blob: fcc94e7b40864afcb72170f19d832f8117e517fc [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>
Jakub Jelinek4732efbe2005-09-06 15:16:25 -070055#include <asm/futex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Ingo Molnarc87e2832006-06-27 02:54:58 -070057#include "rtmutex_common.h"
58
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
60
61/*
Ingo Molnarc87e2832006-06-27 02:54:58 -070062 * Priority Inheritance state:
63 */
64struct futex_pi_state {
65 /*
66 * list of 'owned' pi_state instances - these have to be
67 * cleaned up in do_exit() if the task exits prematurely:
68 */
69 struct list_head list;
70
71 /*
72 * The PI object:
73 */
74 struct rt_mutex pi_mutex;
75
76 struct task_struct *owner;
77 atomic_t refcount;
78
79 union futex_key key;
80};
81
82/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 * We use this hashed waitqueue instead of a normal wait_queue_t, so
84 * we can wake only the relevant ones (hashed queues may be shared).
85 *
86 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
Pierre Peifferec92d082007-05-09 02:35:00 -070087 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 * The order of wakup is always to make the first condition true, then
89 * wake up q->waiters, then make the second condition true.
90 */
91struct futex_q {
Pierre Peifferec92d082007-05-09 02:35:00 -070092 struct plist_node list;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 wait_queue_head_t waiters;
94
Ingo Molnare2970f22006-06-27 02:54:47 -070095 /* Which hash list lock to use: */
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 spinlock_t *lock_ptr;
97
Ingo Molnare2970f22006-06-27 02:54:47 -070098 /* Key which the futex is hashed on: */
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 union futex_key key;
100
Ingo Molnare2970f22006-06-27 02:54:47 -0700101 /* For fd, sigio sent using these: */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 int fd;
103 struct file *filp;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700104
105 /* Optional priority inheritance state: */
106 struct futex_pi_state *pi_state;
107 struct task_struct *task;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108};
109
110/*
111 * Split the global futex_lock into every hash list lock.
112 */
113struct futex_hash_bucket {
Pierre Peifferec92d082007-05-09 02:35:00 -0700114 spinlock_t lock;
115 struct plist_head chain;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116};
117
118static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS];
119
120/* Futex-fs vfsmount entry: */
121static struct vfsmount *futex_mnt;
122
123/*
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700124 * Take mm->mmap_sem, when futex is shared
125 */
126static inline void futex_lock_mm(struct rw_semaphore *fshared)
127{
128 if (fshared)
129 down_read(fshared);
130}
131
132/*
133 * Release mm->mmap_sem, when the futex is shared
134 */
135static inline void futex_unlock_mm(struct rw_semaphore *fshared)
136{
137 if (fshared)
138 up_read(fshared);
139}
140
141/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 * We hash on the keys returned from get_futex_key (see below).
143 */
144static struct futex_hash_bucket *hash_futex(union futex_key *key)
145{
146 u32 hash = jhash2((u32*)&key->both.word,
147 (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
148 key->both.offset);
149 return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)];
150}
151
152/*
153 * Return 1 if two futex_keys are equal, 0 otherwise.
154 */
155static inline int match_futex(union futex_key *key1, union futex_key *key2)
156{
157 return (key1->both.word == key2->both.word
158 && key1->both.ptr == key2->both.ptr
159 && key1->both.offset == key2->both.offset);
160}
161
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700162/**
163 * get_futex_key - Get parameters which are the keys for a futex.
164 * @uaddr: virtual address of the futex
165 * @shared: NULL for a PROCESS_PRIVATE futex,
166 * &current->mm->mmap_sem for a PROCESS_SHARED futex
167 * @key: address where result is stored.
168 *
169 * Returns a negative error code or 0
170 * The key words are stored in *key on success.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 *
Josef "Jeff" Sipekf3a43f32006-12-08 02:36:43 -0800172 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 * offset_within_page). For private mappings, it's (uaddr, current->mm).
174 * We can usually work out the index without swapping in the page.
175 *
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700176 * fshared is NULL for PROCESS_PRIVATE futexes
177 * For other futexes, it points to &current->mm->mmap_sem and
178 * caller must have taken the reader lock. but NOT any spinlocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 */
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700180int get_futex_key(u32 __user *uaddr, struct rw_semaphore *fshared,
181 union futex_key *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182{
Ingo Molnare2970f22006-06-27 02:54:47 -0700183 unsigned long address = (unsigned long)uaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 struct mm_struct *mm = current->mm;
185 struct vm_area_struct *vma;
186 struct page *page;
187 int err;
188
189 /*
190 * The futex address must be "naturally" aligned.
191 */
Ingo Molnare2970f22006-06-27 02:54:47 -0700192 key->both.offset = address % PAGE_SIZE;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700193 if (unlikely((address % sizeof(u32)) != 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 return -EINVAL;
Ingo Molnare2970f22006-06-27 02:54:47 -0700195 address -= key->both.offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
197 /*
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700198 * PROCESS_PRIVATE futexes are fast.
199 * As the mm cannot disappear under us and the 'key' only needs
200 * virtual address, we dont even have to find the underlying vma.
201 * Note : We do have to check 'uaddr' is a valid user address,
202 * but access_ok() should be faster than find_vma()
203 */
204 if (!fshared) {
205 if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))))
206 return -EFAULT;
207 key->private.mm = mm;
208 key->private.address = address;
209 return 0;
210 }
211 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 * The futex is hashed differently depending on whether
213 * it's in a shared or private mapping. So check vma first.
214 */
Ingo Molnare2970f22006-06-27 02:54:47 -0700215 vma = find_extend_vma(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 if (unlikely(!vma))
217 return -EFAULT;
218
219 /*
220 * Permissions.
221 */
222 if (unlikely((vma->vm_flags & (VM_IO|VM_READ)) != VM_READ))
223 return (vma->vm_flags & VM_IO) ? -EPERM : -EACCES;
224
225 /*
226 * Private mappings are handled in a simple way.
227 *
228 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
229 * it's a read-only handle, it's expected that futexes attach to
230 * the object not the particular process. Therefore we use
231 * VM_MAYSHARE here, not VM_SHARED which is restricted to shared
232 * mappings of _writable_ handles.
233 */
234 if (likely(!(vma->vm_flags & VM_MAYSHARE))) {
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700235 key->both.offset |= FUT_OFF_MMSHARED; /* reference taken on mm */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 key->private.mm = mm;
Ingo Molnare2970f22006-06-27 02:54:47 -0700237 key->private.address = address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 return 0;
239 }
240
241 /*
242 * Linear file mappings are also simple.
243 */
Josef "Jeff" Sipekf3a43f32006-12-08 02:36:43 -0800244 key->shared.inode = vma->vm_file->f_path.dentry->d_inode;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700245 key->both.offset |= FUT_OFF_INODE; /* inode-based key. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 if (likely(!(vma->vm_flags & VM_NONLINEAR))) {
Ingo Molnare2970f22006-06-27 02:54:47 -0700247 key->shared.pgoff = (((address - vma->vm_start) >> PAGE_SHIFT)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 + vma->vm_pgoff);
249 return 0;
250 }
251
252 /*
253 * We could walk the page table to read the non-linear
254 * pte, and get the page index without fetching the page
255 * from swap. But that's a lot of code to duplicate here
256 * for a rare case, so we simply fetch the page.
257 */
Ingo Molnare2970f22006-06-27 02:54:47 -0700258 err = get_user_pages(current, mm, address, 1, 0, 0, &page, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 if (err >= 0) {
260 key->shared.pgoff =
261 page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
262 put_page(page);
263 return 0;
264 }
265 return err;
266}
Rusty Russell9adef582007-05-08 00:26:42 -0700267EXPORT_SYMBOL_GPL(get_futex_key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
269/*
270 * Take a reference to the resource addressed by a key.
271 * Can be called while holding spinlocks.
272 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 */
Rusty Russell9adef582007-05-08 00:26:42 -0700274inline void get_futex_key_refs(union futex_key *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700276 if (key->both.ptr == 0)
277 return;
278 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
279 case FUT_OFF_INODE:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 atomic_inc(&key->shared.inode->i_count);
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700281 break;
282 case FUT_OFF_MMSHARED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 atomic_inc(&key->private.mm->mm_count);
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700284 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 }
286}
Rusty Russell9adef582007-05-08 00:26:42 -0700287EXPORT_SYMBOL_GPL(get_futex_key_refs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
289/*
290 * Drop a reference to the resource addressed by a key.
291 * The hash bucket spinlock must not be held.
292 */
Rusty Russell9adef582007-05-08 00:26:42 -0700293void drop_futex_key_refs(union futex_key *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700295 if (key->both.ptr == 0)
296 return;
297 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
298 case FUT_OFF_INODE:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 iput(key->shared.inode);
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700300 break;
301 case FUT_OFF_MMSHARED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 mmdrop(key->private.mm);
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700303 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 }
305}
Rusty Russell9adef582007-05-08 00:26:42 -0700306EXPORT_SYMBOL_GPL(drop_futex_key_refs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700308static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval)
309{
310 u32 curval;
311
312 pagefault_disable();
313 curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval);
314 pagefault_enable();
315
316 return curval;
317}
318
319static int get_futex_value_locked(u32 *dest, u32 __user *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320{
321 int ret;
322
Peter Zijlstraa8663742006-12-06 20:32:20 -0800323 pagefault_disable();
Ingo Molnare2970f22006-06-27 02:54:47 -0700324 ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
Peter Zijlstraa8663742006-12-06 20:32:20 -0800325 pagefault_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
327 return ret ? -EFAULT : 0;
328}
329
330/*
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700331 * Fault handling.
332 * if fshared is non NULL, current->mm->mmap_sem is already held
Ingo Molnarc87e2832006-06-27 02:54:58 -0700333 */
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700334static int futex_handle_fault(unsigned long address,
335 struct rw_semaphore *fshared, int attempt)
Ingo Molnarc87e2832006-06-27 02:54:58 -0700336{
337 struct vm_area_struct * vma;
338 struct mm_struct *mm = current->mm;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700339 int ret = -EFAULT;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700340
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700341 if (attempt > 2)
342 return ret;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700343
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700344 if (!fshared)
345 down_read(&mm->mmap_sem);
346 vma = find_vma(mm, address);
347 if (vma && address >= vma->vm_start &&
348 (vma->vm_flags & VM_WRITE)) {
Nick Piggin83c54072007-07-19 01:47:05 -0700349 int fault;
350 fault = handle_mm_fault(mm, vma, address, 1);
351 if (unlikely((fault & VM_FAULT_ERROR))) {
352#if 0
353 /* XXX: let's do this when we verify it is OK */
354 if (ret & VM_FAULT_OOM)
355 ret = -ENOMEM;
356#endif
357 } else {
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700358 ret = 0;
Nick Piggin83c54072007-07-19 01:47:05 -0700359 if (fault & VM_FAULT_MAJOR)
360 current->maj_flt++;
361 else
362 current->min_flt++;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700363 }
Ingo Molnarc87e2832006-06-27 02:54:58 -0700364 }
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700365 if (!fshared)
366 up_read(&mm->mmap_sem);
367 return ret;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700368}
369
370/*
371 * PI code:
372 */
373static int refill_pi_state_cache(void)
374{
375 struct futex_pi_state *pi_state;
376
377 if (likely(current->pi_state_cache))
378 return 0;
379
Burman Yan4668edc2006-12-06 20:38:51 -0800380 pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700381
382 if (!pi_state)
383 return -ENOMEM;
384
Ingo Molnarc87e2832006-06-27 02:54:58 -0700385 INIT_LIST_HEAD(&pi_state->list);
386 /* pi_mutex gets initialized later */
387 pi_state->owner = NULL;
388 atomic_set(&pi_state->refcount, 1);
389
390 current->pi_state_cache = pi_state;
391
392 return 0;
393}
394
395static struct futex_pi_state * alloc_pi_state(void)
396{
397 struct futex_pi_state *pi_state = current->pi_state_cache;
398
399 WARN_ON(!pi_state);
400 current->pi_state_cache = NULL;
401
402 return pi_state;
403}
404
405static void free_pi_state(struct futex_pi_state *pi_state)
406{
407 if (!atomic_dec_and_test(&pi_state->refcount))
408 return;
409
410 /*
411 * If pi_state->owner is NULL, the owner is most probably dying
412 * and has cleaned up the pi_state already
413 */
414 if (pi_state->owner) {
415 spin_lock_irq(&pi_state->owner->pi_lock);
416 list_del_init(&pi_state->list);
417 spin_unlock_irq(&pi_state->owner->pi_lock);
418
419 rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
420 }
421
422 if (current->pi_state_cache)
423 kfree(pi_state);
424 else {
425 /*
426 * pi_state->list is already empty.
427 * clear pi_state->owner.
428 * refcount is at 0 - put it back to 1.
429 */
430 pi_state->owner = NULL;
431 atomic_set(&pi_state->refcount, 1);
432 current->pi_state_cache = pi_state;
433 }
434}
435
436/*
437 * Look up the task based on what TID userspace gave us.
438 * We dont trust it.
439 */
440static struct task_struct * futex_find_get_task(pid_t pid)
441{
442 struct task_struct *p;
443
Oleg Nesterovd359b542006-09-29 02:00:55 -0700444 rcu_read_lock();
Ingo Molnarc87e2832006-06-27 02:54:58 -0700445 p = find_task_by_pid(pid);
Thomas Gleixnera06381f2007-06-23 11:48:40 +0200446
447 if (!p || ((current->euid != p->euid) && (current->euid != p->uid)))
448 p = ERR_PTR(-ESRCH);
449 else
450 get_task_struct(p);
451
Oleg Nesterovd359b542006-09-29 02:00:55 -0700452 rcu_read_unlock();
Ingo Molnarc87e2832006-06-27 02:54:58 -0700453
454 return p;
455}
456
457/*
458 * This task is holding PI mutexes at exit time => bad.
459 * Kernel cleans up PI-state, but userspace is likely hosed.
460 * (Robust-futex cleanup is separate and might save the day for userspace.)
461 */
462void exit_pi_state_list(struct task_struct *curr)
463{
Ingo Molnarc87e2832006-06-27 02:54:58 -0700464 struct list_head *next, *head = &curr->pi_state_list;
465 struct futex_pi_state *pi_state;
Ingo Molnar627371d2006-07-29 05:16:20 +0200466 struct futex_hash_bucket *hb;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700467 union futex_key key;
468
469 /*
470 * We are a ZOMBIE and nobody can enqueue itself on
471 * pi_state_list anymore, but we have to be careful
Ingo Molnar627371d2006-07-29 05:16:20 +0200472 * versus waiters unqueueing themselves:
Ingo Molnarc87e2832006-06-27 02:54:58 -0700473 */
474 spin_lock_irq(&curr->pi_lock);
475 while (!list_empty(head)) {
476
477 next = head->next;
478 pi_state = list_entry(next, struct futex_pi_state, list);
479 key = pi_state->key;
Ingo Molnar627371d2006-07-29 05:16:20 +0200480 hb = hash_futex(&key);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700481 spin_unlock_irq(&curr->pi_lock);
482
Ingo Molnarc87e2832006-06-27 02:54:58 -0700483 spin_lock(&hb->lock);
484
485 spin_lock_irq(&curr->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200486 /*
487 * We dropped the pi-lock, so re-check whether this
488 * task still owns the PI-state:
489 */
Ingo Molnarc87e2832006-06-27 02:54:58 -0700490 if (head->next != next) {
491 spin_unlock(&hb->lock);
492 continue;
493 }
494
Ingo Molnarc87e2832006-06-27 02:54:58 -0700495 WARN_ON(pi_state->owner != curr);
Ingo Molnar627371d2006-07-29 05:16:20 +0200496 WARN_ON(list_empty(&pi_state->list));
497 list_del_init(&pi_state->list);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700498 pi_state->owner = NULL;
499 spin_unlock_irq(&curr->pi_lock);
500
501 rt_mutex_unlock(&pi_state->pi_mutex);
502
503 spin_unlock(&hb->lock);
504
505 spin_lock_irq(&curr->pi_lock);
506 }
507 spin_unlock_irq(&curr->pi_lock);
508}
509
510static int
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700511lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
512 union futex_key *key, struct futex_pi_state **ps)
Ingo Molnarc87e2832006-06-27 02:54:58 -0700513{
514 struct futex_pi_state *pi_state = NULL;
515 struct futex_q *this, *next;
Pierre Peifferec92d082007-05-09 02:35:00 -0700516 struct plist_head *head;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700517 struct task_struct *p;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700518 pid_t pid = uval & FUTEX_TID_MASK;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700519
520 head = &hb->chain;
521
Pierre Peifferec92d082007-05-09 02:35:00 -0700522 plist_for_each_entry_safe(this, next, head, list) {
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700523 if (match_futex(&this->key, key)) {
Ingo Molnarc87e2832006-06-27 02:54:58 -0700524 /*
525 * Another waiter already exists - bump up
526 * the refcount and return its pi_state:
527 */
528 pi_state = this->pi_state;
Thomas Gleixner06a9ec22006-07-10 04:44:30 -0700529 /*
530 * Userspace might have messed up non PI and PI futexes
531 */
532 if (unlikely(!pi_state))
533 return -EINVAL;
534
Ingo Molnar627371d2006-07-29 05:16:20 +0200535 WARN_ON(!atomic_read(&pi_state->refcount));
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700536 WARN_ON(pid && pi_state->owner &&
537 pi_state->owner->pid != pid);
Ingo Molnar627371d2006-07-29 05:16:20 +0200538
Ingo Molnarc87e2832006-06-27 02:54:58 -0700539 atomic_inc(&pi_state->refcount);
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700540 *ps = pi_state;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700541
542 return 0;
543 }
544 }
545
546 /*
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200547 * We are the first waiter - try to look up the real owner and attach
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700548 * the new pi_state to it, but bail out when TID = 0
Ingo Molnarc87e2832006-06-27 02:54:58 -0700549 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700550 if (!pid)
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200551 return -ESRCH;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700552 p = futex_find_get_task(pid);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700553 if (IS_ERR(p))
554 return PTR_ERR(p);
555
556 /*
557 * We need to look at the task state flags to figure out,
558 * whether the task is exiting. To protect against the do_exit
559 * change of the task flags, we do this protected by
560 * p->pi_lock:
561 */
562 spin_lock_irq(&p->pi_lock);
563 if (unlikely(p->flags & PF_EXITING)) {
564 /*
565 * The task is on the way out. When PF_EXITPIDONE is
566 * set, we know that the task has finished the
567 * cleanup:
568 */
569 int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
570
571 spin_unlock_irq(&p->pi_lock);
572 put_task_struct(p);
573 return ret;
574 }
Ingo Molnarc87e2832006-06-27 02:54:58 -0700575
576 pi_state = alloc_pi_state();
577
578 /*
579 * Initialize the pi_mutex in locked state and make 'p'
580 * the owner of it:
581 */
582 rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
583
584 /* Store the key for possible exit cleanups: */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700585 pi_state->key = *key;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700586
Ingo Molnar627371d2006-07-29 05:16:20 +0200587 WARN_ON(!list_empty(&pi_state->list));
Ingo Molnarc87e2832006-06-27 02:54:58 -0700588 list_add(&pi_state->list, &p->pi_state_list);
589 pi_state->owner = p;
590 spin_unlock_irq(&p->pi_lock);
591
592 put_task_struct(p);
593
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700594 *ps = pi_state;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700595
596 return 0;
597}
598
599/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 * The hash bucket lock must be held when this is called.
601 * Afterwards, the futex_q must not be accessed.
602 */
603static void wake_futex(struct futex_q *q)
604{
Pierre Peifferec92d082007-05-09 02:35:00 -0700605 plist_del(&q->list, &q->list.plist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 if (q->filp)
607 send_sigio(&q->filp->f_owner, q->fd, POLL_IN);
608 /*
609 * The lock in wake_up_all() is a crucial memory barrier after the
Pierre Peifferec92d082007-05-09 02:35:00 -0700610 * plist_del() and also before assigning to q->lock_ptr.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 */
612 wake_up_all(&q->waiters);
613 /*
614 * The waiting task can free the futex_q as soon as this is written,
615 * without taking any locks. This must come last.
Andrew Morton8e311082005-12-23 19:54:46 -0800616 *
617 * A memory barrier is required here to prevent the following store
618 * to lock_ptr from getting ahead of the wakeup. Clearing the lock
619 * at the end of wake_up_all() does not prevent this store from
620 * moving.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 */
Ralf Baechleccdea2f2006-12-06 20:40:26 -0800622 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 q->lock_ptr = NULL;
624}
625
Ingo Molnarc87e2832006-06-27 02:54:58 -0700626static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
627{
628 struct task_struct *new_owner;
629 struct futex_pi_state *pi_state = this->pi_state;
630 u32 curval, newval;
631
632 if (!pi_state)
633 return -EINVAL;
634
Ingo Molnar217788672007-03-16 13:38:31 -0800635 spin_lock(&pi_state->pi_mutex.wait_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700636 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
637
638 /*
639 * This happens when we have stolen the lock and the original
640 * pending owner did not enqueue itself back on the rt_mutex.
641 * Thats not a tragedy. We know that way, that a lock waiter
642 * is on the fly. We make the futex_q waiter the pending owner.
643 */
644 if (!new_owner)
645 new_owner = this->task;
646
647 /*
648 * We pass it to the next owner. (The WAITERS bit is always
649 * kept enabled while there is PI state around. We must also
650 * preserve the owner died bit.)
651 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200652 if (!(uval & FUTEX_OWNER_DIED)) {
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700653 int ret = 0;
654
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200655 newval = FUTEX_WAITERS | new_owner->pid;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700656
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700657 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700658
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200659 if (curval == -EFAULT)
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700660 ret = -EFAULT;
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200661 if (curval != uval)
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700662 ret = -EINVAL;
663 if (ret) {
664 spin_unlock(&pi_state->pi_mutex.wait_lock);
665 return ret;
666 }
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200667 }
Ingo Molnarc87e2832006-06-27 02:54:58 -0700668
Ingo Molnar627371d2006-07-29 05:16:20 +0200669 spin_lock_irq(&pi_state->owner->pi_lock);
670 WARN_ON(list_empty(&pi_state->list));
671 list_del_init(&pi_state->list);
672 spin_unlock_irq(&pi_state->owner->pi_lock);
673
674 spin_lock_irq(&new_owner->pi_lock);
675 WARN_ON(!list_empty(&pi_state->list));
Ingo Molnarc87e2832006-06-27 02:54:58 -0700676 list_add(&pi_state->list, &new_owner->pi_state_list);
677 pi_state->owner = new_owner;
Ingo Molnar627371d2006-07-29 05:16:20 +0200678 spin_unlock_irq(&new_owner->pi_lock);
679
Ingo Molnar217788672007-03-16 13:38:31 -0800680 spin_unlock(&pi_state->pi_mutex.wait_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700681 rt_mutex_unlock(&pi_state->pi_mutex);
682
683 return 0;
684}
685
686static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
687{
688 u32 oldval;
689
690 /*
691 * There is no waiter, so we unlock the futex. The owner died
692 * bit has not to be preserved here. We are the owner:
693 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700694 oldval = cmpxchg_futex_value_locked(uaddr, uval, 0);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700695
696 if (oldval == -EFAULT)
697 return oldval;
698 if (oldval != uval)
699 return -EAGAIN;
700
701 return 0;
702}
703
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704/*
Ingo Molnar8b8f3192006-07-03 00:25:05 -0700705 * Express the locking dependencies for lockdep:
706 */
707static inline void
708double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
709{
710 if (hb1 <= hb2) {
711 spin_lock(&hb1->lock);
712 if (hb1 < hb2)
713 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
714 } else { /* hb1 > hb2 */
715 spin_lock(&hb2->lock);
716 spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
717 }
718}
719
720/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 * Wake up all waiters hashed on the physical page that is mapped
722 * to this virtual address:
723 */
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700724static int futex_wake(u32 __user *uaddr, struct rw_semaphore *fshared,
725 int nr_wake)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
Ingo Molnare2970f22006-06-27 02:54:47 -0700727 struct futex_hash_bucket *hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 struct futex_q *this, *next;
Pierre Peifferec92d082007-05-09 02:35:00 -0700729 struct plist_head *head;
Ingo Molnare2970f22006-06-27 02:54:47 -0700730 union futex_key key;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 int ret;
732
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700733 futex_lock_mm(fshared);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700735 ret = get_futex_key(uaddr, fshared, &key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 if (unlikely(ret != 0))
737 goto out;
738
Ingo Molnare2970f22006-06-27 02:54:47 -0700739 hb = hash_futex(&key);
740 spin_lock(&hb->lock);
741 head = &hb->chain;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742
Pierre Peifferec92d082007-05-09 02:35:00 -0700743 plist_for_each_entry_safe(this, next, head, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 if (match_futex (&this->key, &key)) {
Ingo Molnared6f7b12006-07-01 04:35:46 -0700745 if (this->pi_state) {
746 ret = -EINVAL;
747 break;
748 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 wake_futex(this);
750 if (++ret >= nr_wake)
751 break;
752 }
753 }
754
Ingo Molnare2970f22006-06-27 02:54:47 -0700755 spin_unlock(&hb->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756out:
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700757 futex_unlock_mm(fshared);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 return ret;
759}
760
761/*
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700762 * Wake up all waiters hashed on the physical page that is mapped
763 * to this virtual address:
764 */
Ingo Molnare2970f22006-06-27 02:54:47 -0700765static int
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700766futex_wake_op(u32 __user *uaddr1, struct rw_semaphore *fshared,
767 u32 __user *uaddr2,
Ingo Molnare2970f22006-06-27 02:54:47 -0700768 int nr_wake, int nr_wake2, int op)
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700769{
770 union futex_key key1, key2;
Ingo Molnare2970f22006-06-27 02:54:47 -0700771 struct futex_hash_bucket *hb1, *hb2;
Pierre Peifferec92d082007-05-09 02:35:00 -0700772 struct plist_head *head;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700773 struct futex_q *this, *next;
774 int ret, op_ret, attempt = 0;
775
776retryfull:
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700777 futex_lock_mm(fshared);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700778
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700779 ret = get_futex_key(uaddr1, fshared, &key1);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700780 if (unlikely(ret != 0))
781 goto out;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700782 ret = get_futex_key(uaddr2, fshared, &key2);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700783 if (unlikely(ret != 0))
784 goto out;
785
Ingo Molnare2970f22006-06-27 02:54:47 -0700786 hb1 = hash_futex(&key1);
787 hb2 = hash_futex(&key2);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700788
789retry:
Ingo Molnar8b8f3192006-07-03 00:25:05 -0700790 double_lock_hb(hb1, hb2);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700791
Ingo Molnare2970f22006-06-27 02:54:47 -0700792 op_ret = futex_atomic_op_inuser(op, uaddr2);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700793 if (unlikely(op_ret < 0)) {
Ingo Molnare2970f22006-06-27 02:54:47 -0700794 u32 dummy;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700795
Ingo Molnare2970f22006-06-27 02:54:47 -0700796 spin_unlock(&hb1->lock);
797 if (hb1 != hb2)
798 spin_unlock(&hb2->lock);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700799
David Howells7ee1dd32006-01-06 00:11:44 -0800800#ifndef CONFIG_MMU
Ingo Molnare2970f22006-06-27 02:54:47 -0700801 /*
802 * we don't get EFAULT from MMU faults if we don't have an MMU,
803 * but we might get them from range checking
804 */
David Howells7ee1dd32006-01-06 00:11:44 -0800805 ret = op_ret;
806 goto out;
807#endif
808
David Gibson796f8d92005-11-07 00:59:33 -0800809 if (unlikely(op_ret != -EFAULT)) {
810 ret = op_ret;
811 goto out;
812 }
813
Ingo Molnare2970f22006-06-27 02:54:47 -0700814 /*
815 * futex_atomic_op_inuser needs to both read and write
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700816 * *(int __user *)uaddr2, but we can't modify it
817 * non-atomically. Therefore, if get_user below is not
818 * enough, we need to handle the fault ourselves, while
Ingo Molnare2970f22006-06-27 02:54:47 -0700819 * still holding the mmap_sem.
820 */
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700821 if (attempt++) {
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700822 ret = futex_handle_fault((unsigned long)uaddr2,
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700823 fshared, attempt);
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700824 if (ret)
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700825 goto out;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700826 goto retry;
827 }
828
Ingo Molnare2970f22006-06-27 02:54:47 -0700829 /*
830 * If we would have faulted, release mmap_sem,
831 * fault it in and start all over again.
832 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700833 futex_unlock_mm(fshared);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700834
Ingo Molnare2970f22006-06-27 02:54:47 -0700835 ret = get_user(dummy, uaddr2);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700836 if (ret)
837 return ret;
838
839 goto retryfull;
840 }
841
Ingo Molnare2970f22006-06-27 02:54:47 -0700842 head = &hb1->chain;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700843
Pierre Peifferec92d082007-05-09 02:35:00 -0700844 plist_for_each_entry_safe(this, next, head, list) {
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700845 if (match_futex (&this->key, &key1)) {
846 wake_futex(this);
847 if (++ret >= nr_wake)
848 break;
849 }
850 }
851
852 if (op_ret > 0) {
Ingo Molnare2970f22006-06-27 02:54:47 -0700853 head = &hb2->chain;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700854
855 op_ret = 0;
Pierre Peifferec92d082007-05-09 02:35:00 -0700856 plist_for_each_entry_safe(this, next, head, list) {
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700857 if (match_futex (&this->key, &key2)) {
858 wake_futex(this);
859 if (++op_ret >= nr_wake2)
860 break;
861 }
862 }
863 ret += op_ret;
864 }
865
Ingo Molnare2970f22006-06-27 02:54:47 -0700866 spin_unlock(&hb1->lock);
867 if (hb1 != hb2)
868 spin_unlock(&hb2->lock);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700869out:
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700870 futex_unlock_mm(fshared);
871
Jakub Jelinek4732efbe2005-09-06 15:16:25 -0700872 return ret;
873}
874
875/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 * Requeue all waiters hashed on one physical page to another
877 * physical page.
878 */
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700879static int futex_requeue(u32 __user *uaddr1, struct rw_semaphore *fshared,
880 u32 __user *uaddr2,
Ingo Molnare2970f22006-06-27 02:54:47 -0700881 int nr_wake, int nr_requeue, u32 *cmpval)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882{
883 union futex_key key1, key2;
Ingo Molnare2970f22006-06-27 02:54:47 -0700884 struct futex_hash_bucket *hb1, *hb2;
Pierre Peifferec92d082007-05-09 02:35:00 -0700885 struct plist_head *head1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 struct futex_q *this, *next;
887 int ret, drop_count = 0;
888
889 retry:
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700890 futex_lock_mm(fshared);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700892 ret = get_futex_key(uaddr1, fshared, &key1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if (unlikely(ret != 0))
894 goto out;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700895 ret = get_futex_key(uaddr2, fshared, &key2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 if (unlikely(ret != 0))
897 goto out;
898
Ingo Molnare2970f22006-06-27 02:54:47 -0700899 hb1 = hash_futex(&key1);
900 hb2 = hash_futex(&key2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Ingo Molnar8b8f3192006-07-03 00:25:05 -0700902 double_lock_hb(hb1, hb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
Ingo Molnare2970f22006-06-27 02:54:47 -0700904 if (likely(cmpval != NULL)) {
905 u32 curval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
Ingo Molnare2970f22006-06-27 02:54:47 -0700907 ret = get_futex_value_locked(&curval, uaddr1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
909 if (unlikely(ret)) {
Ingo Molnare2970f22006-06-27 02:54:47 -0700910 spin_unlock(&hb1->lock);
911 if (hb1 != hb2)
912 spin_unlock(&hb2->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
Ingo Molnare2970f22006-06-27 02:54:47 -0700914 /*
915 * If we would have faulted, release mmap_sem, fault
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 * it in and start all over again.
917 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700918 futex_unlock_mm(fshared);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Ingo Molnare2970f22006-06-27 02:54:47 -0700920 ret = get_user(curval, uaddr1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
922 if (!ret)
923 goto retry;
924
925 return ret;
926 }
Ingo Molnare2970f22006-06-27 02:54:47 -0700927 if (curval != *cmpval) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 ret = -EAGAIN;
929 goto out_unlock;
930 }
931 }
932
Ingo Molnare2970f22006-06-27 02:54:47 -0700933 head1 = &hb1->chain;
Pierre Peifferec92d082007-05-09 02:35:00 -0700934 plist_for_each_entry_safe(this, next, head1, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 if (!match_futex (&this->key, &key1))
936 continue;
937 if (++ret <= nr_wake) {
938 wake_futex(this);
939 } else {
Sebastien Dugue59e0e0a2006-06-27 02:55:03 -0700940 /*
941 * If key1 and key2 hash to the same bucket, no need to
942 * requeue.
943 */
944 if (likely(head1 != &hb2->chain)) {
Pierre Peifferec92d082007-05-09 02:35:00 -0700945 plist_del(&this->list, &hb1->chain);
946 plist_add(&this->list, &hb2->chain);
Sebastien Dugue59e0e0a2006-06-27 02:55:03 -0700947 this->lock_ptr = &hb2->lock;
Pierre Peifferec92d082007-05-09 02:35:00 -0700948#ifdef CONFIG_DEBUG_PI_LIST
949 this->list.plist.lock = &hb2->lock;
950#endif
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700951 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 this->key = key2;
Rusty Russell9adef582007-05-08 00:26:42 -0700953 get_futex_key_refs(&key2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 drop_count++;
955
956 if (ret - nr_wake >= nr_requeue)
957 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 }
959 }
960
961out_unlock:
Ingo Molnare2970f22006-06-27 02:54:47 -0700962 spin_unlock(&hb1->lock);
963 if (hb1 != hb2)
964 spin_unlock(&hb2->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Rusty Russell9adef582007-05-08 00:26:42 -0700966 /* drop_futex_key_refs() must be called outside the spinlocks. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 while (--drop_count >= 0)
Rusty Russell9adef582007-05-08 00:26:42 -0700968 drop_futex_key_refs(&key1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
970out:
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700971 futex_unlock_mm(fshared);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 return ret;
973}
974
975/* The key must be already stored in q->key. */
976static inline struct futex_hash_bucket *
977queue_lock(struct futex_q *q, int fd, struct file *filp)
978{
Ingo Molnare2970f22006-06-27 02:54:47 -0700979 struct futex_hash_bucket *hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
981 q->fd = fd;
982 q->filp = filp;
983
984 init_waitqueue_head(&q->waiters);
985
Rusty Russell9adef582007-05-08 00:26:42 -0700986 get_futex_key_refs(&q->key);
Ingo Molnare2970f22006-06-27 02:54:47 -0700987 hb = hash_futex(&q->key);
988 q->lock_ptr = &hb->lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
Ingo Molnare2970f22006-06-27 02:54:47 -0700990 spin_lock(&hb->lock);
991 return hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992}
993
Ingo Molnare2970f22006-06-27 02:54:47 -0700994static inline void __queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995{
Pierre Peifferec92d082007-05-09 02:35:00 -0700996 int prio;
997
998 /*
999 * The priority used to register this element is
1000 * - either the real thread-priority for the real-time threads
1001 * (i.e. threads with a priority lower than MAX_RT_PRIO)
1002 * - or MAX_RT_PRIO for non-RT threads.
1003 * Thus, all RT-threads are woken first in priority order, and
1004 * the others are woken last, in FIFO order.
1005 */
1006 prio = min(current->normal_prio, MAX_RT_PRIO);
1007
1008 plist_node_init(&q->list, prio);
1009#ifdef CONFIG_DEBUG_PI_LIST
1010 q->list.plist.lock = &hb->lock;
1011#endif
1012 plist_add(&q->list, &hb->chain);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001013 q->task = current;
Ingo Molnare2970f22006-06-27 02:54:47 -07001014 spin_unlock(&hb->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015}
1016
1017static inline void
Ingo Molnare2970f22006-06-27 02:54:47 -07001018queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019{
Ingo Molnare2970f22006-06-27 02:54:47 -07001020 spin_unlock(&hb->lock);
Rusty Russell9adef582007-05-08 00:26:42 -07001021 drop_futex_key_refs(&q->key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022}
1023
1024/*
1025 * queue_me and unqueue_me must be called as a pair, each
1026 * exactly once. They are called with the hashed spinlock held.
1027 */
1028
1029/* The key must be already stored in q->key. */
1030static void queue_me(struct futex_q *q, int fd, struct file *filp)
1031{
Ingo Molnare2970f22006-06-27 02:54:47 -07001032 struct futex_hash_bucket *hb;
1033
1034 hb = queue_lock(q, fd, filp);
1035 __queue_me(q, hb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036}
1037
1038/* Return 1 if we were still queued (ie. 0 means we were woken) */
1039static int unqueue_me(struct futex_q *q)
1040{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 spinlock_t *lock_ptr;
Ingo Molnare2970f22006-06-27 02:54:47 -07001042 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
1044 /* In the common case we don't take the spinlock, which is nice. */
1045 retry:
1046 lock_ptr = q->lock_ptr;
Christian Borntraegere91467e2006-08-05 12:13:52 -07001047 barrier();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 if (lock_ptr != 0) {
1049 spin_lock(lock_ptr);
1050 /*
1051 * q->lock_ptr can change between reading it and
1052 * spin_lock(), causing us to take the wrong lock. This
1053 * corrects the race condition.
1054 *
1055 * Reasoning goes like this: if we have the wrong lock,
1056 * q->lock_ptr must have changed (maybe several times)
1057 * between reading it and the spin_lock(). It can
1058 * change again after the spin_lock() but only if it was
1059 * already changed before the spin_lock(). It cannot,
1060 * however, change back to the original value. Therefore
1061 * we can detect whether we acquired the correct lock.
1062 */
1063 if (unlikely(lock_ptr != q->lock_ptr)) {
1064 spin_unlock(lock_ptr);
1065 goto retry;
1066 }
Pierre Peifferec92d082007-05-09 02:35:00 -07001067 WARN_ON(plist_node_empty(&q->list));
1068 plist_del(&q->list, &q->list.plist);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001069
1070 BUG_ON(q->pi_state);
1071
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 spin_unlock(lock_ptr);
1073 ret = 1;
1074 }
1075
Rusty Russell9adef582007-05-08 00:26:42 -07001076 drop_futex_key_refs(&q->key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 return ret;
1078}
1079
Ingo Molnarc87e2832006-06-27 02:54:58 -07001080/*
1081 * PI futexes can not be requeued and must remove themself from the
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001082 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1083 * and dropped here.
Ingo Molnarc87e2832006-06-27 02:54:58 -07001084 */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001085static void unqueue_me_pi(struct futex_q *q)
Ingo Molnarc87e2832006-06-27 02:54:58 -07001086{
Pierre Peifferec92d082007-05-09 02:35:00 -07001087 WARN_ON(plist_node_empty(&q->list));
1088 plist_del(&q->list, &q->list.plist);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001089
1090 BUG_ON(!q->pi_state);
1091 free_pi_state(q->pi_state);
1092 q->pi_state = NULL;
1093
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001094 spin_unlock(q->lock_ptr);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001095
Rusty Russell9adef582007-05-08 00:26:42 -07001096 drop_futex_key_refs(&q->key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001097}
1098
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001099/*
1100 * Fixup the pi_state owner with current.
1101 *
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001102 * Must be called with hash bucket lock held and mm->sem held for non
1103 * private futexes.
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001104 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001105static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001106 struct task_struct *curr)
1107{
1108 u32 newtid = curr->pid | FUTEX_WAITERS;
1109 struct futex_pi_state *pi_state = q->pi_state;
1110 u32 uval, curval, newval;
1111 int ret;
1112
1113 /* Owner died? */
1114 if (pi_state->owner != NULL) {
1115 spin_lock_irq(&pi_state->owner->pi_lock);
1116 WARN_ON(list_empty(&pi_state->list));
1117 list_del_init(&pi_state->list);
1118 spin_unlock_irq(&pi_state->owner->pi_lock);
1119 } else
1120 newtid |= FUTEX_OWNER_DIED;
1121
1122 pi_state->owner = curr;
1123
1124 spin_lock_irq(&curr->pi_lock);
1125 WARN_ON(!list_empty(&pi_state->list));
1126 list_add(&pi_state->list, &curr->pi_state_list);
1127 spin_unlock_irq(&curr->pi_lock);
1128
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001129 /*
1130 * We own it, so we have to replace the pending owner
1131 * TID. This must be atomic as we have preserve the
1132 * owner died bit here.
1133 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001134 ret = get_futex_value_locked(&uval, uaddr);
1135
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001136 while (!ret) {
1137 newval = (uval & FUTEX_OWNER_DIED) | newtid;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001138
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001139 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001140
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001141 if (curval == -EFAULT)
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001142 ret = -EFAULT;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001143 if (curval == uval)
1144 break;
1145 uval = curval;
1146 }
1147 return ret;
1148}
1149
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001150/*
1151 * In case we must use restart_block to restart a futex_wait,
1152 * we encode in the 'arg3' shared capability
1153 */
1154#define ARG3_SHARED 1
1155
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001156static long futex_wait_restart(struct restart_block *restart);
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001157
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001158static int futex_wait(u32 __user *uaddr, struct rw_semaphore *fshared,
1159 u32 val, ktime_t *abs_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160{
Ingo Molnarc87e2832006-06-27 02:54:58 -07001161 struct task_struct *curr = current;
1162 DECLARE_WAITQUEUE(wait, curr);
Ingo Molnare2970f22006-06-27 02:54:47 -07001163 struct futex_hash_bucket *hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 struct futex_q q;
Ingo Molnare2970f22006-06-27 02:54:47 -07001165 u32 uval;
1166 int ret;
Thomas Gleixnerbd197232007-06-17 21:11:10 +02001167 struct hrtimer_sleeper t;
Pierre Peifferc19384b2007-05-09 02:35:02 -07001168 int rem = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
Ingo Molnarc87e2832006-06-27 02:54:58 -07001170 q.pi_state = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 retry:
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001172 futex_lock_mm(fshared);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001174 ret = get_futex_key(uaddr, fshared, &q.key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 if (unlikely(ret != 0))
1176 goto out_release_sem;
1177
Ingo Molnare2970f22006-06-27 02:54:47 -07001178 hb = queue_lock(&q, -1, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179
1180 /*
1181 * Access the page AFTER the futex is queued.
1182 * Order is important:
1183 *
1184 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1185 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1186 *
1187 * The basic logical guarantee of a futex is that it blocks ONLY
1188 * if cond(var) is known to be true at the time of blocking, for
1189 * any cond. If we queued after testing *uaddr, that would open
1190 * a race condition where we could block indefinitely with
1191 * cond(var) false, which would violate the guarantee.
1192 *
1193 * A consequence is that futex_wait() can return zero and absorb
1194 * a wakeup when *uaddr != val on entry to the syscall. This is
1195 * rare, but normal.
1196 *
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001197 * for shared futexes, we hold the mmap semaphore, so the mapping
1198 * cannot have changed since we looked it up in get_futex_key.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 */
Ingo Molnare2970f22006-06-27 02:54:47 -07001200 ret = get_futex_value_locked(&uval, uaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
1202 if (unlikely(ret)) {
Ingo Molnare2970f22006-06-27 02:54:47 -07001203 queue_unlock(&q, hb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
Ingo Molnare2970f22006-06-27 02:54:47 -07001205 /*
1206 * If we would have faulted, release mmap_sem, fault it in and
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 * start all over again.
1208 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001209 futex_unlock_mm(fshared);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
Ingo Molnare2970f22006-06-27 02:54:47 -07001211 ret = get_user(uval, uaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
1213 if (!ret)
1214 goto retry;
1215 return ret;
1216 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001217 ret = -EWOULDBLOCK;
1218 if (uval != val)
1219 goto out_unlock_release_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220
1221 /* Only actually queue if *uaddr contained val. */
Ingo Molnare2970f22006-06-27 02:54:47 -07001222 __queue_me(&q, hb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223
1224 /*
1225 * Now the futex is queued and we have checked the data, we
1226 * don't want to hold mmap_sem while we sleep.
Ingo Molnarc87e2832006-06-27 02:54:58 -07001227 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001228 futex_unlock_mm(fshared);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
1230 /*
1231 * There might have been scheduling since the queue_me(), as we
1232 * cannot hold a spinlock across the get_user() in case it
1233 * faults, and we cannot just set TASK_INTERRUPTIBLE state when
1234 * queueing ourselves into the futex hash. This code thus has to
1235 * rely on the futex_wake() code removing us from hash when it
1236 * wakes us up.
1237 */
1238
1239 /* add_wait_queue is the barrier after __set_current_state. */
1240 __set_current_state(TASK_INTERRUPTIBLE);
1241 add_wait_queue(&q.waiters, &wait);
1242 /*
Pierre Peifferec92d082007-05-09 02:35:00 -07001243 * !plist_node_empty() is safe here without any lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 * q.lock_ptr != 0 is not safe, because of ordering against wakeup.
1245 */
Pierre Peifferec92d082007-05-09 02:35:00 -07001246 if (likely(!plist_node_empty(&q.list))) {
Pierre Peifferc19384b2007-05-09 02:35:02 -07001247 if (!abs_time)
1248 schedule();
1249 else {
1250 hrtimer_init(&t.timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1251 hrtimer_init_sleeper(&t, current);
1252 t.timer.expires = *abs_time;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001253
Pierre Peifferc19384b2007-05-09 02:35:02 -07001254 hrtimer_start(&t.timer, t.timer.expires, HRTIMER_MODE_ABS);
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001255
Pierre Peifferc19384b2007-05-09 02:35:02 -07001256 /*
1257 * the timer could have already expired, in which
1258 * case current would be flagged for rescheduling.
1259 * Don't bother calling schedule.
1260 */
1261 if (likely(t.task))
1262 schedule();
1263
1264 hrtimer_cancel(&t.timer);
1265
1266 /* Flag if a timeout occured */
1267 rem = (t.task == NULL);
1268 }
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001269 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 __set_current_state(TASK_RUNNING);
1271
1272 /*
1273 * NOTE: we don't remove ourselves from the waitqueue because
1274 * we are the only user of it.
1275 */
1276
1277 /* If we were woken (and unqueued), we succeeded, whatever. */
1278 if (!unqueue_me(&q))
1279 return 0;
Pierre Peifferc19384b2007-05-09 02:35:02 -07001280 if (rem)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 return -ETIMEDOUT;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001282
Ingo Molnare2970f22006-06-27 02:54:47 -07001283 /*
1284 * We expect signal_pending(current), but another thread may
1285 * have handled it for us already.
1286 */
Pierre Peifferc19384b2007-05-09 02:35:02 -07001287 if (!abs_time)
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001288 return -ERESTARTSYS;
1289 else {
1290 struct restart_block *restart;
1291 restart = &current_thread_info()->restart_block;
1292 restart->fn = futex_wait_restart;
1293 restart->arg0 = (unsigned long)uaddr;
1294 restart->arg1 = (unsigned long)val;
Pierre Peifferc19384b2007-05-09 02:35:02 -07001295 restart->arg2 = (unsigned long)abs_time;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001296 restart->arg3 = 0;
1297 if (fshared)
1298 restart->arg3 |= ARG3_SHARED;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001299 return -ERESTART_RESTARTBLOCK;
1300 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
Ingo Molnarc87e2832006-06-27 02:54:58 -07001302 out_unlock_release_sem:
1303 queue_unlock(&q, hb);
1304
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 out_release_sem:
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001306 futex_unlock_mm(fshared);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001307 return ret;
1308}
1309
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001310
1311static long futex_wait_restart(struct restart_block *restart)
1312{
1313 u32 __user *uaddr = (u32 __user *)restart->arg0;
1314 u32 val = (u32)restart->arg1;
Pierre Peifferc19384b2007-05-09 02:35:02 -07001315 ktime_t *abs_time = (ktime_t *)restart->arg2;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001316 struct rw_semaphore *fshared = NULL;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001317
1318 restart->fn = do_no_restart_syscall;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001319 if (restart->arg3 & ARG3_SHARED)
1320 fshared = &current->mm->mmap_sem;
1321 return (long)futex_wait(uaddr, fshared, val, abs_time);
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001322}
1323
1324
Ingo Molnarc87e2832006-06-27 02:54:58 -07001325/*
1326 * Userspace tried a 0 -> TID atomic transition of the futex value
1327 * and failed. The kernel side here does the whole locking operation:
1328 * if there are waiters then it will block, it does PI, etc. (Due to
1329 * races the kernel might see a 0 value of the futex too.)
1330 */
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001331static int futex_lock_pi(u32 __user *uaddr, struct rw_semaphore *fshared,
1332 int detect, ktime_t *time, int trylock)
Ingo Molnarc87e2832006-06-27 02:54:58 -07001333{
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001334 struct hrtimer_sleeper timeout, *to = NULL;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001335 struct task_struct *curr = current;
1336 struct futex_hash_bucket *hb;
1337 u32 uval, newval, curval;
1338 struct futex_q q;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001339 int ret, lock_taken, ownerdied = 0, attempt = 0;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001340
1341 if (refill_pi_state_cache())
1342 return -ENOMEM;
1343
Pierre Peifferc19384b2007-05-09 02:35:02 -07001344 if (time) {
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001345 to = &timeout;
Thomas Gleixnerc9cb2e32007-02-16 01:27:49 -08001346 hrtimer_init(&to->timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001347 hrtimer_init_sleeper(to, current);
Pierre Peifferc19384b2007-05-09 02:35:02 -07001348 to->timer.expires = *time;
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001349 }
1350
Ingo Molnarc87e2832006-06-27 02:54:58 -07001351 q.pi_state = NULL;
1352 retry:
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001353 futex_lock_mm(fshared);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001354
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001355 ret = get_futex_key(uaddr, fshared, &q.key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001356 if (unlikely(ret != 0))
1357 goto out_release_sem;
1358
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001359 retry_unlocked:
Ingo Molnarc87e2832006-06-27 02:54:58 -07001360 hb = queue_lock(&q, -1, NULL);
1361
1362 retry_locked:
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001363 ret = lock_taken = 0;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001364
Ingo Molnarc87e2832006-06-27 02:54:58 -07001365 /*
1366 * To avoid races, we attempt to take the lock here again
1367 * (by doing a 0 -> TID atomic cmpxchg), while holding all
1368 * the locks. It will most likely not succeed.
1369 */
1370 newval = current->pid;
1371
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001372 curval = cmpxchg_futex_value_locked(uaddr, 0, newval);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001373
1374 if (unlikely(curval == -EFAULT))
1375 goto uaddr_faulted;
1376
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001377 /*
1378 * Detect deadlocks. In case of REQUEUE_PI this is a valid
1379 * situation and we return success to user space.
1380 */
Ingo Molnarc87e2832006-06-27 02:54:58 -07001381 if (unlikely((curval & FUTEX_TID_MASK) == current->pid)) {
Thomas Gleixnerbd197232007-06-17 21:11:10 +02001382 ret = -EDEADLK;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001383 goto out_unlock_release_sem;
1384 }
1385
1386 /*
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001387 * Surprise - we got the lock. Just return to userspace:
Ingo Molnarc87e2832006-06-27 02:54:58 -07001388 */
1389 if (unlikely(!curval))
1390 goto out_unlock_release_sem;
1391
1392 uval = curval;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001393
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001394 /*
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001395 * Set the WAITERS flag, so the owner will know it has someone
1396 * to wake at next unlock
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001397 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001398 newval = curval | FUTEX_WAITERS;
1399
1400 /*
1401 * There are two cases, where a futex might have no owner (the
Thomas Gleixnerbd197232007-06-17 21:11:10 +02001402 * owner TID is 0): OWNER_DIED. We take over the futex in this
1403 * case. We also do an unconditional take over, when the owner
1404 * of the futex died.
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001405 *
1406 * This is safe as we are protected by the hash bucket lock !
1407 */
1408 if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) {
Thomas Gleixnerbd197232007-06-17 21:11:10 +02001409 /* Keep the OWNER_DIED bit */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001410 newval = (curval & ~FUTEX_TID_MASK) | current->pid;
1411 ownerdied = 0;
1412 lock_taken = 1;
1413 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001414
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001415 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001416
1417 if (unlikely(curval == -EFAULT))
1418 goto uaddr_faulted;
1419 if (unlikely(curval != uval))
1420 goto retry_locked;
1421
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001422 /*
Thomas Gleixnerbd197232007-06-17 21:11:10 +02001423 * We took the lock due to owner died take over.
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001424 */
Thomas Gleixnerbd197232007-06-17 21:11:10 +02001425 if (unlikely(lock_taken))
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001426 goto out_unlock_release_sem;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001427
Ingo Molnarc87e2832006-06-27 02:54:58 -07001428 /*
1429 * We dont have the lock. Look up the PI state (or create it if
1430 * we are the first waiter):
1431 */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001432 ret = lookup_pi_state(uval, hb, &q.key, &q.pi_state);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001433
1434 if (unlikely(ret)) {
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001435 switch (ret) {
Ingo Molnarc87e2832006-06-27 02:54:58 -07001436
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001437 case -EAGAIN:
1438 /*
1439 * Task is exiting and we just wait for the
1440 * exit to complete.
1441 */
1442 queue_unlock(&q, hb);
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001443 futex_unlock_mm(fshared);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001444 cond_resched();
1445 goto retry;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001446
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001447 case -ESRCH:
1448 /*
1449 * No owner found for this futex. Check if the
1450 * OWNER_DIED bit is set to figure out whether
1451 * this is a robust futex or not.
1452 */
1453 if (get_futex_value_locked(&curval, uaddr))
Ingo Molnarc87e2832006-06-27 02:54:58 -07001454 goto uaddr_faulted;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001455
1456 /*
1457 * We simply start over in case of a robust
1458 * futex. The code above will take the futex
1459 * and return happy.
1460 */
1461 if (curval & FUTEX_OWNER_DIED) {
1462 ownerdied = 1;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001463 goto retry_locked;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001464 }
1465 default:
1466 goto out_unlock_release_sem;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001467 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001468 }
1469
1470 /*
1471 * Only actually queue now that the atomic ops are done:
1472 */
1473 __queue_me(&q, hb);
1474
1475 /*
1476 * Now the futex is queued and we have checked the data, we
1477 * don't want to hold mmap_sem while we sleep.
1478 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001479 futex_unlock_mm(fshared);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001480
1481 WARN_ON(!q.pi_state);
1482 /*
1483 * Block on the PI mutex:
1484 */
1485 if (!trylock)
1486 ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1);
1487 else {
1488 ret = rt_mutex_trylock(&q.pi_state->pi_mutex);
1489 /* Fixup the trylock return value: */
1490 ret = ret ? 0 : -EWOULDBLOCK;
1491 }
1492
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001493 futex_lock_mm(fshared);
Vernon Mauerya99e4e42006-07-01 04:35:42 -07001494 spin_lock(q.lock_ptr);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001495
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001496 if (!ret) {
1497 /*
1498 * Got the lock. We might not be the anticipated owner
1499 * if we did a lock-steal - fix up the PI-state in
1500 * that case:
1501 */
1502 if (q.pi_state->owner != curr)
1503 ret = fixup_pi_state_owner(uaddr, &q, curr);
1504 } else {
Ingo Molnarc87e2832006-06-27 02:54:58 -07001505 /*
1506 * Catch the rare case, where the lock was released
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001507 * when we were on the way back before we locked the
1508 * hash bucket.
Ingo Molnarc87e2832006-06-27 02:54:58 -07001509 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001510 if (q.pi_state->owner == curr &&
1511 rt_mutex_trylock(&q.pi_state->pi_mutex)) {
1512 ret = 0;
1513 } else {
1514 /*
1515 * Paranoia check. If we did not take the lock
1516 * in the trylock above, then we should not be
1517 * the owner of the rtmutex, neither the real
1518 * nor the pending one:
1519 */
1520 if (rt_mutex_owner(&q.pi_state->pi_mutex) == curr)
1521 printk(KERN_ERR "futex_lock_pi: ret = %d "
1522 "pi-mutex: %p pi-state %p\n", ret,
1523 q.pi_state->pi_mutex.owner,
1524 q.pi_state->owner);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001525 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001526 }
1527
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001528 /* Unqueue and drop the lock */
1529 unqueue_me_pi(&q);
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001530 futex_unlock_mm(fshared);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001531
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001532 return ret != -EINTR ? ret : -ERESTARTNOINTR;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001533
1534 out_unlock_release_sem:
1535 queue_unlock(&q, hb);
1536
1537 out_release_sem:
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001538 futex_unlock_mm(fshared);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001539 return ret;
1540
1541 uaddr_faulted:
1542 /*
1543 * We have to r/w *(int __user *)uaddr, but we can't modify it
1544 * non-atomically. Therefore, if get_user below is not
1545 * enough, we need to handle the fault ourselves, while
1546 * still holding the mmap_sem.
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001547 *
1548 * ... and hb->lock. :-) --ANK
Ingo Molnarc87e2832006-06-27 02:54:58 -07001549 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001550 queue_unlock(&q, hb);
1551
Ingo Molnarc87e2832006-06-27 02:54:58 -07001552 if (attempt++) {
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001553 ret = futex_handle_fault((unsigned long)uaddr, fshared,
1554 attempt);
1555 if (ret)
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001556 goto out_release_sem;
1557 goto retry_unlocked;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001558 }
1559
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001560 futex_unlock_mm(fshared);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001561
1562 ret = get_user(uval, uaddr);
1563 if (!ret && (uval != -EFAULT))
1564 goto retry;
1565
1566 return ret;
1567}
1568
1569/*
Ingo Molnarc87e2832006-06-27 02:54:58 -07001570 * Userspace attempted a TID -> 0 atomic transition, and failed.
1571 * This is the in-kernel slowpath: we look up the PI state (if any),
1572 * and do the rt-mutex unlock.
1573 */
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001574static int futex_unlock_pi(u32 __user *uaddr, struct rw_semaphore *fshared)
Ingo Molnarc87e2832006-06-27 02:54:58 -07001575{
1576 struct futex_hash_bucket *hb;
1577 struct futex_q *this, *next;
1578 u32 uval;
Pierre Peifferec92d082007-05-09 02:35:00 -07001579 struct plist_head *head;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001580 union futex_key key;
1581 int ret, attempt = 0;
1582
1583retry:
1584 if (get_user(uval, uaddr))
1585 return -EFAULT;
1586 /*
1587 * We release only a lock we actually own:
1588 */
1589 if ((uval & FUTEX_TID_MASK) != current->pid)
1590 return -EPERM;
1591 /*
1592 * First take all the futex related locks:
1593 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001594 futex_lock_mm(fshared);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001595
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001596 ret = get_futex_key(uaddr, fshared, &key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001597 if (unlikely(ret != 0))
1598 goto out;
1599
1600 hb = hash_futex(&key);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001601retry_unlocked:
Ingo Molnarc87e2832006-06-27 02:54:58 -07001602 spin_lock(&hb->lock);
1603
Ingo Molnarc87e2832006-06-27 02:54:58 -07001604 /*
1605 * To avoid races, try to do the TID -> 0 atomic transition
1606 * again. If it succeeds then we can return without waking
1607 * anyone else up:
1608 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001609 if (!(uval & FUTEX_OWNER_DIED))
1610 uval = cmpxchg_futex_value_locked(uaddr, current->pid, 0);
1611
Ingo Molnarc87e2832006-06-27 02:54:58 -07001612
1613 if (unlikely(uval == -EFAULT))
1614 goto pi_faulted;
1615 /*
1616 * Rare case: we managed to release the lock atomically,
1617 * no need to wake anyone else up:
1618 */
1619 if (unlikely(uval == current->pid))
1620 goto out_unlock;
1621
1622 /*
1623 * Ok, other tasks may need to be woken up - check waiters
1624 * and do the wakeup if necessary:
1625 */
1626 head = &hb->chain;
1627
Pierre Peifferec92d082007-05-09 02:35:00 -07001628 plist_for_each_entry_safe(this, next, head, list) {
Ingo Molnarc87e2832006-06-27 02:54:58 -07001629 if (!match_futex (&this->key, &key))
1630 continue;
1631 ret = wake_futex_pi(uaddr, uval, this);
1632 /*
1633 * The atomic access to the futex value
1634 * generated a pagefault, so retry the
1635 * user-access and the wakeup:
1636 */
1637 if (ret == -EFAULT)
1638 goto pi_faulted;
1639 goto out_unlock;
1640 }
1641 /*
1642 * No waiters - kernel unlocks the futex:
1643 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001644 if (!(uval & FUTEX_OWNER_DIED)) {
1645 ret = unlock_futex_pi(uaddr, uval);
1646 if (ret == -EFAULT)
1647 goto pi_faulted;
1648 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001649
1650out_unlock:
1651 spin_unlock(&hb->lock);
1652out:
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001653 futex_unlock_mm(fshared);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001654
1655 return ret;
1656
1657pi_faulted:
1658 /*
1659 * We have to r/w *(int __user *)uaddr, but we can't modify it
1660 * non-atomically. Therefore, if get_user below is not
1661 * enough, we need to handle the fault ourselves, while
1662 * still holding the mmap_sem.
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001663 *
1664 * ... and hb->lock. --ANK
Ingo Molnarc87e2832006-06-27 02:54:58 -07001665 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001666 spin_unlock(&hb->lock);
1667
Ingo Molnarc87e2832006-06-27 02:54:58 -07001668 if (attempt++) {
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001669 ret = futex_handle_fault((unsigned long)uaddr, fshared,
1670 attempt);
1671 if (ret)
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001672 goto out;
john stultz187226f2007-08-22 14:01:10 -07001673 uval = 0;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001674 goto retry_unlocked;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001675 }
1676
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001677 futex_unlock_mm(fshared);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001678
1679 ret = get_user(uval, uaddr);
1680 if (!ret && (uval != -EFAULT))
1681 goto retry;
1682
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 return ret;
1684}
1685
1686static int futex_close(struct inode *inode, struct file *filp)
1687{
1688 struct futex_q *q = filp->private_data;
1689
1690 unqueue_me(q);
1691 kfree(q);
Ingo Molnare2970f22006-06-27 02:54:47 -07001692
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 return 0;
1694}
1695
1696/* This is one-shot: once it's gone off you need a new fd */
1697static unsigned int futex_poll(struct file *filp,
1698 struct poll_table_struct *wait)
1699{
1700 struct futex_q *q = filp->private_data;
1701 int ret = 0;
1702
1703 poll_wait(filp, &q->waiters, wait);
1704
1705 /*
Pierre Peifferec92d082007-05-09 02:35:00 -07001706 * plist_node_empty() is safe here without any lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 * q->lock_ptr != 0 is not safe, because of ordering against wakeup.
1708 */
Pierre Peifferec92d082007-05-09 02:35:00 -07001709 if (plist_node_empty(&q->list))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 ret = POLLIN | POLLRDNORM;
1711
1712 return ret;
1713}
1714
Helge Deller15ad7cd2006-12-06 20:40:36 -08001715static const struct file_operations futex_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 .release = futex_close,
1717 .poll = futex_poll,
1718};
1719
1720/*
1721 * Signal allows caller to avoid the race which would occur if they
1722 * set the sigio stuff up afterwards.
1723 */
Ingo Molnare2970f22006-06-27 02:54:47 -07001724static int futex_fd(u32 __user *uaddr, int signal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725{
1726 struct futex_q *q;
1727 struct file *filp;
1728 int ret, err;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001729 struct rw_semaphore *fshared;
Andrew Morton19c6b6e2006-11-02 22:07:17 -08001730 static unsigned long printk_interval;
1731
1732 if (printk_timed_ratelimit(&printk_interval, 60 * 60 * 1000)) {
1733 printk(KERN_WARNING "Process `%s' used FUTEX_FD, which "
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001734 "will be removed from the kernel in June 2007\n",
1735 current->comm);
Andrew Morton19c6b6e2006-11-02 22:07:17 -08001736 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
1738 ret = -EINVAL;
Jesper Juhl7ed20e12005-05-01 08:59:14 -07001739 if (!valid_signal(signal))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 goto out;
1741
1742 ret = get_unused_fd();
1743 if (ret < 0)
1744 goto out;
1745 filp = get_empty_filp();
1746 if (!filp) {
1747 put_unused_fd(ret);
1748 ret = -ENFILE;
1749 goto out;
1750 }
1751 filp->f_op = &futex_fops;
Josef "Jeff" Sipekf3a43f32006-12-08 02:36:43 -08001752 filp->f_path.mnt = mntget(futex_mnt);
1753 filp->f_path.dentry = dget(futex_mnt->mnt_root);
1754 filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755
1756 if (signal) {
Eric W. Biederman609d7fa2006-10-02 02:17:15 -07001757 err = __f_setown(filp, task_pid(current), PIDTYPE_PID, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 if (err < 0) {
Pekka Enberg39ed3fd2005-09-06 15:17:44 -07001759 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 }
1761 filp->f_owner.signum = signal;
1762 }
1763
1764 q = kmalloc(sizeof(*q), GFP_KERNEL);
1765 if (!q) {
Pekka Enberg39ed3fd2005-09-06 15:17:44 -07001766 err = -ENOMEM;
1767 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001769 q->pi_state = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001771 fshared = &current->mm->mmap_sem;
1772 down_read(fshared);
1773 err = get_futex_key(uaddr, fshared, &q->key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
1775 if (unlikely(err != 0)) {
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001776 up_read(fshared);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 kfree(q);
Pekka Enberg39ed3fd2005-09-06 15:17:44 -07001778 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 }
1780
1781 /*
1782 * queue_me() must be called before releasing mmap_sem, because
1783 * key->shared.inode needs to be referenced while holding it.
1784 */
1785 filp->private_data = q;
1786
1787 queue_me(q, ret, filp);
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001788 up_read(fshared);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
1790 /* Now we map fd to filp, so userspace can access it */
1791 fd_install(ret, filp);
1792out:
1793 return ret;
Pekka Enberg39ed3fd2005-09-06 15:17:44 -07001794error:
1795 put_unused_fd(ret);
1796 put_filp(filp);
1797 ret = err;
1798 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799}
1800
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001801/*
1802 * Support for robust futexes: the kernel cleans up held futexes at
1803 * thread exit time.
1804 *
1805 * Implementation: user-space maintains a per-thread list of locks it
1806 * is holding. Upon do_exit(), the kernel carefully walks this list,
1807 * and marks all locks that are owned by this thread with the
Ingo Molnarc87e2832006-06-27 02:54:58 -07001808 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001809 * always manipulated with the lock held, so the list is private and
1810 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
1811 * field, to allow the kernel to clean up if the thread dies after
1812 * acquiring the lock, but just before it could have added itself to
1813 * the list. There can only be one such pending lock.
1814 */
1815
1816/**
1817 * sys_set_robust_list - set the robust-futex list head of a task
1818 * @head: pointer to the list-head
1819 * @len: length of the list-head, as userspace expects
1820 */
1821asmlinkage long
1822sys_set_robust_list(struct robust_list_head __user *head,
1823 size_t len)
1824{
1825 /*
1826 * The kernel knows only one size for now:
1827 */
1828 if (unlikely(len != sizeof(*head)))
1829 return -EINVAL;
1830
1831 current->robust_list = head;
1832
1833 return 0;
1834}
1835
1836/**
1837 * sys_get_robust_list - get the robust-futex list head of a task
1838 * @pid: pid of the process [zero for current task]
1839 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
1840 * @len_ptr: pointer to a length field, the kernel fills in the header size
1841 */
1842asmlinkage long
Al Viroba46df92006-10-10 22:46:07 +01001843sys_get_robust_list(int pid, struct robust_list_head __user * __user *head_ptr,
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001844 size_t __user *len_ptr)
1845{
Al Viroba46df92006-10-10 22:46:07 +01001846 struct robust_list_head __user *head;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001847 unsigned long ret;
1848
1849 if (!pid)
1850 head = current->robust_list;
1851 else {
1852 struct task_struct *p;
1853
1854 ret = -ESRCH;
Oleg Nesterovaaa2a972006-09-29 02:00:55 -07001855 rcu_read_lock();
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001856 p = find_task_by_pid(pid);
1857 if (!p)
1858 goto err_unlock;
1859 ret = -EPERM;
1860 if ((current->euid != p->euid) && (current->euid != p->uid) &&
1861 !capable(CAP_SYS_PTRACE))
1862 goto err_unlock;
1863 head = p->robust_list;
Oleg Nesterovaaa2a972006-09-29 02:00:55 -07001864 rcu_read_unlock();
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001865 }
1866
1867 if (put_user(sizeof(*head), len_ptr))
1868 return -EFAULT;
1869 return put_user(head, head_ptr);
1870
1871err_unlock:
Oleg Nesterovaaa2a972006-09-29 02:00:55 -07001872 rcu_read_unlock();
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001873
1874 return ret;
1875}
1876
1877/*
1878 * Process a futex-list entry, check whether it's owned by the
1879 * dying task, and do notification if so:
1880 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001881int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001882{
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001883 u32 uval, nval, mval;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001884
Ingo Molnar8f17d3a2006-03-27 01:16:27 -08001885retry:
1886 if (get_user(uval, uaddr))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001887 return -1;
1888
Ingo Molnar8f17d3a2006-03-27 01:16:27 -08001889 if ((uval & FUTEX_TID_MASK) == curr->pid) {
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001890 /*
1891 * Ok, this dying thread is truly holding a futex
1892 * of interest. Set the OWNER_DIED bit atomically
1893 * via cmpxchg, and if the value had FUTEX_WAITERS
1894 * set, wake up a waiter (if any). (We have to do a
1895 * futex_wake() even if OWNER_DIED is already set -
1896 * to handle the rare but possible case of recursive
1897 * thread-death.) The rest of the cleanup is done in
1898 * userspace.
1899 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001900 mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
1901 nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval);
1902
Ingo Molnarc87e2832006-06-27 02:54:58 -07001903 if (nval == -EFAULT)
1904 return -1;
1905
1906 if (nval != uval)
Ingo Molnar8f17d3a2006-03-27 01:16:27 -08001907 goto retry;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001908
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001909 /*
1910 * Wake robust non-PI futexes here. The wakeup of
1911 * PI futexes happens in exit_pi_state():
1912 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001913 if (!pi && (uval & FUTEX_WAITERS))
Eric Dumazet34f01cc2007-05-09 02:35:04 -07001914 futex_wake(uaddr, &curr->mm->mmap_sem, 1);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001915 }
1916 return 0;
1917}
1918
1919/*
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001920 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
1921 */
1922static inline int fetch_robust_entry(struct robust_list __user **entry,
Al Viroba46df92006-10-10 22:46:07 +01001923 struct robust_list __user * __user *head,
1924 int *pi)
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001925{
1926 unsigned long uentry;
1927
Al Viroba46df92006-10-10 22:46:07 +01001928 if (get_user(uentry, (unsigned long __user *)head))
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001929 return -EFAULT;
1930
Al Viroba46df92006-10-10 22:46:07 +01001931 *entry = (void __user *)(uentry & ~1UL);
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001932 *pi = uentry & 1;
1933
1934 return 0;
1935}
1936
1937/*
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001938 * Walk curr->robust_list (very carefully, it's a userspace list!)
1939 * and mark any locks found there dead, and notify any waiters.
1940 *
1941 * We silently return on any sign of list-walking problem.
1942 */
1943void exit_robust_list(struct task_struct *curr)
1944{
1945 struct robust_list_head __user *head = curr->robust_list;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001946 struct robust_list __user *entry, *next_entry, *pending;
1947 unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001948 unsigned long futex_offset;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001949 int rc;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001950
1951 /*
1952 * Fetch the list head (which was registered earlier, via
1953 * sys_set_robust_list()):
1954 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001955 if (fetch_robust_entry(&entry, &head->list.next, &pi))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001956 return;
1957 /*
1958 * Fetch the relative futex offset:
1959 */
1960 if (get_user(futex_offset, &head->futex_offset))
1961 return;
1962 /*
1963 * Fetch any possibly pending lock-add first, and handle it
1964 * if it exists:
1965 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001966 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001967 return;
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001968
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001969 next_entry = NULL; /* avoid warning with gcc */
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001970 while (entry != &head->list) {
1971 /*
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001972 * Fetch the next entry in the list before calling
1973 * handle_futex_death:
1974 */
1975 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
1976 /*
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001977 * A pending lock might already be on the list, so
Ingo Molnarc87e2832006-06-27 02:54:58 -07001978 * don't process it twice:
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001979 */
1980 if (entry != pending)
Al Viroba46df92006-10-10 22:46:07 +01001981 if (handle_futex_death((void __user *)entry + futex_offset,
Ingo Molnare3f2dde2006-07-29 05:17:57 +02001982 curr, pi))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001983 return;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001984 if (rc)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001985 return;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001986 entry = next_entry;
1987 pi = next_pi;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08001988 /*
1989 * Avoid excessively long or circular lists:
1990 */
1991 if (!--limit)
1992 break;
1993
1994 cond_resched();
1995 }
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07001996
1997 if (pending)
1998 handle_futex_death((void __user *)pending + futex_offset,
1999 curr, pip);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002000}
2001
Pierre Peifferc19384b2007-05-09 02:35:02 -07002002long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
Ingo Molnare2970f22006-06-27 02:54:47 -07002003 u32 __user *uaddr2, u32 val2, u32 val3)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004{
2005 int ret;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002006 int cmd = op & FUTEX_CMD_MASK;
2007 struct rw_semaphore *fshared = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002009 if (!(op & FUTEX_PRIVATE_FLAG))
2010 fshared = &current->mm->mmap_sem;
2011
2012 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 case FUTEX_WAIT:
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002014 ret = futex_wait(uaddr, fshared, val, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 break;
2016 case FUTEX_WAKE:
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002017 ret = futex_wake(uaddr, fshared, val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 break;
2019 case FUTEX_FD:
2020 /* non-zero val means F_SETOWN(getpid()) & F_SETSIG(val) */
2021 ret = futex_fd(uaddr, val);
2022 break;
2023 case FUTEX_REQUEUE:
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002024 ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 break;
2026 case FUTEX_CMP_REQUEUE:
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002027 ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 break;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07002029 case FUTEX_WAKE_OP:
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002030 ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07002031 break;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002032 case FUTEX_LOCK_PI:
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002033 ret = futex_lock_pi(uaddr, fshared, val, timeout, 0);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002034 break;
2035 case FUTEX_UNLOCK_PI:
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002036 ret = futex_unlock_pi(uaddr, fshared);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002037 break;
2038 case FUTEX_TRYLOCK_PI:
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002039 ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002040 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 default:
2042 ret = -ENOSYS;
2043 }
2044 return ret;
2045}
2046
2047
Ingo Molnare2970f22006-06-27 02:54:47 -07002048asmlinkage long sys_futex(u32 __user *uaddr, int op, u32 val,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 struct timespec __user *utime, u32 __user *uaddr2,
Ingo Molnare2970f22006-06-27 02:54:47 -07002050 u32 val3)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051{
Pierre Peifferc19384b2007-05-09 02:35:02 -07002052 struct timespec ts;
2053 ktime_t t, *tp = NULL;
Ingo Molnare2970f22006-06-27 02:54:47 -07002054 u32 val2 = 0;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002055 int cmd = op & FUTEX_CMD_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002057 if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI)) {
Pierre Peifferc19384b2007-05-09 02:35:02 -07002058 if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 return -EFAULT;
Pierre Peifferc19384b2007-05-09 02:35:02 -07002060 if (!timespec_valid(&ts))
Thomas Gleixner9741ef962006-03-31 02:31:32 -08002061 return -EINVAL;
Pierre Peifferc19384b2007-05-09 02:35:02 -07002062
2063 t = timespec_to_ktime(ts);
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002064 if (cmd == FUTEX_WAIT)
Pierre Peifferc19384b2007-05-09 02:35:02 -07002065 t = ktime_add(ktime_get(), t);
2066 tp = &t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 }
2068 /*
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002069 * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE.
Andreas Schwabf54f0982007-07-31 00:38:51 -07002070 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 */
Andreas Schwabf54f0982007-07-31 00:38:51 -07002072 if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
2073 cmd == FUTEX_WAKE_OP)
Ingo Molnare2970f22006-06-27 02:54:47 -07002074 val2 = (u32) (unsigned long) utime;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075
Pierre Peifferc19384b2007-05-09 02:35:02 -07002076 return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077}
2078
David Howells454e2392006-06-23 02:02:57 -07002079static int futexfs_get_sb(struct file_system_type *fs_type,
2080 int flags, const char *dev_name, void *data,
2081 struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082{
David Howells454e2392006-06-23 02:02:57 -07002083 return get_sb_pseudo(fs_type, "futex", NULL, 0xBAD1DEA, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084}
2085
2086static struct file_system_type futex_fs_type = {
2087 .name = "futexfs",
2088 .get_sb = futexfs_get_sb,
2089 .kill_sb = kill_anon_super,
2090};
2091
2092static int __init init(void)
2093{
Akinobu Mita95362fa2006-12-06 20:39:03 -08002094 int i = register_filesystem(&futex_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095
Akinobu Mita95362fa2006-12-06 20:39:03 -08002096 if (i)
2097 return i;
2098
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 futex_mnt = kern_mount(&futex_fs_type);
Akinobu Mita95362fa2006-12-06 20:39:03 -08002100 if (IS_ERR(futex_mnt)) {
2101 unregister_filesystem(&futex_fs_type);
2102 return PTR_ERR(futex_mnt);
2103 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
2105 for (i = 0; i < ARRAY_SIZE(futex_queues); i++) {
Pierre Peifferec92d082007-05-09 02:35:00 -07002106 plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 spin_lock_init(&futex_queues[i].lock);
2108 }
2109 return 0;
2110}
2111__initcall(init);