blob: b26dcfc02c9489b3ca00bfd076f2208bb4964366 [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>
Clark Williams8bd75c72013-02-07 09:47:07 -060063#include <linux/sched/rt.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070064
Jakub Jelinek4732efbe2005-09-06 15:16:25 -070065#include <asm/futex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Ingo Molnarc87e2832006-06-27 02:54:58 -070067#include "rtmutex_common.h"
68
Thomas Gleixnera0c1e902008-02-23 15:23:57 -080069int __read_mostly futex_cmpxchg_enabled;
70
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
72
73/*
Darren Hartb41277d2010-11-08 13:10:09 -080074 * Futex flags used to encode options to functions and preserve them across
75 * restarts.
76 */
77#define FLAGS_SHARED 0x01
78#define FLAGS_CLOCKRT 0x02
79#define FLAGS_HAS_TIMEOUT 0x04
80
81/*
Ingo Molnarc87e2832006-06-27 02:54:58 -070082 * Priority Inheritance state:
83 */
84struct futex_pi_state {
85 /*
86 * list of 'owned' pi_state instances - these have to be
87 * cleaned up in do_exit() if the task exits prematurely:
88 */
89 struct list_head list;
90
91 /*
92 * The PI object:
93 */
94 struct rt_mutex pi_mutex;
95
96 struct task_struct *owner;
97 atomic_t refcount;
98
99 union futex_key key;
100};
101
Darren Hartd8d88fb2009-09-21 22:30:30 -0700102/**
103 * struct futex_q - The hashed futex queue entry, one per waiting task
Randy Dunlapfb62db22010-10-13 11:02:34 -0700104 * @list: priority-sorted list of tasks waiting on this futex
Darren Hartd8d88fb2009-09-21 22:30:30 -0700105 * @task: the task waiting on the futex
106 * @lock_ptr: the hash bucket lock
107 * @key: the key the futex is hashed on
108 * @pi_state: optional priority inheritance state
109 * @rt_waiter: rt_waiter storage for use with requeue_pi
110 * @requeue_pi_key: the requeue_pi target futex key
111 * @bitset: bitset for the optional bitmasked wakeup
112 *
113 * We use this hashed waitqueue, instead of a normal wait_queue_t, so
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 * we can wake only the relevant ones (hashed queues may be shared).
115 *
116 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
Pierre Peifferec92d082007-05-09 02:35:00 -0700117 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
Randy Dunlapfb62db22010-10-13 11:02:34 -0700118 * The order of wakeup is always to make the first condition true, then
Darren Hartd8d88fb2009-09-21 22:30:30 -0700119 * the second.
120 *
121 * PI futexes are typically woken before they are removed from the hash list via
122 * the rt_mutex code. See unqueue_me_pi().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 */
124struct futex_q {
Pierre Peifferec92d082007-05-09 02:35:00 -0700125 struct plist_node list;
Darren Hartd8d88fb2009-09-21 22:30:30 -0700126
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200127 struct task_struct *task;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 spinlock_t *lock_ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 union futex_key key;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700130 struct futex_pi_state *pi_state;
Darren Hart52400ba2009-04-03 13:40:49 -0700131 struct rt_mutex_waiter *rt_waiter;
Darren Hart84bc4af2009-08-13 17:36:53 -0700132 union futex_key *requeue_pi_key;
Thomas Gleixnercd6899852008-02-01 17:45:14 +0100133 u32 bitset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134};
135
Darren Hart5bdb05f2010-11-08 13:40:28 -0800136static const struct futex_q futex_q_init = {
137 /* list gets initialized in queue_me()*/
138 .key = FUTEX_KEY_INIT,
139 .bitset = FUTEX_BITSET_MATCH_ANY
140};
141
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142/*
Darren Hartb2d09942009-03-12 00:55:37 -0700143 * Hash buckets are shared by all the futex_keys that hash to the same
144 * location. Each key may have multiple futex_q structures, one for each task
145 * waiting on a futex.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 */
147struct futex_hash_bucket {
Pierre Peifferec92d082007-05-09 02:35:00 -0700148 spinlock_t lock;
149 struct plist_head chain;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150};
151
152static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS];
153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154/*
155 * We hash on the keys returned from get_futex_key (see below).
156 */
157static struct futex_hash_bucket *hash_futex(union futex_key *key)
158{
159 u32 hash = jhash2((u32*)&key->both.word,
160 (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
161 key->both.offset);
162 return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)];
163}
164
165/*
166 * Return 1 if two futex_keys are equal, 0 otherwise.
167 */
168static inline int match_futex(union futex_key *key1, union futex_key *key2)
169{
Darren Hart2bc87202009-10-14 10:12:39 -0700170 return (key1 && key2
171 && key1->both.word == key2->both.word
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 && key1->both.ptr == key2->both.ptr
173 && key1->both.offset == key2->both.offset);
174}
175
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200176/*
177 * Take a reference to the resource addressed by a key.
178 * Can be called while holding spinlocks.
179 *
180 */
181static void get_futex_key_refs(union futex_key *key)
182{
183 if (!key->both.ptr)
184 return;
185
186 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
187 case FUT_OFF_INODE:
Al Viro7de9c6ee2010-10-23 11:11:40 -0400188 ihold(key->shared.inode);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200189 break;
190 case FUT_OFF_MMSHARED:
191 atomic_inc(&key->private.mm->mm_count);
192 break;
193 }
194}
195
196/*
197 * Drop a reference to the resource addressed by a key.
198 * The hash bucket spinlock must not be held.
199 */
200static void drop_futex_key_refs(union futex_key *key)
201{
Darren Hart90621c42008-12-29 19:43:21 -0800202 if (!key->both.ptr) {
203 /* If we're here then we tried to put a key we failed to get */
204 WARN_ON_ONCE(1);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200205 return;
Darren Hart90621c42008-12-29 19:43:21 -0800206 }
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200207
208 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
209 case FUT_OFF_INODE:
210 iput(key->shared.inode);
211 break;
212 case FUT_OFF_MMSHARED:
213 mmdrop(key->private.mm);
214 break;
215 }
216}
217
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700218/**
Darren Hartd96ee562009-09-21 22:30:22 -0700219 * get_futex_key() - Get parameters which are the keys for a futex
220 * @uaddr: virtual address of the futex
221 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
222 * @key: address where result is stored.
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500223 * @rw: mapping needs to be read/write (values: VERIFY_READ,
224 * VERIFY_WRITE)
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700225 *
Randy Dunlap6c23cbb2013-03-05 10:00:24 -0800226 * Return: a negative error code or 0
227 *
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700228 * The key words are stored in *key on success.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 *
Al Viro6131ffa2013-02-27 16:59:05 -0500230 * For shared mappings, it's (page->index, file_inode(vma->vm_file),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 * offset_within_page). For private mappings, it's (uaddr, current->mm).
232 * We can usually work out the index without swapping in the page.
233 *
Darren Hartb2d09942009-03-12 00:55:37 -0700234 * lock_page() might sleep, the caller should not hold a spinlock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 */
Thomas Gleixner64d13042009-05-18 21:20:10 +0200236static int
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500237get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
Ingo Molnare2970f22006-06-27 02:54:47 -0700239 unsigned long address = (unsigned long)uaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 struct mm_struct *mm = current->mm;
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800241 struct page *page, *page_head;
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500242 int err, ro = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
244 /*
245 * The futex address must be "naturally" aligned.
246 */
Ingo Molnare2970f22006-06-27 02:54:47 -0700247 key->both.offset = address % PAGE_SIZE;
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700248 if (unlikely((address % sizeof(u32)) != 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 return -EINVAL;
Ingo Molnare2970f22006-06-27 02:54:47 -0700250 address -= key->both.offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252 /*
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700253 * PROCESS_PRIVATE futexes are fast.
254 * As the mm cannot disappear under us and the 'key' only needs
255 * virtual address, we dont even have to find the underlying vma.
256 * Note : We do have to check 'uaddr' is a valid user address,
257 * but access_ok() should be faster than find_vma()
258 */
259 if (!fshared) {
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +0900260 if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))))
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700261 return -EFAULT;
262 key->private.mm = mm;
263 key->private.address = address;
Peter Zijlstra42569c32008-09-30 12:33:07 +0200264 get_futex_key_refs(key);
Eric Dumazet34f01cc2007-05-09 02:35:04 -0700265 return 0;
266 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200268again:
KOSAKI Motohiro7485d0d2010-01-05 16:32:43 +0900269 err = get_user_pages_fast(address, 1, 1, &page);
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500270 /*
271 * If write access is not required (eg. FUTEX_WAIT), try
272 * and get read-only access.
273 */
274 if (err == -EFAULT && rw == VERIFY_READ) {
275 err = get_user_pages_fast(address, 1, 0, &page);
276 ro = 1;
277 }
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200278 if (err < 0)
279 return err;
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500280 else
281 err = 0;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200282
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800283#ifdef CONFIG_TRANSPARENT_HUGEPAGE
284 page_head = page;
285 if (unlikely(PageTail(page))) {
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200286 put_page(page);
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800287 /* serialize against __split_huge_page_splitting() */
288 local_irq_disable();
289 if (likely(__get_user_pages_fast(address, 1, 1, &page) == 1)) {
290 page_head = compound_head(page);
291 /*
292 * page_head is valid pointer but we must pin
293 * it before taking the PG_lock and/or
294 * PG_compound_lock. The moment we re-enable
295 * irqs __split_huge_page_splitting() can
296 * return and the head page can be freed from
297 * under us. We can't take the PG_lock and/or
298 * PG_compound_lock on a page that could be
299 * freed from under us.
300 */
301 if (page != page_head) {
302 get_page(page_head);
303 put_page(page);
304 }
305 local_irq_enable();
306 } else {
307 local_irq_enable();
308 goto again;
309 }
310 }
311#else
312 page_head = compound_head(page);
313 if (page != page_head) {
314 get_page(page_head);
315 put_page(page);
316 }
317#endif
318
319 lock_page(page_head);
Hugh Dickinse6780f72011-12-31 11:44:01 -0800320
321 /*
322 * If page_head->mapping is NULL, then it cannot be a PageAnon
323 * page; but it might be the ZERO_PAGE or in the gate area or
324 * in a special mapping (all cases which we are happy to fail);
325 * or it may have been a good file page when get_user_pages_fast
326 * found it, but truncated or holepunched or subjected to
327 * invalidate_complete_page2 before we got the page lock (also
328 * cases which we are happy to fail). And we hold a reference,
329 * so refcount care in invalidate_complete_page's remove_mapping
330 * prevents drop_caches from setting mapping to NULL beneath us.
331 *
332 * The case we do have to guard against is when memory pressure made
333 * shmem_writepage move it from filecache to swapcache beneath us:
334 * an unlikely race, but we do need to retry for page_head->mapping.
335 */
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800336 if (!page_head->mapping) {
Hugh Dickinse6780f72011-12-31 11:44:01 -0800337 int shmem_swizzled = PageSwapCache(page_head);
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800338 unlock_page(page_head);
339 put_page(page_head);
Hugh Dickinse6780f72011-12-31 11:44:01 -0800340 if (shmem_swizzled)
341 goto again;
342 return -EFAULT;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200343 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
345 /*
346 * Private mappings are handled in a simple way.
347 *
348 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
349 * it's a read-only handle, it's expected that futexes attach to
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200350 * the object not the particular process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 */
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800352 if (PageAnon(page_head)) {
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500353 /*
354 * A RO anonymous page will never change and thus doesn't make
355 * sense for futex operations.
356 */
357 if (ro) {
358 err = -EFAULT;
359 goto out;
360 }
361
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200362 key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 key->private.mm = mm;
Ingo Molnare2970f22006-06-27 02:54:47 -0700364 key->private.address = address;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200365 } else {
366 key->both.offset |= FUT_OFF_INODE; /* inode-based key */
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800367 key->shared.inode = page_head->mapping->host;
368 key->shared.pgoff = page_head->index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 }
370
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200371 get_futex_key_refs(key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500373out:
Andrea Arcangelia5b338f2011-01-13 15:46:34 -0800374 unlock_page(page_head);
375 put_page(page_head);
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500376 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377}
378
Thomas Gleixnerae791a22010-11-10 13:30:36 +0100379static inline void put_futex_key(union futex_key *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380{
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200381 drop_futex_key_refs(key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382}
383
Darren Hartd96ee562009-09-21 22:30:22 -0700384/**
385 * fault_in_user_writeable() - Fault in user address and verify RW access
Thomas Gleixnerd0725992009-06-11 23:15:43 +0200386 * @uaddr: pointer to faulting user space address
387 *
388 * Slow path to fixup the fault we just took in the atomic write
389 * access to @uaddr.
390 *
Randy Dunlapfb62db22010-10-13 11:02:34 -0700391 * We have no generic implementation of a non-destructive write to the
Thomas Gleixnerd0725992009-06-11 23:15:43 +0200392 * user address. We know that we faulted in the atomic pagefault
393 * disabled section so we can as well avoid the #PF overhead by
394 * calling get_user_pages() right away.
395 */
396static int fault_in_user_writeable(u32 __user *uaddr)
397{
Andi Kleen722d0172009-12-08 13:19:42 +0100398 struct mm_struct *mm = current->mm;
399 int ret;
400
401 down_read(&mm->mmap_sem);
Benjamin Herrenschmidt2efaca92011-07-25 17:12:32 -0700402 ret = fixup_user_fault(current, mm, (unsigned long)uaddr,
403 FAULT_FLAG_WRITE);
Andi Kleen722d0172009-12-08 13:19:42 +0100404 up_read(&mm->mmap_sem);
405
Thomas Gleixnerd0725992009-06-11 23:15:43 +0200406 return ret < 0 ? ret : 0;
407}
408
Darren Hart4b1c4862009-04-03 13:39:42 -0700409/**
410 * futex_top_waiter() - Return the highest priority waiter on a futex
Darren Hartd96ee562009-09-21 22:30:22 -0700411 * @hb: the hash bucket the futex_q's reside in
412 * @key: the futex key (to distinguish it from other futex futex_q's)
Darren Hart4b1c4862009-04-03 13:39:42 -0700413 *
414 * Must be called with the hb lock held.
415 */
416static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb,
417 union futex_key *key)
418{
419 struct futex_q *this;
420
421 plist_for_each_entry(this, &hb->chain, list) {
422 if (match_futex(&this->key, key))
423 return this;
424 }
425 return NULL;
426}
427
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800428static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr,
429 u32 uval, u32 newval)
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700430{
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800431 int ret;
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700432
433 pagefault_disable();
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800434 ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval);
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700435 pagefault_enable();
436
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800437 return ret;
Thomas Gleixner36cf3b52007-07-15 23:41:20 -0700438}
439
440static int get_futex_value_locked(u32 *dest, u32 __user *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441{
442 int ret;
443
Peter Zijlstraa8663742006-12-06 20:32:20 -0800444 pagefault_disable();
Ingo Molnare2970f22006-06-27 02:54:47 -0700445 ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
Peter Zijlstraa8663742006-12-06 20:32:20 -0800446 pagefault_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
448 return ret ? -EFAULT : 0;
449}
450
Ingo Molnarc87e2832006-06-27 02:54:58 -0700451
452/*
453 * PI code:
454 */
455static int refill_pi_state_cache(void)
456{
457 struct futex_pi_state *pi_state;
458
459 if (likely(current->pi_state_cache))
460 return 0;
461
Burman Yan4668edc2006-12-06 20:38:51 -0800462 pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700463
464 if (!pi_state)
465 return -ENOMEM;
466
Ingo Molnarc87e2832006-06-27 02:54:58 -0700467 INIT_LIST_HEAD(&pi_state->list);
468 /* pi_mutex gets initialized later */
469 pi_state->owner = NULL;
470 atomic_set(&pi_state->refcount, 1);
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200471 pi_state->key = FUTEX_KEY_INIT;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700472
473 current->pi_state_cache = pi_state;
474
475 return 0;
476}
477
478static struct futex_pi_state * alloc_pi_state(void)
479{
480 struct futex_pi_state *pi_state = current->pi_state_cache;
481
482 WARN_ON(!pi_state);
483 current->pi_state_cache = NULL;
484
485 return pi_state;
486}
487
488static void free_pi_state(struct futex_pi_state *pi_state)
489{
490 if (!atomic_dec_and_test(&pi_state->refcount))
491 return;
492
493 /*
494 * If pi_state->owner is NULL, the owner is most probably dying
495 * and has cleaned up the pi_state already
496 */
497 if (pi_state->owner) {
Thomas Gleixner1d615482009-11-17 14:54:03 +0100498 raw_spin_lock_irq(&pi_state->owner->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700499 list_del_init(&pi_state->list);
Thomas Gleixner1d615482009-11-17 14:54:03 +0100500 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700501
502 rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
503 }
504
505 if (current->pi_state_cache)
506 kfree(pi_state);
507 else {
508 /*
509 * pi_state->list is already empty.
510 * clear pi_state->owner.
511 * refcount is at 0 - put it back to 1.
512 */
513 pi_state->owner = NULL;
514 atomic_set(&pi_state->refcount, 1);
515 current->pi_state_cache = pi_state;
516 }
517}
518
519/*
520 * Look up the task based on what TID userspace gave us.
521 * We dont trust it.
522 */
523static struct task_struct * futex_find_get_task(pid_t pid)
524{
525 struct task_struct *p;
526
Oleg Nesterovd359b542006-09-29 02:00:55 -0700527 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700528 p = find_task_by_vpid(pid);
Michal Hocko7a0ea092010-06-30 09:51:19 +0200529 if (p)
530 get_task_struct(p);
Thomas Gleixnera06381f2007-06-23 11:48:40 +0200531
Oleg Nesterovd359b542006-09-29 02:00:55 -0700532 rcu_read_unlock();
Ingo Molnarc87e2832006-06-27 02:54:58 -0700533
534 return p;
535}
536
537/*
538 * This task is holding PI mutexes at exit time => bad.
539 * Kernel cleans up PI-state, but userspace is likely hosed.
540 * (Robust-futex cleanup is separate and might save the day for userspace.)
541 */
542void exit_pi_state_list(struct task_struct *curr)
543{
Ingo Molnarc87e2832006-06-27 02:54:58 -0700544 struct list_head *next, *head = &curr->pi_state_list;
545 struct futex_pi_state *pi_state;
Ingo Molnar627371d2006-07-29 05:16:20 +0200546 struct futex_hash_bucket *hb;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200547 union futex_key key = FUTEX_KEY_INIT;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700548
Thomas Gleixnera0c1e902008-02-23 15:23:57 -0800549 if (!futex_cmpxchg_enabled)
550 return;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700551 /*
552 * We are a ZOMBIE and nobody can enqueue itself on
553 * pi_state_list anymore, but we have to be careful
Ingo Molnar627371d2006-07-29 05:16:20 +0200554 * versus waiters unqueueing themselves:
Ingo Molnarc87e2832006-06-27 02:54:58 -0700555 */
Thomas Gleixner1d615482009-11-17 14:54:03 +0100556 raw_spin_lock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700557 while (!list_empty(head)) {
558
559 next = head->next;
560 pi_state = list_entry(next, struct futex_pi_state, list);
561 key = pi_state->key;
Ingo Molnar627371d2006-07-29 05:16:20 +0200562 hb = hash_futex(&key);
Thomas Gleixner1d615482009-11-17 14:54:03 +0100563 raw_spin_unlock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700564
Ingo Molnarc87e2832006-06-27 02:54:58 -0700565 spin_lock(&hb->lock);
566
Thomas Gleixner1d615482009-11-17 14:54:03 +0100567 raw_spin_lock_irq(&curr->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200568 /*
569 * We dropped the pi-lock, so re-check whether this
570 * task still owns the PI-state:
571 */
Ingo Molnarc87e2832006-06-27 02:54:58 -0700572 if (head->next != next) {
573 spin_unlock(&hb->lock);
574 continue;
575 }
576
Ingo Molnarc87e2832006-06-27 02:54:58 -0700577 WARN_ON(pi_state->owner != curr);
Ingo Molnar627371d2006-07-29 05:16:20 +0200578 WARN_ON(list_empty(&pi_state->list));
579 list_del_init(&pi_state->list);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700580 pi_state->owner = NULL;
Thomas Gleixner1d615482009-11-17 14:54:03 +0100581 raw_spin_unlock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700582
583 rt_mutex_unlock(&pi_state->pi_mutex);
584
585 spin_unlock(&hb->lock);
586
Thomas Gleixner1d615482009-11-17 14:54:03 +0100587 raw_spin_lock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700588 }
Thomas Gleixner1d615482009-11-17 14:54:03 +0100589 raw_spin_unlock_irq(&curr->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700590}
591
592static int
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700593lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
594 union futex_key *key, struct futex_pi_state **ps)
Ingo Molnarc87e2832006-06-27 02:54:58 -0700595{
596 struct futex_pi_state *pi_state = NULL;
597 struct futex_q *this, *next;
Pierre Peifferec92d082007-05-09 02:35:00 -0700598 struct plist_head *head;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700599 struct task_struct *p;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700600 pid_t pid = uval & FUTEX_TID_MASK;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700601
602 head = &hb->chain;
603
Pierre Peifferec92d082007-05-09 02:35:00 -0700604 plist_for_each_entry_safe(this, next, head, list) {
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700605 if (match_futex(&this->key, key)) {
Ingo Molnarc87e2832006-06-27 02:54:58 -0700606 /*
607 * Another waiter already exists - bump up
608 * the refcount and return its pi_state:
609 */
610 pi_state = this->pi_state;
Thomas Gleixner06a9ec22006-07-10 04:44:30 -0700611 /*
Randy Dunlapfb62db22010-10-13 11:02:34 -0700612 * Userspace might have messed up non-PI and PI futexes
Thomas Gleixner06a9ec22006-07-10 04:44:30 -0700613 */
614 if (unlikely(!pi_state))
615 return -EINVAL;
616
Ingo Molnar627371d2006-07-29 05:16:20 +0200617 WARN_ON(!atomic_read(&pi_state->refcount));
Thomas Gleixner59647b62010-02-03 09:33:05 +0100618
619 /*
620 * When pi_state->owner is NULL then the owner died
621 * and another waiter is on the fly. pi_state->owner
622 * is fixed up by the task which acquires
623 * pi_state->rt_mutex.
624 *
625 * We do not check for pid == 0 which can happen when
626 * the owner died and robust_list_exit() cleared the
627 * TID.
628 */
629 if (pid && pi_state->owner) {
630 /*
631 * Bail out if user space manipulated the
632 * futex value.
633 */
634 if (pid != task_pid_vnr(pi_state->owner))
635 return -EINVAL;
636 }
Ingo Molnar627371d2006-07-29 05:16:20 +0200637
Ingo Molnarc87e2832006-06-27 02:54:58 -0700638 atomic_inc(&pi_state->refcount);
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700639 *ps = pi_state;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700640
641 return 0;
642 }
643 }
644
645 /*
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200646 * We are the first waiter - try to look up the real owner and attach
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700647 * the new pi_state to it, but bail out when TID = 0
Ingo Molnarc87e2832006-06-27 02:54:58 -0700648 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700649 if (!pid)
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200650 return -ESRCH;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700651 p = futex_find_get_task(pid);
Michal Hocko7a0ea092010-06-30 09:51:19 +0200652 if (!p)
653 return -ESRCH;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700654
655 /*
656 * We need to look at the task state flags to figure out,
657 * whether the task is exiting. To protect against the do_exit
658 * change of the task flags, we do this protected by
659 * p->pi_lock:
660 */
Thomas Gleixner1d615482009-11-17 14:54:03 +0100661 raw_spin_lock_irq(&p->pi_lock);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700662 if (unlikely(p->flags & PF_EXITING)) {
663 /*
664 * The task is on the way out. When PF_EXITPIDONE is
665 * set, we know that the task has finished the
666 * cleanup:
667 */
668 int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
669
Thomas Gleixner1d615482009-11-17 14:54:03 +0100670 raw_spin_unlock_irq(&p->pi_lock);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700671 put_task_struct(p);
672 return ret;
673 }
Ingo Molnarc87e2832006-06-27 02:54:58 -0700674
675 pi_state = alloc_pi_state();
676
677 /*
678 * Initialize the pi_mutex in locked state and make 'p'
679 * the owner of it:
680 */
681 rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
682
683 /* Store the key for possible exit cleanups: */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700684 pi_state->key = *key;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700685
Ingo Molnar627371d2006-07-29 05:16:20 +0200686 WARN_ON(!list_empty(&pi_state->list));
Ingo Molnarc87e2832006-06-27 02:54:58 -0700687 list_add(&pi_state->list, &p->pi_state_list);
688 pi_state->owner = p;
Thomas Gleixner1d615482009-11-17 14:54:03 +0100689 raw_spin_unlock_irq(&p->pi_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700690
691 put_task_struct(p);
692
Pierre Peifferd0aa7a72007-05-09 02:35:02 -0700693 *ps = pi_state;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700694
695 return 0;
696}
697
Darren Hart1a520842009-04-03 13:39:52 -0700698/**
Darren Hartd96ee562009-09-21 22:30:22 -0700699 * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
Darren Hartbab5bc92009-04-07 23:23:50 -0700700 * @uaddr: the pi futex user address
701 * @hb: the pi futex hash bucket
702 * @key: the futex key associated with uaddr and hb
703 * @ps: the pi_state pointer where we store the result of the
704 * lookup
705 * @task: the task to perform the atomic lock work for. This will
706 * be "current" except in the case of requeue pi.
707 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
Darren Hart1a520842009-04-03 13:39:52 -0700708 *
Randy Dunlap6c23cbb2013-03-05 10:00:24 -0800709 * Return:
710 * 0 - ready to wait;
711 * 1 - acquired the lock;
Darren Hart1a520842009-04-03 13:39:52 -0700712 * <0 - error
713 *
714 * The hb->lock and futex_key refs shall be held by the caller.
715 */
716static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb,
717 union futex_key *key,
718 struct futex_pi_state **ps,
Darren Hartbab5bc92009-04-07 23:23:50 -0700719 struct task_struct *task, int set_waiters)
Darren Hart1a520842009-04-03 13:39:52 -0700720{
Thomas Gleixner59fa6242012-10-23 22:29:38 +0200721 int lock_taken, ret, force_take = 0;
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +0100722 u32 uval, newval, curval, vpid = task_pid_vnr(task);
Darren Hart1a520842009-04-03 13:39:52 -0700723
724retry:
725 ret = lock_taken = 0;
726
727 /*
728 * To avoid races, we attempt to take the lock here again
729 * (by doing a 0 -> TID atomic cmpxchg), while holding all
730 * the locks. It will most likely not succeed.
731 */
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +0100732 newval = vpid;
Darren Hartbab5bc92009-04-07 23:23:50 -0700733 if (set_waiters)
734 newval |= FUTEX_WAITERS;
Darren Hart1a520842009-04-03 13:39:52 -0700735
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800736 if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval)))
Darren Hart1a520842009-04-03 13:39:52 -0700737 return -EFAULT;
738
739 /*
740 * Detect deadlocks.
741 */
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +0100742 if ((unlikely((curval & FUTEX_TID_MASK) == vpid)))
Darren Hart1a520842009-04-03 13:39:52 -0700743 return -EDEADLK;
744
745 /*
746 * Surprise - we got the lock. Just return to userspace:
747 */
748 if (unlikely(!curval))
749 return 1;
750
751 uval = curval;
752
753 /*
754 * Set the FUTEX_WAITERS flag, so the owner will know it has someone
755 * to wake at the next unlock.
756 */
757 newval = curval | FUTEX_WAITERS;
758
759 /*
Thomas Gleixner59fa6242012-10-23 22:29:38 +0200760 * Should we force take the futex? See below.
Darren Hart1a520842009-04-03 13:39:52 -0700761 */
Thomas Gleixner59fa6242012-10-23 22:29:38 +0200762 if (unlikely(force_take)) {
763 /*
764 * Keep the OWNER_DIED and the WAITERS bit and set the
765 * new TID value.
766 */
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +0100767 newval = (curval & ~FUTEX_TID_MASK) | vpid;
Thomas Gleixner59fa6242012-10-23 22:29:38 +0200768 force_take = 0;
Darren Hart1a520842009-04-03 13:39:52 -0700769 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 /*
Thomas Gleixner59fa6242012-10-23 22:29:38 +0200778 * We took the lock due to forced take over.
Darren Hart1a520842009-04-03 13:39:52 -0700779 */
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 /*
Thomas Gleixner59fa6242012-10-23 22:29:38 +0200793 * We failed to find an owner for this
794 * futex. So we have no pi_state to block
795 * on. This can happen in two cases:
796 *
797 * 1) The owner died
798 * 2) A stale FUTEX_WAITERS bit
799 *
800 * Re-read the futex value.
Darren Hart1a520842009-04-03 13:39:52 -0700801 */
802 if (get_futex_value_locked(&curval, uaddr))
803 return -EFAULT;
804
805 /*
Thomas Gleixner59fa6242012-10-23 22:29:38 +0200806 * If the owner died or we have a stale
807 * WAITERS bit the owner TID in the user space
808 * futex is 0.
Darren Hart1a520842009-04-03 13:39:52 -0700809 */
Thomas Gleixner59fa6242012-10-23 22:29:38 +0200810 if (!(curval & FUTEX_TID_MASK)) {
811 force_take = 1;
Darren Hart1a520842009-04-03 13:39:52 -0700812 goto retry;
813 }
814 default:
815 break;
816 }
817 }
818
819 return ret;
820}
821
Lai Jiangshan2e129782010-12-22 14:18:50 +0800822/**
823 * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket
824 * @q: The futex_q to unqueue
825 *
826 * The q->lock_ptr must not be NULL and must be held by the caller.
827 */
828static void __unqueue_futex(struct futex_q *q)
829{
830 struct futex_hash_bucket *hb;
831
Steven Rostedt29096202011-03-17 15:21:07 -0400832 if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr))
833 || WARN_ON(plist_node_empty(&q->list)))
Lai Jiangshan2e129782010-12-22 14:18:50 +0800834 return;
835
836 hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock);
837 plist_del(&q->list, &hb->chain);
838}
839
Ingo Molnarc87e2832006-06-27 02:54:58 -0700840/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 * The hash bucket lock must be held when this is called.
842 * Afterwards, the futex_q must not be accessed.
843 */
844static void wake_futex(struct futex_q *q)
845{
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200846 struct task_struct *p = q->task;
847
Darren Hartaa109902012-11-26 16:29:56 -0800848 if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n"))
849 return;
850
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200851 /*
852 * We set q->lock_ptr = NULL _before_ we wake up the task. If
Randy Dunlapfb62db22010-10-13 11:02:34 -0700853 * a non-futex wake up happens on another CPU then the task
854 * might exit and p would dereference a non-existing task
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200855 * struct. Prevent this by holding a reference on p across the
856 * wake up.
857 */
858 get_task_struct(p);
859
Lai Jiangshan2e129782010-12-22 14:18:50 +0800860 __unqueue_futex(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 /*
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200862 * The waiting task can free the futex_q as soon as
863 * q->lock_ptr = NULL is written, without taking any locks. A
864 * memory barrier is required here to prevent the following
865 * store to lock_ptr from getting ahead of the plist_del.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 */
Ralf Baechleccdea2f2006-12-06 20:40:26 -0800867 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 q->lock_ptr = NULL;
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +0200869
870 wake_up_state(p, TASK_NORMAL);
871 put_task_struct(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872}
873
Ingo Molnarc87e2832006-06-27 02:54:58 -0700874static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
875{
876 struct task_struct *new_owner;
877 struct futex_pi_state *pi_state = this->pi_state;
Vitaliy Ivanov7cfdaf32011-07-07 15:10:31 +0300878 u32 uninitialized_var(curval), newval;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700879
880 if (!pi_state)
881 return -EINVAL;
882
Thomas Gleixner51246bf2010-02-02 11:40:27 +0100883 /*
884 * If current does not own the pi_state then the futex is
885 * inconsistent and user space fiddled with the futex value.
886 */
887 if (pi_state->owner != current)
888 return -EINVAL;
889
Thomas Gleixnerd209d742009-11-17 18:22:11 +0100890 raw_spin_lock(&pi_state->pi_mutex.wait_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700891 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
892
893 /*
Steven Rostedtf123c982011-01-06 15:08:29 -0500894 * It is possible that the next waiter (the one that brought
895 * this owner to the kernel) timed out and is no longer
896 * waiting on the lock.
Ingo Molnarc87e2832006-06-27 02:54:58 -0700897 */
898 if (!new_owner)
899 new_owner = this->task;
900
901 /*
902 * We pass it to the next owner. (The WAITERS bit is always
903 * kept enabled while there is PI state around. We must also
904 * preserve the owner died bit.)
905 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200906 if (!(uval & FUTEX_OWNER_DIED)) {
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700907 int ret = 0;
908
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700909 newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700910
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800911 if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700912 ret = -EFAULT;
Thomas Gleixnercde898f2007-12-05 15:46:09 +0100913 else if (curval != uval)
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700914 ret = -EINVAL;
915 if (ret) {
Thomas Gleixnerd209d742009-11-17 18:22:11 +0100916 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700917 return ret;
918 }
Ingo Molnare3f2dde2006-07-29 05:17:57 +0200919 }
Ingo Molnarc87e2832006-06-27 02:54:58 -0700920
Thomas Gleixner1d615482009-11-17 14:54:03 +0100921 raw_spin_lock_irq(&pi_state->owner->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200922 WARN_ON(list_empty(&pi_state->list));
923 list_del_init(&pi_state->list);
Thomas Gleixner1d615482009-11-17 14:54:03 +0100924 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200925
Thomas Gleixner1d615482009-11-17 14:54:03 +0100926 raw_spin_lock_irq(&new_owner->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200927 WARN_ON(!list_empty(&pi_state->list));
Ingo Molnarc87e2832006-06-27 02:54:58 -0700928 list_add(&pi_state->list, &new_owner->pi_state_list);
929 pi_state->owner = new_owner;
Thomas Gleixner1d615482009-11-17 14:54:03 +0100930 raw_spin_unlock_irq(&new_owner->pi_lock);
Ingo Molnar627371d2006-07-29 05:16:20 +0200931
Thomas Gleixnerd209d742009-11-17 18:22:11 +0100932 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700933 rt_mutex_unlock(&pi_state->pi_mutex);
934
935 return 0;
936}
937
938static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
939{
Vitaliy Ivanov7cfdaf32011-07-07 15:10:31 +0300940 u32 uninitialized_var(oldval);
Ingo Molnarc87e2832006-06-27 02:54:58 -0700941
942 /*
943 * There is no waiter, so we unlock the futex. The owner died
944 * bit has not to be preserved here. We are the owner:
945 */
Michel Lespinasse37a9d912011-03-10 18:48:51 -0800946 if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0))
947 return -EFAULT;
Ingo Molnarc87e2832006-06-27 02:54:58 -0700948 if (oldval != uval)
949 return -EAGAIN;
950
951 return 0;
952}
953
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954/*
Ingo Molnar8b8f3192006-07-03 00:25:05 -0700955 * Express the locking dependencies for lockdep:
956 */
957static inline void
958double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
959{
960 if (hb1 <= hb2) {
961 spin_lock(&hb1->lock);
962 if (hb1 < hb2)
963 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
964 } else { /* hb1 > hb2 */
965 spin_lock(&hb2->lock);
966 spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
967 }
968}
969
Darren Hart5eb3dc62009-03-12 00:55:52 -0700970static inline void
971double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
972{
Darren Hartf061d352009-03-12 15:11:18 -0700973 spin_unlock(&hb1->lock);
Ingo Molnar88f502f2009-03-13 10:32:07 +0100974 if (hb1 != hb2)
975 spin_unlock(&hb2->lock);
Darren Hart5eb3dc62009-03-12 00:55:52 -0700976}
977
Ingo Molnar8b8f3192006-07-03 00:25:05 -0700978/*
Darren Hartb2d09942009-03-12 00:55:37 -0700979 * Wake up waiters matching bitset queued on this futex (uaddr).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 */
Darren Hartb41277d2010-11-08 13:10:09 -0800981static int
982futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983{
Ingo Molnare2970f22006-06-27 02:54:47 -0700984 struct futex_hash_bucket *hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 struct futex_q *this, *next;
Pierre Peifferec92d082007-05-09 02:35:00 -0700986 struct plist_head *head;
Peter Zijlstra38d47c12008-09-26 19:32:20 +0200987 union futex_key key = FUTEX_KEY_INIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 int ret;
989
Thomas Gleixnercd6899852008-02-01 17:45:14 +0100990 if (!bitset)
991 return -EINVAL;
992
Shawn Bohrer9ea71502011-06-30 11:21:32 -0500993 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 if (unlikely(ret != 0))
995 goto out;
996
Ingo Molnare2970f22006-06-27 02:54:47 -0700997 hb = hash_futex(&key);
998 spin_lock(&hb->lock);
999 head = &hb->chain;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Pierre Peifferec92d082007-05-09 02:35:00 -07001001 plist_for_each_entry_safe(this, next, head, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 if (match_futex (&this->key, &key)) {
Darren Hart52400ba2009-04-03 13:40:49 -07001003 if (this->pi_state || this->rt_waiter) {
Ingo Molnared6f7b12006-07-01 04:35:46 -07001004 ret = -EINVAL;
1005 break;
1006 }
Thomas Gleixnercd6899852008-02-01 17:45:14 +01001007
1008 /* Check if one of the bits is set in both bitsets */
1009 if (!(this->bitset & bitset))
1010 continue;
1011
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 wake_futex(this);
1013 if (++ret >= nr_wake)
1014 break;
1015 }
1016 }
1017
Ingo Molnare2970f22006-06-27 02:54:47 -07001018 spin_unlock(&hb->lock);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001019 put_futex_key(&key);
Darren Hart42d35d42008-12-29 15:49:53 -08001020out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 return ret;
1022}
1023
1024/*
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001025 * Wake up all waiters hashed on the physical page that is mapped
1026 * to this virtual address:
1027 */
Ingo Molnare2970f22006-06-27 02:54:47 -07001028static int
Darren Hartb41277d2010-11-08 13:10:09 -08001029futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
Ingo Molnare2970f22006-06-27 02:54:47 -07001030 int nr_wake, int nr_wake2, int op)
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001031{
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001032 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
Ingo Molnare2970f22006-06-27 02:54:47 -07001033 struct futex_hash_bucket *hb1, *hb2;
Pierre Peifferec92d082007-05-09 02:35:00 -07001034 struct plist_head *head;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001035 struct futex_q *this, *next;
Darren Harte4dc5b72009-03-12 00:56:13 -07001036 int ret, op_ret;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001037
Darren Harte4dc5b72009-03-12 00:56:13 -07001038retry:
Shawn Bohrer9ea71502011-06-30 11:21:32 -05001039 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001040 if (unlikely(ret != 0))
1041 goto out;
Shawn Bohrer9ea71502011-06-30 11:21:32 -05001042 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001043 if (unlikely(ret != 0))
Darren Hart42d35d42008-12-29 15:49:53 -08001044 goto out_put_key1;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001045
Ingo Molnare2970f22006-06-27 02:54:47 -07001046 hb1 = hash_futex(&key1);
1047 hb2 = hash_futex(&key2);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001048
Darren Harte4dc5b72009-03-12 00:56:13 -07001049retry_private:
Thomas Gleixnereaaea802009-10-04 09:34:17 +02001050 double_lock_hb(hb1, hb2);
Ingo Molnare2970f22006-06-27 02:54:47 -07001051 op_ret = futex_atomic_op_inuser(op, uaddr2);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001052 if (unlikely(op_ret < 0)) {
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001053
Darren Hart5eb3dc62009-03-12 00:55:52 -07001054 double_unlock_hb(hb1, hb2);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001055
David Howells7ee1dd32006-01-06 00:11:44 -08001056#ifndef CONFIG_MMU
Ingo Molnare2970f22006-06-27 02:54:47 -07001057 /*
1058 * we don't get EFAULT from MMU faults if we don't have an MMU,
1059 * but we might get them from range checking
1060 */
David Howells7ee1dd32006-01-06 00:11:44 -08001061 ret = op_ret;
Darren Hart42d35d42008-12-29 15:49:53 -08001062 goto out_put_keys;
David Howells7ee1dd32006-01-06 00:11:44 -08001063#endif
1064
David Gibson796f8d92005-11-07 00:59:33 -08001065 if (unlikely(op_ret != -EFAULT)) {
1066 ret = op_ret;
Darren Hart42d35d42008-12-29 15:49:53 -08001067 goto out_put_keys;
David Gibson796f8d92005-11-07 00:59:33 -08001068 }
1069
Thomas Gleixnerd0725992009-06-11 23:15:43 +02001070 ret = fault_in_user_writeable(uaddr2);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001071 if (ret)
Darren Hartde87fcc2009-03-12 00:55:46 -07001072 goto out_put_keys;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001073
Darren Hartb41277d2010-11-08 13:10:09 -08001074 if (!(flags & FLAGS_SHARED))
Darren Harte4dc5b72009-03-12 00:56:13 -07001075 goto retry_private;
1076
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001077 put_futex_key(&key2);
1078 put_futex_key(&key1);
Darren Harte4dc5b72009-03-12 00:56:13 -07001079 goto retry;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001080 }
1081
Ingo Molnare2970f22006-06-27 02:54:47 -07001082 head = &hb1->chain;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001083
Pierre Peifferec92d082007-05-09 02:35:00 -07001084 plist_for_each_entry_safe(this, next, head, list) {
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001085 if (match_futex (&this->key, &key1)) {
Darren Hartaa109902012-11-26 16:29:56 -08001086 if (this->pi_state || this->rt_waiter) {
1087 ret = -EINVAL;
1088 goto out_unlock;
1089 }
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001090 wake_futex(this);
1091 if (++ret >= nr_wake)
1092 break;
1093 }
1094 }
1095
1096 if (op_ret > 0) {
Ingo Molnare2970f22006-06-27 02:54:47 -07001097 head = &hb2->chain;
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001098
1099 op_ret = 0;
Pierre Peifferec92d082007-05-09 02:35:00 -07001100 plist_for_each_entry_safe(this, next, head, list) {
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001101 if (match_futex (&this->key, &key2)) {
Darren Hartaa109902012-11-26 16:29:56 -08001102 if (this->pi_state || this->rt_waiter) {
1103 ret = -EINVAL;
1104 goto out_unlock;
1105 }
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001106 wake_futex(this);
1107 if (++op_ret >= nr_wake2)
1108 break;
1109 }
1110 }
1111 ret += op_ret;
1112 }
1113
Darren Hartaa109902012-11-26 16:29:56 -08001114out_unlock:
Darren Hart5eb3dc62009-03-12 00:55:52 -07001115 double_unlock_hb(hb1, hb2);
Darren Hart42d35d42008-12-29 15:49:53 -08001116out_put_keys:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001117 put_futex_key(&key2);
Darren Hart42d35d42008-12-29 15:49:53 -08001118out_put_key1:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001119 put_futex_key(&key1);
Darren Hart42d35d42008-12-29 15:49:53 -08001120out:
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07001121 return ret;
1122}
1123
Darren Hart9121e472009-04-03 13:40:31 -07001124/**
1125 * requeue_futex() - Requeue a futex_q from one hb to another
1126 * @q: the futex_q to requeue
1127 * @hb1: the source hash_bucket
1128 * @hb2: the target hash_bucket
1129 * @key2: the new key for the requeued futex_q
1130 */
1131static inline
1132void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1,
1133 struct futex_hash_bucket *hb2, union futex_key *key2)
1134{
1135
1136 /*
1137 * If key1 and key2 hash to the same bucket, no need to
1138 * requeue.
1139 */
1140 if (likely(&hb1->chain != &hb2->chain)) {
1141 plist_del(&q->list, &hb1->chain);
1142 plist_add(&q->list, &hb2->chain);
1143 q->lock_ptr = &hb2->lock;
Darren Hart9121e472009-04-03 13:40:31 -07001144 }
1145 get_futex_key_refs(key2);
1146 q->key = *key2;
1147}
1148
Darren Hart52400ba2009-04-03 13:40:49 -07001149/**
1150 * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
Darren Hartd96ee562009-09-21 22:30:22 -07001151 * @q: the futex_q
1152 * @key: the key of the requeue target futex
1153 * @hb: the hash_bucket of the requeue target futex
Darren Hart52400ba2009-04-03 13:40:49 -07001154 *
1155 * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1156 * target futex if it is uncontended or via a lock steal. Set the futex_q key
1157 * to the requeue target futex so the waiter can detect the wakeup on the right
1158 * futex, but remove it from the hb and NULL the rt_waiter so it can detect
Darren Hartbeda2c72009-08-09 15:34:39 -07001159 * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1160 * to protect access to the pi_state to fixup the owner later. Must be called
1161 * with both q->lock_ptr and hb->lock held.
Darren Hart52400ba2009-04-03 13:40:49 -07001162 */
1163static inline
Darren Hartbeda2c72009-08-09 15:34:39 -07001164void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key,
1165 struct futex_hash_bucket *hb)
Darren Hart52400ba2009-04-03 13:40:49 -07001166{
Darren Hart52400ba2009-04-03 13:40:49 -07001167 get_futex_key_refs(key);
1168 q->key = *key;
1169
Lai Jiangshan2e129782010-12-22 14:18:50 +08001170 __unqueue_futex(q);
Darren Hart52400ba2009-04-03 13:40:49 -07001171
1172 WARN_ON(!q->rt_waiter);
1173 q->rt_waiter = NULL;
1174
Darren Hartbeda2c72009-08-09 15:34:39 -07001175 q->lock_ptr = &hb->lock;
Darren Hartbeda2c72009-08-09 15:34:39 -07001176
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02001177 wake_up_state(q->task, TASK_NORMAL);
Darren Hart52400ba2009-04-03 13:40:49 -07001178}
1179
1180/**
1181 * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
Darren Hartbab5bc92009-04-07 23:23:50 -07001182 * @pifutex: the user address of the to futex
1183 * @hb1: the from futex hash bucket, must be locked by the caller
1184 * @hb2: the to futex hash bucket, must be locked by the caller
1185 * @key1: the from futex key
1186 * @key2: the to futex key
1187 * @ps: address to store the pi_state pointer
1188 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
Darren Hart52400ba2009-04-03 13:40:49 -07001189 *
1190 * Try and get the lock on behalf of the top waiter if we can do it atomically.
Darren Hartbab5bc92009-04-07 23:23:50 -07001191 * Wake the top waiter if we succeed. If the caller specified set_waiters,
1192 * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1193 * hb1 and hb2 must be held by the caller.
Darren Hart52400ba2009-04-03 13:40:49 -07001194 *
Randy Dunlap6c23cbb2013-03-05 10:00:24 -08001195 * Return:
1196 * 0 - failed to acquire the lock atomically;
1197 * 1 - acquired the lock;
Darren Hart52400ba2009-04-03 13:40:49 -07001198 * <0 - error
1199 */
1200static int futex_proxy_trylock_atomic(u32 __user *pifutex,
1201 struct futex_hash_bucket *hb1,
1202 struct futex_hash_bucket *hb2,
1203 union futex_key *key1, union futex_key *key2,
Darren Hartbab5bc92009-04-07 23:23:50 -07001204 struct futex_pi_state **ps, int set_waiters)
Darren Hart52400ba2009-04-03 13:40:49 -07001205{
Darren Hartbab5bc92009-04-07 23:23:50 -07001206 struct futex_q *top_waiter = NULL;
Darren Hart52400ba2009-04-03 13:40:49 -07001207 u32 curval;
1208 int ret;
1209
1210 if (get_futex_value_locked(&curval, pifutex))
1211 return -EFAULT;
1212
Darren Hartbab5bc92009-04-07 23:23:50 -07001213 /*
1214 * Find the top_waiter and determine if there are additional waiters.
1215 * If the caller intends to requeue more than 1 waiter to pifutex,
1216 * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1217 * as we have means to handle the possible fault. If not, don't set
1218 * the bit unecessarily as it will force the subsequent unlock to enter
1219 * the kernel.
1220 */
Darren Hart52400ba2009-04-03 13:40:49 -07001221 top_waiter = futex_top_waiter(hb1, key1);
1222
1223 /* There are no waiters, nothing for us to do. */
1224 if (!top_waiter)
1225 return 0;
1226
Darren Hart84bc4af2009-08-13 17:36:53 -07001227 /* Ensure we requeue to the expected futex. */
1228 if (!match_futex(top_waiter->requeue_pi_key, key2))
1229 return -EINVAL;
1230
Darren Hart52400ba2009-04-03 13:40:49 -07001231 /*
Darren Hartbab5bc92009-04-07 23:23:50 -07001232 * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1233 * the contended case or if set_waiters is 1. The pi_state is returned
1234 * in ps in contended cases.
Darren Hart52400ba2009-04-03 13:40:49 -07001235 */
Darren Hartbab5bc92009-04-07 23:23:50 -07001236 ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task,
1237 set_waiters);
Darren Hart52400ba2009-04-03 13:40:49 -07001238 if (ret == 1)
Darren Hartbeda2c72009-08-09 15:34:39 -07001239 requeue_pi_wake_futex(top_waiter, key2, hb2);
Darren Hart52400ba2009-04-03 13:40:49 -07001240
1241 return ret;
1242}
1243
1244/**
1245 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
Randy Dunlapfb62db22010-10-13 11:02:34 -07001246 * @uaddr1: source futex user address
Darren Hartb41277d2010-11-08 13:10:09 -08001247 * @flags: futex flags (FLAGS_SHARED, etc.)
Randy Dunlapfb62db22010-10-13 11:02:34 -07001248 * @uaddr2: target futex user address
1249 * @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1250 * @nr_requeue: number of waiters to requeue (0-INT_MAX)
1251 * @cmpval: @uaddr1 expected value (or %NULL)
1252 * @requeue_pi: if we are attempting to requeue from a non-pi futex to a
Darren Hartb41277d2010-11-08 13:10:09 -08001253 * pi futex (pi to pi requeue is not supported)
Darren Hart52400ba2009-04-03 13:40:49 -07001254 *
1255 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1256 * uaddr2 atomically on behalf of the top waiter.
1257 *
Randy Dunlap6c23cbb2013-03-05 10:00:24 -08001258 * Return:
1259 * >=0 - on success, the number of tasks requeued or woken;
Darren Hart52400ba2009-04-03 13:40:49 -07001260 * <0 - on error
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 */
Darren Hartb41277d2010-11-08 13:10:09 -08001262static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
1263 u32 __user *uaddr2, int nr_wake, int nr_requeue,
1264 u32 *cmpval, int requeue_pi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265{
Peter Zijlstra38d47c12008-09-26 19:32:20 +02001266 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
Darren Hart52400ba2009-04-03 13:40:49 -07001267 int drop_count = 0, task_count = 0, ret;
1268 struct futex_pi_state *pi_state = NULL;
Ingo Molnare2970f22006-06-27 02:54:47 -07001269 struct futex_hash_bucket *hb1, *hb2;
Pierre Peifferec92d082007-05-09 02:35:00 -07001270 struct plist_head *head1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 struct futex_q *this, *next;
Darren Hart52400ba2009-04-03 13:40:49 -07001272 u32 curval2;
1273
1274 if (requeue_pi) {
1275 /*
1276 * requeue_pi requires a pi_state, try to allocate it now
1277 * without any locks in case it fails.
1278 */
1279 if (refill_pi_state_cache())
1280 return -ENOMEM;
1281 /*
1282 * requeue_pi must wake as many tasks as it can, up to nr_wake
1283 * + nr_requeue, since it acquires the rt_mutex prior to
1284 * returning to userspace, so as to not leave the rt_mutex with
1285 * waiters and no owner. However, second and third wake-ups
1286 * cannot be predicted as they involve race conditions with the
1287 * first wake and a fault while looking up the pi_state. Both
1288 * pthread_cond_signal() and pthread_cond_broadcast() should
1289 * use nr_wake=1.
1290 */
1291 if (nr_wake != 1)
1292 return -EINVAL;
1293 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
Darren Hart42d35d42008-12-29 15:49:53 -08001295retry:
Darren Hart52400ba2009-04-03 13:40:49 -07001296 if (pi_state != NULL) {
1297 /*
1298 * We will have to lookup the pi_state again, so free this one
1299 * to keep the accounting correct.
1300 */
1301 free_pi_state(pi_state);
1302 pi_state = NULL;
1303 }
1304
Shawn Bohrer9ea71502011-06-30 11:21:32 -05001305 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 if (unlikely(ret != 0))
1307 goto out;
Shawn Bohrer9ea71502011-06-30 11:21:32 -05001308 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2,
1309 requeue_pi ? VERIFY_WRITE : VERIFY_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 if (unlikely(ret != 0))
Darren Hart42d35d42008-12-29 15:49:53 -08001311 goto out_put_key1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
Ingo Molnare2970f22006-06-27 02:54:47 -07001313 hb1 = hash_futex(&key1);
1314 hb2 = hash_futex(&key2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
Darren Harte4dc5b72009-03-12 00:56:13 -07001316retry_private:
Ingo Molnar8b8f3192006-07-03 00:25:05 -07001317 double_lock_hb(hb1, hb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318
Ingo Molnare2970f22006-06-27 02:54:47 -07001319 if (likely(cmpval != NULL)) {
1320 u32 curval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
Ingo Molnare2970f22006-06-27 02:54:47 -07001322 ret = get_futex_value_locked(&curval, uaddr1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
1324 if (unlikely(ret)) {
Darren Hart5eb3dc62009-03-12 00:55:52 -07001325 double_unlock_hb(hb1, hb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
Darren Harte4dc5b72009-03-12 00:56:13 -07001327 ret = get_user(curval, uaddr1);
1328 if (ret)
1329 goto out_put_keys;
1330
Darren Hartb41277d2010-11-08 13:10:09 -08001331 if (!(flags & FLAGS_SHARED))
Darren Harte4dc5b72009-03-12 00:56:13 -07001332 goto retry_private;
1333
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001334 put_futex_key(&key2);
1335 put_futex_key(&key1);
Darren Harte4dc5b72009-03-12 00:56:13 -07001336 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 }
Ingo Molnare2970f22006-06-27 02:54:47 -07001338 if (curval != *cmpval) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 ret = -EAGAIN;
1340 goto out_unlock;
1341 }
1342 }
1343
Darren Hart52400ba2009-04-03 13:40:49 -07001344 if (requeue_pi && (task_count - nr_wake < nr_requeue)) {
Darren Hartbab5bc92009-04-07 23:23:50 -07001345 /*
1346 * Attempt to acquire uaddr2 and wake the top waiter. If we
1347 * intend to requeue waiters, force setting the FUTEX_WAITERS
1348 * bit. We force this here where we are able to easily handle
1349 * faults rather in the requeue loop below.
1350 */
Darren Hart52400ba2009-04-03 13:40:49 -07001351 ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1,
Darren Hartbab5bc92009-04-07 23:23:50 -07001352 &key2, &pi_state, nr_requeue);
Darren Hart52400ba2009-04-03 13:40:49 -07001353
1354 /*
1355 * At this point the top_waiter has either taken uaddr2 or is
1356 * waiting on it. If the former, then the pi_state will not
1357 * exist yet, look it up one more time to ensure we have a
1358 * reference to it.
1359 */
1360 if (ret == 1) {
1361 WARN_ON(pi_state);
Darren Hart89061d32009-10-15 15:30:48 -07001362 drop_count++;
Darren Hart52400ba2009-04-03 13:40:49 -07001363 task_count++;
1364 ret = get_futex_value_locked(&curval2, uaddr2);
1365 if (!ret)
1366 ret = lookup_pi_state(curval2, hb2, &key2,
1367 &pi_state);
1368 }
1369
1370 switch (ret) {
1371 case 0:
1372 break;
1373 case -EFAULT:
1374 double_unlock_hb(hb1, hb2);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001375 put_futex_key(&key2);
1376 put_futex_key(&key1);
Thomas Gleixnerd0725992009-06-11 23:15:43 +02001377 ret = fault_in_user_writeable(uaddr2);
Darren Hart52400ba2009-04-03 13:40:49 -07001378 if (!ret)
1379 goto retry;
1380 goto out;
1381 case -EAGAIN:
1382 /* The owner was exiting, try again. */
1383 double_unlock_hb(hb1, hb2);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001384 put_futex_key(&key2);
1385 put_futex_key(&key1);
Darren Hart52400ba2009-04-03 13:40:49 -07001386 cond_resched();
1387 goto retry;
1388 default:
1389 goto out_unlock;
1390 }
1391 }
1392
Ingo Molnare2970f22006-06-27 02:54:47 -07001393 head1 = &hb1->chain;
Pierre Peifferec92d082007-05-09 02:35:00 -07001394 plist_for_each_entry_safe(this, next, head1, list) {
Darren Hart52400ba2009-04-03 13:40:49 -07001395 if (task_count - nr_wake >= nr_requeue)
1396 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397
Darren Hart52400ba2009-04-03 13:40:49 -07001398 if (!match_futex(&this->key, &key1))
1399 continue;
1400
Darren Hart392741e2009-08-07 15:20:48 -07001401 /*
1402 * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
1403 * be paired with each other and no other futex ops.
Darren Hartaa109902012-11-26 16:29:56 -08001404 *
1405 * We should never be requeueing a futex_q with a pi_state,
1406 * which is awaiting a futex_unlock_pi().
Darren Hart392741e2009-08-07 15:20:48 -07001407 */
1408 if ((requeue_pi && !this->rt_waiter) ||
Darren Hartaa109902012-11-26 16:29:56 -08001409 (!requeue_pi && this->rt_waiter) ||
1410 this->pi_state) {
Darren Hart392741e2009-08-07 15:20:48 -07001411 ret = -EINVAL;
1412 break;
1413 }
Darren Hart52400ba2009-04-03 13:40:49 -07001414
1415 /*
1416 * Wake nr_wake waiters. For requeue_pi, if we acquired the
1417 * lock, we already woke the top_waiter. If not, it will be
1418 * woken by futex_unlock_pi().
1419 */
1420 if (++task_count <= nr_wake && !requeue_pi) {
1421 wake_futex(this);
1422 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 }
Darren Hart52400ba2009-04-03 13:40:49 -07001424
Darren Hart84bc4af2009-08-13 17:36:53 -07001425 /* Ensure we requeue to the expected futex for requeue_pi. */
1426 if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) {
1427 ret = -EINVAL;
1428 break;
1429 }
1430
Darren Hart52400ba2009-04-03 13:40:49 -07001431 /*
1432 * Requeue nr_requeue waiters and possibly one more in the case
1433 * of requeue_pi if we couldn't acquire the lock atomically.
1434 */
1435 if (requeue_pi) {
1436 /* Prepare the waiter to take the rt_mutex. */
1437 atomic_inc(&pi_state->refcount);
1438 this->pi_state = pi_state;
1439 ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex,
1440 this->rt_waiter,
1441 this->task, 1);
1442 if (ret == 1) {
1443 /* We got the lock. */
Darren Hartbeda2c72009-08-09 15:34:39 -07001444 requeue_pi_wake_futex(this, &key2, hb2);
Darren Hart89061d32009-10-15 15:30:48 -07001445 drop_count++;
Darren Hart52400ba2009-04-03 13:40:49 -07001446 continue;
1447 } else if (ret) {
1448 /* -EDEADLK */
1449 this->pi_state = NULL;
1450 free_pi_state(pi_state);
1451 goto out_unlock;
1452 }
1453 }
1454 requeue_futex(this, hb1, hb2, &key2);
1455 drop_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 }
1457
1458out_unlock:
Darren Hart5eb3dc62009-03-12 00:55:52 -07001459 double_unlock_hb(hb1, hb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
Darren Hartcd84a422009-04-02 14:19:38 -07001461 /*
1462 * drop_futex_key_refs() must be called outside the spinlocks. During
1463 * the requeue we moved futex_q's from the hash bucket at key1 to the
1464 * one at key2 and updated their key pointer. We no longer need to
1465 * hold the references to key1.
1466 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 while (--drop_count >= 0)
Rusty Russell9adef58b2007-05-08 00:26:42 -07001468 drop_futex_key_refs(&key1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469
Darren Hart42d35d42008-12-29 15:49:53 -08001470out_put_keys:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001471 put_futex_key(&key2);
Darren Hart42d35d42008-12-29 15:49:53 -08001472out_put_key1:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001473 put_futex_key(&key1);
Darren Hart42d35d42008-12-29 15:49:53 -08001474out:
Darren Hart52400ba2009-04-03 13:40:49 -07001475 if (pi_state != NULL)
1476 free_pi_state(pi_state);
1477 return ret ? ret : task_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478}
1479
1480/* The key must be already stored in q->key. */
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01001481static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
Namhyung Kim15e408c2010-09-14 21:43:48 +09001482 __acquires(&hb->lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483{
Ingo Molnare2970f22006-06-27 02:54:47 -07001484 struct futex_hash_bucket *hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
Ingo Molnare2970f22006-06-27 02:54:47 -07001486 hb = hash_futex(&q->key);
1487 q->lock_ptr = &hb->lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
Ingo Molnare2970f22006-06-27 02:54:47 -07001489 spin_lock(&hb->lock);
1490 return hb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491}
1492
Darren Hartd40d65c2009-09-21 22:30:15 -07001493static inline void
1494queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
Namhyung Kim15e408c2010-09-14 21:43:48 +09001495 __releases(&hb->lock)
Darren Hartd40d65c2009-09-21 22:30:15 -07001496{
1497 spin_unlock(&hb->lock);
Darren Hartd40d65c2009-09-21 22:30:15 -07001498}
1499
1500/**
1501 * queue_me() - Enqueue the futex_q on the futex_hash_bucket
1502 * @q: The futex_q to enqueue
1503 * @hb: The destination hash bucket
1504 *
1505 * The hb->lock must be held by the caller, and is released here. A call to
1506 * queue_me() is typically paired with exactly one call to unqueue_me(). The
1507 * exceptions involve the PI related operations, which may use unqueue_me_pi()
1508 * or nothing if the unqueue is done as part of the wake process and the unqueue
1509 * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
1510 * an example).
1511 */
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01001512static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
Namhyung Kim15e408c2010-09-14 21:43:48 +09001513 __releases(&hb->lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514{
Pierre Peifferec92d082007-05-09 02:35:00 -07001515 int prio;
1516
1517 /*
1518 * The priority used to register this element is
1519 * - either the real thread-priority for the real-time threads
1520 * (i.e. threads with a priority lower than MAX_RT_PRIO)
1521 * - or MAX_RT_PRIO for non-RT threads.
1522 * Thus, all RT-threads are woken first in priority order, and
1523 * the others are woken last, in FIFO order.
1524 */
1525 prio = min(current->normal_prio, MAX_RT_PRIO);
1526
1527 plist_node_init(&q->list, prio);
Pierre Peifferec92d082007-05-09 02:35:00 -07001528 plist_add(&q->list, &hb->chain);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001529 q->task = current;
Ingo Molnare2970f22006-06-27 02:54:47 -07001530 spin_unlock(&hb->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531}
1532
Darren Hartd40d65c2009-09-21 22:30:15 -07001533/**
1534 * unqueue_me() - Remove the futex_q from its futex_hash_bucket
1535 * @q: The futex_q to unqueue
1536 *
1537 * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must
1538 * be paired with exactly one earlier call to queue_me().
1539 *
Randy Dunlap6c23cbb2013-03-05 10:00:24 -08001540 * Return:
1541 * 1 - if the futex_q was still queued (and we removed unqueued it);
Darren Hartd40d65c2009-09-21 22:30:15 -07001542 * 0 - if the futex_q was already removed by the waking thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544static int unqueue_me(struct futex_q *q)
1545{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 spinlock_t *lock_ptr;
Ingo Molnare2970f22006-06-27 02:54:47 -07001547 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
1549 /* In the common case we don't take the spinlock, which is nice. */
Darren Hart42d35d42008-12-29 15:49:53 -08001550retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 lock_ptr = q->lock_ptr;
Christian Borntraegere91467e2006-08-05 12:13:52 -07001552 barrier();
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001553 if (lock_ptr != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 spin_lock(lock_ptr);
1555 /*
1556 * q->lock_ptr can change between reading it and
1557 * spin_lock(), causing us to take the wrong lock. This
1558 * corrects the race condition.
1559 *
1560 * Reasoning goes like this: if we have the wrong lock,
1561 * q->lock_ptr must have changed (maybe several times)
1562 * between reading it and the spin_lock(). It can
1563 * change again after the spin_lock() but only if it was
1564 * already changed before the spin_lock(). It cannot,
1565 * however, change back to the original value. Therefore
1566 * we can detect whether we acquired the correct lock.
1567 */
1568 if (unlikely(lock_ptr != q->lock_ptr)) {
1569 spin_unlock(lock_ptr);
1570 goto retry;
1571 }
Lai Jiangshan2e129782010-12-22 14:18:50 +08001572 __unqueue_futex(q);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001573
1574 BUG_ON(q->pi_state);
1575
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 spin_unlock(lock_ptr);
1577 ret = 1;
1578 }
1579
Rusty Russell9adef58b2007-05-08 00:26:42 -07001580 drop_futex_key_refs(&q->key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 return ret;
1582}
1583
Ingo Molnarc87e2832006-06-27 02:54:58 -07001584/*
1585 * PI futexes can not be requeued and must remove themself from the
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001586 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1587 * and dropped here.
Ingo Molnarc87e2832006-06-27 02:54:58 -07001588 */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001589static void unqueue_me_pi(struct futex_q *q)
Namhyung Kim15e408c2010-09-14 21:43:48 +09001590 __releases(q->lock_ptr)
Ingo Molnarc87e2832006-06-27 02:54:58 -07001591{
Lai Jiangshan2e129782010-12-22 14:18:50 +08001592 __unqueue_futex(q);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001593
1594 BUG_ON(!q->pi_state);
1595 free_pi_state(q->pi_state);
1596 q->pi_state = NULL;
1597
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001598 spin_unlock(q->lock_ptr);
Ingo Molnarc87e2832006-06-27 02:54:58 -07001599}
1600
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001601/*
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001602 * Fixup the pi_state owner with the new owner.
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001603 *
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001604 * Must be called with hash bucket lock held and mm->sem held for non
1605 * private futexes.
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001606 */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001607static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001608 struct task_struct *newowner)
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001609{
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001610 u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001611 struct futex_pi_state *pi_state = q->pi_state;
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001612 struct task_struct *oldowner = pi_state->owner;
Vitaliy Ivanov7cfdaf32011-07-07 15:10:31 +03001613 u32 uval, uninitialized_var(curval), newval;
Darren Harte4dc5b72009-03-12 00:56:13 -07001614 int ret;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001615
1616 /* Owner died? */
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001617 if (!pi_state->owner)
1618 newtid |= FUTEX_OWNER_DIED;
1619
1620 /*
1621 * We are here either because we stole the rtmutex from the
Lai Jiangshan81612392011-01-14 17:09:41 +08001622 * previous highest priority waiter or we are the highest priority
1623 * waiter but failed to get the rtmutex the first time.
1624 * We have to replace the newowner TID in the user space variable.
1625 * This must be atomic as we have to preserve the owner died bit here.
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001626 *
Darren Hartb2d09942009-03-12 00:55:37 -07001627 * Note: We write the user space value _before_ changing the pi_state
1628 * because we can fault here. Imagine swapped out pages or a fork
1629 * that marked all the anonymous memory readonly for cow.
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001630 *
1631 * Modifying pi_state _before_ the user space value would
1632 * leave the pi_state in an inconsistent state when we fault
1633 * here, because we need to drop the hash bucket lock to
1634 * handle the fault. This might be observed in the PID check
1635 * in lookup_pi_state.
1636 */
1637retry:
1638 if (get_futex_value_locked(&uval, uaddr))
1639 goto handle_fault;
1640
1641 while (1) {
1642 newval = (uval & FUTEX_OWNER_DIED) | newtid;
1643
Michel Lespinasse37a9d912011-03-10 18:48:51 -08001644 if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001645 goto handle_fault;
1646 if (curval == uval)
1647 break;
1648 uval = curval;
1649 }
1650
1651 /*
1652 * We fixed up user space. Now we need to fix the pi_state
1653 * itself.
1654 */
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001655 if (pi_state->owner != NULL) {
Thomas Gleixner1d615482009-11-17 14:54:03 +01001656 raw_spin_lock_irq(&pi_state->owner->pi_lock);
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001657 WARN_ON(list_empty(&pi_state->list));
1658 list_del_init(&pi_state->list);
Thomas Gleixner1d615482009-11-17 14:54:03 +01001659 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001660 }
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001661
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001662 pi_state->owner = newowner;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001663
Thomas Gleixner1d615482009-11-17 14:54:03 +01001664 raw_spin_lock_irq(&newowner->pi_lock);
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001665 WARN_ON(!list_empty(&pi_state->list));
Thomas Gleixnercdf71a12008-01-08 19:47:38 +01001666 list_add(&pi_state->list, &newowner->pi_state_list);
Thomas Gleixner1d615482009-11-17 14:54:03 +01001667 raw_spin_unlock_irq(&newowner->pi_lock);
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001668 return 0;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001669
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001670 /*
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001671 * To handle the page fault we need to drop the hash bucket
Lai Jiangshan81612392011-01-14 17:09:41 +08001672 * lock here. That gives the other task (either the highest priority
1673 * waiter itself or the task which stole the rtmutex) the
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001674 * chance to try the fixup of the pi_state. So once we are
1675 * back from handling the fault we need to check the pi_state
1676 * after reacquiring the hash bucket lock and before trying to
1677 * do another fixup. When the fixup has been done already we
1678 * simply return.
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001679 */
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001680handle_fault:
1681 spin_unlock(q->lock_ptr);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001682
Thomas Gleixnerd0725992009-06-11 23:15:43 +02001683 ret = fault_in_user_writeable(uaddr);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001684
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001685 spin_lock(q->lock_ptr);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07001686
Thomas Gleixner1b7558e2008-06-23 11:21:58 +02001687 /*
1688 * Check if someone else fixed it for us:
1689 */
1690 if (pi_state->owner != oldowner)
1691 return 0;
1692
1693 if (ret)
1694 return ret;
1695
1696 goto retry;
Pierre Peifferd0aa7a72007-05-09 02:35:02 -07001697}
1698
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001699static long futex_wait_restart(struct restart_block *restart);
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07001700
Darren Hartca5f9522009-04-03 13:39:33 -07001701/**
Darren Hartdd973992009-04-03 13:40:02 -07001702 * fixup_owner() - Post lock pi_state and corner case management
1703 * @uaddr: user address of the futex
Darren Hartdd973992009-04-03 13:40:02 -07001704 * @q: futex_q (contains pi_state and access to the rt_mutex)
1705 * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
1706 *
1707 * After attempting to lock an rt_mutex, this function is called to cleanup
1708 * the pi_state owner as well as handle race conditions that may allow us to
1709 * acquire the lock. Must be called with the hb lock held.
1710 *
Randy Dunlap6c23cbb2013-03-05 10:00:24 -08001711 * Return:
1712 * 1 - success, lock taken;
1713 * 0 - success, lock not taken;
Darren Hartdd973992009-04-03 13:40:02 -07001714 * <0 - on error (-EFAULT)
1715 */
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001716static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked)
Darren Hartdd973992009-04-03 13:40:02 -07001717{
1718 struct task_struct *owner;
1719 int ret = 0;
1720
1721 if (locked) {
1722 /*
1723 * Got the lock. We might not be the anticipated owner if we
1724 * did a lock-steal - fix up the PI-state in that case:
1725 */
1726 if (q->pi_state->owner != current)
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001727 ret = fixup_pi_state_owner(uaddr, q, current);
Darren Hartdd973992009-04-03 13:40:02 -07001728 goto out;
1729 }
1730
1731 /*
1732 * Catch the rare case, where the lock was released when we were on the
1733 * way back before we locked the hash bucket.
1734 */
1735 if (q->pi_state->owner == current) {
1736 /*
1737 * Try to get the rt_mutex now. This might fail as some other
1738 * task acquired the rt_mutex after we removed ourself from the
1739 * rt_mutex waiters list.
1740 */
1741 if (rt_mutex_trylock(&q->pi_state->pi_mutex)) {
1742 locked = 1;
1743 goto out;
1744 }
1745
1746 /*
1747 * pi_state is incorrect, some other task did a lock steal and
1748 * we returned due to timeout or signal without taking the
Lai Jiangshan81612392011-01-14 17:09:41 +08001749 * rt_mutex. Too late.
Darren Hartdd973992009-04-03 13:40:02 -07001750 */
Lai Jiangshan81612392011-01-14 17:09:41 +08001751 raw_spin_lock(&q->pi_state->pi_mutex.wait_lock);
Darren Hartdd973992009-04-03 13:40:02 -07001752 owner = rt_mutex_owner(&q->pi_state->pi_mutex);
Lai Jiangshan81612392011-01-14 17:09:41 +08001753 if (!owner)
1754 owner = rt_mutex_next_owner(&q->pi_state->pi_mutex);
1755 raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001756 ret = fixup_pi_state_owner(uaddr, q, owner);
Darren Hartdd973992009-04-03 13:40:02 -07001757 goto out;
1758 }
1759
1760 /*
1761 * Paranoia check. If we did not take the lock, then we should not be
Lai Jiangshan81612392011-01-14 17:09:41 +08001762 * the owner of the rt_mutex.
Darren Hartdd973992009-04-03 13:40:02 -07001763 */
1764 if (rt_mutex_owner(&q->pi_state->pi_mutex) == current)
1765 printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p "
1766 "pi-state %p\n", ret,
1767 q->pi_state->pi_mutex.owner,
1768 q->pi_state->owner);
1769
1770out:
1771 return ret ? ret : locked;
1772}
1773
1774/**
Darren Hartca5f9522009-04-03 13:39:33 -07001775 * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
1776 * @hb: the futex hash bucket, must be locked by the caller
1777 * @q: the futex_q to queue up on
1778 * @timeout: the prepared hrtimer_sleeper, or null for no timeout
Darren Hartca5f9522009-04-03 13:39:33 -07001779 */
1780static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02001781 struct hrtimer_sleeper *timeout)
Darren Hartca5f9522009-04-03 13:39:33 -07001782{
Darren Hart9beba3c2009-09-24 11:54:47 -07001783 /*
1784 * The task state is guaranteed to be set before another task can
1785 * wake it. set_current_state() is implemented using set_mb() and
1786 * queue_me() calls spin_unlock() upon completion, both serializing
1787 * access to the hash list and forcing another memory barrier.
1788 */
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02001789 set_current_state(TASK_INTERRUPTIBLE);
Darren Hart0729e192009-09-21 22:30:38 -07001790 queue_me(q, hb);
Darren Hartca5f9522009-04-03 13:39:33 -07001791
1792 /* Arm the timer */
1793 if (timeout) {
1794 hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
1795 if (!hrtimer_active(&timeout->timer))
1796 timeout->task = NULL;
1797 }
1798
1799 /*
Darren Hart0729e192009-09-21 22:30:38 -07001800 * If we have been removed from the hash list, then another task
1801 * has tried to wake us, and we can skip the call to schedule().
Darren Hartca5f9522009-04-03 13:39:33 -07001802 */
1803 if (likely(!plist_node_empty(&q->list))) {
1804 /*
1805 * If the timer has already expired, current will already be
1806 * flagged for rescheduling. Only call schedule if there
1807 * is no timeout, or if it has yet to expire.
1808 */
1809 if (!timeout || timeout->task)
1810 schedule();
1811 }
1812 __set_current_state(TASK_RUNNING);
1813}
1814
Darren Hartf8010732009-04-03 13:40:40 -07001815/**
1816 * futex_wait_setup() - Prepare to wait on a futex
1817 * @uaddr: the futex userspace address
1818 * @val: the expected value
Darren Hartb41277d2010-11-08 13:10:09 -08001819 * @flags: futex flags (FLAGS_SHARED, etc.)
Darren Hartf8010732009-04-03 13:40:40 -07001820 * @q: the associated futex_q
1821 * @hb: storage for hash_bucket pointer to be returned to caller
1822 *
1823 * Setup the futex_q and locate the hash_bucket. Get the futex value and
1824 * compare it with the expected value. Handle atomic faults internally.
1825 * Return with the hb lock held and a q.key reference on success, and unlocked
1826 * with no q.key reference on failure.
1827 *
Randy Dunlap6c23cbb2013-03-05 10:00:24 -08001828 * Return:
1829 * 0 - uaddr contains val and hb has been locked;
Bart Van Asscheca4a04c2011-07-17 09:01:00 +02001830 * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked
Darren Hartf8010732009-04-03 13:40:40 -07001831 */
Darren Hartb41277d2010-11-08 13:10:09 -08001832static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
Darren Hartf8010732009-04-03 13:40:40 -07001833 struct futex_q *q, struct futex_hash_bucket **hb)
1834{
1835 u32 uval;
1836 int ret;
1837
1838 /*
1839 * Access the page AFTER the hash-bucket is locked.
1840 * Order is important:
1841 *
1842 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1843 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1844 *
1845 * The basic logical guarantee of a futex is that it blocks ONLY
1846 * if cond(var) is known to be true at the time of blocking, for
Michel Lespinasse8fe8f542011-03-06 18:07:50 -08001847 * any cond. If we locked the hash-bucket after testing *uaddr, that
1848 * would open a race condition where we could block indefinitely with
Darren Hartf8010732009-04-03 13:40:40 -07001849 * cond(var) false, which would violate the guarantee.
1850 *
Michel Lespinasse8fe8f542011-03-06 18:07:50 -08001851 * On the other hand, we insert q and release the hash-bucket only
1852 * after testing *uaddr. This guarantees that futex_wait() will NOT
1853 * absorb a wakeup if *uaddr does not match the desired values
1854 * while the syscall executes.
Darren Hartf8010732009-04-03 13:40:40 -07001855 */
1856retry:
Shawn Bohrer9ea71502011-06-30 11:21:32 -05001857 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ);
Darren Hartf8010732009-04-03 13:40:40 -07001858 if (unlikely(ret != 0))
Darren Harta5a2a0c2009-04-10 09:50:05 -07001859 return ret;
Darren Hartf8010732009-04-03 13:40:40 -07001860
1861retry_private:
1862 *hb = queue_lock(q);
1863
1864 ret = get_futex_value_locked(&uval, uaddr);
1865
1866 if (ret) {
1867 queue_unlock(q, *hb);
1868
1869 ret = get_user(uval, uaddr);
1870 if (ret)
1871 goto out;
1872
Darren Hartb41277d2010-11-08 13:10:09 -08001873 if (!(flags & FLAGS_SHARED))
Darren Hartf8010732009-04-03 13:40:40 -07001874 goto retry_private;
1875
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001876 put_futex_key(&q->key);
Darren Hartf8010732009-04-03 13:40:40 -07001877 goto retry;
1878 }
1879
1880 if (uval != val) {
1881 queue_unlock(q, *hb);
1882 ret = -EWOULDBLOCK;
1883 }
1884
1885out:
1886 if (ret)
Thomas Gleixnerae791a22010-11-10 13:30:36 +01001887 put_futex_key(&q->key);
Darren Hartf8010732009-04-03 13:40:40 -07001888 return ret;
1889}
1890
Darren Hartb41277d2010-11-08 13:10:09 -08001891static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val,
1892 ktime_t *abs_time, u32 bitset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893{
Darren Hartca5f9522009-04-03 13:39:33 -07001894 struct hrtimer_sleeper timeout, *to = NULL;
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001895 struct restart_block *restart;
Ingo Molnare2970f22006-06-27 02:54:47 -07001896 struct futex_hash_bucket *hb;
Darren Hart5bdb05f2010-11-08 13:40:28 -08001897 struct futex_q q = futex_q_init;
Ingo Molnare2970f22006-06-27 02:54:47 -07001898 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899
Thomas Gleixnercd6899852008-02-01 17:45:14 +01001900 if (!bitset)
1901 return -EINVAL;
Thomas Gleixnercd6899852008-02-01 17:45:14 +01001902 q.bitset = bitset;
Darren Hartca5f9522009-04-03 13:39:33 -07001903
1904 if (abs_time) {
1905 to = &timeout;
1906
Darren Hartb41277d2010-11-08 13:10:09 -08001907 hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
1908 CLOCK_REALTIME : CLOCK_MONOTONIC,
1909 HRTIMER_MODE_ABS);
Darren Hartca5f9522009-04-03 13:39:33 -07001910 hrtimer_init_sleeper(to, current);
1911 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
1912 current->timer_slack_ns);
1913 }
1914
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02001915retry:
Darren Hart7ada8762010-10-17 08:35:04 -07001916 /*
1917 * Prepare to wait on uaddr. On success, holds hb lock and increments
1918 * q.key refs.
1919 */
Darren Hartb41277d2010-11-08 13:10:09 -08001920 ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
Darren Hartf8010732009-04-03 13:40:40 -07001921 if (ret)
Darren Hart42d35d42008-12-29 15:49:53 -08001922 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
Darren Hartca5f9522009-04-03 13:39:33 -07001924 /* queue_me and wait for wakeup, timeout, or a signal. */
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02001925 futex_wait_queue_me(hb, &q, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926
1927 /* If we were woken (and unqueued), we succeeded, whatever. */
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001928 ret = 0;
Darren Hart7ada8762010-10-17 08:35:04 -07001929 /* unqueue_me() drops q.key ref */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 if (!unqueue_me(&q))
Darren Hart7ada8762010-10-17 08:35:04 -07001931 goto out;
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001932 ret = -ETIMEDOUT;
Darren Hartca5f9522009-04-03 13:39:33 -07001933 if (to && !to->task)
Darren Hart7ada8762010-10-17 08:35:04 -07001934 goto out;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001935
Ingo Molnare2970f22006-06-27 02:54:47 -07001936 /*
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02001937 * We expect signal_pending(current), but we might be the
1938 * victim of a spurious wakeup as well.
Ingo Molnare2970f22006-06-27 02:54:47 -07001939 */
Darren Hart7ada8762010-10-17 08:35:04 -07001940 if (!signal_pending(current))
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02001941 goto retry;
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02001942
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001943 ret = -ERESTARTSYS;
Pierre Peifferc19384b2007-05-09 02:35:02 -07001944 if (!abs_time)
Darren Hart7ada8762010-10-17 08:35:04 -07001945 goto out;
Steven Rostedtce6bd422007-12-05 15:46:09 +01001946
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001947 restart = &current_thread_info()->restart_block;
1948 restart->fn = futex_wait_restart;
Namhyung Kima3c74c52010-09-14 21:43:47 +09001949 restart->futex.uaddr = uaddr;
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001950 restart->futex.val = val;
1951 restart->futex.time = abs_time->tv64;
1952 restart->futex.bitset = bitset;
Darren Hart0cd9c642011-04-14 15:41:57 -07001953 restart->futex.flags = flags | FLAGS_HAS_TIMEOUT;
Peter Zijlstra2fff78c2009-02-11 18:10:10 +01001954
1955 ret = -ERESTART_RESTARTBLOCK;
1956
Darren Hart42d35d42008-12-29 15:49:53 -08001957out:
Darren Hartca5f9522009-04-03 13:39:33 -07001958 if (to) {
1959 hrtimer_cancel(&to->timer);
1960 destroy_hrtimer_on_stack(&to->timer);
1961 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07001962 return ret;
1963}
1964
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001965
1966static long futex_wait_restart(struct restart_block *restart)
1967{
Namhyung Kima3c74c52010-09-14 21:43:47 +09001968 u32 __user *uaddr = restart->futex.uaddr;
Darren Harta72188d2009-04-03 13:40:22 -07001969 ktime_t t, *tp = NULL;
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001970
Darren Harta72188d2009-04-03 13:40:22 -07001971 if (restart->futex.flags & FLAGS_HAS_TIMEOUT) {
1972 t.tv64 = restart->futex.time;
1973 tp = &t;
1974 }
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001975 restart->fn = do_no_restart_syscall;
Darren Hartb41277d2010-11-08 13:10:09 -08001976
1977 return (long)futex_wait(uaddr, restart->futex.flags,
1978 restart->futex.val, tp, restart->futex.bitset);
Nick Piggin72c1bbf2007-05-08 00:26:43 -07001979}
1980
1981
Ingo Molnarc87e2832006-06-27 02:54:58 -07001982/*
1983 * Userspace tried a 0 -> TID atomic transition of the futex value
1984 * and failed. The kernel side here does the whole locking operation:
1985 * if there are waiters then it will block, it does PI, etc. (Due to
1986 * races the kernel might see a 0 value of the futex too.)
1987 */
Darren Hartb41277d2010-11-08 13:10:09 -08001988static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect,
1989 ktime_t *time, int trylock)
Ingo Molnarc87e2832006-06-27 02:54:58 -07001990{
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07001991 struct hrtimer_sleeper timeout, *to = NULL;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001992 struct futex_hash_bucket *hb;
Darren Hart5bdb05f2010-11-08 13:40:28 -08001993 struct futex_q q = futex_q_init;
Darren Hartdd973992009-04-03 13:40:02 -07001994 int res, ret;
Ingo Molnarc87e2832006-06-27 02:54:58 -07001995
1996 if (refill_pi_state_cache())
1997 return -ENOMEM;
1998
Pierre Peifferc19384b2007-05-09 02:35:02 -07001999 if (time) {
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07002000 to = &timeout;
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07002001 hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME,
2002 HRTIMER_MODE_ABS);
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07002003 hrtimer_init_sleeper(to, current);
Arjan van de Vencc584b22008-09-01 15:02:30 -07002004 hrtimer_set_expires(&to->timer, *time);
Thomas Gleixnerc5780e92006-09-08 09:47:15 -07002005 }
2006
Darren Hart42d35d42008-12-29 15:49:53 -08002007retry:
Shawn Bohrer9ea71502011-06-30 11:21:32 -05002008 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002009 if (unlikely(ret != 0))
Darren Hart42d35d42008-12-29 15:49:53 -08002010 goto out;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002011
Darren Harte4dc5b72009-03-12 00:56:13 -07002012retry_private:
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01002013 hb = queue_lock(&q);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002014
Darren Hartbab5bc92009-04-07 23:23:50 -07002015 ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002016 if (unlikely(ret)) {
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002017 switch (ret) {
Darren Hart1a520842009-04-03 13:39:52 -07002018 case 1:
2019 /* We got the lock. */
2020 ret = 0;
2021 goto out_unlock_put_key;
2022 case -EFAULT:
2023 goto uaddr_faulted;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002024 case -EAGAIN:
2025 /*
2026 * Task is exiting and we just wait for the
2027 * exit to complete.
2028 */
2029 queue_unlock(&q, hb);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002030 put_futex_key(&q.key);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002031 cond_resched();
2032 goto retry;
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002033 default:
Darren Hart42d35d42008-12-29 15:49:53 -08002034 goto out_unlock_put_key;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002035 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07002036 }
2037
2038 /*
2039 * Only actually queue now that the atomic ops are done:
2040 */
Eric Sesterhenn82af7ac2008-01-25 10:40:46 +01002041 queue_me(&q, hb);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002042
Ingo Molnarc87e2832006-06-27 02:54:58 -07002043 WARN_ON(!q.pi_state);
2044 /*
2045 * Block on the PI mutex:
2046 */
2047 if (!trylock)
2048 ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1);
2049 else {
2050 ret = rt_mutex_trylock(&q.pi_state->pi_mutex);
2051 /* Fixup the trylock return value: */
2052 ret = ret ? 0 : -EWOULDBLOCK;
2053 }
2054
Vernon Mauerya99e4e42006-07-01 04:35:42 -07002055 spin_lock(q.lock_ptr);
Darren Hartdd973992009-04-03 13:40:02 -07002056 /*
2057 * Fixup the pi_state owner and possibly acquire the lock if we
2058 * haven't already.
2059 */
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002060 res = fixup_owner(uaddr, &q, !ret);
Darren Hartdd973992009-04-03 13:40:02 -07002061 /*
2062 * If fixup_owner() returned an error, proprogate that. If it acquired
2063 * the lock, clear our -ETIMEDOUT or -EINTR.
2064 */
2065 if (res)
2066 ret = (res < 0) ? res : 0;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002067
Darren Harte8f63862009-03-12 00:56:06 -07002068 /*
Darren Hartdd973992009-04-03 13:40:02 -07002069 * If fixup_owner() faulted and was unable to handle the fault, unlock
2070 * it and return the fault to userspace.
Darren Harte8f63862009-03-12 00:56:06 -07002071 */
2072 if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current))
2073 rt_mutex_unlock(&q.pi_state->pi_mutex);
2074
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002075 /* Unqueue and drop the lock */
2076 unqueue_me_pi(&q);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002077
Mikael Pettersson5ecb01c2010-01-23 22:36:29 +01002078 goto out_put_key;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002079
Darren Hart42d35d42008-12-29 15:49:53 -08002080out_unlock_put_key:
Ingo Molnarc87e2832006-06-27 02:54:58 -07002081 queue_unlock(&q, hb);
2082
Darren Hart42d35d42008-12-29 15:49:53 -08002083out_put_key:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002084 put_futex_key(&q.key);
Darren Hart42d35d42008-12-29 15:49:53 -08002085out:
Thomas Gleixner237fc6e2008-04-30 00:55:04 -07002086 if (to)
2087 destroy_hrtimer_on_stack(&to->timer);
Darren Hartdd973992009-04-03 13:40:02 -07002088 return ret != -EINTR ? ret : -ERESTARTNOINTR;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002089
Darren Hart42d35d42008-12-29 15:49:53 -08002090uaddr_faulted:
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002091 queue_unlock(&q, hb);
2092
Thomas Gleixnerd0725992009-06-11 23:15:43 +02002093 ret = fault_in_user_writeable(uaddr);
Darren Harte4dc5b72009-03-12 00:56:13 -07002094 if (ret)
2095 goto out_put_key;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002096
Darren Hartb41277d2010-11-08 13:10:09 -08002097 if (!(flags & FLAGS_SHARED))
Darren Harte4dc5b72009-03-12 00:56:13 -07002098 goto retry_private;
2099
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002100 put_futex_key(&q.key);
Darren Harte4dc5b72009-03-12 00:56:13 -07002101 goto retry;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002102}
2103
2104/*
Ingo Molnarc87e2832006-06-27 02:54:58 -07002105 * Userspace attempted a TID -> 0 atomic transition, and failed.
2106 * This is the in-kernel slowpath: we look up the PI state (if any),
2107 * and do the rt-mutex unlock.
2108 */
Darren Hartb41277d2010-11-08 13:10:09 -08002109static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags)
Ingo Molnarc87e2832006-06-27 02:54:58 -07002110{
2111 struct futex_hash_bucket *hb;
2112 struct futex_q *this, *next;
Pierre Peifferec92d082007-05-09 02:35:00 -07002113 struct plist_head *head;
Peter Zijlstra38d47c12008-09-26 19:32:20 +02002114 union futex_key key = FUTEX_KEY_INIT;
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +01002115 u32 uval, vpid = task_pid_vnr(current);
Darren Harte4dc5b72009-03-12 00:56:13 -07002116 int ret;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002117
2118retry:
2119 if (get_user(uval, uaddr))
2120 return -EFAULT;
2121 /*
2122 * We release only a lock we actually own:
2123 */
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +01002124 if ((uval & FUTEX_TID_MASK) != vpid)
Ingo Molnarc87e2832006-06-27 02:54:58 -07002125 return -EPERM;
Ingo Molnarc87e2832006-06-27 02:54:58 -07002126
Shawn Bohrer9ea71502011-06-30 11:21:32 -05002127 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002128 if (unlikely(ret != 0))
2129 goto out;
2130
2131 hb = hash_futex(&key);
2132 spin_lock(&hb->lock);
2133
Ingo Molnarc87e2832006-06-27 02:54:58 -07002134 /*
2135 * To avoid races, try to do the TID -> 0 atomic transition
2136 * again. If it succeeds then we can return without waking
2137 * anyone else up:
2138 */
Michel Lespinasse37a9d912011-03-10 18:48:51 -08002139 if (!(uval & FUTEX_OWNER_DIED) &&
2140 cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0))
Ingo Molnarc87e2832006-06-27 02:54:58 -07002141 goto pi_faulted;
2142 /*
2143 * Rare case: we managed to release the lock atomically,
2144 * no need to wake anyone else up:
2145 */
Thomas Gleixnerc0c9ed12011-03-11 11:51:22 +01002146 if (unlikely(uval == vpid))
Ingo Molnarc87e2832006-06-27 02:54:58 -07002147 goto out_unlock;
2148
2149 /*
2150 * Ok, other tasks may need to be woken up - check waiters
2151 * and do the wakeup if necessary:
2152 */
2153 head = &hb->chain;
2154
Pierre Peifferec92d082007-05-09 02:35:00 -07002155 plist_for_each_entry_safe(this, next, head, list) {
Ingo Molnarc87e2832006-06-27 02:54:58 -07002156 if (!match_futex (&this->key, &key))
2157 continue;
2158 ret = wake_futex_pi(uaddr, uval, this);
2159 /*
2160 * The atomic access to the futex value
2161 * generated a pagefault, so retry the
2162 * user-access and the wakeup:
2163 */
2164 if (ret == -EFAULT)
2165 goto pi_faulted;
2166 goto out_unlock;
2167 }
2168 /*
2169 * No waiters - kernel unlocks the futex:
2170 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002171 if (!(uval & FUTEX_OWNER_DIED)) {
2172 ret = unlock_futex_pi(uaddr, uval);
2173 if (ret == -EFAULT)
2174 goto pi_faulted;
2175 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07002176
2177out_unlock:
2178 spin_unlock(&hb->lock);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002179 put_futex_key(&key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002180
Darren Hart42d35d42008-12-29 15:49:53 -08002181out:
Ingo Molnarc87e2832006-06-27 02:54:58 -07002182 return ret;
2183
2184pi_faulted:
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -07002185 spin_unlock(&hb->lock);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002186 put_futex_key(&key);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002187
Thomas Gleixnerd0725992009-06-11 23:15:43 +02002188 ret = fault_in_user_writeable(uaddr);
Darren Hartb5686362008-12-18 15:06:34 -08002189 if (!ret)
Ingo Molnarc87e2832006-06-27 02:54:58 -07002190 goto retry;
2191
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 return ret;
2193}
2194
Darren Hart52400ba2009-04-03 13:40:49 -07002195/**
2196 * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2197 * @hb: the hash_bucket futex_q was original enqueued on
2198 * @q: the futex_q woken while waiting to be requeued
2199 * @key2: the futex_key of the requeue target futex
2200 * @timeout: the timeout associated with the wait (NULL if none)
2201 *
2202 * Detect if the task was woken on the initial futex as opposed to the requeue
2203 * target futex. If so, determine if it was a timeout or a signal that caused
2204 * the wakeup and return the appropriate error code to the caller. Must be
2205 * called with the hb lock held.
2206 *
Randy Dunlap6c23cbb2013-03-05 10:00:24 -08002207 * Return:
2208 * 0 = no early wakeup detected;
2209 * <0 = -ETIMEDOUT or -ERESTARTNOINTR
Darren Hart52400ba2009-04-03 13:40:49 -07002210 */
2211static inline
2212int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb,
2213 struct futex_q *q, union futex_key *key2,
2214 struct hrtimer_sleeper *timeout)
2215{
2216 int ret = 0;
2217
2218 /*
2219 * With the hb lock held, we avoid races while we process the wakeup.
2220 * We only need to hold hb (and not hb2) to ensure atomicity as the
2221 * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2222 * It can't be requeued from uaddr2 to something else since we don't
2223 * support a PI aware source futex for requeue.
2224 */
2225 if (!match_futex(&q->key, key2)) {
2226 WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr));
2227 /*
2228 * We were woken prior to requeue by a timeout or a signal.
2229 * Unqueue the futex_q and determine which it was.
2230 */
Lai Jiangshan2e129782010-12-22 14:18:50 +08002231 plist_del(&q->list, &hb->chain);
Darren Hart52400ba2009-04-03 13:40:49 -07002232
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02002233 /* Handle spurious wakeups gracefully */
Thomas Gleixner11df6dd2009-10-28 20:26:48 +01002234 ret = -EWOULDBLOCK;
Darren Hart52400ba2009-04-03 13:40:49 -07002235 if (timeout && !timeout->task)
2236 ret = -ETIMEDOUT;
Thomas Gleixnerd58e6572009-10-13 20:40:43 +02002237 else if (signal_pending(current))
Thomas Gleixner1c840c12009-05-20 09:22:40 +02002238 ret = -ERESTARTNOINTR;
Darren Hart52400ba2009-04-03 13:40:49 -07002239 }
2240 return ret;
2241}
2242
2243/**
2244 * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
Darren Hart56ec1602009-09-21 22:29:59 -07002245 * @uaddr: the futex we initially wait on (non-pi)
Darren Hartb41277d2010-11-08 13:10:09 -08002246 * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be
Darren Hart52400ba2009-04-03 13:40:49 -07002247 * the same type, no requeueing from private to shared, etc.
2248 * @val: the expected value of uaddr
2249 * @abs_time: absolute timeout
Darren Hart56ec1602009-09-21 22:29:59 -07002250 * @bitset: 32 bit wakeup bitset set by userspace, defaults to all
Darren Hart52400ba2009-04-03 13:40:49 -07002251 * @uaddr2: the pi futex we will take prior to returning to user-space
2252 *
2253 * The caller will wait on uaddr and will be requeued by futex_requeue() to
Darren Hart6f7b0a22012-07-20 11:53:31 -07002254 * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake
2255 * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to
2256 * userspace. This ensures the rt_mutex maintains an owner when it has waiters;
2257 * without one, the pi logic would not know which task to boost/deboost, if
2258 * there was a need to.
Darren Hart52400ba2009-04-03 13:40:49 -07002259 *
2260 * We call schedule in futex_wait_queue_me() when we enqueue and return there
Randy Dunlap6c23cbb2013-03-05 10:00:24 -08002261 * via the following--
Darren Hart52400ba2009-04-03 13:40:49 -07002262 * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
Darren Hartcc6db4e2009-07-31 16:20:10 -07002263 * 2) wakeup on uaddr2 after a requeue
2264 * 3) signal
2265 * 4) timeout
Darren Hart52400ba2009-04-03 13:40:49 -07002266 *
Darren Hartcc6db4e2009-07-31 16:20:10 -07002267 * If 3, cleanup and return -ERESTARTNOINTR.
Darren Hart52400ba2009-04-03 13:40:49 -07002268 *
2269 * If 2, we may then block on trying to take the rt_mutex and return via:
2270 * 5) successful lock
2271 * 6) signal
2272 * 7) timeout
2273 * 8) other lock acquisition failure
2274 *
Darren Hartcc6db4e2009-07-31 16:20:10 -07002275 * If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
Darren Hart52400ba2009-04-03 13:40:49 -07002276 *
2277 * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2278 *
Randy Dunlap6c23cbb2013-03-05 10:00:24 -08002279 * Return:
2280 * 0 - On success;
Darren Hart52400ba2009-04-03 13:40:49 -07002281 * <0 - On error
2282 */
Darren Hartb41277d2010-11-08 13:10:09 -08002283static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
Darren Hart52400ba2009-04-03 13:40:49 -07002284 u32 val, ktime_t *abs_time, u32 bitset,
Darren Hartb41277d2010-11-08 13:10:09 -08002285 u32 __user *uaddr2)
Darren Hart52400ba2009-04-03 13:40:49 -07002286{
2287 struct hrtimer_sleeper timeout, *to = NULL;
2288 struct rt_mutex_waiter rt_waiter;
2289 struct rt_mutex *pi_mutex = NULL;
Darren Hart52400ba2009-04-03 13:40:49 -07002290 struct futex_hash_bucket *hb;
Darren Hart5bdb05f2010-11-08 13:40:28 -08002291 union futex_key key2 = FUTEX_KEY_INIT;
2292 struct futex_q q = futex_q_init;
Darren Hart52400ba2009-04-03 13:40:49 -07002293 int res, ret;
Darren Hart52400ba2009-04-03 13:40:49 -07002294
Darren Hart6f7b0a22012-07-20 11:53:31 -07002295 if (uaddr == uaddr2)
2296 return -EINVAL;
2297
Darren Hart52400ba2009-04-03 13:40:49 -07002298 if (!bitset)
2299 return -EINVAL;
2300
2301 if (abs_time) {
2302 to = &timeout;
Darren Hartb41277d2010-11-08 13:10:09 -08002303 hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
2304 CLOCK_REALTIME : CLOCK_MONOTONIC,
2305 HRTIMER_MODE_ABS);
Darren Hart52400ba2009-04-03 13:40:49 -07002306 hrtimer_init_sleeper(to, current);
2307 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
2308 current->timer_slack_ns);
2309 }
2310
2311 /*
2312 * The waiter is allocated on our stack, manipulated by the requeue
2313 * code while we sleep on uaddr.
2314 */
2315 debug_rt_mutex_init_waiter(&rt_waiter);
2316 rt_waiter.task = NULL;
2317
Shawn Bohrer9ea71502011-06-30 11:21:32 -05002318 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
Darren Hart52400ba2009-04-03 13:40:49 -07002319 if (unlikely(ret != 0))
2320 goto out;
2321
Darren Hart84bc4af2009-08-13 17:36:53 -07002322 q.bitset = bitset;
2323 q.rt_waiter = &rt_waiter;
2324 q.requeue_pi_key = &key2;
2325
Darren Hart7ada8762010-10-17 08:35:04 -07002326 /*
2327 * Prepare to wait on uaddr. On success, increments q.key (key1) ref
2328 * count.
2329 */
Darren Hartb41277d2010-11-08 13:10:09 -08002330 ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
Thomas Gleixnerc8b15a72009-05-20 09:18:50 +02002331 if (ret)
2332 goto out_key2;
Darren Hart52400ba2009-04-03 13:40:49 -07002333
2334 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
Thomas Gleixnerf1a11e02009-05-05 19:21:40 +02002335 futex_wait_queue_me(hb, &q, to);
Darren Hart52400ba2009-04-03 13:40:49 -07002336
2337 spin_lock(&hb->lock);
2338 ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
2339 spin_unlock(&hb->lock);
2340 if (ret)
2341 goto out_put_keys;
2342
2343 /*
2344 * In order for us to be here, we know our q.key == key2, and since
2345 * we took the hb->lock above, we also know that futex_requeue() has
2346 * completed and we no longer have to concern ourselves with a wakeup
Darren Hart7ada8762010-10-17 08:35:04 -07002347 * race with the atomic proxy lock acquisition by the requeue code. The
2348 * futex_requeue dropped our key1 reference and incremented our key2
2349 * reference count.
Darren Hart52400ba2009-04-03 13:40:49 -07002350 */
2351
2352 /* Check if the requeue code acquired the second futex for us. */
2353 if (!q.rt_waiter) {
2354 /*
2355 * Got the lock. We might not be the anticipated owner if we
2356 * did a lock-steal - fix up the PI-state in that case.
2357 */
2358 if (q.pi_state && (q.pi_state->owner != current)) {
2359 spin_lock(q.lock_ptr);
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002360 ret = fixup_pi_state_owner(uaddr2, &q, current);
Darren Hart52400ba2009-04-03 13:40:49 -07002361 spin_unlock(q.lock_ptr);
2362 }
2363 } else {
2364 /*
2365 * We have been woken up by futex_unlock_pi(), a timeout, or a
2366 * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2367 * the pi_state.
2368 */
Darren Hartf27071c2012-07-20 11:53:30 -07002369 WARN_ON(!q.pi_state);
Darren Hart52400ba2009-04-03 13:40:49 -07002370 pi_mutex = &q.pi_state->pi_mutex;
2371 ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1);
2372 debug_rt_mutex_free_waiter(&rt_waiter);
2373
2374 spin_lock(q.lock_ptr);
2375 /*
2376 * Fixup the pi_state owner and possibly acquire the lock if we
2377 * haven't already.
2378 */
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002379 res = fixup_owner(uaddr2, &q, !ret);
Darren Hart52400ba2009-04-03 13:40:49 -07002380 /*
2381 * If fixup_owner() returned an error, proprogate that. If it
Darren Hart56ec1602009-09-21 22:29:59 -07002382 * acquired the lock, clear -ETIMEDOUT or -EINTR.
Darren Hart52400ba2009-04-03 13:40:49 -07002383 */
2384 if (res)
2385 ret = (res < 0) ? res : 0;
2386
2387 /* Unqueue and drop the lock. */
2388 unqueue_me_pi(&q);
2389 }
2390
2391 /*
2392 * If fixup_pi_state_owner() faulted and was unable to handle the
2393 * fault, unlock the rt_mutex and return the fault to userspace.
2394 */
2395 if (ret == -EFAULT) {
Darren Hartb6070a82012-07-20 11:53:29 -07002396 if (pi_mutex && rt_mutex_owner(pi_mutex) == current)
Darren Hart52400ba2009-04-03 13:40:49 -07002397 rt_mutex_unlock(pi_mutex);
2398 } else if (ret == -EINTR) {
Darren Hart52400ba2009-04-03 13:40:49 -07002399 /*
Darren Hartcc6db4e2009-07-31 16:20:10 -07002400 * We've already been requeued, but cannot restart by calling
2401 * futex_lock_pi() directly. We could restart this syscall, but
2402 * it would detect that the user space "val" changed and return
2403 * -EWOULDBLOCK. Save the overhead of the restart and return
2404 * -EWOULDBLOCK directly.
Darren Hart52400ba2009-04-03 13:40:49 -07002405 */
Thomas Gleixner20708872009-05-19 23:04:59 +02002406 ret = -EWOULDBLOCK;
Darren Hart52400ba2009-04-03 13:40:49 -07002407 }
2408
2409out_put_keys:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002410 put_futex_key(&q.key);
Thomas Gleixnerc8b15a72009-05-20 09:18:50 +02002411out_key2:
Thomas Gleixnerae791a22010-11-10 13:30:36 +01002412 put_futex_key(&key2);
Darren Hart52400ba2009-04-03 13:40:49 -07002413
2414out:
2415 if (to) {
2416 hrtimer_cancel(&to->timer);
2417 destroy_hrtimer_on_stack(&to->timer);
2418 }
2419 return ret;
2420}
2421
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002422/*
2423 * Support for robust futexes: the kernel cleans up held futexes at
2424 * thread exit time.
2425 *
2426 * Implementation: user-space maintains a per-thread list of locks it
2427 * is holding. Upon do_exit(), the kernel carefully walks this list,
2428 * and marks all locks that are owned by this thread with the
Ingo Molnarc87e2832006-06-27 02:54:58 -07002429 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002430 * always manipulated with the lock held, so the list is private and
2431 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2432 * field, to allow the kernel to clean up if the thread dies after
2433 * acquiring the lock, but just before it could have added itself to
2434 * the list. There can only be one such pending lock.
2435 */
2436
2437/**
Darren Hartd96ee562009-09-21 22:30:22 -07002438 * sys_set_robust_list() - Set the robust-futex list head of a task
2439 * @head: pointer to the list-head
2440 * @len: length of the list-head, as userspace expects
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002441 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01002442SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head,
2443 size_t, len)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002444{
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002445 if (!futex_cmpxchg_enabled)
2446 return -ENOSYS;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002447 /*
2448 * The kernel knows only one size for now:
2449 */
2450 if (unlikely(len != sizeof(*head)))
2451 return -EINVAL;
2452
2453 current->robust_list = head;
2454
2455 return 0;
2456}
2457
2458/**
Darren Hartd96ee562009-09-21 22:30:22 -07002459 * sys_get_robust_list() - Get the robust-futex list head of a task
2460 * @pid: pid of the process [zero for current task]
2461 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2462 * @len_ptr: pointer to a length field, the kernel fills in the header size
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002463 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01002464SYSCALL_DEFINE3(get_robust_list, int, pid,
2465 struct robust_list_head __user * __user *, head_ptr,
2466 size_t __user *, len_ptr)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002467{
Al Viroba46df92006-10-10 22:46:07 +01002468 struct robust_list_head __user *head;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002469 unsigned long ret;
Kees Cookbdbb7762012-03-19 16:12:53 -07002470 struct task_struct *p;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002471
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002472 if (!futex_cmpxchg_enabled)
2473 return -ENOSYS;
2474
Kees Cookbdbb7762012-03-19 16:12:53 -07002475 rcu_read_lock();
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002476
Kees Cookbdbb7762012-03-19 16:12:53 -07002477 ret = -ESRCH;
2478 if (!pid)
2479 p = current;
2480 else {
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07002481 p = find_task_by_vpid(pid);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002482 if (!p)
2483 goto err_unlock;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002484 }
2485
Kees Cookbdbb7762012-03-19 16:12:53 -07002486 ret = -EPERM;
2487 if (!ptrace_may_access(p, PTRACE_MODE_READ))
2488 goto err_unlock;
2489
2490 head = p->robust_list;
2491 rcu_read_unlock();
2492
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002493 if (put_user(sizeof(*head), len_ptr))
2494 return -EFAULT;
2495 return put_user(head, head_ptr);
2496
2497err_unlock:
Oleg Nesterovaaa2a972006-09-29 02:00:55 -07002498 rcu_read_unlock();
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002499
2500 return ret;
2501}
2502
2503/*
2504 * Process a futex-list entry, check whether it's owned by the
2505 * dying task, and do notification if so:
2506 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002507int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002508{
Vitaliy Ivanov7cfdaf32011-07-07 15:10:31 +03002509 u32 uval, uninitialized_var(nval), mval;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002510
Ingo Molnar8f17d3a2006-03-27 01:16:27 -08002511retry:
2512 if (get_user(uval, uaddr))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002513 return -1;
2514
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002515 if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) {
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002516 /*
2517 * Ok, this dying thread is truly holding a futex
2518 * of interest. Set the OWNER_DIED bit atomically
2519 * via cmpxchg, and if the value had FUTEX_WAITERS
2520 * set, wake up a waiter (if any). (We have to do a
2521 * futex_wake() even if OWNER_DIED is already set -
2522 * to handle the rare but possible case of recursive
2523 * thread-death.) The rest of the cleanup is done in
2524 * userspace.
2525 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002526 mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
Thomas Gleixner6e0aa9f2011-03-14 10:34:35 +01002527 /*
2528 * We are not holding a lock here, but we want to have
2529 * the pagefault_disable/enable() protection because
2530 * we want to handle the fault gracefully. If the
2531 * access fails we try to fault in the futex with R/W
2532 * verification via get_user_pages. get_user() above
2533 * does not guarantee R/W access. If that fails we
2534 * give up and leave the futex locked.
2535 */
2536 if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) {
2537 if (fault_in_user_writeable(uaddr))
2538 return -1;
2539 goto retry;
2540 }
Ingo Molnarc87e2832006-06-27 02:54:58 -07002541 if (nval != uval)
Ingo Molnar8f17d3a2006-03-27 01:16:27 -08002542 goto retry;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002543
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002544 /*
2545 * Wake robust non-PI futexes here. The wakeup of
2546 * PI futexes happens in exit_pi_state():
2547 */
Thomas Gleixner36cf3b52007-07-15 23:41:20 -07002548 if (!pi && (uval & FUTEX_WAITERS))
Peter Zijlstrac2f9f202008-09-26 19:32:23 +02002549 futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002550 }
2551 return 0;
2552}
2553
2554/*
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002555 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2556 */
2557static inline int fetch_robust_entry(struct robust_list __user **entry,
Al Viroba46df92006-10-10 22:46:07 +01002558 struct robust_list __user * __user *head,
Namhyung Kim1dcc41b2010-09-14 21:43:46 +09002559 unsigned int *pi)
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002560{
2561 unsigned long uentry;
2562
Al Viroba46df92006-10-10 22:46:07 +01002563 if (get_user(uentry, (unsigned long __user *)head))
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002564 return -EFAULT;
2565
Al Viroba46df92006-10-10 22:46:07 +01002566 *entry = (void __user *)(uentry & ~1UL);
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002567 *pi = uentry & 1;
2568
2569 return 0;
2570}
2571
2572/*
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002573 * Walk curr->robust_list (very carefully, it's a userspace list!)
2574 * and mark any locks found there dead, and notify any waiters.
2575 *
2576 * We silently return on any sign of list-walking problem.
2577 */
2578void exit_robust_list(struct task_struct *curr)
2579{
2580 struct robust_list_head __user *head = curr->robust_list;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002581 struct robust_list __user *entry, *next_entry, *pending;
Darren Hart4c115e92010-11-04 15:00:00 -04002582 unsigned int limit = ROBUST_LIST_LIMIT, pi, pip;
2583 unsigned int uninitialized_var(next_pi);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002584 unsigned long futex_offset;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002585 int rc;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002586
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002587 if (!futex_cmpxchg_enabled)
2588 return;
2589
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002590 /*
2591 * Fetch the list head (which was registered earlier, via
2592 * sys_set_robust_list()):
2593 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002594 if (fetch_robust_entry(&entry, &head->list.next, &pi))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002595 return;
2596 /*
2597 * Fetch the relative futex offset:
2598 */
2599 if (get_user(futex_offset, &head->futex_offset))
2600 return;
2601 /*
2602 * Fetch any possibly pending lock-add first, and handle it
2603 * if it exists:
2604 */
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002605 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002606 return;
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002607
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002608 next_entry = NULL; /* avoid warning with gcc */
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002609 while (entry != &head->list) {
2610 /*
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002611 * Fetch the next entry in the list before calling
2612 * handle_futex_death:
2613 */
2614 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
2615 /*
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002616 * A pending lock might already be on the list, so
Ingo Molnarc87e2832006-06-27 02:54:58 -07002617 * don't process it twice:
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002618 */
2619 if (entry != pending)
Al Viroba46df92006-10-10 22:46:07 +01002620 if (handle_futex_death((void __user *)entry + futex_offset,
Ingo Molnare3f2dde2006-07-29 05:17:57 +02002621 curr, pi))
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002622 return;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002623 if (rc)
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002624 return;
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002625 entry = next_entry;
2626 pi = next_pi;
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002627 /*
2628 * Avoid excessively long or circular lists:
2629 */
2630 if (!--limit)
2631 break;
2632
2633 cond_resched();
2634 }
Martin Schwidefsky9f96cb12007-10-01 01:20:13 -07002635
2636 if (pending)
2637 handle_futex_death((void __user *)pending + futex_offset,
2638 curr, pip);
Ingo Molnar0771dfe2006-03-27 01:16:22 -08002639}
2640
Pierre Peifferc19384b2007-05-09 02:35:02 -07002641long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
Ingo Molnare2970f22006-06-27 02:54:47 -07002642 u32 __user *uaddr2, u32 val2, u32 val3)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643{
Thomas Gleixner81b40532012-02-15 12:17:09 +01002644 int cmd = op & FUTEX_CMD_MASK;
Darren Hartb41277d2010-11-08 13:10:09 -08002645 unsigned int flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002647 if (!(op & FUTEX_PRIVATE_FLAG))
Darren Hartb41277d2010-11-08 13:10:09 -08002648 flags |= FLAGS_SHARED;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002649
Darren Hartb41277d2010-11-08 13:10:09 -08002650 if (op & FUTEX_CLOCK_REALTIME) {
2651 flags |= FLAGS_CLOCKRT;
2652 if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
2653 return -ENOSYS;
2654 }
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002655
2656 switch (cmd) {
Thomas Gleixner59263b52012-02-15 12:08:34 +01002657 case FUTEX_LOCK_PI:
2658 case FUTEX_UNLOCK_PI:
2659 case FUTEX_TRYLOCK_PI:
2660 case FUTEX_WAIT_REQUEUE_PI:
2661 case FUTEX_CMP_REQUEUE_PI:
2662 if (!futex_cmpxchg_enabled)
2663 return -ENOSYS;
2664 }
2665
2666 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 case FUTEX_WAIT:
Thomas Gleixnercd6899852008-02-01 17:45:14 +01002668 val3 = FUTEX_BITSET_MATCH_ANY;
2669 case FUTEX_WAIT_BITSET:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002670 return futex_wait(uaddr, flags, val, timeout, val3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 case FUTEX_WAKE:
Thomas Gleixnercd6899852008-02-01 17:45:14 +01002672 val3 = FUTEX_BITSET_MATCH_ANY;
2673 case FUTEX_WAKE_BITSET:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002674 return futex_wake(uaddr, flags, val, val3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 case FUTEX_REQUEUE:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002676 return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 case FUTEX_CMP_REQUEUE:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002678 return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0);
Jakub Jelinek4732efbe2005-09-06 15:16:25 -07002679 case FUTEX_WAKE_OP:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002680 return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002681 case FUTEX_LOCK_PI:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002682 return futex_lock_pi(uaddr, flags, val, timeout, 0);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002683 case FUTEX_UNLOCK_PI:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002684 return futex_unlock_pi(uaddr, flags);
Ingo Molnarc87e2832006-06-27 02:54:58 -07002685 case FUTEX_TRYLOCK_PI:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002686 return futex_lock_pi(uaddr, flags, 0, timeout, 1);
Darren Hart52400ba2009-04-03 13:40:49 -07002687 case FUTEX_WAIT_REQUEUE_PI:
2688 val3 = FUTEX_BITSET_MATCH_ANY;
Thomas Gleixner81b40532012-02-15 12:17:09 +01002689 return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3,
2690 uaddr2);
Darren Hart52400ba2009-04-03 13:40:49 -07002691 case FUTEX_CMP_REQUEUE_PI:
Thomas Gleixner81b40532012-02-15 12:17:09 +01002692 return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 }
Thomas Gleixner81b40532012-02-15 12:17:09 +01002694 return -ENOSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695}
2696
2697
Heiko Carstens17da2bd2009-01-14 14:14:10 +01002698SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
2699 struct timespec __user *, utime, u32 __user *, uaddr2,
2700 u32, val3)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701{
Pierre Peifferc19384b2007-05-09 02:35:02 -07002702 struct timespec ts;
2703 ktime_t t, *tp = NULL;
Ingo Molnare2970f22006-06-27 02:54:47 -07002704 u32 val2 = 0;
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002705 int cmd = op & FUTEX_CMD_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706
Thomas Gleixnercd6899852008-02-01 17:45:14 +01002707 if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
Darren Hart52400ba2009-04-03 13:40:49 -07002708 cmd == FUTEX_WAIT_BITSET ||
2709 cmd == FUTEX_WAIT_REQUEUE_PI)) {
Pierre Peifferc19384b2007-05-09 02:35:02 -07002710 if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 return -EFAULT;
Pierre Peifferc19384b2007-05-09 02:35:02 -07002712 if (!timespec_valid(&ts))
Thomas Gleixner9741ef962006-03-31 02:31:32 -08002713 return -EINVAL;
Pierre Peifferc19384b2007-05-09 02:35:02 -07002714
2715 t = timespec_to_ktime(ts);
Eric Dumazet34f01cc2007-05-09 02:35:04 -07002716 if (cmd == FUTEX_WAIT)
Thomas Gleixner5a7780e2008-02-13 09:20:43 +01002717 t = ktime_add_safe(ktime_get(), t);
Pierre Peifferc19384b2007-05-09 02:35:02 -07002718 tp = &t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719 }
2720 /*
Darren Hart52400ba2009-04-03 13:40:49 -07002721 * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
Andreas Schwabf54f0982007-07-31 00:38:51 -07002722 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 */
Andreas Schwabf54f0982007-07-31 00:38:51 -07002724 if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
Darren Hartba9c22f2009-04-20 22:22:22 -07002725 cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP)
Ingo Molnare2970f22006-06-27 02:54:47 -07002726 val2 = (u32) (unsigned long) utime;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727
Pierre Peifferc19384b2007-05-09 02:35:02 -07002728 return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729}
2730
Benjamin Herrenschmidtf6d107f2008-03-27 14:52:15 +11002731static int __init futex_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732{
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002733 u32 curval;
Thomas Gleixner3e4ab742008-02-23 15:23:55 -08002734 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002736 /*
2737 * This will fail and we want it. Some arch implementations do
2738 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2739 * functionality. We want to know that before we call in any
2740 * of the complex code paths. Also we want to prevent
2741 * registration of robust lists in that case. NULL is
2742 * guaranteed to fault and we get -EFAULT on functional
Randy Dunlapfb62db22010-10-13 11:02:34 -07002743 * implementation, the non-functional ones will return
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002744 * -ENOSYS.
2745 */
Michel Lespinasse37a9d912011-03-10 18:48:51 -08002746 if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT)
Thomas Gleixnera0c1e902008-02-23 15:23:57 -08002747 futex_cmpxchg_enabled = 1;
2748
Thomas Gleixner3e4ab742008-02-23 15:23:55 -08002749 for (i = 0; i < ARRAY_SIZE(futex_queues); i++) {
Dima Zavin732375c2011-07-07 17:27:59 -07002750 plist_head_init(&futex_queues[i].chain);
Thomas Gleixner3e4ab742008-02-23 15:23:55 -08002751 spin_lock_init(&futex_queues[i].lock);
2752 }
2753
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 return 0;
2755}
Benjamin Herrenschmidtf6d107f2008-03-27 14:52:15 +11002756__initcall(futex_init);