15 +11002042__initcall(futex_init);