blob: e2b0fb9a0b3b3d0d5871ee37bf08618a6bf052e7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Fast Userspace Mutexes (which I call "Futexes!").
3 * (C) Rusty Russell, IBM 2002
4 *
5 * Generalized futexes, futex requeueing, misc fixes by Ingo Molnar
6 * (C) Copyright 2003 Red Hat Inc, All Rights Reserved
7 *
8 * Removed page pinning, fix privately mapped COW pages and other cleanups
9 * (C) Copyright 2003, 2004 Jamie Lokier
10 *
Ingo Molnar0771dfe2006-03-27 01:16:22 -080011 * Robust futex support started by Ingo Molnar
12 * (C) Copyright 2006 Red Hat Inc, All Rights Reserved
13 * Thanks to Thomas Gleixner for suggestions, analysis and fixes.
14 *
Ingo Molnarc87e2832006-06-27 02:54:58 -070015 * PI-futex support started by Ingo Molnar and Thomas Gleixner
16 * Copyright (C) 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
17 * Copyright (C) 2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
18 *
Eric Dumazet34f01cc2007-05-09 02:35:04 -070019 * PRIVATE futexes by Eric Dumazet
20 * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com>
21 *
Darren Hart52400ba2009-04-03 13:40:49 -070022 * Requeue-PI support by Darren Hart <dvhltc@us.ibm.com>
23 * Copyright (C) IBM Corporation, 2009
24 * Thanks to Thomas Gleixner for conceptual design and careful reviews.
25 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070026 * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly
27 * enough at me, Linus for the original (flawed) idea, Matthew
28 * Kirkwood for proof-of-concept implementation.
29 *
30 * "The futexes are also cursed."
31 * "But they come in a choice of three flavours!"
32 *
33 * This program is free software; you can redistribute it and/or modify
34 * it under the terms of the GNU General Public License as published by
35 * the Free Software Foundation; either version 2 of the License, or
36 * (at your option) any later version.
37 *
38 * This program is distributed in the hope that it will be useful,
39 * but WITHOUT ANY WARRANTY; without even the implied warranty of
40 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
41 * GNU General Public License for more details.
42 *
43 * You should have received a copy of the GNU General Public License
44 * along with this program; if not, write to the Free Software
45 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
46 */
47#include <linux/slab.h>
48#include <linux/poll.h>
49#include <linux/fs.h>
50#include <linux/file.h>
51#include <linux/jhash.h>
52#include <linux/init.h>
53#include <linux/futex.h>
54#include <linux/mount.h>
55#include <linux/pagemap.h>
56#include <linux/syscalls.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070057#include <linux/signal.h>
Paul Gortmaker9984de12011-05-23 14:51:41 -040058#include <linux/export.h>
Andrey Mirkinfd5eea42007-10-16 23:30:13 -070059#include <linux/magic.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070060#include <linux/pid.h>
61#include <linux/nsproxy.h>
Kees Cookbdbb7762012-03-19 16:12:53 -070062#include <linux/ptrace.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070063
Jakub Jelinek4732efbe2005-09-06 15:16:25 -070064#include <asm/futex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Ingo Molnarc87e2832006-06-27 02:54:58 -070066#include "rtmutex_common.h"
67
Thomas Gleixnera0c1e902008-02-23 15:23:57 -080068int __read_mostly futex_cmpxchg_enabled;
69
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
71
72/*
Darren Hartb41277d2010-11-08 13:10:09 -080073 * Futex flags used to encode options to functions and preserve them across
74 * restarts.
75 */
76#define FLAGS_SHARED 0x01
77#define FLAGS_CLOCKRT 0x02
78#define FLAGS_HAS_TIMEOUT 0x04
79
80/*
Ingo Molnarc87e2832006-06-27 02:54:58 -070081 * Priority Inheritance state:
82 */
83struct futex_pi_state {
84 /*
85 * list of 'owned' pi_state instances - these have to be
86 * cleaned up in do_exit() if the task exits prematurely:
87 */
88 struct list_head list;
89
90 /*
91 * The PI object:
92 */
93 struct rt_mutex pi_mutex;
94
95 struct task_struct *owner;
96 atomic_t refcount;
97
98 union futex_key key;
99};
100
Darren Hartd8d88fb2009-09-21 22:30:30 -0700101/**
102 * struct futex_q - The hashed futex queue entry, one per waiting task
Randy Dunlapfb62db22010-10-13 11:02:34 -0700103 * @list: priority-sorted list of tasks waiting on this futex
Darren Hartd8d88fb2009-09-21 22:30:30 -0700104 * @task: the task waiting on the futex
105 * @lock_ptr: the hash bucket lock
106 * @key: the key the futex is hashed on
107 * @pi_state: optional priority inheritance state
108 * @rt_waiter: rt_waiter storage for use with requeue_pi
109 * @requeue_pi_key: the requeue_pi target futex key
110 * @bitset: bitset for the optional bitmasked wakeup
111 *
112 * We use this hashed waitqueue, instead of a normal wait_queue_t, so
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 * we can wake only the relevant ones (hashed queues may be shared).
114 *
115 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
Pierre Peifferec92d082007-05-09 02:35:00 -0700116 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
Randy Dunlapfb62db22010-10-13 11:02:34 -0700117 * The order of wakeup is always to make the first condition true, then
Darren Hartd8d88fb2009-09-21 22:30:30 -0700118 * the second.
119 *
120 * PI futexes are typically woken before they are removed from the hash list via
121 * the rt_mutex code. See unqueue_me_pi().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 */
123struct futex_q {
Pierre Peifferec92d082007-05-09 02:35:00 -0700124 struct plist_node list;
Darren Hartd8d88fb2009-09-21 22:30:30 -0700125
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200126 struct task_struct *task;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 spinlock_t *lock_ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 union futex_key key;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700129 struct futex_pi_state *pi_state;
Darren Hart52400ba2009-04-03 13:40:49 -0700130 struct rt_mutex_waiter *rt_waiter;
Darren Hart84bc4af2009-08-13 17:36:53 -0700131 union futex_key *requeue_pi_key;
Thomas Gleixnercd689982008-02-01 17:45:14 +0100132 u32 bitset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133};
134
Darren Hart5bdb05f2010-11-08 13:40:28 -0800135static const struct futex_q futex_q_init = {
136 /* list gets initialized in queue_me()*/
137 .key = FUTEX_KEY_INIT,
138 .bitset = FUTEX_BITSET_MATCH_ANY
139};
140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141/*
Darren Hartb2d09942009-03-12 00:55:37 -0700142 * Hash buckets are shared by all the futex_keys that hash to the same
143 * location. Each key may have multiple futex_q structures, one for each task
144 * waiting on a futex.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 */
146struct futex_hash_bucket {
Pierre Peifferec92d082007-05-09 02:35:00 -0700147 spinlock_t lock;
148 struct plist_head chain;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149};
150
151static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS];
152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153/*
154 * We hash on the keys returned from get_futex_key (see below).
155 */
156static struct futex_hash_bucket *hash_futex(union futex_key *key)
157{
158 u32 hash = jhash2((u32*)&key->both.word,
159 (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
160 key->both.offset);
161 return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)];
162}
163
164/*
165 * Return 1 if two futex_keys are equal, 0 otherwise.
166 */
167static inline int match_futex(union futex_key *key1, union futex_key *key2)
168{
Darren Hart2bc87202009-10-14 10:12:39 -0700169 return (key1 && key2
170 && key1->both.word == key2->both.word
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 && key1->both.ptr == key2->both.ptr
172 && key1->both.offset == key2->both.offset);
173}
174
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200175/*
176 * Take a reference to the resource addressed by a key.
177 * Can be called while holding spinlocks.
178 *
179 */
180static void get_futex_key_refs(union futex_key *key)
181{
182 if (!key->both.ptr)
183 return;
184
185 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
186 case FUT_OFF_INODE:
Al Viro7de9c6ee2010-10-23 11:11:40 -0400187 ihold(key->shared.inode);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200188 break;
189 case FUT_OFF_MMSHARED:
190 atomic_inc(&key->private.mm->mm_count);
191 break;
192 }
193}
194
195/*
196 * Drop a reference to the resource addressed by a key.
197 * The hash bucket spinlock must not be held.
198 */
199static void drop_futex_key_refs(union futex_key *key)
200{
Darren Hart90621c42008-12-29 19:43:21 -0800201 if (!key->both.ptr) {
202 /* If we're here then we tried to put a key we failed to get */
203 WARN_ON_ONCE(1);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200204 return;
Darren Hart90621c42008-12-29 19:43:21 -0800205 }
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200206
207 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
208 case FUT_OFF_INODE:
209 iput(key->shared.inode);
210 break;
211 case FUT_OFF_MMSHARED:
212 mmdrop(key->private.mm);
213 break;
214 }
215}
216
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700217/**
Darren Hartd96ee562009-09-21 22:30:22 -0700218 * get_futex_key() - Get parameters which are the keys for a futex
219 * @uaddr: virtual address of the futex
220 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
221 * @key: address where result is stored.
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500222 * @rw: mapping needs to be read/write (values: VERIFY_READ,
223 * VERIFY_WRITE)
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700224 *
225 * Returns a negative error code or 0
226 * The key words are stored in *key on success.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 *
Josef "Jeff" Sipekf3a43f32006-12-08 02:36:43 -0800228 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 * offset_within_page). For private mappings, it's (uaddr, current->mm).
230 * We can usually work out the index without swapping in the page.
231 *
Darren Hartb2d09942009-03-12 00:55:37 -0700232 * lock_page() might sleep, the caller should not hold a spinlock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 */
Thomas Gleixner64d13042009-05-18 21:20:10 +0200234static int
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500235get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236{
Ingo Molnare2970f22006-06-27 02:54:47 -0700237 unsigned long address = (unsigned long)uaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 struct mm_struct *mm = current->mm;
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800239 struct page *page, *page_head;
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500240 int err, ro = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
242 /*
243 * The futex address must be "naturally" aligned.
244 */
Ingo Molnare2970f22006-06-27 02:54:47 -0700245 key->both.offset = address % PAGE_SIZE;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700246 if (unlikely((address % sizeof(u32)) != 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 return -EINVAL;
Ingo Molnare2970f22006-06-27 02:54:47 -0700248 address -= key->both.offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250 /*
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700251 * PROCESS_PRIVATE futexes are fast.
252 * As the mm cannot disappear under us and the 'key' only needs
253 * virtual address, we dont even have to find the underlying vma.
254 * Note : We do have to check 'uaddr' is a valid user address,
255 * but access_ok() should be faster than find_vma()
256 */
257 if (!fshared) {
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +0900258 if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))))
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700259 return -EFAULT;
260 key->private.mm = mm;
261 key->private.address = address;
Peter Zijlstra42569c32008-09-30 12:33:07 +0200262 get_futex_key_refs(key);
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700263 return 0;
264 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200266again:
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +0900267 err = get_user_pages_fast(address, 1, 1, &page);
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500268 /*
269 * If write access is not required (eg. FUTEX_WAIT), try
270 * and get read-only access.
271 */
272 if (err == -EFAULT && rw == VERIFY_READ) {
273 err = get_user_pages_fast(address, 1, 0, &page);
274 ro = 1;
275 }
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200276 if (err < 0)
277 return err;
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500278 else
279 err = 0;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200280
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800281#ifdef CONFIG_TRANSPARENT_HUGEPAGE
282 page_head = page;
283 if (unlikely(PageTail(page))) {
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200284 put_page(page);
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800285 /* serialize against __split_huge_page_splitting() */
286 local_irq_disable();
287 if (likely(__get_user_pages_fast(address, 1, 1, &page) == 1)) {
288 page_head = compound_head(page);
289 /*
290 * page_head is valid pointer but we must pin
291 * it before taking the PG_lock and/or
292 * PG_compound_lock. The moment we re-enable
293 * irqs __split_huge_page_splitting() can
294 * return and the head page can be freed from
295 * under us. We can't take the PG_lock and/or
296 * PG_compound_lock on a page that could be
297 * freed from under us.
298 */
299 if (page != page_head) {
300 get_page(page_head);
301 put_page(page);
302 }
303 local_irq_enable();
304 } else {
305 local_irq_enable();
306 goto again;
307 }
308 }
309#else
310 page_head = compound_head(page);
311 if (page != page_head) {
312 get_page(page_head);
313 put_page(page);
314 }
315#endif
316
317 lock_page(page_head);
Hugh Dickinse6780f72011-12-31 11:44:01 -0800318
319 /*
320 * If page_head->mapping is NULL, then it cannot be a PageAnon
321 * page; but it might be the ZERO_PAGE or in the gate area or
322 * in a special mapping (all cases which we are happy to fail);
323 * or it may have been a good file page when get_user_pages_fast
324 * found it, but truncated or holepunched or subjected to
325 * invalidate_complete_page2 before we got the page lock (also
326 * cases which we are happy to fail). And we hold a reference,
327 * so refcount care in invalidate_complete_page's remove_mapping
328 * prevents drop_caches from setting mapping to NULL beneath us.
329 *
330 * The case we do have to guard against is when memory pressure made
331 * shmem_writepage move it from filecache to swapcache beneath us:
332 * an unlikely race, but we do need to retry for page_head->mapping.
333 */
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800334 if (!page_head->mapping) {
Hugh Dickinse6780f72011-12-31 11:44:01 -0800335 int shmem_swizzled = PageSwapCache(page_head);
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800336 unlock_page(page_head);
337 put_page(page_head);
Hugh Dickinse6780f72011-12-31 11:44:01 -0800338 if (shmem_swizzled)
339 goto again;
340 return -EFAULT;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200341 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
343 /*
344 * Private mappings are handled in a simple way.
345 *
346 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
347 * it's a read-only handle, it's expected that futexes attach to
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200348 * the object not the particular process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 */
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800350 if (PageAnon(page_head)) {
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500351 /*
352 * A RO anonymous page will never change and thus doesn't make
353 * sense for futex operations.
354 */
355 if (ro) {
356 err = -EFAULT;
357 goto out;
358 }
359
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200360 key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 key->private.mm = mm;
Ingo Molnare2970f22006-06-27 02:54:47 -0700362 key->private.address = address;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200363 } else {
364 key->both.offset |= FUT_OFF_INODE; /* inode-based key */
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800365 key->shared.inode = page_head->mapping->host;
366 key->shared.pgoff = page_head->index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 }
368
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200369 get_futex_key_refs(key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500371out:
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800372 unlock_page(page_head);
373 put_page(page_head);
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500374 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375}
376
Thomas Gleixnerae791a22010-11-10 13:30:36 +0100377static inline void put_futex_key(union futex_key *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378{
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200379 drop_futex_key_refs(key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380}
381
Darren Hartd96ee562009-09-21 22:30:22 -0700382/**
383 * fault_in_user_writeable() - Fault in user address and verify RW access
Thomas Gleixnerd0725992009-06-11 23:15:43 +0200384 * @uaddr: pointer to faulting user space address
385 *
386 * Slow path to fixup the fault we just took in the atomic write
387 * access to @uaddr.
388 *
Randy Dunlapfb62db22010-10-13 11:02:34 -0700389 * We have no generic implementation of a non-destructive write to the
Thomas Gleixnerd0725992009-06-11 23:15:43 +0200390 * user address. We know that we faulted in the atomic pagefault
391 * disabled section so we can as well avoid the #PF overhead by
392 * calling get_user_pages() right away.
393 */
394static int fault_in_user_writeable(u32 __user *uaddr)
395{
Andi Kleen722d0172009-12-08 13:19:42 +0100396 struct mm_struct *mm = current->mm;
397 int ret;
398
399 down_read(&mm->mmap_sem);
Benjamin Herrenschmidt2efaca92011-07-25 17:12:32 -0700400 ret = fixup_user_fault(current, mm, (unsigned long)uaddr,
401 FAULT_FLAG_WRITE);
Andi Kleen722d0172009-12-08 13:19:42 +0100402 up_read(&mm->mmap_sem);
403
Thomas Gleixnerd0725992009-06-11 23:15:43 +0200404 return ret < 0 ? ret : 0;
405}
406
Darren Hart4b1c4862009-04-03 13:39:42 -0700407/**
408 * futex_top_waiter() - Return the highest priority waiter on a futex
Darren Hartd96ee562009-09-21 22:30:22 -0700409 * @hb: the hash bucket the futex_q's reside in
410 * @key: the futex key (to distinguish it from other futex futex_q's)
Darren Hart4b1c4862009-04-03 13:39:42 -0700411 *
412 * Must be called with the hb lock held.
413 */
414static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb,
415 union futex_key *key)
416{
417 struct futex_q *this;
418
419 plist_for_each_entry(this, &hb->chain, list) {
420 if (match_futex(&this->key, key))
421 return this;
422 }
423 return NULL;
424}
425
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800426static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr,
427 u32 uval, u32 newval)
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700428{
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800429 int ret;
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700430
431 pagefault_disable();
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800432 ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval);
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700433 pagefault_enable();
434
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800435 return ret;
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700436}
437
438static int get_futex_value_locked(u32 *dest, u32 __user *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439{
440 int ret;
441
Peter Zijlstraa8663742006-12-06 20:32:20 -0800442 pagefault_disable();
Ingo Molnare2970f22006-06-27 02:54:47 -0700443 ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
Peter Zijlstraa8663742006-12-06 20:32:20 -0800444 pagefault_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
446 return ret ? -EFAULT : 0;
447}
448
Ingo Molnarc87e2832006-06-27 02:54:58 -0700449
450/*
451 * PI code:
452 */
453static int refill_pi_state_cache(void)
454{
455 struct futex_pi_state *pi_state;
456
457 if (likely(current->pi_state_cache))
458 return 0;
459
Burman Yan4668edc2006-12-06 20:38:51 -0800460 pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700461
462 if (!pi_state)
463 return -ENOMEM;
464
Ingo Molnarc87e2832006-06-27 02:54:58 -0700465 INIT_LIST_HEAD(&pi_state->list);
466 /* pi_mutex gets initialized later */
467 pi_state->owner = NULL;
468 atomic_set(&pi_state->refcount, 1);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200469 pi_state->key = FUTEX_KEY_INIT;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700470
471 current->pi_state_cache = pi_state;
472
473 return 0;
474}
475
476static struct futex_pi_state * alloc_pi_state(void)
477{
478 struct futex_pi_state *pi_state = current->pi_state_cache;
479
480 WARN_ON(!pi_state);
481 current->pi_state_cache = NULL;
482
483 return pi_state;
484}
485
486static void free_pi_state(struct futex_pi_state *pi_state)
487{
488 if (!atomic_dec_and_test(&pi_state->refcount))
489 return;
490
491 /*
492 * If pi_state->owner is NULL, the owner is most probably dying
493 * and has cleaned up the pi_state already
494 */
495 if (pi_state->owner) {
Thomas Gleixner1d615482009-11-17 14:54:03 +0100496 raw_spin_lock_irq(&pi_state->owner->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700497 list_del_init(&pi_state->list);
Thomas Gleixner1d615482009-11-17 14:54:03 +0100498 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700499
500 rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
501 }
502
503 if (current->pi_state_cache)
504 kfree(pi_state);
505 else {
506 /*
507 * pi_state->list is already empty.
508 * clear pi_state->owner.
509 * refcount is at 0 - put it back to 1.
510 */
511 pi_state->owner = NULL;
512 atomic_set(&pi_state->refcount, 1);
513 current->pi_state_cache = pi_state;
514 }
515}
516
517/*
518 * Look up the task based on what TID userspace gave us.
519 * We dont trust it.
520 */
521static struct task_struct * futex_find_get_task(pid_t pid)
522{
523 struct task_struct *p;
524
Oleg Nesterovd359b542006-09-29 02:00:55 -0700525 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700526 p = find_task_by_vpid(pid);
Michal Hocko7a0ea092010-06-30 09:51:19 +0200527 if (p)
528 get_task_struct(p);
Thomas Gleixnera06381f2007-06-23 11:48:40 +0200529
Oleg Nesterovd359b542006-09-29 02:00:55 -0700530 rcu_read_unlock();
Ingo Molnarc87e2832006-06-27 02:54:58 -0700531
532 return p;
533}
534
535/*
536 * This task is holding PI mutexes at exit time => bad.
537 * Kernel cleans up PI-state, but userspace is likely hosed.
538 * (Robust-futex cleanup is separate and might save the day for userspace.)
539 */
540void exit_pi_state_list(struct task_struct *curr)
541{
Ingo Molnarc87e2832006-06-27 02:54:58 -0700542 struct list_head *next, *head = &curr->pi_state_list;
543 struct futex_pi_state *pi_state;
Ingo Molnar627371d2006-07-29 05:16:20 +0200544 struct futex_hash_bucket *hb;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200545 union futex_key key = FUTEX_KEY_INIT;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700546
Thomas Gleixnera0c1e902008-02-23 15:23:57 -0800547 if (!futex_cmpxchg_enabled)
548 return;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700549 /*
550 * We are a ZOMBIE and nobody can enqueue itself on
551 * pi_state_list anymore, but we have to be careful
Ingo Molnar627371d2006-07-29 05:16:20 +0200552 * versus waiters unqueueing themselves:
Ingo Molnarc87e2832006-06-27 02:54:58 -0700553 */
Thomas Gleixner1d615482009-11-17 14:54:03 +0100554 raw_spin_lock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700555 while (!list_empty(head)) {
556
557 next = head->next;
558 pi_state = list_entry(next, struct futex_pi_state, list);
559 key = pi_state->key;
Ingo Molnar627371d2006-07-29 05:16:20 +0200560 hb = hash_futex(&key);
Thomas Gleixner1d615482009-11-17 14:54:03 +0100561 raw_spin_unlock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700562
Ingo Molnarc87e2832006-06-27 02:54:58 -0700563 spin_lock(&hb->lock);
564
Thomas Gleixner1d615482009-11-17 14:54:03 +0100565 raw_spin_lock_irq(&curr->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200566 /*
567 * We dropped the pi-lock, so re-check whether this
568 * task still owns the PI-state:
569 */
Ingo Molnarc87e2832006-06-27 02:54:58 -0700570 if (head->next != next) {
571 spin_unlock(&hb->lock);
572 continue;
573 }
574
Ingo Molnarc87e2832006-06-27 02:54:58 -0700575 WARN_ON(pi_state->owner != curr);
Ingo Molnar627371d2006-07-29 05:16:20 +0200576 WARN_ON(list_empty(&pi_state->list));
577 list_del_init(&pi_state->list);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700578 pi_state->owner = NULL;
Thomas Gleixner1d615482009-11-17 14:54:03 +0100579 raw_spin_unlock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700580
581 rt_mutex_unlock(&pi_state->pi_mutex);
582
583 spin_unlock(&hb->lock);
584
Thomas Gleixner1d615482009-11-17 14:54:03 +0100585 raw_spin_lock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700586 }
Thomas Gleixner1d615482009-11-17 14:54:03 +0100587 raw_spin_unlock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700588}
589
590static int
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700591lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
592 union futex_key *key, struct futex_pi_state **ps)
Ingo Molnarc87e2832006-06-27 02:54:58 -0700593{
594 struct futex_pi_state *pi_state = NULL;
595 struct futex_q *this, *next;
Pierre Peifferec92d082007-05-09 02:35:00 -0700596 struct plist_head *head;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700597 struct task_struct *p;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700598 pid_t pid = uval & FUTEX_TID_MASK;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700599
600 head = &hb->chain;
601
Pierre Peifferec92d082007-05-09 02:35:00 -0700602 plist_for_each_entry_safe(this, next, head, list) {
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700603 if (match_futex(&this->key, key)) {
Ingo Molnarc87e2832006-06-27 02:54:58 -0700604 /*
605 * Another waiter already exists - bump up
606 * the refcount and return its pi_state:
607 */
608 pi_state = this->pi_state;
Thomas Gleixner06a9ec22006-07-10 04:44:30 -0700609 /*
Randy Dunlapfb62db22010-10-13 11:02:34 -0700610 * Userspace might have messed up non-PI and PI futexes
Thomas Gleixner06a9ec22006-07-10 04:44:30 -0700611 */
612 if (unlikely(!pi_state))
613 return -EINVAL;
614
Ingo Molnar627371d2006-07-29 05:16:20 +0200615 WARN_ON(!atomic_read(&pi_state->refcount));
Thomas Gleixner59647b62010-02-03 09:33:05 +0100616
617 /*
618 * When pi_state->owner is NULL then the owner died
619 * and another waiter is on the fly. pi_state->owner
620 * is fixed up by the task which acquires
621 * pi_state->rt_mutex.
622 *
623 * We do not check for pid == 0 which can happen when
624 * the owner died and robust_list_exit() cleared the
625 * TID.
626 */
627 if (pid && pi_state->owner) {
628 /*
629 * Bail out if user space manipulated the
630 * futex value.
631 */
632 if (pid != task_pid_vnr(pi_state->owner))
633 return -EINVAL;
634 }
Ingo Molnar627371d2006-07-29 05:16:20 +0200635
Ingo Molnarc87e2832006-06-27 02:54:58 -0700636 atomic_inc(&pi_state->refcount);
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700637 *ps = pi_state;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700638
639 return 0;
640 }
641 }
642
643 /*
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200644 * We are the first waiter - try to look up the real owner and attach
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700645 * the new pi_state to it, but bail out when TID = 0
Ingo Molnarc87e2832006-06-27 02:54:58 -0700646 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700647 if (!pid)
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200648 return -ESRCH;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700649 p = futex_find_get_task(pid);
Michal Hocko7a0ea092010-06-30 09:51:19 +0200650 if (!p)
651 return -ESRCH;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700652
653 /*
654 * We need to look at the task state flags to figure out,
655 * whether the task is exiting. To protect against the do_exit
656 * change of the task flags, we do this protected by
657 * p->pi_lock:
658 */
Thomas Gleixner1d615482009-11-17 14:54:03 +0100659 raw_spin_lock_irq(&p->pi_lock);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700660 if (unlikely(p->flags & PF_EXITING)) {
661 /*
662 * The task is on the way out. When PF_EXITPIDONE is
663 * set, we know that the task has finished the
664 * cleanup:
665 */
666 int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
667
Thomas Gleixner1d615482009-11-17 14:54:03 +0100668 raw_spin_unlock_irq(&p->pi_lock);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700669 put_task_struct(p);
670 return ret;
671 }
Ingo Molnarc87e2832006-06-27 02:54:58 -0700672
673 pi_state = alloc_pi_state();
674
675 /*
676 * Initialize the pi_mutex in locked state and make 'p'
677 * the owner of it:
678 */
679 rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
680
681 /* Store the key for possible exit cleanups: */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700682 pi_state->key = *key;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700683
Ingo Molnar627371d2006-07-29 05:16:20 +0200684 WARN_ON(!list_empty(&pi_state->list));
Ingo Molnarc87e2832006-06-27 02:54:58 -0700685 list_add(&pi_state->list, &p->pi_state_list);
686 pi_state->owner = p;
Thomas Gleixner1d615482009-11-17 14:54:03 +0100687 raw_spin_unlock_irq(&p->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700688
689 put_task_struct(p);
690
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700691 *ps = pi_state;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700692
693 return 0;
694}
695
Darren Hart1a520842009-04-03 13:39:52 -0700696/**
Darren Hartd96ee562009-09-21 22:30:22 -0700697 * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
Darren Hartbab5bc92009-04-07 23:23:50 -0700698 * @uaddr: the pi futex user address
699 * @hb: the pi futex hash bucket
700 * @key: the futex key associated with uaddr and hb
701 * @ps: the pi_state pointer where we store the result of the
702 * lookup
703 * @task: the task to perform the atomic lock work for. This will
704 * be "current" except in the case of requeue pi.
705 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
Darren Hart1a520842009-04-03 13:39:52 -0700706 *
707 * Returns:
708 * 0 - ready to wait
709 * 1 - acquired the lock
710 * <0 - error
711 *
712 * The hb->lock and futex_key refs shall be held by the caller.
713 */
714static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb,
715 union futex_key *key,
716 struct futex_pi_state **ps,
Darren Hartbab5bc92009-04-07 23:23:50 -0700717 struct task_struct *task, int set_waiters)
Darren Hart1a520842009-04-03 13:39:52 -0700718{
719 int lock_taken, ret, ownerdied = 0;
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +0100720 u32 uval, newval, curval, vpid = task_pid_vnr(task);
Darren Hart1a520842009-04-03 13:39:52 -0700721
722retry:
723 ret = lock_taken = 0;
724
725 /*
726 * To avoid races, we attempt to take the lock here again
727 * (by doing a 0 -> TID atomic cmpxchg), while holding all
728 * the locks. It will most likely not succeed.
729 */
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +0100730 newval = vpid;
Darren Hartbab5bc92009-04-07 23:23:50 -0700731 if (set_waiters)
732 newval |= FUTEX_WAITERS;
Darren Hart1a520842009-04-03 13:39:52 -0700733
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800734 if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval)))
Darren Hart1a520842009-04-03 13:39:52 -0700735 return -EFAULT;
736
737 /*
738 * Detect deadlocks.
739 */
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +0100740 if ((unlikely((curval & FUTEX_TID_MASK) == vpid)))
Darren Hart1a520842009-04-03 13:39:52 -0700741 return -EDEADLK;
742
743 /*
744 * Surprise - we got the lock. Just return to userspace:
745 */
746 if (unlikely(!curval))
747 return 1;
748
749 uval = curval;
750
751 /*
752 * Set the FUTEX_WAITERS flag, so the owner will know it has someone
753 * to wake at the next unlock.
754 */
755 newval = curval | FUTEX_WAITERS;
756
757 /*
758 * There are two cases, where a futex might have no owner (the
759 * owner TID is 0): OWNER_DIED. We take over the futex in this
760 * case. We also do an unconditional take over, when the owner
761 * of the futex died.
762 *
763 * This is safe as we are protected by the hash bucket lock !
764 */
765 if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) {
766 /* Keep the OWNER_DIED bit */
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +0100767 newval = (curval & ~FUTEX_TID_MASK) | vpid;
Darren Hart1a520842009-04-03 13:39:52 -0700768 ownerdied = 0;
769 lock_taken = 1;
770 }
771
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800772 if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)))
Darren Hart1a520842009-04-03 13:39:52 -0700773 return -EFAULT;
774 if (unlikely(curval != uval))
775 goto retry;
776
777 /*
778 * We took the lock due to owner died take over.
779 */
780 if (unlikely(lock_taken))
781 return 1;
782
783 /*
784 * We dont have the lock. Look up the PI state (or create it if
785 * we are the first waiter):
786 */
787 ret = lookup_pi_state(uval, hb, key, ps);
788
789 if (unlikely(ret)) {
790 switch (ret) {
791 case -ESRCH:
792 /*
793 * No owner found for this futex. Check if the
794 * OWNER_DIED bit is set to figure out whether
795 * this is a robust futex or not.
796 */
797 if (get_futex_value_locked(&curval, uaddr))
798 return -EFAULT;
799
800 /*
801 * We simply start over in case of a robust
802 * futex. The code above will take the futex
803 * and return happy.
804 */
805 if (curval & FUTEX_OWNER_DIED) {
806 ownerdied = 1;
807 goto retry;
808 }
809 default:
810 break;
811 }
812 }
813
814 return ret;
815}
816
Lai Jiangshan2e129782010-12-22 14:18:50 +0800817/**
818 * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket
819 * @q: The futex_q to unqueue
820 *
821 * The q->lock_ptr must not be NULL and must be held by the caller.
822 */
823static void __unqueue_futex(struct futex_q *q)
824{
825 struct futex_hash_bucket *hb;
826
Steven Rostedt29096202011-03-17 15:21:07 -0400827 if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr))
828 || WARN_ON(plist_node_empty(&q->list)))
Lai Jiangshan2e129782010-12-22 14:18:50 +0800829 return;
830
831 hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock);
832 plist_del(&q->list, &hb->chain);
833}
834
Ingo Molnarc87e2832006-06-27 02:54:58 -0700835/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 * The hash bucket lock must be held when this is called.
837 * Afterwards, the futex_q must not be accessed.
838 */
839static void wake_futex(struct futex_q *q)
840{
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200841 struct task_struct *p = q->task;
842
843 /*
844 * We set q->lock_ptr = NULL _before_ we wake up the task. If
Randy Dunlapfb62db22010-10-13 11:02:34 -0700845 * a non-futex wake up happens on another CPU then the task
846 * might exit and p would dereference a non-existing task
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200847 * struct. Prevent this by holding a reference on p across the
848 * wake up.
849 */
850 get_task_struct(p);
851
Lai Jiangshan2e129782010-12-22 14:18:50 +0800852 __unqueue_futex(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 /*
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200854 * The waiting task can free the futex_q as soon as
855 * q->lock_ptr = NULL is written, without taking any locks. A
856 * memory barrier is required here to prevent the following
857 * store to lock_ptr from getting ahead of the plist_del.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 */
Ralf Baechleccdea2f2006-12-06 20:40:26 -0800859 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 q->lock_ptr = NULL;
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200861
862 wake_up_state(p, TASK_NORMAL);
863 put_task_struct(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864}
865
Ingo Molnarc87e2832006-06-27 02:54:58 -0700866static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
867{
868 struct task_struct *new_owner;
869 struct futex_pi_state *pi_state = this->pi_state;
Vitaliy Ivanov7cfdaf32011-07-07 15:10:31 +0300870 u32 uninitialized_var(curval), newval;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700871
872 if (!pi_state)
873 return -EINVAL;
874
Thomas Gleixner51246bf2010-02-02 11:40:27 +0100875 /*
876 * If current does not own the pi_state then the futex is
877 * inconsistent and user space fiddled with the futex value.
878 */
879 if (pi_state->owner != current)
880 return -EINVAL;
881
Thomas Gleixnerd209d742009-11-17 18:22:11 +0100882 raw_spin_lock(&pi_state->pi_mutex.wait_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700883 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
884
885 /*
Steven Rostedtf123c982011-01-06 15:08:29 -0500886 * It is possible that the next waiter (the one that brought
887 * this owner to the kernel) timed out and is no longer
888 * waiting on the lock.
Ingo Molnarc87e2832006-06-27 02:54:58 -0700889 */
890 if (!new_owner)
891 new_owner = this->task;
892
893 /*
894 * We pass it to the next owner. (The WAITERS bit is always
895 * kept enabled while there is PI state around. We must also
896 * preserve the owner died bit.)
897 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200898 if (!(uval & FUTEX_OWNER_DIED)) {
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700899 int ret = 0;
900
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700901 newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700902
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800903 if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700904 ret = -EFAULT;
Thomas Gleixnercde898f2007-12-05 15:46:09 +0100905 else if (curval != uval)
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700906 ret = -EINVAL;
907 if (ret) {
Thomas Gleixnerd209d742009-11-17 18:22:11 +0100908 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700909 return ret;
910 }
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200911 }
Ingo Molnarc87e2832006-06-27 02:54:58 -0700912
Thomas Gleixner1d615482009-11-17 14:54:03 +0100913 raw_spin_lock_irq(&pi_state->owner->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200914 WARN_ON(list_empty(&pi_state->list));
915 list_del_init(&pi_state->list);
Thomas Gleixner1d615482009-11-17 14:54:03 +0100916 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200917
Thomas Gleixner1d615482009-11-17 14:54:03 +0100918 raw_spin_lock_irq(&new_owner->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200919 WARN_ON(!list_empty(&pi_state->list));
Ingo Molnarc87e2832006-06-27 02:54:58 -0700920 list_add(&pi_state->list, &new_owner->pi_state_list);
921 pi_state->owner = new_owner;
Thomas Gleixner1d615482009-11-17 14:54:03 +0100922 raw_spin_unlock_irq(&new_owner->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200923
Thomas Gleixnerd209d742009-11-17 18:22:11 +0100924 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700925 rt_mutex_unlock(&pi_state->pi_mutex);
926
927 return 0;
928}
929
930static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
931{
Vitaliy Ivanov7cfdaf32011-07-07 15:10:31 +0300932 u32 uninitialized_var(oldval);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700933
934 /*
935 * There is no waiter, so we unlock the futex. The owner died
936 * bit has not to be preserved here. We are the owner:
937 */
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800938 if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0))
939 return -EFAULT;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700940 if (oldval != uval)
941 return -EAGAIN;
942
943 return 0;
944}
945
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946/*
Ingo Molnar8b8f3192006-07-03 00:25:05 -0700947 * Express the locking dependencies for lockdep:
948 */
949static inline void
950double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
951{
952 if (hb1 <= hb2) {
953 spin_lock(&hb1->lock);
954 if (hb1 < hb2)
955 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
956 } else { /* hb1 > hb2 */
957 spin_lock(&hb2->lock);
958 spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
959 }
960}
961
Darren Hart5eb3dc62009-03-12 00:55:52 -0700962static inline void
963double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
964{
Darren Hartf061d352009-03-12 15:11:18 -0700965 spin_unlock(&hb1->lock);
Ingo Molnar88f502f2009-03-13 10:32:07 +0100966 if (hb1 != hb2)
967 spin_unlock(&hb2->lock);
Darren Hart5eb3dc62009-03-12 00:55:52 -0700968}
969
Ingo Molnar8b8f3192006-07-03 00:25:05 -0700970/*
Darren Hartb2d09942009-03-12 00:55:37 -0700971 * Wake up waiters matching bitset queued on this futex (uaddr).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 */
Darren Hartb41277d2010-11-08 13:10:09 -0800973static int
974futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975{
Ingo Molnare2970f22006-06-27 02:54:47 -0700976 struct futex_hash_bucket *hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 struct futex_q *this, *next;
Pierre Peifferec92d082007-05-09 02:35:00 -0700978 struct plist_head *head;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200979 union futex_key key = FUTEX_KEY_INIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 int ret;
981
Thomas Gleixnercd689982008-02-01 17:45:14 +0100982 if (!bitset)
983 return -EINVAL;
984
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500985 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 if (unlikely(ret != 0))
987 goto out;
988
Ingo Molnare2970f22006-06-27 02:54:47 -0700989 hb = hash_futex(&key);
990 spin_lock(&hb->lock);
991 head = &hb->chain;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
Pierre Peifferec92d082007-05-09 02:35:00 -0700993 plist_for_each_entry_safe(this, next, head, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 if (match_futex (&this->key, &key)) {
Darren Hart52400ba2009-04-03 13:40:49 -0700995 if (this->pi_state || this->rt_waiter) {
Ingo Molnared6f7b12006-07-01 04:35:46 -0700996 ret = -EINVAL;
997 break;
998 }
Thomas Gleixnercd689982008-02-01 17:45:14 +0100999
1000 /* Check if one of the bits is set in both bitsets */
1001 if (!(this->bitset & bitset))
1002 continue;
1003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 wake_futex(this);
1005 if (++ret >= nr_wake)
1006 break;
1007 }
1008 }
1009
Ingo Molnare2970f22006-06-27 02:54:47 -07001010 spin_unlock(&hb->lock);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001011 put_futex_key(&key);
Darren Hart42d35d42008-12-29 15:49:53 -08001012out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 return ret;
1014}
1015
1016/*
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001017 * Wake up all waiters hashed on the physical page that is mapped
1018 * to this virtual address:
1019 */
Ingo Molnare2970f22006-06-27 02:54:47 -07001020static int
Darren Hartb41277d2010-11-08 13:10:09 -08001021futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
Ingo Molnare2970f22006-06-27 02:54:47 -07001022 int nr_wake, int nr_wake2, int op)
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001023{
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001024 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
Ingo Molnare2970f22006-06-27 02:54:47 -07001025 struct futex_hash_bucket *hb1, *hb2;
Pierre Peifferec92d082007-05-09 02:35:00 -07001026 struct plist_head *head;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001027 struct futex_q *this, *next;
Darren Harte4dc5b72009-03-12 00:56:13 -07001028 int ret, op_ret;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001029
Darren Harte4dc5b72009-03-12 00:56:13 -07001030retry:
Shawn Bohrer9ea71502011-06-30 11:21:32 -05001031 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001032 if (unlikely(ret != 0))
1033 goto out;
Shawn Bohrer9ea71502011-06-30 11:21:32 -05001034 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001035 if (unlikely(ret != 0))
Darren Hart42d35d42008-12-29 15:49:53 -08001036 goto out_put_key1;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001037
Ingo Molnare2970f22006-06-27 02:54:47 -07001038 hb1 = hash_futex(&key1);
1039 hb2 = hash_futex(&key2);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001040
Darren Harte4dc5b72009-03-12 00:56:13 -07001041retry_private:
Thomas Gleixnereaaea802009-10-04 09:34:17 +02001042 double_lock_hb(hb1, hb2);
Ingo Molnare2970f22006-06-27 02:54:47 -07001043 op_ret = futex_atomic_op_inuser(op, uaddr2);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001044 if (unlikely(op_ret < 0)) {
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001045
Darren Hart5eb3dc62009-03-12 00:55:52 -07001046 double_unlock_hb(hb1, hb2);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001047
David Howells7ee1dd32006-01-06 00:11:44 -08001048#ifndef CONFIG_MMU
Ingo Molnare2970f22006-06-27 02:54:47 -07001049 /*
1050 * we don't get EFAULT from MMU faults if we don't have an MMU,
1051 * but we might get them from range checking
1052 */
David Howells7ee1dd32006-01-06 00:11:44 -08001053 ret = op_ret;
Darren Hart42d35d42008-12-29 15:49:53 -08001054 goto out_put_keys;
David Howells7ee1dd32006-01-06 00:11:44 -08001055#endif
1056
David Gibson796f8d92005-11-07 00:59:33 -08001057 if (unlikely(op_ret != -EFAULT)) {
1058 ret = op_ret;
Darren Hart42d35d42008-12-29 15:49:53 -08001059 goto out_put_keys;
David Gibson796f8d92005-11-07 00:59:33 -08001060 }
1061
Thomas Gleixnerd0725992009-06-11 23:15:43 +02001062 ret = fault_in_user_writeable(uaddr2);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001063 if (ret)
Darren Hartde87fcc2009-03-12 00:55:46 -07001064 goto out_put_keys;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001065
Darren Hartb41277d2010-11-08 13:10:09 -08001066 if (!(flags & FLAGS_SHARED))
Darren Harte4dc5b72009-03-12 00:56:13 -07001067 goto retry_private;
1068
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001069 put_futex_key(&key2);
1070 put_futex_key(&key1);
Darren Harte4dc5b72009-03-12 00:56:13 -07001071 goto retry;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001072 }
1073
Ingo Molnare2970f22006-06-27 02:54:47 -07001074 head = &hb1->chain;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001075
Pierre Peifferec92d082007-05-09 02:35:00 -07001076 plist_for_each_entry_safe(this, next, head, list) {
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001077 if (match_futex (&this->key, &key1)) {
1078 wake_futex(this);
1079 if (++ret >= nr_wake)
1080 break;
1081 }
1082 }
1083
1084 if (op_ret > 0) {
Ingo Molnare2970f22006-06-27 02:54:47 -07001085 head = &hb2->chain;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001086
1087 op_ret = 0;
Pierre Peifferec92d082007-05-09 02:35:00 -07001088 plist_for_each_entry_safe(this, next, head, list) {
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001089 if (match_futex (&this->key, &key2)) {
1090 wake_futex(this);
1091 if (++op_ret >= nr_wake2)
1092 break;
1093 }
1094 }
1095 ret += op_ret;
1096 }
1097
Darren Hart5eb3dc62009-03-12 00:55:52 -07001098 double_unlock_hb(hb1, hb2);
Darren Hart42d35d42008-12-29 15:49:53 -08001099out_put_keys:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001100 put_futex_key(&key2);
Darren Hart42d35d42008-12-29 15:49:53 -08001101out_put_key1:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001102 put_futex_key(&key1);
Darren Hart42d35d42008-12-29 15:49:53 -08001103out:
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001104 return ret;
1105}
1106
Darren Hart9121e472009-04-03 13:40:31 -07001107/**
1108 * requeue_futex() - Requeue a futex_q from one hb to another
1109 * @q: the futex_q to requeue
1110 * @hb1: the source hash_bucket
1111 * @hb2: the target hash_bucket
1112 * @key2: the new key for the requeued futex_q
1113 */
1114static inline
1115void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1,
1116 struct futex_hash_bucket *hb2, union futex_key *key2)
1117{
1118
1119 /*
1120 * If key1 and key2 hash to the same bucket, no need to
1121 * requeue.
1122 */
1123 if (likely(&hb1->chain != &hb2->chain)) {
1124 plist_del(&q->list, &hb1->chain);
1125 plist_add(&q->list, &hb2->chain);
1126 q->lock_ptr = &hb2->lock;
Darren Hart9121e472009-04-03 13:40:31 -07001127 }
1128 get_futex_key_refs(key2);
1129 q->key = *key2;
1130}
1131
Darren Hart52400ba2009-04-03 13:40:49 -07001132/**
1133 * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
Darren Hartd96ee562009-09-21 22:30:22 -07001134 * @q: the futex_q
1135 * @key: the key of the requeue target futex
1136 * @hb: the hash_bucket of the requeue target futex
Darren Hart52400ba2009-04-03 13:40:49 -07001137 *
1138 * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1139 * target futex if it is uncontended or via a lock steal. Set the futex_q key
1140 * to the requeue target futex so the waiter can detect the wakeup on the right
1141 * futex, but remove it from the hb and NULL the rt_waiter so it can detect
Darren Hartbeda2c72009-08-09 15:34:39 -07001142 * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1143 * to protect access to the pi_state to fixup the owner later. Must be called
1144 * with both q->lock_ptr and hb->lock held.
Darren Hart52400ba2009-04-03 13:40:49 -07001145 */
1146static inline
Darren Hartbeda2c72009-08-09 15:34:39 -07001147void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key,
1148 struct futex_hash_bucket *hb)
Darren Hart52400ba2009-04-03 13:40:49 -07001149{
Darren Hart52400ba2009-04-03 13:40:49 -07001150 get_futex_key_refs(key);
1151 q->key = *key;
1152
Lai Jiangshan2e129782010-12-22 14:18:50 +08001153 __unqueue_futex(q);
Darren Hart52400ba2009-04-03 13:40:49 -07001154
1155 WARN_ON(!q->rt_waiter);
1156 q->rt_waiter = NULL;
1157
Darren Hartbeda2c72009-08-09 15:34:39 -07001158 q->lock_ptr = &hb->lock;
Darren Hartbeda2c72009-08-09 15:34:39 -07001159
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02001160 wake_up_state(q->task, TASK_NORMAL);
Darren Hart52400ba2009-04-03 13:40:49 -07001161}
1162
1163/**
1164 * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
Darren Hartbab5bc92009-04-07 23:23:50 -07001165 * @pifutex: the user address of the to futex
1166 * @hb1: the from futex hash bucket, must be locked by the caller
1167 * @hb2: the to futex hash bucket, must be locked by the caller
1168 * @key1: the from futex key
1169 * @key2: the to futex key
1170 * @ps: address to store the pi_state pointer
1171 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
Darren Hart52400ba2009-04-03 13:40:49 -07001172 *
1173 * Try and get the lock on behalf of the top waiter if we can do it atomically.
Darren Hartbab5bc92009-04-07 23:23:50 -07001174 * Wake the top waiter if we succeed. If the caller specified set_waiters,
1175 * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1176 * hb1 and hb2 must be held by the caller.
Darren Hart52400ba2009-04-03 13:40:49 -07001177 *
1178 * Returns:
1179 * 0 - failed to acquire the lock atomicly
1180 * 1 - acquired the lock
1181 * <0 - error
1182 */
1183static int futex_proxy_trylock_atomic(u32 __user *pifutex,
1184 struct futex_hash_bucket *hb1,
1185 struct futex_hash_bucket *hb2,
1186 union futex_key *key1, union futex_key *key2,
Darren Hartbab5bc92009-04-07 23:23:50 -07001187 struct futex_pi_state **ps, int set_waiters)
Darren Hart52400ba2009-04-03 13:40:49 -07001188{
Darren Hartbab5bc92009-04-07 23:23:50 -07001189 struct futex_q *top_waiter = NULL;
Darren Hart52400ba2009-04-03 13:40:49 -07001190 u32 curval;
1191 int ret;
1192
1193 if (get_futex_value_locked(&curval, pifutex))
1194 return -EFAULT;
1195
Darren Hartbab5bc92009-04-07 23:23:50 -07001196 /*
1197 * Find the top_waiter and determine if there are additional waiters.
1198 * If the caller intends to requeue more than 1 waiter to pifutex,
1199 * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1200 * as we have means to handle the possible fault. If not, don't set
1201 * the bit unecessarily as it will force the subsequent unlock to enter
1202 * the kernel.
1203 */
Darren Hart52400ba2009-04-03 13:40:49 -07001204 top_waiter = futex_top_waiter(hb1, key1);
1205
1206 /* There are no waiters, nothing for us to do. */
1207 if (!top_waiter)
1208 return 0;
1209
Darren Hart84bc4af2009-08-13 17:36:53 -07001210 /* Ensure we requeue to the expected futex. */
1211 if (!match_futex(top_waiter->requeue_pi_key, key2))
1212 return -EINVAL;
1213
Darren Hart52400ba2009-04-03 13:40:49 -07001214 /*
Darren Hartbab5bc92009-04-07 23:23:50 -07001215 * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1216 * the contended case or if set_waiters is 1. The pi_state is returned
1217 * in ps in contended cases.
Darren Hart52400ba2009-04-03 13:40:49 -07001218 */
Darren Hartbab5bc92009-04-07 23:23:50 -07001219 ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task,
1220 set_waiters);
Darren Hart52400ba2009-04-03 13:40:49 -07001221 if (ret == 1)
Darren Hartbeda2c72009-08-09 15:34:39 -07001222 requeue_pi_wake_futex(top_waiter, key2, hb2);
Darren Hart52400ba2009-04-03 13:40:49 -07001223
1224 return ret;
1225}
1226
1227/**
1228 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
Randy Dunlapfb62db22010-10-13 11:02:34 -07001229 * @uaddr1: source futex user address
Darren Hartb41277d2010-11-08 13:10:09 -08001230 * @flags: futex flags (FLAGS_SHARED, etc.)
Randy Dunlapfb62db22010-10-13 11:02:34 -07001231 * @uaddr2: target futex user address
1232 * @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1233 * @nr_requeue: number of waiters to requeue (0-INT_MAX)
1234 * @cmpval: @uaddr1 expected value (or %NULL)
1235 * @requeue_pi: if we are attempting to requeue from a non-pi futex to a
Darren Hartb41277d2010-11-08 13:10:09 -08001236 * pi futex (pi to pi requeue is not supported)
Darren Hart52400ba2009-04-03 13:40:49 -07001237 *
1238 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1239 * uaddr2 atomically on behalf of the top waiter.
1240 *
1241 * Returns:
1242 * >=0 - on success, the number of tasks requeued or woken
1243 * <0 - on error
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 */
Darren Hartb41277d2010-11-08 13:10:09 -08001245static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
1246 u32 __user *uaddr2, int nr_wake, int nr_requeue,
1247 u32 *cmpval, int requeue_pi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248{
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001249 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
Darren Hart52400ba2009-04-03 13:40:49 -07001250 int drop_count = 0, task_count = 0, ret;
1251 struct futex_pi_state *pi_state = NULL;
Ingo Molnare2970f22006-06-27 02:54:47 -07001252 struct futex_hash_bucket *hb1, *hb2;
Pierre Peifferec92d082007-05-09 02:35:00 -07001253 struct plist_head *head1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 struct futex_q *this, *next;
Darren Hart52400ba2009-04-03 13:40:49 -07001255 u32 curval2;
1256
1257 if (requeue_pi) {
1258 /*
1259 * requeue_pi requires a pi_state, try to allocate it now
1260 * without any locks in case it fails.
1261 */
1262 if (refill_pi_state_cache())
1263 return -ENOMEM;
1264 /*
1265 * requeue_pi must wake as many tasks as it can, up to nr_wake
1266 * + nr_requeue, since it acquires the rt_mutex prior to
1267 * returning to userspace, so as to not leave the rt_mutex with
1268 * waiters and no owner. However, second and third wake-ups
1269 * cannot be predicted as they involve race conditions with the
1270 * first wake and a fault while looking up the pi_state. Both
1271 * pthread_cond_signal() and pthread_cond_broadcast() should
1272 * use nr_wake=1.
1273 */
1274 if (nr_wake != 1)
1275 return -EINVAL;
1276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
Darren Hart42d35d42008-12-29 15:49:53 -08001278retry:
Darren Hart52400ba2009-04-03 13:40:49 -07001279 if (pi_state != NULL) {
1280 /*
1281 * We will have to lookup the pi_state again, so free this one
1282 * to keep the accounting correct.
1283 */
1284 free_pi_state(pi_state);
1285 pi_state = NULL;
1286 }
1287
Shawn Bohrer9ea71502011-06-30 11:21:32 -05001288 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 if (unlikely(ret != 0))
1290 goto out;
Shawn Bohrer9ea71502011-06-30 11:21:32 -05001291 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2,
1292 requeue_pi ? VERIFY_WRITE : VERIFY_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 if (unlikely(ret != 0))
Darren Hart42d35d42008-12-29 15:49:53 -08001294 goto out_put_key1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
Ingo Molnare2970f22006-06-27 02:54:47 -07001296 hb1 = hash_futex(&key1);
1297 hb2 = hash_futex(&key2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
Darren Harte4dc5b72009-03-12 00:56:13 -07001299retry_private:
Ingo Molnar8b8f3192006-07-03 00:25:05 -07001300 double_lock_hb(hb1, hb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
Ingo Molnare2970f22006-06-27 02:54:47 -07001302 if (likely(cmpval != NULL)) {
1303 u32 curval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Ingo Molnare2970f22006-06-27 02:54:47 -07001305 ret = get_futex_value_locked(&curval, uaddr1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
1307 if (unlikely(ret)) {
Darren Hart5eb3dc62009-03-12 00:55:52 -07001308 double_unlock_hb(hb1, hb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
Darren Harte4dc5b72009-03-12 00:56:13 -07001310 ret = get_user(curval, uaddr1);
1311 if (ret)
1312 goto out_put_keys;
1313
Darren Hartb41277d2010-11-08 13:10:09 -08001314 if (!(flags & FLAGS_SHARED))
Darren Harte4dc5b72009-03-12 00:56:13 -07001315 goto retry_private;
1316
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001317 put_futex_key(&key2);
1318 put_futex_key(&key1);
Darren Harte4dc5b72009-03-12 00:56:13 -07001319 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 }
Ingo Molnare2970f22006-06-27 02:54:47 -07001321 if (curval != *cmpval) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 ret = -EAGAIN;
1323 goto out_unlock;
1324 }
1325 }
1326
Darren Hart52400ba2009-04-03 13:40:49 -07001327 if (requeue_pi && (task_count - nr_wake < nr_requeue)) {
Darren Hartbab5bc92009-04-07 23:23:50 -07001328 /*
1329 * Attempt to acquire uaddr2 and wake the top waiter. If we
1330 * intend to requeue waiters, force setting the FUTEX_WAITERS
1331 * bit. We force this here where we are able to easily handle
1332 * faults rather in the requeue loop below.
1333 */
Darren Hart52400ba2009-04-03 13:40:49 -07001334 ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1,
Darren Hartbab5bc92009-04-07 23:23:50 -07001335 &key2, &pi_state, nr_requeue);
Darren Hart52400ba2009-04-03 13:40:49 -07001336
1337 /*
1338 * At this point the top_waiter has either taken uaddr2 or is
1339 * waiting on it. If the former, then the pi_state will not
1340 * exist yet, look it up one more time to ensure we have a
1341 * reference to it.
1342 */
1343 if (ret == 1) {
1344 WARN_ON(pi_state);
Darren Hart89061d32009-10-15 15:30:48 -07001345 drop_count++;
Darren Hart52400ba2009-04-03 13:40:49 -07001346 task_count++;
1347 ret = get_futex_value_locked(&curval2, uaddr2);
1348 if (!ret)
1349 ret = lookup_pi_state(curval2, hb2, &key2,
1350 &pi_state);
1351 }
1352
1353 switch (ret) {
1354 case 0:
1355 break;
1356 case -EFAULT:
1357 double_unlock_hb(hb1, hb2);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001358 put_futex_key(&key2);
1359 put_futex_key(&key1);
Thomas Gleixnerd0725992009-06-11 23:15:43 +02001360 ret = fault_in_user_writeable(uaddr2);
Darren Hart52400ba2009-04-03 13:40:49 -07001361 if (!ret)
1362 goto retry;
1363 goto out;
1364 case -EAGAIN:
1365 /* The owner was exiting, try again. */
1366 double_unlock_hb(hb1, hb2);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001367 put_futex_key(&key2);
1368 put_futex_key(&key1);
Darren Hart52400ba2009-04-03 13:40:49 -07001369 cond_resched();
1370 goto retry;
1371 default:
1372 goto out_unlock;
1373 }
1374 }
1375
Ingo Molnare2970f22006-06-27 02:54:47 -07001376 head1 = &hb1->chain;
Pierre Peifferec92d082007-05-09 02:35:00 -07001377 plist_for_each_entry_safe(this, next, head1, list) {
Darren Hart52400ba2009-04-03 13:40:49 -07001378 if (task_count - nr_wake >= nr_requeue)
1379 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
Darren Hart52400ba2009-04-03 13:40:49 -07001381 if (!match_futex(&this->key, &key1))
1382 continue;
1383
Darren Hart392741e2009-08-07 15:20:48 -07001384 /*
1385 * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
1386 * be paired with each other and no other futex ops.
1387 */
1388 if ((requeue_pi && !this->rt_waiter) ||
1389 (!requeue_pi && this->rt_waiter)) {
1390 ret = -EINVAL;
1391 break;
1392 }
Darren Hart52400ba2009-04-03 13:40:49 -07001393
1394 /*
1395 * Wake nr_wake waiters. For requeue_pi, if we acquired the
1396 * lock, we already woke the top_waiter. If not, it will be
1397 * woken by futex_unlock_pi().
1398 */
1399 if (++task_count <= nr_wake && !requeue_pi) {
1400 wake_futex(this);
1401 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 }
Darren Hart52400ba2009-04-03 13:40:49 -07001403
Darren Hart84bc4af2009-08-13 17:36:53 -07001404 /* Ensure we requeue to the expected futex for requeue_pi. */
1405 if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) {
1406 ret = -EINVAL;
1407 break;
1408 }
1409
Darren Hart52400ba2009-04-03 13:40:49 -07001410 /*
1411 * Requeue nr_requeue waiters and possibly one more in the case
1412 * of requeue_pi if we couldn't acquire the lock atomically.
1413 */
1414 if (requeue_pi) {
1415 /* Prepare the waiter to take the rt_mutex. */
1416 atomic_inc(&pi_state->refcount);
1417 this->pi_state = pi_state;
1418 ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex,
1419 this->rt_waiter,
1420 this->task, 1);
1421 if (ret == 1) {
1422 /* We got the lock. */
Darren Hartbeda2c72009-08-09 15:34:39 -07001423 requeue_pi_wake_futex(this, &key2, hb2);
Darren Hart89061d32009-10-15 15:30:48 -07001424 drop_count++;
Darren Hart52400ba2009-04-03 13:40:49 -07001425 continue;
1426 } else if (ret) {
1427 /* -EDEADLK */
1428 this->pi_state = NULL;
1429 free_pi_state(pi_state);
1430 goto out_unlock;
1431 }
1432 }
1433 requeue_futex(this, hb1, hb2, &key2);
1434 drop_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 }
1436
1437out_unlock:
Darren Hart5eb3dc62009-03-12 00:55:52 -07001438 double_unlock_hb(hb1, hb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
Darren Hartcd84a422009-04-02 14:19:38 -07001440 /*
1441 * drop_futex_key_refs() must be called outside the spinlocks. During
1442 * the requeue we moved futex_q's from the hash bucket at key1 to the
1443 * one at key2 and updated their key pointer. We no longer need to
1444 * hold the references to key1.
1445 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 while (--drop_count >= 0)
Rusty Russell9adef582007-05-08 00:26:42 -07001447 drop_futex_key_refs(&key1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
Darren Hart42d35d42008-12-29 15:49:53 -08001449out_put_keys:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001450 put_futex_key(&key2);
Darren Hart42d35d42008-12-29 15:49:53 -08001451out_put_key1:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001452 put_futex_key(&key1);
Darren Hart42d35d42008-12-29 15:49:53 -08001453out:
Darren Hart52400ba2009-04-03 13:40:49 -07001454 if (pi_state != NULL)
1455 free_pi_state(pi_state);
1456 return ret ? ret : task_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457}
1458
1459/* The key must be already stored in q->key. */
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01001460static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
Namhyung Kim15e408c2010-09-14 21:43:48 +09001461 __acquires(&hb->lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462{
Ingo Molnare2970f22006-06-27 02:54:47 -07001463 struct futex_hash_bucket *hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
Ingo Molnare2970f22006-06-27 02:54:47 -07001465 hb = hash_futex(&q->key);
1466 q->lock_ptr = &hb->lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467
Ingo Molnare2970f22006-06-27 02:54:47 -07001468 spin_lock(&hb->lock);
1469 return hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470}
1471
Darren Hartd40d65c2009-09-21 22:30:15 -07001472static inline void
1473queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
Namhyung Kim15e408c2010-09-14 21:43:48 +09001474 __releases(&hb->lock)
Darren Hartd40d65c2009-09-21 22:30:15 -07001475{
1476 spin_unlock(&hb->lock);
Darren Hartd40d65c2009-09-21 22:30:15 -07001477}
1478
1479/**
1480 * queue_me() - Enqueue the futex_q on the futex_hash_bucket
1481 * @q: The futex_q to enqueue
1482 * @hb: The destination hash bucket
1483 *
1484 * The hb->lock must be held by the caller, and is released here. A call to
1485 * queue_me() is typically paired with exactly one call to unqueue_me(). The
1486 * exceptions involve the PI related operations, which may use unqueue_me_pi()
1487 * or nothing if the unqueue is done as part of the wake process and the unqueue
1488 * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
1489 * an example).
1490 */
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01001491static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
Namhyung Kim15e408c2010-09-14 21:43:48 +09001492 __releases(&hb->lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493{
Pierre Peifferec92d082007-05-09 02:35:00 -07001494 int prio;
1495
1496 /*
1497 * The priority used to register this element is
1498 * - either the real thread-priority for the real-time threads
1499 * (i.e. threads with a priority lower than MAX_RT_PRIO)
1500 * - or MAX_RT_PRIO for non-RT threads.
1501 * Thus, all RT-threads are woken first in priority order, and
1502 * the others are woken last, in FIFO order.
1503 */
1504 prio = min(current->normal_prio, MAX_RT_PRIO);
1505
1506 plist_node_init(&q->list, prio);
Pierre Peifferec92d082007-05-09 02:35:00 -07001507 plist_add(&q->list, &hb->chain);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001508 q->task = current;
Ingo Molnare2970f22006-06-27 02:54:47 -07001509 spin_unlock(&hb->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510}
1511
Darren Hartd40d65c2009-09-21 22:30:15 -07001512/**
1513 * unqueue_me() - Remove the futex_q from its futex_hash_bucket
1514 * @q: The futex_q to unqueue
1515 *
1516 * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must
1517 * be paired with exactly one earlier call to queue_me().
1518 *
1519 * Returns:
1520 * 1 - if the futex_q was still queued (and we removed unqueued it)
1521 * 0 - if the futex_q was already removed by the waking thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523static int unqueue_me(struct futex_q *q)
1524{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 spinlock_t *lock_ptr;
Ingo Molnare2970f22006-06-27 02:54:47 -07001526 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
1528 /* In the common case we don't take the spinlock, which is nice. */
Darren Hart42d35d42008-12-29 15:49:53 -08001529retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 lock_ptr = q->lock_ptr;
Christian Borntraegere91467e2006-08-05 12:13:52 -07001531 barrier();
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001532 if (lock_ptr != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 spin_lock(lock_ptr);
1534 /*
1535 * q->lock_ptr can change between reading it and
1536 * spin_lock(), causing us to take the wrong lock. This
1537 * corrects the race condition.
1538 *
1539 * Reasoning goes like this: if we have the wrong lock,
1540 * q->lock_ptr must have changed (maybe several times)
1541 * between reading it and the spin_lock(). It can
1542 * change again after the spin_lock() but only if it was
1543 * already changed before the spin_lock(). It cannot,
1544 * however, change back to the original value. Therefore
1545 * we can detect whether we acquired the correct lock.
1546 */
1547 if (unlikely(lock_ptr != q->lock_ptr)) {
1548 spin_unlock(lock_ptr);
1549 goto retry;
1550 }
Lai Jiangshan2e129782010-12-22 14:18:50 +08001551 __unqueue_futex(q);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001552
1553 BUG_ON(q->pi_state);
1554
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 spin_unlock(lock_ptr);
1556 ret = 1;
1557 }
1558
Rusty Russell9adef582007-05-08 00:26:42 -07001559 drop_futex_key_refs(&q->key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 return ret;
1561}
1562
Ingo Molnarc87e2832006-06-27 02:54:58 -07001563/*
1564 * PI futexes can not be requeued and must remove themself from the
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001565 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1566 * and dropped here.
Ingo Molnarc87e2832006-06-27 02:54:58 -07001567 */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001568static void unqueue_me_pi(struct futex_q *q)
Namhyung Kim15e408c2010-09-14 21:43:48 +09001569 __releases(q->lock_ptr)
Ingo Molnarc87e2832006-06-27 02:54:58 -07001570{
Lai Jiangshan2e129782010-12-22 14:18:50 +08001571 __unqueue_futex(q);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001572
1573 BUG_ON(!q->pi_state);
1574 free_pi_state(q->pi_state);
1575 q->pi_state = NULL;
1576
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001577 spin_unlock(q->lock_ptr);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001578}
1579
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001580/*
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001581 * Fixup the pi_state owner with the new owner.
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001582 *
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001583 * Must be called with hash bucket lock held and mm->sem held for non
1584 * private futexes.
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001585 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001586static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001587 struct task_struct *newowner)
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001588{
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001589 u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001590 struct futex_pi_state *pi_state = q->pi_state;
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001591 struct task_struct *oldowner = pi_state->owner;
Vitaliy Ivanov7cfdaf32011-07-07 15:10:31 +03001592 u32 uval, uninitialized_var(curval), newval;
Darren Harte4dc5b72009-03-12 00:56:13 -07001593 int ret;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001594
1595 /* Owner died? */
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001596 if (!pi_state->owner)
1597 newtid |= FUTEX_OWNER_DIED;
1598
1599 /*
1600 * We are here either because we stole the rtmutex from the
Lai Jiangshan81612392011-01-14 17:09:41 +08001601 * previous highest priority waiter or we are the highest priority
1602 * waiter but failed to get the rtmutex the first time.
1603 * We have to replace the newowner TID in the user space variable.
1604 * This must be atomic as we have to preserve the owner died bit here.
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001605 *
Darren Hartb2d09942009-03-12 00:55:37 -07001606 * Note: We write the user space value _before_ changing the pi_state
1607 * because we can fault here. Imagine swapped out pages or a fork
1608 * that marked all the anonymous memory readonly for cow.
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001609 *
1610 * Modifying pi_state _before_ the user space value would
1611 * leave the pi_state in an inconsistent state when we fault
1612 * here, because we need to drop the hash bucket lock to
1613 * handle the fault. This might be observed in the PID check
1614 * in lookup_pi_state.
1615 */
1616retry:
1617 if (get_futex_value_locked(&uval, uaddr))
1618 goto handle_fault;
1619
1620 while (1) {
1621 newval = (uval & FUTEX_OWNER_DIED) | newtid;
1622
Michel Lespinasse37a9d912011-03-10 18:48:51 -08001623 if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001624 goto handle_fault;
1625 if (curval == uval)
1626 break;
1627 uval = curval;
1628 }
1629
1630 /*
1631 * We fixed up user space. Now we need to fix the pi_state
1632 * itself.
1633 */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001634 if (pi_state->owner != NULL) {
Thomas Gleixner1d615482009-11-17 14:54:03 +01001635 raw_spin_lock_irq(&pi_state->owner->pi_lock);
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001636 WARN_ON(list_empty(&pi_state->list));
1637 list_del_init(&pi_state->list);
Thomas Gleixner1d615482009-11-17 14:54:03 +01001638 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001639 }
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001640
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001641 pi_state->owner = newowner;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001642
Thomas Gleixner1d615482009-11-17 14:54:03 +01001643 raw_spin_lock_irq(&newowner->pi_lock);
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001644 WARN_ON(!list_empty(&pi_state->list));
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001645 list_add(&pi_state->list, &newowner->pi_state_list);
Thomas Gleixner1d615482009-11-17 14:54:03 +01001646 raw_spin_unlock_irq(&newowner->pi_lock);
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001647 return 0;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001648
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001649 /*
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001650 * To handle the page fault we need to drop the hash bucket
Lai Jiangshan81612392011-01-14 17:09:41 +08001651 * lock here. That gives the other task (either the highest priority
1652 * waiter itself or the task which stole the rtmutex) the
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001653 * chance to try the fixup of the pi_state. So once we are
1654 * back from handling the fault we need to check the pi_state
1655 * after reacquiring the hash bucket lock and before trying to
1656 * do another fixup. When the fixup has been done already we
1657 * simply return.
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001658 */
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001659handle_fault:
1660 spin_unlock(q->lock_ptr);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001661
Thomas Gleixnerd0725992009-06-11 23:15:43 +02001662 ret = fault_in_user_writeable(uaddr);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001663
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001664 spin_lock(q->lock_ptr);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001665
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001666 /*
1667 * Check if someone else fixed it for us:
1668 */
1669 if (pi_state->owner != oldowner)
1670 return 0;
1671
1672 if (ret)
1673 return ret;
1674
1675 goto retry;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001676}
1677
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001678static long futex_wait_restart(struct restart_block *restart);
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001679
Darren Hartca5f9522009-04-03 13:39:33 -07001680/**
Darren Hartdd973992009-04-03 13:40:02 -07001681 * fixup_owner() - Post lock pi_state and corner case management
1682 * @uaddr: user address of the futex
Darren Hartdd973992009-04-03 13:40:02 -07001683 * @q: futex_q (contains pi_state and access to the rt_mutex)
1684 * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
1685 *
1686 * After attempting to lock an rt_mutex, this function is called to cleanup
1687 * the pi_state owner as well as handle race conditions that may allow us to
1688 * acquire the lock. Must be called with the hb lock held.
1689 *
1690 * Returns:
1691 * 1 - success, lock taken
1692 * 0 - success, lock not taken
1693 * <0 - on error (-EFAULT)
1694 */
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001695static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked)
Darren Hartdd973992009-04-03 13:40:02 -07001696{
1697 struct task_struct *owner;
1698 int ret = 0;
1699
1700 if (locked) {
1701 /*
1702 * Got the lock. We might not be the anticipated owner if we
1703 * did a lock-steal - fix up the PI-state in that case:
1704 */
1705 if (q->pi_state->owner != current)
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001706 ret = fixup_pi_state_owner(uaddr, q, current);
Darren Hartdd973992009-04-03 13:40:02 -07001707 goto out;
1708 }
1709
1710 /*
1711 * Catch the rare case, where the lock was released when we were on the
1712 * way back before we locked the hash bucket.
1713 */
1714 if (q->pi_state->owner == current) {
1715 /*
1716 * Try to get the rt_mutex now. This might fail as some other
1717 * task acquired the rt_mutex after we removed ourself from the
1718 * rt_mutex waiters list.
1719 */
1720 if (rt_mutex_trylock(&q->pi_state->pi_mutex)) {
1721 locked = 1;
1722 goto out;
1723 }
1724
1725 /*
1726 * pi_state is incorrect, some other task did a lock steal and
1727 * we returned due to timeout or signal without taking the
Lai Jiangshan81612392011-01-14 17:09:41 +08001728 * rt_mutex. Too late.
Darren Hartdd973992009-04-03 13:40:02 -07001729 */
Lai Jiangshan81612392011-01-14 17:09:41 +08001730 raw_spin_lock(&q->pi_state->pi_mutex.wait_lock);
Darren Hartdd973992009-04-03 13:40:02 -07001731 owner = rt_mutex_owner(&q->pi_state->pi_mutex);
Lai Jiangshan81612392011-01-14 17:09:41 +08001732 if (!owner)
1733 owner = rt_mutex_next_owner(&q->pi_state->pi_mutex);
1734 raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001735 ret = fixup_pi_state_owner(uaddr, q, owner);
Darren Hartdd973992009-04-03 13:40:02 -07001736 goto out;
1737 }
1738
1739 /*
1740 * Paranoia check. If we did not take the lock, then we should not be
Lai Jiangshan81612392011-01-14 17:09:41 +08001741 * the owner of the rt_mutex.
Darren Hartdd973992009-04-03 13:40:02 -07001742 */
1743 if (rt_mutex_owner(&q->pi_state->pi_mutex) == current)
1744 printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p "
1745 "pi-state %p\n", ret,
1746 q->pi_state->pi_mutex.owner,
1747 q->pi_state->owner);
1748
1749out:
1750 return ret ? ret : locked;
1751}
1752
1753/**
Darren Hartca5f9522009-04-03 13:39:33 -07001754 * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
1755 * @hb: the futex hash bucket, must be locked by the caller
1756 * @q: the futex_q to queue up on
1757 * @timeout: the prepared hrtimer_sleeper, or null for no timeout
Darren Hartca5f9522009-04-03 13:39:33 -07001758 */
1759static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02001760 struct hrtimer_sleeper *timeout)
Darren Hartca5f9522009-04-03 13:39:33 -07001761{
Darren Hart9beba3c2009-09-24 11:54:47 -07001762 /*
1763 * The task state is guaranteed to be set before another task can
1764 * wake it. set_current_state() is implemented using set_mb() and
1765 * queue_me() calls spin_unlock() upon completion, both serializing
1766 * access to the hash list and forcing another memory barrier.
1767 */
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02001768 set_current_state(TASK_INTERRUPTIBLE);
Darren Hart0729e192009-09-21 22:30:38 -07001769 queue_me(q, hb);
Darren Hartca5f9522009-04-03 13:39:33 -07001770
1771 /* Arm the timer */
1772 if (timeout) {
1773 hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
1774 if (!hrtimer_active(&timeout->timer))
1775 timeout->task = NULL;
1776 }
1777
1778 /*
Darren Hart0729e192009-09-21 22:30:38 -07001779 * If we have been removed from the hash list, then another task
1780 * has tried to wake us, and we can skip the call to schedule().
Darren Hartca5f9522009-04-03 13:39:33 -07001781 */
1782 if (likely(!plist_node_empty(&q->list))) {
1783 /*
1784 * If the timer has already expired, current will already be
1785 * flagged for rescheduling. Only call schedule if there
1786 * is no timeout, or if it has yet to expire.
1787 */
1788 if (!timeout || timeout->task)
1789 schedule();
1790 }
1791 __set_current_state(TASK_RUNNING);
1792}
1793
Darren Hartf8010732009-04-03 13:40:40 -07001794/**
1795 * futex_wait_setup() - Prepare to wait on a futex
1796 * @uaddr: the futex userspace address
1797 * @val: the expected value
Darren Hartb41277d2010-11-08 13:10:09 -08001798 * @flags: futex flags (FLAGS_SHARED, etc.)
Darren Hartf8010732009-04-03 13:40:40 -07001799 * @q: the associated futex_q
1800 * @hb: storage for hash_bucket pointer to be returned to caller
1801 *
1802 * Setup the futex_q and locate the hash_bucket. Get the futex value and
1803 * compare it with the expected value. Handle atomic faults internally.
1804 * Return with the hb lock held and a q.key reference on success, and unlocked
1805 * with no q.key reference on failure.
1806 *
1807 * Returns:
1808 * 0 - uaddr contains val and hb has been locked
Bart Van Asscheca4a04c2011-07-17 09:01:00 +02001809 * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked
Darren Hartf8010732009-04-03 13:40:40 -07001810 */
Darren Hartb41277d2010-11-08 13:10:09 -08001811static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
Darren Hartf8010732009-04-03 13:40:40 -07001812 struct futex_q *q, struct futex_hash_bucket **hb)
1813{
1814 u32 uval;
1815 int ret;
1816
1817 /*
1818 * Access the page AFTER the hash-bucket is locked.
1819 * Order is important:
1820 *
1821 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1822 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1823 *
1824 * The basic logical guarantee of a futex is that it blocks ONLY
1825 * if cond(var) is known to be true at the time of blocking, for
Michel Lespinasse8fe8f542011-03-06 18:07:50 -08001826 * any cond. If we locked the hash-bucket after testing *uaddr, that
1827 * would open a race condition where we could block indefinitely with
Darren Hartf8010732009-04-03 13:40:40 -07001828 * cond(var) false, which would violate the guarantee.
1829 *
Michel Lespinasse8fe8f542011-03-06 18:07:50 -08001830 * On the other hand, we insert q and release the hash-bucket only
1831 * after testing *uaddr. This guarantees that futex_wait() will NOT
1832 * absorb a wakeup if *uaddr does not match the desired values
1833 * while the syscall executes.
Darren Hartf8010732009-04-03 13:40:40 -07001834 */
1835retry:
Shawn Bohrer9ea71502011-06-30 11:21:32 -05001836 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ);
Darren Hartf8010732009-04-03 13:40:40 -07001837 if (unlikely(ret != 0))
Darren Harta5a2a0c2009-04-10 09:50:05 -07001838 return ret;
Darren Hartf8010732009-04-03 13:40:40 -07001839
1840retry_private:
1841 *hb = queue_lock(q);
1842
1843 ret = get_futex_value_locked(&uval, uaddr);
1844
1845 if (ret) {
1846 queue_unlock(q, *hb);
1847
1848 ret = get_user(uval, uaddr);
1849 if (ret)
1850 goto out;
1851
Darren Hartb41277d2010-11-08 13:10:09 -08001852 if (!(flags & FLAGS_SHARED))
Darren Hartf8010732009-04-03 13:40:40 -07001853 goto retry_private;
1854
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001855 put_futex_key(&q->key);
Darren Hartf8010732009-04-03 13:40:40 -07001856 goto retry;
1857 }
1858
1859 if (uval != val) {
1860 queue_unlock(q, *hb);
1861 ret = -EWOULDBLOCK;
1862 }
1863
1864out:
1865 if (ret)
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001866 put_futex_key(&q->key);
Darren Hartf8010732009-04-03 13:40:40 -07001867 return ret;
1868}
1869
Darren Hartb41277d2010-11-08 13:10:09 -08001870static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val,
1871 ktime_t *abs_time, u32 bitset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872{
Darren Hartca5f9522009-04-03 13:39:33 -07001873 struct hrtimer_sleeper timeout, *to = NULL;
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001874 struct restart_block *restart;
Ingo Molnare2970f22006-06-27 02:54:47 -07001875 struct futex_hash_bucket *hb;
Darren Hart5bdb05f2010-11-08 13:40:28 -08001876 struct futex_q q = futex_q_init;
Ingo Molnare2970f22006-06-27 02:54:47 -07001877 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878
Thomas Gleixnercd689982008-02-01 17:45:14 +01001879 if (!bitset)
1880 return -EINVAL;
Thomas Gleixnercd689982008-02-01 17:45:14 +01001881 q.bitset = bitset;
Darren Hartca5f9522009-04-03 13:39:33 -07001882
1883 if (abs_time) {
1884 to = &timeout;
1885
Darren Hartb41277d2010-11-08 13:10:09 -08001886 hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
1887 CLOCK_REALTIME : CLOCK_MONOTONIC,
1888 HRTIMER_MODE_ABS);
Darren Hartca5f9522009-04-03 13:39:33 -07001889 hrtimer_init_sleeper(to, current);
1890 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
1891 current->timer_slack_ns);
1892 }
1893
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02001894retry:
Darren Hart7ada8762010-10-17 08:35:04 -07001895 /*
1896 * Prepare to wait on uaddr. On success, holds hb lock and increments
1897 * q.key refs.
1898 */
Darren Hartb41277d2010-11-08 13:10:09 -08001899 ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
Darren Hartf8010732009-04-03 13:40:40 -07001900 if (ret)
Darren Hart42d35d42008-12-29 15:49:53 -08001901 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902
Darren Hartca5f9522009-04-03 13:39:33 -07001903 /* queue_me and wait for wakeup, timeout, or a signal. */
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02001904 futex_wait_queue_me(hb, &q, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
1906 /* If we were woken (and unqueued), we succeeded, whatever. */
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001907 ret = 0;
Darren Hart7ada8762010-10-17 08:35:04 -07001908 /* unqueue_me() drops q.key ref */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 if (!unqueue_me(&q))
Darren Hart7ada8762010-10-17 08:35:04 -07001910 goto out;
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001911 ret = -ETIMEDOUT;
Darren Hartca5f9522009-04-03 13:39:33 -07001912 if (to && !to->task)
Darren Hart7ada8762010-10-17 08:35:04 -07001913 goto out;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001914
Ingo Molnare2970f22006-06-27 02:54:47 -07001915 /*
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02001916 * We expect signal_pending(current), but we might be the
1917 * victim of a spurious wakeup as well.
Ingo Molnare2970f22006-06-27 02:54:47 -07001918 */
Darren Hart7ada8762010-10-17 08:35:04 -07001919 if (!signal_pending(current))
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02001920 goto retry;
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02001921
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001922 ret = -ERESTARTSYS;
Pierre Peifferc19384b2007-05-09 02:35:02 -07001923 if (!abs_time)
Darren Hart7ada8762010-10-17 08:35:04 -07001924 goto out;
Steven Rostedtce6bd422007-12-05 15:46:09 +01001925
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001926 restart = &current_thread_info()->restart_block;
1927 restart->fn = futex_wait_restart;
Namhyung Kima3c74c52010-09-14 21:43:47 +09001928 restart->futex.uaddr = uaddr;
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001929 restart->futex.val = val;
1930 restart->futex.time = abs_time->tv64;
1931 restart->futex.bitset = bitset;
Darren Hart0cd9c642011-04-14 15:41:57 -07001932 restart->futex.flags = flags | FLAGS_HAS_TIMEOUT;
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001933
1934 ret = -ERESTART_RESTARTBLOCK;
1935
Darren Hart42d35d42008-12-29 15:49:53 -08001936out:
Darren Hartca5f9522009-04-03 13:39:33 -07001937 if (to) {
1938 hrtimer_cancel(&to->timer);
1939 destroy_hrtimer_on_stack(&to->timer);
1940 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001941 return ret;
1942}
1943
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001944
1945static long futex_wait_restart(struct restart_block *restart)
1946{
Namhyung Kima3c74c52010-09-14 21:43:47 +09001947 u32 __user *uaddr = restart->futex.uaddr;
Darren Harta72188d2009-04-03 13:40:22 -07001948 ktime_t t, *tp = NULL;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001949
Darren Harta72188d2009-04-03 13:40:22 -07001950 if (restart->futex.flags & FLAGS_HAS_TIMEOUT) {
1951 t.tv64 = restart->futex.time;
1952 tp = &t;
1953 }
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001954 restart->fn = do_no_restart_syscall;
Darren Hartb41277d2010-11-08 13:10:09 -08001955
1956 return (long)futex_wait(uaddr, restart->futex.flags,
1957 restart->futex.val, tp, restart->futex.bitset);
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001958}
1959
1960
Ingo Molnarc87e2832006-06-27 02:54:58 -07001961/*
1962 * Userspace tried a 0 -> TID atomic transition of the futex value
1963 * and failed. The kernel side here does the whole locking operation:
1964 * if there are waiters then it will block, it does PI, etc. (Due to
1965 * races the kernel might see a 0 value of the futex too.)
1966 */
Darren Hartb41277d2010-11-08 13:10:09 -08001967static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect,
1968 ktime_t *time, int trylock)
Ingo Molnarc87e2832006-06-27 02:54:58 -07001969{
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001970 struct hrtimer_sleeper timeout, *to = NULL;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001971 struct futex_hash_bucket *hb;
Darren Hart5bdb05f2010-11-08 13:40:28 -08001972 struct futex_q q = futex_q_init;
Darren Hartdd973992009-04-03 13:40:02 -07001973 int res, ret;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001974
1975 if (refill_pi_state_cache())
1976 return -ENOMEM;
1977
Pierre Peifferc19384b2007-05-09 02:35:02 -07001978 if (time) {
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001979 to = &timeout;
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07001980 hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME,
1981 HRTIMER_MODE_ABS);
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001982 hrtimer_init_sleeper(to, current);
Arjan van de Vencc584b22008-09-01 15:02:30 -07001983 hrtimer_set_expires(&to->timer, *time);
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001984 }
1985
Darren Hart42d35d42008-12-29 15:49:53 -08001986retry:
Shawn Bohrer9ea71502011-06-30 11:21:32 -05001987 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001988 if (unlikely(ret != 0))
Darren Hart42d35d42008-12-29 15:49:53 -08001989 goto out;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001990
Darren Harte4dc5b72009-03-12 00:56:13 -07001991retry_private:
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01001992 hb = queue_lock(&q);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001993
Darren Hartbab5bc92009-04-07 23:23:50 -07001994 ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001995 if (unlikely(ret)) {
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001996 switch (ret) {
Darren Hart1a520842009-04-03 13:39:52 -07001997 case 1:
1998 /* We got the lock. */
1999 ret = 0;
2000 goto out_unlock_put_key;
2001 case -EFAULT:
2002 goto uaddr_faulted;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002003 case -EAGAIN:
2004 /*
2005 * Task is exiting and we just wait for the
2006 * exit to complete.
2007 */
2008 queue_unlock(&q, hb);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002009 put_futex_key(&q.key);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002010 cond_resched();
2011 goto retry;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002012 default:
Darren Hart42d35d42008-12-29 15:49:53 -08002013 goto out_unlock_put_key;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002014 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07002015 }
2016
2017 /*
2018 * Only actually queue now that the atomic ops are done:
2019 */
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01002020 queue_me(&q, hb);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002021
Ingo Molnarc87e2832006-06-27 02:54:58 -07002022 WARN_ON(!q.pi_state);
2023 /*
2024 * Block on the PI mutex:
2025 */
2026 if (!trylock)
2027 ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1);
2028 else {
2029 ret = rt_mutex_trylock(&q.pi_state->pi_mutex);
2030 /* Fixup the trylock return value: */
2031 ret = ret ? 0 : -EWOULDBLOCK;
2032 }
2033
Vernon Mauerya99e4e42006-07-01 04:35:42 -07002034 spin_lock(q.lock_ptr);
Darren Hartdd973992009-04-03 13:40:02 -07002035 /*
2036 * Fixup the pi_state owner and possibly acquire the lock if we
2037 * haven't already.
2038 */
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002039 res = fixup_owner(uaddr, &q, !ret);
Darren Hartdd973992009-04-03 13:40:02 -07002040 /*
2041 * If fixup_owner() returned an error, proprogate that. If it acquired
2042 * the lock, clear our -ETIMEDOUT or -EINTR.
2043 */
2044 if (res)
2045 ret = (res < 0) ? res : 0;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002046
Darren Harte8f63862009-03-12 00:56:06 -07002047 /*
Darren Hartdd973992009-04-03 13:40:02 -07002048 * If fixup_owner() faulted and was unable to handle the fault, unlock
2049 * it and return the fault to userspace.
Darren Harte8f63862009-03-12 00:56:06 -07002050 */
2051 if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current))
2052 rt_mutex_unlock(&q.pi_state->pi_mutex);
2053
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002054 /* Unqueue and drop the lock */
2055 unqueue_me_pi(&q);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002056
Mikael Pettersson5ecb01c2010-01-23 22:36:29 +01002057 goto out_put_key;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002058
Darren Hart42d35d42008-12-29 15:49:53 -08002059out_unlock_put_key:
Ingo Molnarc87e2832006-06-27 02:54:58 -07002060 queue_unlock(&q, hb);
2061
Darren Hart42d35d42008-12-29 15:49:53 -08002062out_put_key:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002063 put_futex_key(&q.key);
Darren Hart42d35d42008-12-29 15:49:53 -08002064out:
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07002065 if (to)
2066 destroy_hrtimer_on_stack(&to->timer);
Darren Hartdd973992009-04-03 13:40:02 -07002067 return ret != -EINTR ? ret : -ERESTARTNOINTR;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002068
Darren Hart42d35d42008-12-29 15:49:53 -08002069uaddr_faulted:
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002070 queue_unlock(&q, hb);
2071
Thomas Gleixnerd0725992009-06-11 23:15:43 +02002072 ret = fault_in_user_writeable(uaddr);
Darren Harte4dc5b72009-03-12 00:56:13 -07002073 if (ret)
2074 goto out_put_key;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002075
Darren Hartb41277d2010-11-08 13:10:09 -08002076 if (!(flags & FLAGS_SHARED))
Darren Harte4dc5b72009-03-12 00:56:13 -07002077 goto retry_private;
2078
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002079 put_futex_key(&q.key);
Darren Harte4dc5b72009-03-12 00:56:13 -07002080 goto retry;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002081}
2082
2083/*
Ingo Molnarc87e2832006-06-27 02:54:58 -07002084 * Userspace attempted a TID -> 0 atomic transition, and failed.
2085 * This is the in-kernel slowpath: we look up the PI state (if any),
2086 * and do the rt-mutex unlock.
2087 */
Darren Hartb41277d2010-11-08 13:10:09 -08002088static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags)
Ingo Molnarc87e2832006-06-27 02:54:58 -07002089{
2090 struct futex_hash_bucket *hb;
2091 struct futex_q *this, *next;
Pierre Peifferec92d082007-05-09 02:35:00 -07002092 struct plist_head *head;
Peter Zijlstra38d47c12008-09-26 19:32:20 +02002093 union futex_key key = FUTEX_KEY_INIT;
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +01002094 u32 uval, vpid = task_pid_vnr(current);
Darren Harte4dc5b72009-03-12 00:56:13 -07002095 int ret;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002096
2097retry:
2098 if (get_user(uval, uaddr))
2099 return -EFAULT;
2100 /*
2101 * We release only a lock we actually own:
2102 */
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +01002103 if ((uval & FUTEX_TID_MASK) != vpid)
Ingo Molnarc87e2832006-06-27 02:54:58 -07002104 return -EPERM;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002105
Shawn Bohrer9ea71502011-06-30 11:21:32 -05002106 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002107 if (unlikely(ret != 0))
2108 goto out;
2109
2110 hb = hash_futex(&key);
2111 spin_lock(&hb->lock);
2112
Ingo Molnarc87e2832006-06-27 02:54:58 -07002113 /*
2114 * To avoid races, try to do the TID -> 0 atomic transition
2115 * again. If it succeeds then we can return without waking
2116 * anyone else up:
2117 */
Michel Lespinasse37a9d912011-03-10 18:48:51 -08002118 if (!(uval & FUTEX_OWNER_DIED) &&
2119 cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0))
Ingo Molnarc87e2832006-06-27 02:54:58 -07002120 goto pi_faulted;
2121 /*
2122 * Rare case: we managed to release the lock atomically,
2123 * no need to wake anyone else up:
2124 */
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +01002125 if (unlikely(uval == vpid))
Ingo Molnarc87e2832006-06-27 02:54:58 -07002126 goto out_unlock;
2127
2128 /*
2129 * Ok, other tasks may need to be woken up - check waiters
2130 * and do the wakeup if necessary:
2131 */
2132 head = &hb->chain;
2133
Pierre Peifferec92d082007-05-09 02:35:00 -07002134 plist_for_each_entry_safe(this, next, head, list) {
Ingo Molnarc87e2832006-06-27 02:54:58 -07002135 if (!match_futex (&this->key, &key))
2136 continue;
2137 ret = wake_futex_pi(uaddr, uval, this);
2138 /*
2139 * The atomic access to the futex value
2140 * generated a pagefault, so retry the
2141 * user-access and the wakeup:
2142 */
2143 if (ret == -EFAULT)
2144 goto pi_faulted;
2145 goto out_unlock;
2146 }
2147 /*
2148 * No waiters - kernel unlocks the futex:
2149 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002150 if (!(uval & FUTEX_OWNER_DIED)) {
2151 ret = unlock_futex_pi(uaddr, uval);
2152 if (ret == -EFAULT)
2153 goto pi_faulted;
2154 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07002155
2156out_unlock:
2157 spin_unlock(&hb->lock);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002158 put_futex_key(&key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002159
Darren Hart42d35d42008-12-29 15:49:53 -08002160out:
Ingo Molnarc87e2832006-06-27 02:54:58 -07002161 return ret;
2162
2163pi_faulted:
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002164 spin_unlock(&hb->lock);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002165 put_futex_key(&key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002166
Thomas Gleixnerd0725992009-06-11 23:15:43 +02002167 ret = fault_in_user_writeable(uaddr);
Darren Hartb5686362008-12-18 15:06:34 -08002168 if (!ret)
Ingo Molnarc87e2832006-06-27 02:54:58 -07002169 goto retry;
2170
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 return ret;
2172}
2173
Darren Hart52400ba2009-04-03 13:40:49 -07002174/**
2175 * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2176 * @hb: the hash_bucket futex_q was original enqueued on
2177 * @q: the futex_q woken while waiting to be requeued
2178 * @key2: the futex_key of the requeue target futex
2179 * @timeout: the timeout associated with the wait (NULL if none)
2180 *
2181 * Detect if the task was woken on the initial futex as opposed to the requeue
2182 * target futex. If so, determine if it was a timeout or a signal that caused
2183 * the wakeup and return the appropriate error code to the caller. Must be
2184 * called with the hb lock held.
2185 *
2186 * Returns
2187 * 0 - no early wakeup detected
Thomas Gleixner1c840c12009-05-20 09:22:40 +02002188 * <0 - -ETIMEDOUT or -ERESTARTNOINTR
Darren Hart52400ba2009-04-03 13:40:49 -07002189 */
2190static inline
2191int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb,
2192 struct futex_q *q, union futex_key *key2,
2193 struct hrtimer_sleeper *timeout)
2194{
2195 int ret = 0;
2196
2197 /*
2198 * With the hb lock held, we avoid races while we process the wakeup.
2199 * We only need to hold hb (and not hb2) to ensure atomicity as the
2200 * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2201 * It can't be requeued from uaddr2 to something else since we don't
2202 * support a PI aware source futex for requeue.
2203 */
2204 if (!match_futex(&q->key, key2)) {
2205 WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr));
2206 /*
2207 * We were woken prior to requeue by a timeout or a signal.
2208 * Unqueue the futex_q and determine which it was.
2209 */
Lai Jiangshan2e129782010-12-22 14:18:50 +08002210 plist_del(&q->list, &hb->chain);
Darren Hart52400ba2009-04-03 13:40:49 -07002211
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02002212 /* Handle spurious wakeups gracefully */
Thomas Gleixner11df6dd2009-10-28 20:26:48 +01002213 ret = -EWOULDBLOCK;
Darren Hart52400ba2009-04-03 13:40:49 -07002214 if (timeout && !timeout->task)
2215 ret = -ETIMEDOUT;
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02002216 else if (signal_pending(current))
Thomas Gleixner1c840c12009-05-20 09:22:40 +02002217 ret = -ERESTARTNOINTR;
Darren Hart52400ba2009-04-03 13:40:49 -07002218 }
2219 return ret;
2220}
2221
2222/**
2223 * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
Darren Hart56ec1602009-09-21 22:29:59 -07002224 * @uaddr: the futex we initially wait on (non-pi)
Darren Hartb41277d2010-11-08 13:10:09 -08002225 * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be
Darren Hart52400ba2009-04-03 13:40:49 -07002226 * the same type, no requeueing from private to shared, etc.
2227 * @val: the expected value of uaddr
2228 * @abs_time: absolute timeout
Darren Hart56ec1602009-09-21 22:29:59 -07002229 * @bitset: 32 bit wakeup bitset set by userspace, defaults to all
Darren Hart52400ba2009-04-03 13:40:49 -07002230 * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0)
2231 * @uaddr2: the pi futex we will take prior to returning to user-space
2232 *
2233 * The caller will wait on uaddr and will be requeued by futex_requeue() to
2234 * uaddr2 which must be PI aware. Normal wakeup will wake on uaddr2 and
2235 * complete the acquisition of the rt_mutex prior to returning to userspace.
2236 * This ensures the rt_mutex maintains an owner when it has waiters; without
2237 * one, the pi logic wouldn't know which task to boost/deboost, if there was a
2238 * need to.
2239 *
2240 * We call schedule in futex_wait_queue_me() when we enqueue and return there
2241 * via the following:
2242 * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
Darren Hartcc6db4e2009-07-31 16:20:10 -07002243 * 2) wakeup on uaddr2 after a requeue
2244 * 3) signal
2245 * 4) timeout
Darren Hart52400ba2009-04-03 13:40:49 -07002246 *
Darren Hartcc6db4e2009-07-31 16:20:10 -07002247 * If 3, cleanup and return -ERESTARTNOINTR.
Darren Hart52400ba2009-04-03 13:40:49 -07002248 *
2249 * If 2, we may then block on trying to take the rt_mutex and return via:
2250 * 5) successful lock
2251 * 6) signal
2252 * 7) timeout
2253 * 8) other lock acquisition failure
2254 *
Darren Hartcc6db4e2009-07-31 16:20:10 -07002255 * If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
Darren Hart52400ba2009-04-03 13:40:49 -07002256 *
2257 * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2258 *
2259 * Returns:
2260 * 0 - On success
2261 * <0 - On error
2262 */
Darren Hartb41277d2010-11-08 13:10:09 -08002263static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
Darren Hart52400ba2009-04-03 13:40:49 -07002264 u32 val, ktime_t *abs_time, u32 bitset,
Darren Hartb41277d2010-11-08 13:10:09 -08002265 u32 __user *uaddr2)
Darren Hart52400ba2009-04-03 13:40:49 -07002266{
2267 struct hrtimer_sleeper timeout, *to = NULL;
2268 struct rt_mutex_waiter rt_waiter;
2269 struct rt_mutex *pi_mutex = NULL;
Darren Hart52400ba2009-04-03 13:40:49 -07002270 struct futex_hash_bucket *hb;
Darren Hart5bdb05f2010-11-08 13:40:28 -08002271 union futex_key key2 = FUTEX_KEY_INIT;
2272 struct futex_q q = futex_q_init;
Darren Hart52400ba2009-04-03 13:40:49 -07002273 int res, ret;
Darren Hart52400ba2009-04-03 13:40:49 -07002274
2275 if (!bitset)
2276 return -EINVAL;
2277
2278 if (abs_time) {
2279 to = &timeout;
Darren Hartb41277d2010-11-08 13:10:09 -08002280 hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
2281 CLOCK_REALTIME : CLOCK_MONOTONIC,
2282 HRTIMER_MODE_ABS);
Darren Hart52400ba2009-04-03 13:40:49 -07002283 hrtimer_init_sleeper(to, current);
2284 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
2285 current->timer_slack_ns);
2286 }
2287
2288 /*
2289 * The waiter is allocated on our stack, manipulated by the requeue
2290 * code while we sleep on uaddr.
2291 */
2292 debug_rt_mutex_init_waiter(&rt_waiter);
2293 rt_waiter.task = NULL;
2294
Shawn Bohrer9ea71502011-06-30 11:21:32 -05002295 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
Darren Hart52400ba2009-04-03 13:40:49 -07002296 if (unlikely(ret != 0))
2297 goto out;
2298
Darren Hart84bc4af2009-08-13 17:36:53 -07002299 q.bitset = bitset;
2300 q.rt_waiter = &rt_waiter;
2301 q.requeue_pi_key = &key2;
2302
Darren Hart7ada8762010-10-17 08:35:04 -07002303 /*
2304 * Prepare to wait on uaddr. On success, increments q.key (key1) ref
2305 * count.
2306 */
Darren Hartb41277d2010-11-08 13:10:09 -08002307 ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
Thomas Gleixnerc8b15a72009-05-20 09:18:50 +02002308 if (ret)
2309 goto out_key2;
Darren Hart52400ba2009-04-03 13:40:49 -07002310
2311 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02002312 futex_wait_queue_me(hb, &q, to);
Darren Hart52400ba2009-04-03 13:40:49 -07002313
2314 spin_lock(&hb->lock);
2315 ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
2316 spin_unlock(&hb->lock);
2317 if (ret)
2318 goto out_put_keys;
2319
2320 /*
2321 * In order for us to be here, we know our q.key == key2, and since
2322 * we took the hb->lock above, we also know that futex_requeue() has
2323 * completed and we no longer have to concern ourselves with a wakeup
Darren Hart7ada8762010-10-17 08:35:04 -07002324 * race with the atomic proxy lock acquisition by the requeue code. The
2325 * futex_requeue dropped our key1 reference and incremented our key2
2326 * reference count.
Darren Hart52400ba2009-04-03 13:40:49 -07002327 */
2328
2329 /* Check if the requeue code acquired the second futex for us. */
2330 if (!q.rt_waiter) {
2331 /*
2332 * Got the lock. We might not be the anticipated owner if we
2333 * did a lock-steal - fix up the PI-state in that case.
2334 */
2335 if (q.pi_state && (q.pi_state->owner != current)) {
2336 spin_lock(q.lock_ptr);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002337 ret = fixup_pi_state_owner(uaddr2, &q, current);
Darren Hart52400ba2009-04-03 13:40:49 -07002338 spin_unlock(q.lock_ptr);
2339 }
2340 } else {
2341 /*
2342 * We have been woken up by futex_unlock_pi(), a timeout, or a
2343 * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2344 * the pi_state.
2345 */
2346 WARN_ON(!&q.pi_state);
2347 pi_mutex = &q.pi_state->pi_mutex;
2348 ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1);
2349 debug_rt_mutex_free_waiter(&rt_waiter);
2350
2351 spin_lock(q.lock_ptr);
2352 /*
2353 * Fixup the pi_state owner and possibly acquire the lock if we
2354 * haven't already.
2355 */
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002356 res = fixup_owner(uaddr2, &q, !ret);
Darren Hart52400ba2009-04-03 13:40:49 -07002357 /*
2358 * If fixup_owner() returned an error, proprogate that. If it
Darren Hart56ec1602009-09-21 22:29:59 -07002359 * acquired the lock, clear -ETIMEDOUT or -EINTR.
Darren Hart52400ba2009-04-03 13:40:49 -07002360 */
2361 if (res)
2362 ret = (res < 0) ? res : 0;
2363
2364 /* Unqueue and drop the lock. */
2365 unqueue_me_pi(&q);
2366 }
2367
2368 /*
2369 * If fixup_pi_state_owner() faulted and was unable to handle the
2370 * fault, unlock the rt_mutex and return the fault to userspace.
2371 */
2372 if (ret == -EFAULT) {
2373 if (rt_mutex_owner(pi_mutex) == current)
2374 rt_mutex_unlock(pi_mutex);
2375 } else if (ret == -EINTR) {
Darren Hart52400ba2009-04-03 13:40:49 -07002376 /*
Darren Hartcc6db4e2009-07-31 16:20:10 -07002377 * We've already been requeued, but cannot restart by calling
2378 * futex_lock_pi() directly. We could restart this syscall, but
2379 * it would detect that the user space "val" changed and return
2380 * -EWOULDBLOCK. Save the overhead of the restart and return
2381 * -EWOULDBLOCK directly.
Darren Hart52400ba2009-04-03 13:40:49 -07002382 */
Thomas Gleixner20708872009-05-19 23:04:59 +02002383 ret = -EWOULDBLOCK;
Darren Hart52400ba2009-04-03 13:40:49 -07002384 }
2385
2386out_put_keys:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002387 put_futex_key(&q.key);
Thomas Gleixnerc8b15a72009-05-20 09:18:50 +02002388out_key2:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002389 put_futex_key(&key2);
Darren Hart52400ba2009-04-03 13:40:49 -07002390
2391out:
2392 if (to) {
2393 hrtimer_cancel(&to->timer);
2394 destroy_hrtimer_on_stack(&to->timer);
2395 }
2396 return ret;
2397}
2398
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002399/*
2400 * Support for robust futexes: the kernel cleans up held futexes at
2401 * thread exit time.
2402 *
2403 * Implementation: user-space maintains a per-thread list of locks it
2404 * is holding. Upon do_exit(), the kernel carefully walks this list,
2405 * and marks all locks that are owned by this thread with the
Ingo Molnarc87e2832006-06-27 02:54:58 -07002406 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002407 * always manipulated with the lock held, so the list is private and
2408 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2409 * field, to allow the kernel to clean up if the thread dies after
2410 * acquiring the lock, but just before it could have added itself to
2411 * the list. There can only be one such pending lock.
2412 */
2413
2414/**
Darren Hartd96ee562009-09-21 22:30:22 -07002415 * sys_set_robust_list() - Set the robust-futex list head of a task
2416 * @head: pointer to the list-head
2417 * @len: length of the list-head, as userspace expects
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002418 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01002419SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head,
2420 size_t, len)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002421{
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002422 if (!futex_cmpxchg_enabled)
2423 return -ENOSYS;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002424 /*
2425 * The kernel knows only one size for now:
2426 */
2427 if (unlikely(len != sizeof(*head)))
2428 return -EINVAL;
2429
2430 current->robust_list = head;
2431
2432 return 0;
2433}
2434
2435/**
Darren Hartd96ee562009-09-21 22:30:22 -07002436 * sys_get_robust_list() - Get the robust-futex list head of a task
2437 * @pid: pid of the process [zero for current task]
2438 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2439 * @len_ptr: pointer to a length field, the kernel fills in the header size
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002440 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01002441SYSCALL_DEFINE3(get_robust_list, int, pid,
2442 struct robust_list_head __user * __user *, head_ptr,
2443 size_t __user *, len_ptr)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002444{
Al Viroba46df92006-10-10 22:46:07 +01002445 struct robust_list_head __user *head;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002446 unsigned long ret;
Kees Cookbdbb7762012-03-19 16:12:53 -07002447 struct task_struct *p;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002448
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002449 if (!futex_cmpxchg_enabled)
2450 return -ENOSYS;
2451
Kees Cookec0c4272012-03-23 12:08:55 -07002452 WARN_ONCE(1, "deprecated: get_robust_list will be deleted in 2013.\n");
2453
Kees Cookbdbb7762012-03-19 16:12:53 -07002454 rcu_read_lock();
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002455
Kees Cookbdbb7762012-03-19 16:12:53 -07002456 ret = -ESRCH;
2457 if (!pid)
2458 p = current;
2459 else {
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07002460 p = find_task_by_vpid(pid);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002461 if (!p)
2462 goto err_unlock;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002463 }
2464
Kees Cookbdbb7762012-03-19 16:12:53 -07002465 ret = -EPERM;
2466 if (!ptrace_may_access(p, PTRACE_MODE_READ))
2467 goto err_unlock;
2468
2469 head = p->robust_list;
2470 rcu_read_unlock();
2471
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002472 if (put_user(sizeof(*head), len_ptr))
2473 return -EFAULT;
2474 return put_user(head, head_ptr);
2475
2476err_unlock:
Oleg Nesterovaaa2a972006-09-29 02:00:55 -07002477 rcu_read_unlock();
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002478
2479 return ret;
2480}
2481
2482/*
2483 * Process a futex-list entry, check whether it's owned by the
2484 * dying task, and do notification if so:
2485 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002486int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002487{
Vitaliy Ivanov7cfdaf32011-07-07 15:10:31 +03002488 u32 uval, uninitialized_var(nval), mval;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002489
Ingo Molnar8f17d3a2006-03-27 01:16:27 -08002490retry:
2491 if (get_user(uval, uaddr))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002492 return -1;
2493
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002494 if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) {
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002495 /*
2496 * Ok, this dying thread is truly holding a futex
2497 * of interest. Set the OWNER_DIED bit atomically
2498 * via cmpxchg, and if the value had FUTEX_WAITERS
2499 * set, wake up a waiter (if any). (We have to do a
2500 * futex_wake() even if OWNER_DIED is already set -
2501 * to handle the rare but possible case of recursive
2502 * thread-death.) The rest of the cleanup is done in
2503 * userspace.
2504 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002505 mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
Thomas Gleixner6e0aa9f2011-03-14 10:34:35 +01002506 /*
2507 * We are not holding a lock here, but we want to have
2508 * the pagefault_disable/enable() protection because
2509 * we want to handle the fault gracefully. If the
2510 * access fails we try to fault in the futex with R/W
2511 * verification via get_user_pages. get_user() above
2512 * does not guarantee R/W access. If that fails we
2513 * give up and leave the futex locked.
2514 */
2515 if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) {
2516 if (fault_in_user_writeable(uaddr))
2517 return -1;
2518 goto retry;
2519 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07002520 if (nval != uval)
Ingo Molnar8f17d3a2006-03-27 01:16:27 -08002521 goto retry;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002522
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002523 /*
2524 * Wake robust non-PI futexes here. The wakeup of
2525 * PI futexes happens in exit_pi_state():
2526 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07002527 if (!pi && (uval & FUTEX_WAITERS))
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02002528 futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002529 }
2530 return 0;
2531}
2532
2533/*
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002534 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2535 */
2536static inline int fetch_robust_entry(struct robust_list __user **entry,
Al Viroba46df92006-10-10 22:46:07 +01002537 struct robust_list __user * __user *head,
Namhyung Kim1dcc41b2010-09-14 21:43:46 +09002538 unsigned int *pi)
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002539{
2540 unsigned long uentry;
2541
Al Viroba46df92006-10-10 22:46:07 +01002542 if (get_user(uentry, (unsigned long __user *)head))
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002543 return -EFAULT;
2544
Al Viroba46df92006-10-10 22:46:07 +01002545 *entry = (void __user *)(uentry & ~1UL);
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002546 *pi = uentry & 1;
2547
2548 return 0;
2549}
2550
2551/*
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002552 * Walk curr->robust_list (very carefully, it's a userspace list!)
2553 * and mark any locks found there dead, and notify any waiters.
2554 *
2555 * We silently return on any sign of list-walking problem.
2556 */
2557void exit_robust_list(struct task_struct *curr)
2558{
2559 struct robust_list_head __user *head = curr->robust_list;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002560 struct robust_list __user *entry, *next_entry, *pending;
Darren Hart4c115e92010-11-04 15:00:00 -04002561 unsigned int limit = ROBUST_LIST_LIMIT, pi, pip;
2562 unsigned int uninitialized_var(next_pi);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002563 unsigned long futex_offset;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002564 int rc;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002565
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002566 if (!futex_cmpxchg_enabled)
2567 return;
2568
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002569 /*
2570 * Fetch the list head (which was registered earlier, via
2571 * sys_set_robust_list()):
2572 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002573 if (fetch_robust_entry(&entry, &head->list.next, &pi))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002574 return;
2575 /*
2576 * Fetch the relative futex offset:
2577 */
2578 if (get_user(futex_offset, &head->futex_offset))
2579 return;
2580 /*
2581 * Fetch any possibly pending lock-add first, and handle it
2582 * if it exists:
2583 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002584 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002585 return;
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002586
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002587 next_entry = NULL; /* avoid warning with gcc */
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002588 while (entry != &head->list) {
2589 /*
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002590 * Fetch the next entry in the list before calling
2591 * handle_futex_death:
2592 */
2593 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
2594 /*
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002595 * A pending lock might already be on the list, so
Ingo Molnarc87e2832006-06-27 02:54:58 -07002596 * don't process it twice:
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002597 */
2598 if (entry != pending)
Al Viroba46df92006-10-10 22:46:07 +01002599 if (handle_futex_death((void __user *)entry + futex_offset,
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002600 curr, pi))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002601 return;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002602 if (rc)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002603 return;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002604 entry = next_entry;
2605 pi = next_pi;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002606 /*
2607 * Avoid excessively long or circular lists:
2608 */
2609 if (!--limit)
2610 break;
2611
2612 cond_resched();
2613 }
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002614
2615 if (pending)
2616 handle_futex_death((void __user *)pending + futex_offset,
2617 curr, pip);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002618}
2619
Pierre Peifferc19384b2007-05-09 02:35:02 -07002620long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
Ingo Molnare2970f22006-06-27 02:54:47 -07002621 u32 __user *uaddr2, u32 val2, u32 val3)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622{
Thomas Gleixner81b40532012-02-15 12:17:09 +01002623 int cmd = op & FUTEX_CMD_MASK;
Darren Hartb41277d2010-11-08 13:10:09 -08002624 unsigned int flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002626 if (!(op & FUTEX_PRIVATE_FLAG))
Darren Hartb41277d2010-11-08 13:10:09 -08002627 flags |= FLAGS_SHARED;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002628
Darren Hartb41277d2010-11-08 13:10:09 -08002629 if (op & FUTEX_CLOCK_REALTIME) {
2630 flags |= FLAGS_CLOCKRT;
2631 if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
2632 return -ENOSYS;
2633 }
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002634
2635 switch (cmd) {
Thomas Gleixner59263b52012-02-15 12:08:34 +01002636 case FUTEX_LOCK_PI:
2637 case FUTEX_UNLOCK_PI:
2638 case FUTEX_TRYLOCK_PI:
2639 case FUTEX_WAIT_REQUEUE_PI:
2640 case FUTEX_CMP_REQUEUE_PI:
2641 if (!futex_cmpxchg_enabled)
2642 return -ENOSYS;
2643 }
2644
2645 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 case FUTEX_WAIT:
Thomas Gleixnercd689982008-02-01 17:45:14 +01002647 val3 = FUTEX_BITSET_MATCH_ANY;
2648 case FUTEX_WAIT_BITSET:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002649 return futex_wait(uaddr, flags, val, timeout, val3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 case FUTEX_WAKE:
Thomas Gleixnercd689982008-02-01 17:45:14 +01002651 val3 = FUTEX_BITSET_MATCH_ANY;
2652 case FUTEX_WAKE_BITSET:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002653 return futex_wake(uaddr, flags, val, val3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 case FUTEX_REQUEUE:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002655 return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 case FUTEX_CMP_REQUEUE:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002657 return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07002658 case FUTEX_WAKE_OP:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002659 return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002660 case FUTEX_LOCK_PI:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002661 return futex_lock_pi(uaddr, flags, val, timeout, 0);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002662 case FUTEX_UNLOCK_PI:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002663 return futex_unlock_pi(uaddr, flags);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002664 case FUTEX_TRYLOCK_PI:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002665 return futex_lock_pi(uaddr, flags, 0, timeout, 1);
Darren Hart52400ba2009-04-03 13:40:49 -07002666 case FUTEX_WAIT_REQUEUE_PI:
2667 val3 = FUTEX_BITSET_MATCH_ANY;
Thomas Gleixner81b40532012-02-15 12:17:09 +01002668 return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3,
2669 uaddr2);
Darren Hart52400ba2009-04-03 13:40:49 -07002670 case FUTEX_CMP_REQUEUE_PI:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002671 return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 }
Thomas Gleixner81b40532012-02-15 12:17:09 +01002673 return -ENOSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674}
2675
2676
Heiko Carstens17da2bd2009-01-14 14:14:10 +01002677SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
2678 struct timespec __user *, utime, u32 __user *, uaddr2,
2679 u32, val3)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680{
Pierre Peifferc19384b2007-05-09 02:35:02 -07002681 struct timespec ts;
2682 ktime_t t, *tp = NULL;
Ingo Molnare2970f22006-06-27 02:54:47 -07002683 u32 val2 = 0;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002684 int cmd = op & FUTEX_CMD_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685
Thomas Gleixnercd689982008-02-01 17:45:14 +01002686 if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
Darren Hart52400ba2009-04-03 13:40:49 -07002687 cmd == FUTEX_WAIT_BITSET ||
2688 cmd == FUTEX_WAIT_REQUEUE_PI)) {
Pierre Peifferc19384b2007-05-09 02:35:02 -07002689 if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 return -EFAULT;
Pierre Peifferc19384b2007-05-09 02:35:02 -07002691 if (!timespec_valid(&ts))
Thomas Gleixner9741ef962006-03-31 02:31:32 -08002692 return -EINVAL;
Pierre Peifferc19384b2007-05-09 02:35:02 -07002693
2694 t = timespec_to_ktime(ts);
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002695 if (cmd == FUTEX_WAIT)
Thomas Gleixner5a7780e2008-02-13 09:20:43 +01002696 t = ktime_add_safe(ktime_get(), t);
Pierre Peifferc19384b2007-05-09 02:35:02 -07002697 tp = &t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 }
2699 /*
Darren Hart52400ba2009-04-03 13:40:49 -07002700 * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
Andreas Schwabf54f0982007-07-31 00:38:51 -07002701 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 */
Andreas Schwabf54f0982007-07-31 00:38:51 -07002703 if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
Darren Hartba9c22f2009-04-20 22:22:22 -07002704 cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP)
Ingo Molnare2970f22006-06-27 02:54:47 -07002705 val2 = (u32) (unsigned long) utime;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706
Pierre Peifferc19384b2007-05-09 02:35:02 -07002707 return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708}
2709
Benjamin Herrenschmidtf6d107f2008-03-27 14:52:15 +11002710static int __init futex_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711{
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002712 u32 curval;
Thomas Gleixner3e4ab742008-02-23 15:23:55 -08002713 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002715 /*
2716 * This will fail and we want it. Some arch implementations do
2717 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2718 * functionality. We want to know that before we call in any
2719 * of the complex code paths. Also we want to prevent
2720 * registration of robust lists in that case. NULL is
2721 * guaranteed to fault and we get -EFAULT on functional
Randy Dunlapfb62db22010-10-13 11:02:34 -07002722 * implementation, the non-functional ones will return
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002723 * -ENOSYS.
2724 */
Michel Lespinasse37a9d912011-03-10 18:48:51 -08002725 if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT)
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002726 futex_cmpxchg_enabled = 1;
2727
Thomas Gleixner3e4ab742008-02-23 15:23:55 -08002728 for (i = 0; i < ARRAY_SIZE(futex_queues); i++) {
Dima Zavin732375c2011-07-07 17:27:59 -07002729 plist_head_init(&futex_queues[i].chain);
Thomas Gleixner3e4ab742008-02-23 15:23:55 -08002730 spin_lock_init(&futex_queues[i].lock);
2731 }
2732
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 return 0;
2734}
Benjamin Herrenschmidtf6d107f2008-03-27 14:52:15 +11002735__initcall(futex_init);