Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 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 Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 11 | * 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 Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 15 | * 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 Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 19 | * PRIVATE futexes by Eric Dumazet |
| 20 | * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com> |
| 21 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly |
| 23 | * enough at me, Linus for the original (flawed) idea, Matthew |
| 24 | * Kirkwood for proof-of-concept implementation. |
| 25 | * |
| 26 | * "The futexes are also cursed." |
| 27 | * "But they come in a choice of three flavours!" |
| 28 | * |
| 29 | * This program is free software; you can redistribute it and/or modify |
| 30 | * it under the terms of the GNU General Public License as published by |
| 31 | * the Free Software Foundation; either version 2 of the License, or |
| 32 | * (at your option) any later version. |
| 33 | * |
| 34 | * This program is distributed in the hope that it will be useful, |
| 35 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 36 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 37 | * GNU General Public License for more details. |
| 38 | * |
| 39 | * You should have received a copy of the GNU General Public License |
| 40 | * along with this program; if not, write to the Free Software |
| 41 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| 42 | */ |
| 43 | #include <linux/slab.h> |
| 44 | #include <linux/poll.h> |
| 45 | #include <linux/fs.h> |
| 46 | #include <linux/file.h> |
| 47 | #include <linux/jhash.h> |
| 48 | #include <linux/init.h> |
| 49 | #include <linux/futex.h> |
| 50 | #include <linux/mount.h> |
| 51 | #include <linux/pagemap.h> |
| 52 | #include <linux/syscalls.h> |
Jesper Juhl | 7ed20e1 | 2005-05-01 08:59:14 -0700 | [diff] [blame] | 53 | #include <linux/signal.h> |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 54 | #include <linux/module.h> |
Andrey Mirkin | fd5eea4 | 2007-10-16 23:30:13 -0700 | [diff] [blame] | 55 | #include <linux/magic.h> |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 56 | #include <linux/pid.h> |
| 57 | #include <linux/nsproxy.h> |
| 58 | |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 59 | #include <asm/futex.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 60 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 61 | #include "rtmutex_common.h" |
| 62 | |
Thomas Gleixner | a0c1e90 | 2008-02-23 15:23:57 -0800 | [diff] [blame] | 63 | int __read_mostly futex_cmpxchg_enabled; |
| 64 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 65 | #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8) |
| 66 | |
| 67 | /* |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 68 | * Priority Inheritance state: |
| 69 | */ |
| 70 | struct futex_pi_state { |
| 71 | /* |
| 72 | * list of 'owned' pi_state instances - these have to be |
| 73 | * cleaned up in do_exit() if the task exits prematurely: |
| 74 | */ |
| 75 | struct list_head list; |
| 76 | |
| 77 | /* |
| 78 | * The PI object: |
| 79 | */ |
| 80 | struct rt_mutex pi_mutex; |
| 81 | |
| 82 | struct task_struct *owner; |
| 83 | atomic_t refcount; |
| 84 | |
| 85 | union futex_key key; |
| 86 | }; |
| 87 | |
| 88 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | * We use this hashed waitqueue instead of a normal wait_queue_t, so |
| 90 | * we can wake only the relevant ones (hashed queues may be shared). |
| 91 | * |
| 92 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 93 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 94 | * The order of wakup is always to make the first condition true, then |
Darren Hart | 73500ac | 2008-12-17 17:29:56 -0800 | [diff] [blame] | 95 | * wake up q->waiter, then make the second condition true. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | */ |
| 97 | struct futex_q { |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 98 | struct plist_node list; |
Darren Hart | 73500ac | 2008-12-17 17:29:56 -0800 | [diff] [blame] | 99 | /* There can only be a single waiter */ |
| 100 | wait_queue_head_t waiter; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 101 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 102 | /* Which hash list lock to use: */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | spinlock_t *lock_ptr; |
| 104 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 105 | /* Key which the futex is hashed on: */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | union futex_key key; |
| 107 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 108 | /* Optional priority inheritance state: */ |
| 109 | struct futex_pi_state *pi_state; |
| 110 | struct task_struct *task; |
Thomas Gleixner | cd68998 | 2008-02-01 17:45:14 +0100 | [diff] [blame] | 111 | |
| 112 | /* Bitset for the optional bitmasked wakeup */ |
| 113 | u32 bitset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | }; |
| 115 | |
| 116 | /* |
Darren Hart | b2d0994 | 2009-03-12 00:55:37 -0700 | [diff] [blame] | 117 | * Hash buckets are shared by all the futex_keys that hash to the same |
| 118 | * location. Each key may have multiple futex_q structures, one for each task |
| 119 | * waiting on a futex. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 120 | */ |
| 121 | struct futex_hash_bucket { |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 122 | spinlock_t lock; |
| 123 | struct plist_head chain; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | }; |
| 125 | |
| 126 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; |
| 127 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | /* |
| 129 | * We hash on the keys returned from get_futex_key (see below). |
| 130 | */ |
| 131 | static struct futex_hash_bucket *hash_futex(union futex_key *key) |
| 132 | { |
| 133 | u32 hash = jhash2((u32*)&key->both.word, |
| 134 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, |
| 135 | key->both.offset); |
| 136 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; |
| 137 | } |
| 138 | |
| 139 | /* |
| 140 | * Return 1 if two futex_keys are equal, 0 otherwise. |
| 141 | */ |
| 142 | static inline int match_futex(union futex_key *key1, union futex_key *key2) |
| 143 | { |
| 144 | return (key1->both.word == key2->both.word |
| 145 | && key1->both.ptr == key2->both.ptr |
| 146 | && key1->both.offset == key2->both.offset); |
| 147 | } |
| 148 | |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 149 | /* |
| 150 | * Take a reference to the resource addressed by a key. |
| 151 | * Can be called while holding spinlocks. |
| 152 | * |
| 153 | */ |
| 154 | static void get_futex_key_refs(union futex_key *key) |
| 155 | { |
| 156 | if (!key->both.ptr) |
| 157 | return; |
| 158 | |
| 159 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { |
| 160 | case FUT_OFF_INODE: |
| 161 | atomic_inc(&key->shared.inode->i_count); |
| 162 | break; |
| 163 | case FUT_OFF_MMSHARED: |
| 164 | atomic_inc(&key->private.mm->mm_count); |
| 165 | break; |
| 166 | } |
| 167 | } |
| 168 | |
| 169 | /* |
| 170 | * Drop a reference to the resource addressed by a key. |
| 171 | * The hash bucket spinlock must not be held. |
| 172 | */ |
| 173 | static void drop_futex_key_refs(union futex_key *key) |
| 174 | { |
Darren Hart | 90621c4 | 2008-12-29 19:43:21 -0800 | [diff] [blame] | 175 | if (!key->both.ptr) { |
| 176 | /* If we're here then we tried to put a key we failed to get */ |
| 177 | WARN_ON_ONCE(1); |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 178 | return; |
Darren Hart | 90621c4 | 2008-12-29 19:43:21 -0800 | [diff] [blame] | 179 | } |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 180 | |
| 181 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { |
| 182 | case FUT_OFF_INODE: |
| 183 | iput(key->shared.inode); |
| 184 | break; |
| 185 | case FUT_OFF_MMSHARED: |
| 186 | mmdrop(key->private.mm); |
| 187 | break; |
| 188 | } |
| 189 | } |
| 190 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 191 | /** |
| 192 | * get_futex_key - Get parameters which are the keys for a futex. |
| 193 | * @uaddr: virtual address of the futex |
Darren Hart | b2d0994 | 2009-03-12 00:55:37 -0700 | [diff] [blame] | 194 | * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 195 | * @key: address where result is stored. |
| 196 | * |
| 197 | * Returns a negative error code or 0 |
| 198 | * The key words are stored in *key on success. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | * |
Josef "Jeff" Sipek | f3a43f3 | 2006-12-08 02:36:43 -0800 | [diff] [blame] | 200 | * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 201 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
| 202 | * We can usually work out the index without swapping in the page. |
| 203 | * |
Darren Hart | b2d0994 | 2009-03-12 00:55:37 -0700 | [diff] [blame] | 204 | * lock_page() might sleep, the caller should not hold a spinlock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | */ |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 206 | static int get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 207 | { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 208 | unsigned long address = (unsigned long)uaddr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 209 | struct mm_struct *mm = current->mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | struct page *page; |
| 211 | int err; |
| 212 | |
| 213 | /* |
| 214 | * The futex address must be "naturally" aligned. |
| 215 | */ |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 216 | key->both.offset = address % PAGE_SIZE; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 217 | if (unlikely((address % sizeof(u32)) != 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 218 | return -EINVAL; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 219 | address -= key->both.offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 220 | |
| 221 | /* |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 222 | * PROCESS_PRIVATE futexes are fast. |
| 223 | * As the mm cannot disappear under us and the 'key' only needs |
| 224 | * virtual address, we dont even have to find the underlying vma. |
| 225 | * Note : We do have to check 'uaddr' is a valid user address, |
| 226 | * but access_ok() should be faster than find_vma() |
| 227 | */ |
| 228 | if (!fshared) { |
| 229 | if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32)))) |
| 230 | return -EFAULT; |
| 231 | key->private.mm = mm; |
| 232 | key->private.address = address; |
Peter Zijlstra | 42569c3 | 2008-09-30 12:33:07 +0200 | [diff] [blame] | 233 | get_futex_key_refs(key); |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 234 | return 0; |
| 235 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 237 | again: |
Peter Zijlstra | 734b05b | 2008-09-26 19:32:22 +0200 | [diff] [blame] | 238 | err = get_user_pages_fast(address, 1, 0, &page); |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 239 | if (err < 0) |
| 240 | return err; |
| 241 | |
| 242 | lock_page(page); |
| 243 | if (!page->mapping) { |
| 244 | unlock_page(page); |
| 245 | put_page(page); |
| 246 | goto again; |
| 247 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 248 | |
| 249 | /* |
| 250 | * Private mappings are handled in a simple way. |
| 251 | * |
| 252 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if |
| 253 | * it's a read-only handle, it's expected that futexes attach to |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 254 | * the object not the particular process. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 255 | */ |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 256 | if (PageAnon(page)) { |
| 257 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 258 | key->private.mm = mm; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 259 | key->private.address = address; |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 260 | } else { |
| 261 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ |
| 262 | key->shared.inode = page->mapping->host; |
| 263 | key->shared.pgoff = page->index; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 264 | } |
| 265 | |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 266 | get_futex_key_refs(key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 267 | |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 268 | unlock_page(page); |
| 269 | put_page(page); |
| 270 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 271 | } |
| 272 | |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 273 | static inline |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 274 | void put_futex_key(int fshared, union futex_key *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | { |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 276 | drop_futex_key_refs(key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 277 | } |
| 278 | |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 279 | static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval) |
| 280 | { |
| 281 | u32 curval; |
| 282 | |
| 283 | pagefault_disable(); |
| 284 | curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval); |
| 285 | pagefault_enable(); |
| 286 | |
| 287 | return curval; |
| 288 | } |
| 289 | |
| 290 | static int get_futex_value_locked(u32 *dest, u32 __user *from) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 291 | { |
| 292 | int ret; |
| 293 | |
Peter Zijlstra | a866374 | 2006-12-06 20:32:20 -0800 | [diff] [blame] | 294 | pagefault_disable(); |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 295 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
Peter Zijlstra | a866374 | 2006-12-06 20:32:20 -0800 | [diff] [blame] | 296 | pagefault_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 297 | |
| 298 | return ret ? -EFAULT : 0; |
| 299 | } |
| 300 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 301 | |
| 302 | /* |
| 303 | * PI code: |
| 304 | */ |
| 305 | static int refill_pi_state_cache(void) |
| 306 | { |
| 307 | struct futex_pi_state *pi_state; |
| 308 | |
| 309 | if (likely(current->pi_state_cache)) |
| 310 | return 0; |
| 311 | |
Burman Yan | 4668edc | 2006-12-06 20:38:51 -0800 | [diff] [blame] | 312 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 313 | |
| 314 | if (!pi_state) |
| 315 | return -ENOMEM; |
| 316 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 317 | INIT_LIST_HEAD(&pi_state->list); |
| 318 | /* pi_mutex gets initialized later */ |
| 319 | pi_state->owner = NULL; |
| 320 | atomic_set(&pi_state->refcount, 1); |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 321 | pi_state->key = FUTEX_KEY_INIT; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 322 | |
| 323 | current->pi_state_cache = pi_state; |
| 324 | |
| 325 | return 0; |
| 326 | } |
| 327 | |
| 328 | static struct futex_pi_state * alloc_pi_state(void) |
| 329 | { |
| 330 | struct futex_pi_state *pi_state = current->pi_state_cache; |
| 331 | |
| 332 | WARN_ON(!pi_state); |
| 333 | current->pi_state_cache = NULL; |
| 334 | |
| 335 | return pi_state; |
| 336 | } |
| 337 | |
| 338 | static void free_pi_state(struct futex_pi_state *pi_state) |
| 339 | { |
| 340 | if (!atomic_dec_and_test(&pi_state->refcount)) |
| 341 | return; |
| 342 | |
| 343 | /* |
| 344 | * If pi_state->owner is NULL, the owner is most probably dying |
| 345 | * and has cleaned up the pi_state already |
| 346 | */ |
| 347 | if (pi_state->owner) { |
| 348 | spin_lock_irq(&pi_state->owner->pi_lock); |
| 349 | list_del_init(&pi_state->list); |
| 350 | spin_unlock_irq(&pi_state->owner->pi_lock); |
| 351 | |
| 352 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); |
| 353 | } |
| 354 | |
| 355 | if (current->pi_state_cache) |
| 356 | kfree(pi_state); |
| 357 | else { |
| 358 | /* |
| 359 | * pi_state->list is already empty. |
| 360 | * clear pi_state->owner. |
| 361 | * refcount is at 0 - put it back to 1. |
| 362 | */ |
| 363 | pi_state->owner = NULL; |
| 364 | atomic_set(&pi_state->refcount, 1); |
| 365 | current->pi_state_cache = pi_state; |
| 366 | } |
| 367 | } |
| 368 | |
| 369 | /* |
| 370 | * Look up the task based on what TID userspace gave us. |
| 371 | * We dont trust it. |
| 372 | */ |
| 373 | static struct task_struct * futex_find_get_task(pid_t pid) |
| 374 | { |
| 375 | struct task_struct *p; |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 376 | const struct cred *cred = current_cred(), *pcred; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 377 | |
Oleg Nesterov | d359b54 | 2006-09-29 02:00:55 -0700 | [diff] [blame] | 378 | rcu_read_lock(); |
Pavel Emelyanov | 228ebcb | 2007-10-18 23:40:16 -0700 | [diff] [blame] | 379 | p = find_task_by_vpid(pid); |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 380 | if (!p) { |
Thomas Gleixner | a06381f | 2007-06-23 11:48:40 +0200 | [diff] [blame] | 381 | p = ERR_PTR(-ESRCH); |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 382 | } else { |
| 383 | pcred = __task_cred(p); |
| 384 | if (cred->euid != pcred->euid && |
| 385 | cred->euid != pcred->uid) |
| 386 | p = ERR_PTR(-ESRCH); |
| 387 | else |
| 388 | get_task_struct(p); |
| 389 | } |
Thomas Gleixner | a06381f | 2007-06-23 11:48:40 +0200 | [diff] [blame] | 390 | |
Oleg Nesterov | d359b54 | 2006-09-29 02:00:55 -0700 | [diff] [blame] | 391 | rcu_read_unlock(); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 392 | |
| 393 | return p; |
| 394 | } |
| 395 | |
| 396 | /* |
| 397 | * This task is holding PI mutexes at exit time => bad. |
| 398 | * Kernel cleans up PI-state, but userspace is likely hosed. |
| 399 | * (Robust-futex cleanup is separate and might save the day for userspace.) |
| 400 | */ |
| 401 | void exit_pi_state_list(struct task_struct *curr) |
| 402 | { |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 403 | struct list_head *next, *head = &curr->pi_state_list; |
| 404 | struct futex_pi_state *pi_state; |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 405 | struct futex_hash_bucket *hb; |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 406 | union futex_key key = FUTEX_KEY_INIT; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 407 | |
Thomas Gleixner | a0c1e90 | 2008-02-23 15:23:57 -0800 | [diff] [blame] | 408 | if (!futex_cmpxchg_enabled) |
| 409 | return; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 410 | /* |
| 411 | * We are a ZOMBIE and nobody can enqueue itself on |
| 412 | * pi_state_list anymore, but we have to be careful |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 413 | * versus waiters unqueueing themselves: |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 414 | */ |
| 415 | spin_lock_irq(&curr->pi_lock); |
| 416 | while (!list_empty(head)) { |
| 417 | |
| 418 | next = head->next; |
| 419 | pi_state = list_entry(next, struct futex_pi_state, list); |
| 420 | key = pi_state->key; |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 421 | hb = hash_futex(&key); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 422 | spin_unlock_irq(&curr->pi_lock); |
| 423 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 424 | spin_lock(&hb->lock); |
| 425 | |
| 426 | spin_lock_irq(&curr->pi_lock); |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 427 | /* |
| 428 | * We dropped the pi-lock, so re-check whether this |
| 429 | * task still owns the PI-state: |
| 430 | */ |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 431 | if (head->next != next) { |
| 432 | spin_unlock(&hb->lock); |
| 433 | continue; |
| 434 | } |
| 435 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 436 | WARN_ON(pi_state->owner != curr); |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 437 | WARN_ON(list_empty(&pi_state->list)); |
| 438 | list_del_init(&pi_state->list); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 439 | pi_state->owner = NULL; |
| 440 | spin_unlock_irq(&curr->pi_lock); |
| 441 | |
| 442 | rt_mutex_unlock(&pi_state->pi_mutex); |
| 443 | |
| 444 | spin_unlock(&hb->lock); |
| 445 | |
| 446 | spin_lock_irq(&curr->pi_lock); |
| 447 | } |
| 448 | spin_unlock_irq(&curr->pi_lock); |
| 449 | } |
| 450 | |
| 451 | static int |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 452 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
| 453 | union futex_key *key, struct futex_pi_state **ps) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 454 | { |
| 455 | struct futex_pi_state *pi_state = NULL; |
| 456 | struct futex_q *this, *next; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 457 | struct plist_head *head; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 458 | struct task_struct *p; |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 459 | pid_t pid = uval & FUTEX_TID_MASK; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 460 | |
| 461 | head = &hb->chain; |
| 462 | |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 463 | plist_for_each_entry_safe(this, next, head, list) { |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 464 | if (match_futex(&this->key, key)) { |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 465 | /* |
| 466 | * Another waiter already exists - bump up |
| 467 | * the refcount and return its pi_state: |
| 468 | */ |
| 469 | pi_state = this->pi_state; |
Thomas Gleixner | 06a9ec2 | 2006-07-10 04:44:30 -0700 | [diff] [blame] | 470 | /* |
| 471 | * Userspace might have messed up non PI and PI futexes |
| 472 | */ |
| 473 | if (unlikely(!pi_state)) |
| 474 | return -EINVAL; |
| 475 | |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 476 | WARN_ON(!atomic_read(&pi_state->refcount)); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 477 | WARN_ON(pid && pi_state->owner && |
| 478 | pi_state->owner->pid != pid); |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 479 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 480 | atomic_inc(&pi_state->refcount); |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 481 | *ps = pi_state; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 482 | |
| 483 | return 0; |
| 484 | } |
| 485 | } |
| 486 | |
| 487 | /* |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 488 | * We are the first waiter - try to look up the real owner and attach |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 489 | * the new pi_state to it, but bail out when TID = 0 |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 490 | */ |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 491 | if (!pid) |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 492 | return -ESRCH; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 493 | p = futex_find_get_task(pid); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 494 | if (IS_ERR(p)) |
| 495 | return PTR_ERR(p); |
| 496 | |
| 497 | /* |
| 498 | * We need to look at the task state flags to figure out, |
| 499 | * whether the task is exiting. To protect against the do_exit |
| 500 | * change of the task flags, we do this protected by |
| 501 | * p->pi_lock: |
| 502 | */ |
| 503 | spin_lock_irq(&p->pi_lock); |
| 504 | if (unlikely(p->flags & PF_EXITING)) { |
| 505 | /* |
| 506 | * The task is on the way out. When PF_EXITPIDONE is |
| 507 | * set, we know that the task has finished the |
| 508 | * cleanup: |
| 509 | */ |
| 510 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; |
| 511 | |
| 512 | spin_unlock_irq(&p->pi_lock); |
| 513 | put_task_struct(p); |
| 514 | return ret; |
| 515 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 516 | |
| 517 | pi_state = alloc_pi_state(); |
| 518 | |
| 519 | /* |
| 520 | * Initialize the pi_mutex in locked state and make 'p' |
| 521 | * the owner of it: |
| 522 | */ |
| 523 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); |
| 524 | |
| 525 | /* Store the key for possible exit cleanups: */ |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 526 | pi_state->key = *key; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 527 | |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 528 | WARN_ON(!list_empty(&pi_state->list)); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 529 | list_add(&pi_state->list, &p->pi_state_list); |
| 530 | pi_state->owner = p; |
| 531 | spin_unlock_irq(&p->pi_lock); |
| 532 | |
| 533 | put_task_struct(p); |
| 534 | |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 535 | *ps = pi_state; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 536 | |
| 537 | return 0; |
| 538 | } |
| 539 | |
| 540 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 541 | * The hash bucket lock must be held when this is called. |
| 542 | * Afterwards, the futex_q must not be accessed. |
| 543 | */ |
| 544 | static void wake_futex(struct futex_q *q) |
| 545 | { |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 546 | plist_del(&q->list, &q->list.plist); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 547 | /* |
| 548 | * The lock in wake_up_all() is a crucial memory barrier after the |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 549 | * plist_del() and also before assigning to q->lock_ptr. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | */ |
Darren Hart | 73500ac | 2008-12-17 17:29:56 -0800 | [diff] [blame] | 551 | wake_up(&q->waiter); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 552 | /* |
| 553 | * The waiting task can free the futex_q as soon as this is written, |
| 554 | * without taking any locks. This must come last. |
Andrew Morton | 8e31108 | 2005-12-23 19:54:46 -0800 | [diff] [blame] | 555 | * |
Darren Hart | b2d0994 | 2009-03-12 00:55:37 -0700 | [diff] [blame] | 556 | * A memory barrier is required here to prevent the following store to |
| 557 | * lock_ptr from getting ahead of the wakeup. Clearing the lock at the |
| 558 | * end of wake_up() does not prevent this store from moving. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 559 | */ |
Ralf Baechle | ccdea2f | 2006-12-06 20:40:26 -0800 | [diff] [blame] | 560 | smp_wmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 561 | q->lock_ptr = NULL; |
| 562 | } |
| 563 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 564 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
| 565 | { |
| 566 | struct task_struct *new_owner; |
| 567 | struct futex_pi_state *pi_state = this->pi_state; |
| 568 | u32 curval, newval; |
| 569 | |
| 570 | if (!pi_state) |
| 571 | return -EINVAL; |
| 572 | |
Ingo Molnar | 21778867 | 2007-03-16 13:38:31 -0800 | [diff] [blame] | 573 | spin_lock(&pi_state->pi_mutex.wait_lock); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 574 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
| 575 | |
| 576 | /* |
| 577 | * This happens when we have stolen the lock and the original |
| 578 | * pending owner did not enqueue itself back on the rt_mutex. |
| 579 | * Thats not a tragedy. We know that way, that a lock waiter |
| 580 | * is on the fly. We make the futex_q waiter the pending owner. |
| 581 | */ |
| 582 | if (!new_owner) |
| 583 | new_owner = this->task; |
| 584 | |
| 585 | /* |
| 586 | * We pass it to the next owner. (The WAITERS bit is always |
| 587 | * kept enabled while there is PI state around. We must also |
| 588 | * preserve the owner died bit.) |
| 589 | */ |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 590 | if (!(uval & FUTEX_OWNER_DIED)) { |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 591 | int ret = 0; |
| 592 | |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 593 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 594 | |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 595 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 596 | |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 597 | if (curval == -EFAULT) |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 598 | ret = -EFAULT; |
Thomas Gleixner | cde898f | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 599 | else if (curval != uval) |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 600 | ret = -EINVAL; |
| 601 | if (ret) { |
| 602 | spin_unlock(&pi_state->pi_mutex.wait_lock); |
| 603 | return ret; |
| 604 | } |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 605 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 606 | |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 607 | spin_lock_irq(&pi_state->owner->pi_lock); |
| 608 | WARN_ON(list_empty(&pi_state->list)); |
| 609 | list_del_init(&pi_state->list); |
| 610 | spin_unlock_irq(&pi_state->owner->pi_lock); |
| 611 | |
| 612 | spin_lock_irq(&new_owner->pi_lock); |
| 613 | WARN_ON(!list_empty(&pi_state->list)); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 614 | list_add(&pi_state->list, &new_owner->pi_state_list); |
| 615 | pi_state->owner = new_owner; |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 616 | spin_unlock_irq(&new_owner->pi_lock); |
| 617 | |
Ingo Molnar | 21778867 | 2007-03-16 13:38:31 -0800 | [diff] [blame] | 618 | spin_unlock(&pi_state->pi_mutex.wait_lock); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 619 | rt_mutex_unlock(&pi_state->pi_mutex); |
| 620 | |
| 621 | return 0; |
| 622 | } |
| 623 | |
| 624 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) |
| 625 | { |
| 626 | u32 oldval; |
| 627 | |
| 628 | /* |
| 629 | * There is no waiter, so we unlock the futex. The owner died |
| 630 | * bit has not to be preserved here. We are the owner: |
| 631 | */ |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 632 | oldval = cmpxchg_futex_value_locked(uaddr, uval, 0); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 633 | |
| 634 | if (oldval == -EFAULT) |
| 635 | return oldval; |
| 636 | if (oldval != uval) |
| 637 | return -EAGAIN; |
| 638 | |
| 639 | return 0; |
| 640 | } |
| 641 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 642 | /* |
Ingo Molnar | 8b8f319 | 2006-07-03 00:25:05 -0700 | [diff] [blame] | 643 | * Express the locking dependencies for lockdep: |
| 644 | */ |
| 645 | static inline void |
| 646 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) |
| 647 | { |
| 648 | if (hb1 <= hb2) { |
| 649 | spin_lock(&hb1->lock); |
| 650 | if (hb1 < hb2) |
| 651 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); |
| 652 | } else { /* hb1 > hb2 */ |
| 653 | spin_lock(&hb2->lock); |
| 654 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); |
| 655 | } |
| 656 | } |
| 657 | |
Darren Hart | 5eb3dc6 | 2009-03-12 00:55:52 -0700 | [diff] [blame] | 658 | static inline void |
| 659 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) |
| 660 | { |
| 661 | if (hb1 <= hb2) { |
| 662 | spin_unlock(&hb2->lock); |
| 663 | if (hb1 < hb2) |
| 664 | spin_unlock(&hb1->lock); |
| 665 | } else { /* hb1 > hb2 */ |
| 666 | spin_unlock(&hb1->lock); |
| 667 | spin_unlock(&hb2->lock); |
| 668 | } |
| 669 | } |
| 670 | |
Ingo Molnar | 8b8f319 | 2006-07-03 00:25:05 -0700 | [diff] [blame] | 671 | /* |
Darren Hart | b2d0994 | 2009-03-12 00:55:37 -0700 | [diff] [blame] | 672 | * Wake up waiters matching bitset queued on this futex (uaddr). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | */ |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 674 | static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 675 | { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 676 | struct futex_hash_bucket *hb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 677 | struct futex_q *this, *next; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 678 | struct plist_head *head; |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 679 | union futex_key key = FUTEX_KEY_INIT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 680 | int ret; |
| 681 | |
Thomas Gleixner | cd68998 | 2008-02-01 17:45:14 +0100 | [diff] [blame] | 682 | if (!bitset) |
| 683 | return -EINVAL; |
| 684 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 685 | ret = get_futex_key(uaddr, fshared, &key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 686 | if (unlikely(ret != 0)) |
| 687 | goto out; |
| 688 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 689 | hb = hash_futex(&key); |
| 690 | spin_lock(&hb->lock); |
| 691 | head = &hb->chain; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 692 | |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 693 | plist_for_each_entry_safe(this, next, head, list) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 694 | if (match_futex (&this->key, &key)) { |
Ingo Molnar | ed6f7b1 | 2006-07-01 04:35:46 -0700 | [diff] [blame] | 695 | if (this->pi_state) { |
| 696 | ret = -EINVAL; |
| 697 | break; |
| 698 | } |
Thomas Gleixner | cd68998 | 2008-02-01 17:45:14 +0100 | [diff] [blame] | 699 | |
| 700 | /* Check if one of the bits is set in both bitsets */ |
| 701 | if (!(this->bitset & bitset)) |
| 702 | continue; |
| 703 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 704 | wake_futex(this); |
| 705 | if (++ret >= nr_wake) |
| 706 | break; |
| 707 | } |
| 708 | } |
| 709 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 710 | spin_unlock(&hb->lock); |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 711 | put_futex_key(fshared, &key); |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 712 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 713 | return ret; |
| 714 | } |
| 715 | |
| 716 | /* |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 717 | * Wake up all waiters hashed on the physical page that is mapped |
| 718 | * to this virtual address: |
| 719 | */ |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 720 | static int |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 721 | futex_wake_op(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 722 | int nr_wake, int nr_wake2, int op) |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 723 | { |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 724 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 725 | struct futex_hash_bucket *hb1, *hb2; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 726 | struct plist_head *head; |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 727 | struct futex_q *this, *next; |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 728 | int ret, op_ret; |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 729 | |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 730 | retry: |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 731 | ret = get_futex_key(uaddr1, fshared, &key1); |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 732 | if (unlikely(ret != 0)) |
| 733 | goto out; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 734 | ret = get_futex_key(uaddr2, fshared, &key2); |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 735 | if (unlikely(ret != 0)) |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 736 | goto out_put_key1; |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 737 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 738 | hb1 = hash_futex(&key1); |
| 739 | hb2 = hash_futex(&key2); |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 740 | |
Ingo Molnar | 8b8f319 | 2006-07-03 00:25:05 -0700 | [diff] [blame] | 741 | double_lock_hb(hb1, hb2); |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 742 | retry_private: |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 743 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 744 | if (unlikely(op_ret < 0)) { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 745 | u32 dummy; |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 746 | |
Darren Hart | 5eb3dc6 | 2009-03-12 00:55:52 -0700 | [diff] [blame] | 747 | double_unlock_hb(hb1, hb2); |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 748 | |
David Howells | 7ee1dd3 | 2006-01-06 00:11:44 -0800 | [diff] [blame] | 749 | #ifndef CONFIG_MMU |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 750 | /* |
| 751 | * we don't get EFAULT from MMU faults if we don't have an MMU, |
| 752 | * but we might get them from range checking |
| 753 | */ |
David Howells | 7ee1dd3 | 2006-01-06 00:11:44 -0800 | [diff] [blame] | 754 | ret = op_ret; |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 755 | goto out_put_keys; |
David Howells | 7ee1dd3 | 2006-01-06 00:11:44 -0800 | [diff] [blame] | 756 | #endif |
| 757 | |
David Gibson | 796f8d9 | 2005-11-07 00:59:33 -0800 | [diff] [blame] | 758 | if (unlikely(op_ret != -EFAULT)) { |
| 759 | ret = op_ret; |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 760 | goto out_put_keys; |
David Gibson | 796f8d9 | 2005-11-07 00:59:33 -0800 | [diff] [blame] | 761 | } |
| 762 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 763 | ret = get_user(dummy, uaddr2); |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 764 | if (ret) |
Darren Hart | de87fcc | 2009-03-12 00:55:46 -0700 | [diff] [blame] | 765 | goto out_put_keys; |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 766 | |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 767 | if (!fshared) |
| 768 | goto retry_private; |
| 769 | |
Darren Hart | de87fcc | 2009-03-12 00:55:46 -0700 | [diff] [blame] | 770 | put_futex_key(fshared, &key2); |
| 771 | put_futex_key(fshared, &key1); |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 772 | goto retry; |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 773 | } |
| 774 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 775 | head = &hb1->chain; |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 776 | |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 777 | plist_for_each_entry_safe(this, next, head, list) { |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 778 | if (match_futex (&this->key, &key1)) { |
| 779 | wake_futex(this); |
| 780 | if (++ret >= nr_wake) |
| 781 | break; |
| 782 | } |
| 783 | } |
| 784 | |
| 785 | if (op_ret > 0) { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 786 | head = &hb2->chain; |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 787 | |
| 788 | op_ret = 0; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 789 | plist_for_each_entry_safe(this, next, head, list) { |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 790 | if (match_futex (&this->key, &key2)) { |
| 791 | wake_futex(this); |
| 792 | if (++op_ret >= nr_wake2) |
| 793 | break; |
| 794 | } |
| 795 | } |
| 796 | ret += op_ret; |
| 797 | } |
| 798 | |
Darren Hart | 5eb3dc6 | 2009-03-12 00:55:52 -0700 | [diff] [blame] | 799 | double_unlock_hb(hb1, hb2); |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 800 | out_put_keys: |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 801 | put_futex_key(fshared, &key2); |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 802 | out_put_key1: |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 803 | put_futex_key(fshared, &key1); |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 804 | out: |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 805 | return ret; |
| 806 | } |
| 807 | |
| 808 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 809 | * Requeue all waiters hashed on one physical page to another |
| 810 | * physical page. |
| 811 | */ |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 812 | static int futex_requeue(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 813 | int nr_wake, int nr_requeue, u32 *cmpval) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 814 | { |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 815 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 816 | struct futex_hash_bucket *hb1, *hb2; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 817 | struct plist_head *head1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 818 | struct futex_q *this, *next; |
| 819 | int ret, drop_count = 0; |
| 820 | |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 821 | retry: |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 822 | ret = get_futex_key(uaddr1, fshared, &key1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 823 | if (unlikely(ret != 0)) |
| 824 | goto out; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 825 | ret = get_futex_key(uaddr2, fshared, &key2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 826 | if (unlikely(ret != 0)) |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 827 | goto out_put_key1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 828 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 829 | hb1 = hash_futex(&key1); |
| 830 | hb2 = hash_futex(&key2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 831 | |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 832 | retry_private: |
Ingo Molnar | 8b8f319 | 2006-07-03 00:25:05 -0700 | [diff] [blame] | 833 | double_lock_hb(hb1, hb2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 834 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 835 | if (likely(cmpval != NULL)) { |
| 836 | u32 curval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 837 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 838 | ret = get_futex_value_locked(&curval, uaddr1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 839 | |
| 840 | if (unlikely(ret)) { |
Darren Hart | 5eb3dc6 | 2009-03-12 00:55:52 -0700 | [diff] [blame] | 841 | double_unlock_hb(hb1, hb2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 842 | |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 843 | ret = get_user(curval, uaddr1); |
| 844 | if (ret) |
| 845 | goto out_put_keys; |
| 846 | |
| 847 | if (!fshared) |
| 848 | goto retry_private; |
| 849 | |
Darren Hart | de87fcc | 2009-03-12 00:55:46 -0700 | [diff] [blame] | 850 | put_futex_key(fshared, &key2); |
| 851 | put_futex_key(fshared, &key1); |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 852 | goto retry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 853 | } |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 854 | if (curval != *cmpval) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 855 | ret = -EAGAIN; |
| 856 | goto out_unlock; |
| 857 | } |
| 858 | } |
| 859 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 860 | head1 = &hb1->chain; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 861 | plist_for_each_entry_safe(this, next, head1, list) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 862 | if (!match_futex (&this->key, &key1)) |
| 863 | continue; |
| 864 | if (++ret <= nr_wake) { |
| 865 | wake_futex(this); |
| 866 | } else { |
Sebastien Dugue | 59e0e0a | 2006-06-27 02:55:03 -0700 | [diff] [blame] | 867 | /* |
| 868 | * If key1 and key2 hash to the same bucket, no need to |
| 869 | * requeue. |
| 870 | */ |
| 871 | if (likely(head1 != &hb2->chain)) { |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 872 | plist_del(&this->list, &hb1->chain); |
| 873 | plist_add(&this->list, &hb2->chain); |
Sebastien Dugue | 59e0e0a | 2006-06-27 02:55:03 -0700 | [diff] [blame] | 874 | this->lock_ptr = &hb2->lock; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 875 | #ifdef CONFIG_DEBUG_PI_LIST |
| 876 | this->list.plist.lock = &hb2->lock; |
| 877 | #endif |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 878 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 879 | this->key = key2; |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 880 | get_futex_key_refs(&key2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 881 | drop_count++; |
| 882 | |
| 883 | if (ret - nr_wake >= nr_requeue) |
| 884 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 885 | } |
| 886 | } |
| 887 | |
| 888 | out_unlock: |
Darren Hart | 5eb3dc6 | 2009-03-12 00:55:52 -0700 | [diff] [blame] | 889 | double_unlock_hb(hb1, hb2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 890 | |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 891 | /* drop_futex_key_refs() must be called outside the spinlocks. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 892 | while (--drop_count >= 0) |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 893 | drop_futex_key_refs(&key1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 894 | |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 895 | out_put_keys: |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 896 | put_futex_key(fshared, &key2); |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 897 | out_put_key1: |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 898 | put_futex_key(fshared, &key1); |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 899 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 900 | return ret; |
| 901 | } |
| 902 | |
| 903 | /* The key must be already stored in q->key. */ |
Eric Sesterhenn | 82af7ac | 2008-01-25 10:40:46 +0100 | [diff] [blame] | 904 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 905 | { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 906 | struct futex_hash_bucket *hb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 907 | |
Darren Hart | 73500ac | 2008-12-17 17:29:56 -0800 | [diff] [blame] | 908 | init_waitqueue_head(&q->waiter); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 909 | |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 910 | get_futex_key_refs(&q->key); |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 911 | hb = hash_futex(&q->key); |
| 912 | q->lock_ptr = &hb->lock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 913 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 914 | spin_lock(&hb->lock); |
| 915 | return hb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 916 | } |
| 917 | |
Eric Sesterhenn | 82af7ac | 2008-01-25 10:40:46 +0100 | [diff] [blame] | 918 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 919 | { |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 920 | int prio; |
| 921 | |
| 922 | /* |
| 923 | * The priority used to register this element is |
| 924 | * - either the real thread-priority for the real-time threads |
| 925 | * (i.e. threads with a priority lower than MAX_RT_PRIO) |
| 926 | * - or MAX_RT_PRIO for non-RT threads. |
| 927 | * Thus, all RT-threads are woken first in priority order, and |
| 928 | * the others are woken last, in FIFO order. |
| 929 | */ |
| 930 | prio = min(current->normal_prio, MAX_RT_PRIO); |
| 931 | |
| 932 | plist_node_init(&q->list, prio); |
| 933 | #ifdef CONFIG_DEBUG_PI_LIST |
| 934 | q->list.plist.lock = &hb->lock; |
| 935 | #endif |
| 936 | plist_add(&q->list, &hb->chain); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 937 | q->task = current; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 938 | spin_unlock(&hb->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 939 | } |
| 940 | |
| 941 | static inline void |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 942 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 943 | { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 944 | spin_unlock(&hb->lock); |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 945 | drop_futex_key_refs(&q->key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 946 | } |
| 947 | |
| 948 | /* |
| 949 | * queue_me and unqueue_me must be called as a pair, each |
| 950 | * exactly once. They are called with the hashed spinlock held. |
| 951 | */ |
| 952 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 953 | /* Return 1 if we were still queued (ie. 0 means we were woken) */ |
| 954 | static int unqueue_me(struct futex_q *q) |
| 955 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 956 | spinlock_t *lock_ptr; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 957 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 958 | |
| 959 | /* In the common case we don't take the spinlock, which is nice. */ |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 960 | retry: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 961 | lock_ptr = q->lock_ptr; |
Christian Borntraeger | e91467e | 2006-08-05 12:13:52 -0700 | [diff] [blame] | 962 | barrier(); |
Stephen Hemminger | c80544d | 2007-10-18 03:07:05 -0700 | [diff] [blame] | 963 | if (lock_ptr != NULL) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 964 | spin_lock(lock_ptr); |
| 965 | /* |
| 966 | * q->lock_ptr can change between reading it and |
| 967 | * spin_lock(), causing us to take the wrong lock. This |
| 968 | * corrects the race condition. |
| 969 | * |
| 970 | * Reasoning goes like this: if we have the wrong lock, |
| 971 | * q->lock_ptr must have changed (maybe several times) |
| 972 | * between reading it and the spin_lock(). It can |
| 973 | * change again after the spin_lock() but only if it was |
| 974 | * already changed before the spin_lock(). It cannot, |
| 975 | * however, change back to the original value. Therefore |
| 976 | * we can detect whether we acquired the correct lock. |
| 977 | */ |
| 978 | if (unlikely(lock_ptr != q->lock_ptr)) { |
| 979 | spin_unlock(lock_ptr); |
| 980 | goto retry; |
| 981 | } |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 982 | WARN_ON(plist_node_empty(&q->list)); |
| 983 | plist_del(&q->list, &q->list.plist); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 984 | |
| 985 | BUG_ON(q->pi_state); |
| 986 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 987 | spin_unlock(lock_ptr); |
| 988 | ret = 1; |
| 989 | } |
| 990 | |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 991 | drop_futex_key_refs(&q->key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 992 | return ret; |
| 993 | } |
| 994 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 995 | /* |
| 996 | * PI futexes can not be requeued and must remove themself from the |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 997 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
| 998 | * and dropped here. |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 999 | */ |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1000 | static void unqueue_me_pi(struct futex_q *q) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1001 | { |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 1002 | WARN_ON(plist_node_empty(&q->list)); |
| 1003 | plist_del(&q->list, &q->list.plist); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1004 | |
| 1005 | BUG_ON(!q->pi_state); |
| 1006 | free_pi_state(q->pi_state); |
| 1007 | q->pi_state = NULL; |
| 1008 | |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1009 | spin_unlock(q->lock_ptr); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1010 | |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 1011 | drop_futex_key_refs(&q->key); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1012 | } |
| 1013 | |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1014 | /* |
Thomas Gleixner | cdf71a1 | 2008-01-08 19:47:38 +0100 | [diff] [blame] | 1015 | * Fixup the pi_state owner with the new owner. |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1016 | * |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1017 | * Must be called with hash bucket lock held and mm->sem held for non |
| 1018 | * private futexes. |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1019 | */ |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1020 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 1021 | struct task_struct *newowner, int fshared) |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1022 | { |
Thomas Gleixner | cdf71a1 | 2008-01-08 19:47:38 +0100 | [diff] [blame] | 1023 | u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1024 | struct futex_pi_state *pi_state = q->pi_state; |
Thomas Gleixner | 1b7558e | 2008-06-23 11:21:58 +0200 | [diff] [blame] | 1025 | struct task_struct *oldowner = pi_state->owner; |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1026 | u32 uval, curval, newval; |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 1027 | int ret; |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1028 | |
| 1029 | /* Owner died? */ |
Thomas Gleixner | 1b7558e | 2008-06-23 11:21:58 +0200 | [diff] [blame] | 1030 | if (!pi_state->owner) |
| 1031 | newtid |= FUTEX_OWNER_DIED; |
| 1032 | |
| 1033 | /* |
| 1034 | * We are here either because we stole the rtmutex from the |
| 1035 | * pending owner or we are the pending owner which failed to |
| 1036 | * get the rtmutex. We have to replace the pending owner TID |
| 1037 | * in the user space variable. This must be atomic as we have |
| 1038 | * to preserve the owner died bit here. |
| 1039 | * |
Darren Hart | b2d0994 | 2009-03-12 00:55:37 -0700 | [diff] [blame] | 1040 | * Note: We write the user space value _before_ changing the pi_state |
| 1041 | * because we can fault here. Imagine swapped out pages or a fork |
| 1042 | * that marked all the anonymous memory readonly for cow. |
Thomas Gleixner | 1b7558e | 2008-06-23 11:21:58 +0200 | [diff] [blame] | 1043 | * |
| 1044 | * Modifying pi_state _before_ the user space value would |
| 1045 | * leave the pi_state in an inconsistent state when we fault |
| 1046 | * here, because we need to drop the hash bucket lock to |
| 1047 | * handle the fault. This might be observed in the PID check |
| 1048 | * in lookup_pi_state. |
| 1049 | */ |
| 1050 | retry: |
| 1051 | if (get_futex_value_locked(&uval, uaddr)) |
| 1052 | goto handle_fault; |
| 1053 | |
| 1054 | while (1) { |
| 1055 | newval = (uval & FUTEX_OWNER_DIED) | newtid; |
| 1056 | |
| 1057 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
| 1058 | |
| 1059 | if (curval == -EFAULT) |
| 1060 | goto handle_fault; |
| 1061 | if (curval == uval) |
| 1062 | break; |
| 1063 | uval = curval; |
| 1064 | } |
| 1065 | |
| 1066 | /* |
| 1067 | * We fixed up user space. Now we need to fix the pi_state |
| 1068 | * itself. |
| 1069 | */ |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1070 | if (pi_state->owner != NULL) { |
| 1071 | spin_lock_irq(&pi_state->owner->pi_lock); |
| 1072 | WARN_ON(list_empty(&pi_state->list)); |
| 1073 | list_del_init(&pi_state->list); |
| 1074 | spin_unlock_irq(&pi_state->owner->pi_lock); |
Thomas Gleixner | 1b7558e | 2008-06-23 11:21:58 +0200 | [diff] [blame] | 1075 | } |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1076 | |
Thomas Gleixner | cdf71a1 | 2008-01-08 19:47:38 +0100 | [diff] [blame] | 1077 | pi_state->owner = newowner; |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1078 | |
Thomas Gleixner | cdf71a1 | 2008-01-08 19:47:38 +0100 | [diff] [blame] | 1079 | spin_lock_irq(&newowner->pi_lock); |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1080 | WARN_ON(!list_empty(&pi_state->list)); |
Thomas Gleixner | cdf71a1 | 2008-01-08 19:47:38 +0100 | [diff] [blame] | 1081 | list_add(&pi_state->list, &newowner->pi_state_list); |
| 1082 | spin_unlock_irq(&newowner->pi_lock); |
Thomas Gleixner | 1b7558e | 2008-06-23 11:21:58 +0200 | [diff] [blame] | 1083 | return 0; |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1084 | |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1085 | /* |
Thomas Gleixner | 1b7558e | 2008-06-23 11:21:58 +0200 | [diff] [blame] | 1086 | * To handle the page fault we need to drop the hash bucket |
| 1087 | * lock here. That gives the other task (either the pending |
| 1088 | * owner itself or the task which stole the rtmutex) the |
| 1089 | * chance to try the fixup of the pi_state. So once we are |
| 1090 | * back from handling the fault we need to check the pi_state |
| 1091 | * after reacquiring the hash bucket lock and before trying to |
| 1092 | * do another fixup. When the fixup has been done already we |
| 1093 | * simply return. |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1094 | */ |
Thomas Gleixner | 1b7558e | 2008-06-23 11:21:58 +0200 | [diff] [blame] | 1095 | handle_fault: |
| 1096 | spin_unlock(q->lock_ptr); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1097 | |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 1098 | ret = get_user(uval, uaddr); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1099 | |
Thomas Gleixner | 1b7558e | 2008-06-23 11:21:58 +0200 | [diff] [blame] | 1100 | spin_lock(q->lock_ptr); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1101 | |
Thomas Gleixner | 1b7558e | 2008-06-23 11:21:58 +0200 | [diff] [blame] | 1102 | /* |
| 1103 | * Check if someone else fixed it for us: |
| 1104 | */ |
| 1105 | if (pi_state->owner != oldowner) |
| 1106 | return 0; |
| 1107 | |
| 1108 | if (ret) |
| 1109 | return ret; |
| 1110 | |
| 1111 | goto retry; |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1112 | } |
| 1113 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1114 | /* |
| 1115 | * In case we must use restart_block to restart a futex_wait, |
Steven Rostedt | ce6bd42 | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 1116 | * we encode in the 'flags' shared capability |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1117 | */ |
Thomas Gleixner | 1acdac1 | 2008-11-20 10:02:53 -0800 | [diff] [blame] | 1118 | #define FLAGS_SHARED 0x01 |
| 1119 | #define FLAGS_CLOCKRT 0x02 |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1120 | |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1121 | static long futex_wait_restart(struct restart_block *restart); |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1122 | |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 1123 | static int futex_wait(u32 __user *uaddr, int fshared, |
Thomas Gleixner | 1acdac1 | 2008-11-20 10:02:53 -0800 | [diff] [blame] | 1124 | u32 val, ktime_t *abs_time, u32 bitset, int clockrt) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1125 | { |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1126 | struct task_struct *curr = current; |
Peter Zijlstra | 2fff78c | 2009-02-11 18:10:10 +0100 | [diff] [blame] | 1127 | struct restart_block *restart; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1128 | DECLARE_WAITQUEUE(wait, curr); |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1129 | struct futex_hash_bucket *hb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1130 | struct futex_q q; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1131 | u32 uval; |
| 1132 | int ret; |
Thomas Gleixner | bd19723 | 2007-06-17 21:11:10 +0200 | [diff] [blame] | 1133 | struct hrtimer_sleeper t; |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1134 | int rem = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1135 | |
Thomas Gleixner | cd68998 | 2008-02-01 17:45:14 +0100 | [diff] [blame] | 1136 | if (!bitset) |
| 1137 | return -EINVAL; |
| 1138 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1139 | q.pi_state = NULL; |
Thomas Gleixner | cd68998 | 2008-02-01 17:45:14 +0100 | [diff] [blame] | 1140 | q.bitset = bitset; |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1141 | retry: |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 1142 | q.key = FUTEX_KEY_INIT; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1143 | ret = get_futex_key(uaddr, fshared, &q.key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1144 | if (unlikely(ret != 0)) |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1145 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1146 | |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 1147 | retry_private: |
Eric Sesterhenn | 82af7ac | 2008-01-25 10:40:46 +0100 | [diff] [blame] | 1148 | hb = queue_lock(&q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1149 | |
| 1150 | /* |
Darren Hart | b2d0994 | 2009-03-12 00:55:37 -0700 | [diff] [blame] | 1151 | * Access the page AFTER the hash-bucket is locked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1152 | * Order is important: |
| 1153 | * |
| 1154 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); |
| 1155 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } |
| 1156 | * |
| 1157 | * The basic logical guarantee of a futex is that it blocks ONLY |
| 1158 | * if cond(var) is known to be true at the time of blocking, for |
| 1159 | * any cond. If we queued after testing *uaddr, that would open |
| 1160 | * a race condition where we could block indefinitely with |
| 1161 | * cond(var) false, which would violate the guarantee. |
| 1162 | * |
| 1163 | * A consequence is that futex_wait() can return zero and absorb |
| 1164 | * a wakeup when *uaddr != val on entry to the syscall. This is |
| 1165 | * rare, but normal. |
| 1166 | * |
Darren Hart | b2d0994 | 2009-03-12 00:55:37 -0700 | [diff] [blame] | 1167 | * For shared futexes, we hold the mmap semaphore, so the mapping |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1168 | * cannot have changed since we looked it up in get_futex_key. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1169 | */ |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1170 | ret = get_futex_value_locked(&uval, uaddr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1171 | |
| 1172 | if (unlikely(ret)) { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1173 | queue_unlock(&q, hb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1174 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1175 | ret = get_user(uval, uaddr); |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 1176 | if (ret) |
| 1177 | goto out_put_key; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1178 | |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 1179 | if (!fshared) |
| 1180 | goto retry_private; |
| 1181 | |
| 1182 | put_futex_key(fshared, &q.key); |
| 1183 | goto retry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1184 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1185 | ret = -EWOULDBLOCK; |
Peter Zijlstra | 2fff78c | 2009-02-11 18:10:10 +0100 | [diff] [blame] | 1186 | if (unlikely(uval != val)) { |
| 1187 | queue_unlock(&q, hb); |
| 1188 | goto out_put_key; |
| 1189 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1190 | |
| 1191 | /* Only actually queue if *uaddr contained val. */ |
Eric Sesterhenn | 82af7ac | 2008-01-25 10:40:46 +0100 | [diff] [blame] | 1192 | queue_me(&q, hb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1193 | |
| 1194 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1195 | * There might have been scheduling since the queue_me(), as we |
| 1196 | * cannot hold a spinlock across the get_user() in case it |
| 1197 | * faults, and we cannot just set TASK_INTERRUPTIBLE state when |
| 1198 | * queueing ourselves into the futex hash. This code thus has to |
| 1199 | * rely on the futex_wake() code removing us from hash when it |
| 1200 | * wakes us up. |
| 1201 | */ |
| 1202 | |
| 1203 | /* add_wait_queue is the barrier after __set_current_state. */ |
| 1204 | __set_current_state(TASK_INTERRUPTIBLE); |
Darren Hart | 73500ac | 2008-12-17 17:29:56 -0800 | [diff] [blame] | 1205 | add_wait_queue(&q.waiter, &wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1206 | /* |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 1207 | * !plist_node_empty() is safe here without any lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1208 | * q.lock_ptr != 0 is not safe, because of ordering against wakeup. |
| 1209 | */ |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 1210 | if (likely(!plist_node_empty(&q.list))) { |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1211 | if (!abs_time) |
| 1212 | schedule(); |
| 1213 | else { |
Thomas Gleixner | 1acdac1 | 2008-11-20 10:02:53 -0800 | [diff] [blame] | 1214 | hrtimer_init_on_stack(&t.timer, |
| 1215 | clockrt ? CLOCK_REALTIME : |
| 1216 | CLOCK_MONOTONIC, |
| 1217 | HRTIMER_MODE_ABS); |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1218 | hrtimer_init_sleeper(&t, current); |
Darren Hart | 16f4993 | 2009-03-12 00:55:59 -0700 | [diff] [blame] | 1219 | hrtimer_set_expires_range_ns(&t.timer, *abs_time, |
| 1220 | current->timer_slack_ns); |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1221 | |
Arjan van de Ven | cc584b2 | 2008-09-01 15:02:30 -0700 | [diff] [blame] | 1222 | hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS); |
Peter Zijlstra | 3588a08 | 2008-02-01 17:45:13 +0100 | [diff] [blame] | 1223 | if (!hrtimer_active(&t.timer)) |
| 1224 | t.task = NULL; |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1225 | |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1226 | /* |
| 1227 | * the timer could have already expired, in which |
| 1228 | * case current would be flagged for rescheduling. |
| 1229 | * Don't bother calling schedule. |
| 1230 | */ |
| 1231 | if (likely(t.task)) |
| 1232 | schedule(); |
| 1233 | |
| 1234 | hrtimer_cancel(&t.timer); |
| 1235 | |
| 1236 | /* Flag if a timeout occured */ |
| 1237 | rem = (t.task == NULL); |
Thomas Gleixner | 237fc6e | 2008-04-30 00:55:04 -0700 | [diff] [blame] | 1238 | |
| 1239 | destroy_hrtimer_on_stack(&t.timer); |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1240 | } |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1241 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1242 | __set_current_state(TASK_RUNNING); |
| 1243 | |
| 1244 | /* |
| 1245 | * NOTE: we don't remove ourselves from the waitqueue because |
| 1246 | * we are the only user of it. |
| 1247 | */ |
| 1248 | |
| 1249 | /* If we were woken (and unqueued), we succeeded, whatever. */ |
Peter Zijlstra | 2fff78c | 2009-02-11 18:10:10 +0100 | [diff] [blame] | 1250 | ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1251 | if (!unqueue_me(&q)) |
Peter Zijlstra | 2fff78c | 2009-02-11 18:10:10 +0100 | [diff] [blame] | 1252 | goto out_put_key; |
| 1253 | ret = -ETIMEDOUT; |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1254 | if (rem) |
Peter Zijlstra | 2fff78c | 2009-02-11 18:10:10 +0100 | [diff] [blame] | 1255 | goto out_put_key; |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1256 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1257 | /* |
| 1258 | * We expect signal_pending(current), but another thread may |
| 1259 | * have handled it for us already. |
| 1260 | */ |
Peter Zijlstra | 2fff78c | 2009-02-11 18:10:10 +0100 | [diff] [blame] | 1261 | ret = -ERESTARTSYS; |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1262 | if (!abs_time) |
Peter Zijlstra | 2fff78c | 2009-02-11 18:10:10 +0100 | [diff] [blame] | 1263 | goto out_put_key; |
Steven Rostedt | ce6bd42 | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 1264 | |
Peter Zijlstra | 2fff78c | 2009-02-11 18:10:10 +0100 | [diff] [blame] | 1265 | restart = ¤t_thread_info()->restart_block; |
| 1266 | restart->fn = futex_wait_restart; |
| 1267 | restart->futex.uaddr = (u32 *)uaddr; |
| 1268 | restart->futex.val = val; |
| 1269 | restart->futex.time = abs_time->tv64; |
| 1270 | restart->futex.bitset = bitset; |
| 1271 | restart->futex.flags = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1272 | |
Peter Zijlstra | 2fff78c | 2009-02-11 18:10:10 +0100 | [diff] [blame] | 1273 | if (fshared) |
| 1274 | restart->futex.flags |= FLAGS_SHARED; |
| 1275 | if (clockrt) |
| 1276 | restart->futex.flags |= FLAGS_CLOCKRT; |
| 1277 | |
| 1278 | ret = -ERESTART_RESTARTBLOCK; |
| 1279 | |
| 1280 | out_put_key: |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 1281 | put_futex_key(fshared, &q.key); |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1282 | out: |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1283 | return ret; |
| 1284 | } |
| 1285 | |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1286 | |
| 1287 | static long futex_wait_restart(struct restart_block *restart) |
| 1288 | { |
Steven Rostedt | ce6bd42 | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 1289 | u32 __user *uaddr = (u32 __user *)restart->futex.uaddr; |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 1290 | int fshared = 0; |
Steven Rostedt | ce6bd42 | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 1291 | ktime_t t; |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1292 | |
Steven Rostedt | ce6bd42 | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 1293 | t.tv64 = restart->futex.time; |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1294 | restart->fn = do_no_restart_syscall; |
Steven Rostedt | ce6bd42 | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 1295 | if (restart->futex.flags & FLAGS_SHARED) |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 1296 | fshared = 1; |
Thomas Gleixner | cd68998 | 2008-02-01 17:45:14 +0100 | [diff] [blame] | 1297 | return (long)futex_wait(uaddr, fshared, restart->futex.val, &t, |
Thomas Gleixner | 1acdac1 | 2008-11-20 10:02:53 -0800 | [diff] [blame] | 1298 | restart->futex.bitset, |
| 1299 | restart->futex.flags & FLAGS_CLOCKRT); |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1300 | } |
| 1301 | |
| 1302 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1303 | /* |
| 1304 | * Userspace tried a 0 -> TID atomic transition of the futex value |
| 1305 | * and failed. The kernel side here does the whole locking operation: |
| 1306 | * if there are waiters then it will block, it does PI, etc. (Due to |
| 1307 | * races the kernel might see a 0 value of the futex too.) |
| 1308 | */ |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 1309 | static int futex_lock_pi(u32 __user *uaddr, int fshared, |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1310 | int detect, ktime_t *time, int trylock) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1311 | { |
Thomas Gleixner | c5780e9 | 2006-09-08 09:47:15 -0700 | [diff] [blame] | 1312 | struct hrtimer_sleeper timeout, *to = NULL; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1313 | struct task_struct *curr = current; |
| 1314 | struct futex_hash_bucket *hb; |
| 1315 | u32 uval, newval, curval; |
| 1316 | struct futex_q q; |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 1317 | int ret, lock_taken, ownerdied = 0; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1318 | |
| 1319 | if (refill_pi_state_cache()) |
| 1320 | return -ENOMEM; |
| 1321 | |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1322 | if (time) { |
Thomas Gleixner | c5780e9 | 2006-09-08 09:47:15 -0700 | [diff] [blame] | 1323 | to = &timeout; |
Thomas Gleixner | 237fc6e | 2008-04-30 00:55:04 -0700 | [diff] [blame] | 1324 | hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME, |
| 1325 | HRTIMER_MODE_ABS); |
Thomas Gleixner | c5780e9 | 2006-09-08 09:47:15 -0700 | [diff] [blame] | 1326 | hrtimer_init_sleeper(to, current); |
Arjan van de Ven | cc584b2 | 2008-09-01 15:02:30 -0700 | [diff] [blame] | 1327 | hrtimer_set_expires(&to->timer, *time); |
Thomas Gleixner | c5780e9 | 2006-09-08 09:47:15 -0700 | [diff] [blame] | 1328 | } |
| 1329 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1330 | q.pi_state = NULL; |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1331 | retry: |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 1332 | q.key = FUTEX_KEY_INIT; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1333 | ret = get_futex_key(uaddr, fshared, &q.key); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1334 | if (unlikely(ret != 0)) |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1335 | goto out; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1336 | |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 1337 | retry_private: |
Eric Sesterhenn | 82af7ac | 2008-01-25 10:40:46 +0100 | [diff] [blame] | 1338 | hb = queue_lock(&q); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1339 | |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1340 | retry_locked: |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1341 | ret = lock_taken = 0; |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1342 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1343 | /* |
| 1344 | * To avoid races, we attempt to take the lock here again |
| 1345 | * (by doing a 0 -> TID atomic cmpxchg), while holding all |
| 1346 | * the locks. It will most likely not succeed. |
| 1347 | */ |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 1348 | newval = task_pid_vnr(current); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1349 | |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1350 | curval = cmpxchg_futex_value_locked(uaddr, 0, newval); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1351 | |
| 1352 | if (unlikely(curval == -EFAULT)) |
| 1353 | goto uaddr_faulted; |
| 1354 | |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1355 | /* |
| 1356 | * Detect deadlocks. In case of REQUEUE_PI this is a valid |
| 1357 | * situation and we return success to user space. |
| 1358 | */ |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 1359 | if (unlikely((curval & FUTEX_TID_MASK) == task_pid_vnr(current))) { |
Thomas Gleixner | bd19723 | 2007-06-17 21:11:10 +0200 | [diff] [blame] | 1360 | ret = -EDEADLK; |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1361 | goto out_unlock_put_key; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1362 | } |
| 1363 | |
| 1364 | /* |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1365 | * Surprise - we got the lock. Just return to userspace: |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1366 | */ |
| 1367 | if (unlikely(!curval)) |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1368 | goto out_unlock_put_key; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1369 | |
| 1370 | uval = curval; |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1371 | |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1372 | /* |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1373 | * Set the WAITERS flag, so the owner will know it has someone |
| 1374 | * to wake at next unlock |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1375 | */ |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1376 | newval = curval | FUTEX_WAITERS; |
| 1377 | |
| 1378 | /* |
| 1379 | * There are two cases, where a futex might have no owner (the |
Thomas Gleixner | bd19723 | 2007-06-17 21:11:10 +0200 | [diff] [blame] | 1380 | * owner TID is 0): OWNER_DIED. We take over the futex in this |
| 1381 | * case. We also do an unconditional take over, when the owner |
| 1382 | * of the futex died. |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1383 | * |
| 1384 | * This is safe as we are protected by the hash bucket lock ! |
| 1385 | */ |
| 1386 | if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { |
Thomas Gleixner | bd19723 | 2007-06-17 21:11:10 +0200 | [diff] [blame] | 1387 | /* Keep the OWNER_DIED bit */ |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 1388 | newval = (curval & ~FUTEX_TID_MASK) | task_pid_vnr(current); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1389 | ownerdied = 0; |
| 1390 | lock_taken = 1; |
| 1391 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1392 | |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1393 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1394 | |
| 1395 | if (unlikely(curval == -EFAULT)) |
| 1396 | goto uaddr_faulted; |
| 1397 | if (unlikely(curval != uval)) |
| 1398 | goto retry_locked; |
| 1399 | |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1400 | /* |
Thomas Gleixner | bd19723 | 2007-06-17 21:11:10 +0200 | [diff] [blame] | 1401 | * We took the lock due to owner died take over. |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1402 | */ |
Thomas Gleixner | bd19723 | 2007-06-17 21:11:10 +0200 | [diff] [blame] | 1403 | if (unlikely(lock_taken)) |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1404 | goto out_unlock_put_key; |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1405 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1406 | /* |
| 1407 | * We dont have the lock. Look up the PI state (or create it if |
| 1408 | * we are the first waiter): |
| 1409 | */ |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1410 | ret = lookup_pi_state(uval, hb, &q.key, &q.pi_state); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1411 | |
| 1412 | if (unlikely(ret)) { |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1413 | switch (ret) { |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1414 | |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1415 | case -EAGAIN: |
| 1416 | /* |
| 1417 | * Task is exiting and we just wait for the |
| 1418 | * exit to complete. |
| 1419 | */ |
| 1420 | queue_unlock(&q, hb); |
Darren Hart | de87fcc | 2009-03-12 00:55:46 -0700 | [diff] [blame] | 1421 | put_futex_key(fshared, &q.key); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1422 | cond_resched(); |
| 1423 | goto retry; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1424 | |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1425 | case -ESRCH: |
| 1426 | /* |
| 1427 | * No owner found for this futex. Check if the |
| 1428 | * OWNER_DIED bit is set to figure out whether |
| 1429 | * this is a robust futex or not. |
| 1430 | */ |
| 1431 | if (get_futex_value_locked(&curval, uaddr)) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1432 | goto uaddr_faulted; |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1433 | |
| 1434 | /* |
| 1435 | * We simply start over in case of a robust |
| 1436 | * futex. The code above will take the futex |
| 1437 | * and return happy. |
| 1438 | */ |
| 1439 | if (curval & FUTEX_OWNER_DIED) { |
| 1440 | ownerdied = 1; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1441 | goto retry_locked; |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1442 | } |
| 1443 | default: |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1444 | goto out_unlock_put_key; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1445 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1446 | } |
| 1447 | |
| 1448 | /* |
| 1449 | * Only actually queue now that the atomic ops are done: |
| 1450 | */ |
Eric Sesterhenn | 82af7ac | 2008-01-25 10:40:46 +0100 | [diff] [blame] | 1451 | queue_me(&q, hb); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1452 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1453 | WARN_ON(!q.pi_state); |
| 1454 | /* |
| 1455 | * Block on the PI mutex: |
| 1456 | */ |
| 1457 | if (!trylock) |
| 1458 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); |
| 1459 | else { |
| 1460 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); |
| 1461 | /* Fixup the trylock return value: */ |
| 1462 | ret = ret ? 0 : -EWOULDBLOCK; |
| 1463 | } |
| 1464 | |
Vernon Mauery | a99e4e4 | 2006-07-01 04:35:42 -0700 | [diff] [blame] | 1465 | spin_lock(q.lock_ptr); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1466 | |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1467 | if (!ret) { |
| 1468 | /* |
| 1469 | * Got the lock. We might not be the anticipated owner |
| 1470 | * if we did a lock-steal - fix up the PI-state in |
| 1471 | * that case: |
| 1472 | */ |
| 1473 | if (q.pi_state->owner != curr) |
Thomas Gleixner | 1b7558e | 2008-06-23 11:21:58 +0200 | [diff] [blame] | 1474 | ret = fixup_pi_state_owner(uaddr, &q, curr, fshared); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1475 | } else { |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1476 | /* |
| 1477 | * Catch the rare case, where the lock was released |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1478 | * when we were on the way back before we locked the |
| 1479 | * hash bucket. |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1480 | */ |
Thomas Gleixner | cdf71a1 | 2008-01-08 19:47:38 +0100 | [diff] [blame] | 1481 | if (q.pi_state->owner == curr) { |
| 1482 | /* |
| 1483 | * Try to get the rt_mutex now. This might |
| 1484 | * fail as some other task acquired the |
| 1485 | * rt_mutex after we removed ourself from the |
| 1486 | * rt_mutex waiters list. |
| 1487 | */ |
| 1488 | if (rt_mutex_trylock(&q.pi_state->pi_mutex)) |
| 1489 | ret = 0; |
| 1490 | else { |
| 1491 | /* |
| 1492 | * pi_state is incorrect, some other |
| 1493 | * task did a lock steal and we |
| 1494 | * returned due to timeout or signal |
| 1495 | * without taking the rt_mutex. Too |
| 1496 | * late. We can access the |
| 1497 | * rt_mutex_owner without locking, as |
| 1498 | * the other task is now blocked on |
| 1499 | * the hash bucket lock. Fix the state |
| 1500 | * up. |
| 1501 | */ |
| 1502 | struct task_struct *owner; |
| 1503 | int res; |
| 1504 | |
| 1505 | owner = rt_mutex_owner(&q.pi_state->pi_mutex); |
Thomas Gleixner | 1b7558e | 2008-06-23 11:21:58 +0200 | [diff] [blame] | 1506 | res = fixup_pi_state_owner(uaddr, &q, owner, |
| 1507 | fshared); |
Thomas Gleixner | cdf71a1 | 2008-01-08 19:47:38 +0100 | [diff] [blame] | 1508 | |
Thomas Gleixner | cdf71a1 | 2008-01-08 19:47:38 +0100 | [diff] [blame] | 1509 | /* propagate -EFAULT, if the fixup failed */ |
| 1510 | if (res) |
| 1511 | ret = res; |
| 1512 | } |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1513 | } else { |
| 1514 | /* |
| 1515 | * Paranoia check. If we did not take the lock |
| 1516 | * in the trylock above, then we should not be |
| 1517 | * the owner of the rtmutex, neither the real |
| 1518 | * nor the pending one: |
| 1519 | */ |
| 1520 | if (rt_mutex_owner(&q.pi_state->pi_mutex) == curr) |
| 1521 | printk(KERN_ERR "futex_lock_pi: ret = %d " |
| 1522 | "pi-mutex: %p pi-state %p\n", ret, |
| 1523 | q.pi_state->pi_mutex.owner, |
| 1524 | q.pi_state->owner); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1525 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1526 | } |
| 1527 | |
Darren Hart | e8f6386 | 2009-03-12 00:56:06 -0700 | [diff] [blame] | 1528 | /* |
| 1529 | * If fixup_pi_state_owner() faulted and was unable to handle the |
| 1530 | * fault, unlock it and return the fault to userspace. |
| 1531 | */ |
| 1532 | if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) |
| 1533 | rt_mutex_unlock(&q.pi_state->pi_mutex); |
| 1534 | |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1535 | /* Unqueue and drop the lock */ |
| 1536 | unqueue_me_pi(&q); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1537 | |
Thomas Gleixner | 237fc6e | 2008-04-30 00:55:04 -0700 | [diff] [blame] | 1538 | if (to) |
| 1539 | destroy_hrtimer_on_stack(&to->timer); |
Thomas Gleixner | c5780e9 | 2006-09-08 09:47:15 -0700 | [diff] [blame] | 1540 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1541 | |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1542 | out_unlock_put_key: |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1543 | queue_unlock(&q, hb); |
| 1544 | |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1545 | out_put_key: |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 1546 | put_futex_key(fshared, &q.key); |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1547 | out: |
Thomas Gleixner | 237fc6e | 2008-04-30 00:55:04 -0700 | [diff] [blame] | 1548 | if (to) |
| 1549 | destroy_hrtimer_on_stack(&to->timer); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1550 | return ret; |
| 1551 | |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1552 | uaddr_faulted: |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1553 | /* |
Darren Hart | b568636 | 2008-12-18 15:06:34 -0800 | [diff] [blame] | 1554 | * We have to r/w *(int __user *)uaddr, and we have to modify it |
| 1555 | * atomically. Therefore, if we continue to fault after get_user() |
| 1556 | * below, we need to handle the fault ourselves, while still holding |
| 1557 | * the mmap_sem. This can occur if the uaddr is under contention as |
| 1558 | * we have to drop the mmap_sem in order to call get_user(). |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1559 | */ |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1560 | queue_unlock(&q, hb); |
| 1561 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1562 | ret = get_user(uval, uaddr); |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 1563 | if (ret) |
| 1564 | goto out_put_key; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1565 | |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 1566 | if (!fshared) |
| 1567 | goto retry_private; |
| 1568 | |
| 1569 | put_futex_key(fshared, &q.key); |
| 1570 | goto retry; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1571 | } |
| 1572 | |
Darren Hart | de87fcc | 2009-03-12 00:55:46 -0700 | [diff] [blame] | 1573 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1574 | /* |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1575 | * Userspace attempted a TID -> 0 atomic transition, and failed. |
| 1576 | * This is the in-kernel slowpath: we look up the PI state (if any), |
| 1577 | * and do the rt-mutex unlock. |
| 1578 | */ |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 1579 | static int futex_unlock_pi(u32 __user *uaddr, int fshared) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1580 | { |
| 1581 | struct futex_hash_bucket *hb; |
| 1582 | struct futex_q *this, *next; |
| 1583 | u32 uval; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 1584 | struct plist_head *head; |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 1585 | union futex_key key = FUTEX_KEY_INIT; |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 1586 | int ret; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1587 | |
| 1588 | retry: |
| 1589 | if (get_user(uval, uaddr)) |
| 1590 | return -EFAULT; |
| 1591 | /* |
| 1592 | * We release only a lock we actually own: |
| 1593 | */ |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 1594 | if ((uval & FUTEX_TID_MASK) != task_pid_vnr(current)) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1595 | return -EPERM; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1596 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1597 | ret = get_futex_key(uaddr, fshared, &key); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1598 | if (unlikely(ret != 0)) |
| 1599 | goto out; |
| 1600 | |
| 1601 | hb = hash_futex(&key); |
| 1602 | spin_lock(&hb->lock); |
| 1603 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1604 | /* |
| 1605 | * To avoid races, try to do the TID -> 0 atomic transition |
| 1606 | * again. If it succeeds then we can return without waking |
| 1607 | * anyone else up: |
| 1608 | */ |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1609 | if (!(uval & FUTEX_OWNER_DIED)) |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 1610 | uval = cmpxchg_futex_value_locked(uaddr, task_pid_vnr(current), 0); |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1611 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1612 | |
| 1613 | if (unlikely(uval == -EFAULT)) |
| 1614 | goto pi_faulted; |
| 1615 | /* |
| 1616 | * Rare case: we managed to release the lock atomically, |
| 1617 | * no need to wake anyone else up: |
| 1618 | */ |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 1619 | if (unlikely(uval == task_pid_vnr(current))) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1620 | goto out_unlock; |
| 1621 | |
| 1622 | /* |
| 1623 | * Ok, other tasks may need to be woken up - check waiters |
| 1624 | * and do the wakeup if necessary: |
| 1625 | */ |
| 1626 | head = &hb->chain; |
| 1627 | |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 1628 | plist_for_each_entry_safe(this, next, head, list) { |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1629 | if (!match_futex (&this->key, &key)) |
| 1630 | continue; |
| 1631 | ret = wake_futex_pi(uaddr, uval, this); |
| 1632 | /* |
| 1633 | * The atomic access to the futex value |
| 1634 | * generated a pagefault, so retry the |
| 1635 | * user-access and the wakeup: |
| 1636 | */ |
| 1637 | if (ret == -EFAULT) |
| 1638 | goto pi_faulted; |
| 1639 | goto out_unlock; |
| 1640 | } |
| 1641 | /* |
| 1642 | * No waiters - kernel unlocks the futex: |
| 1643 | */ |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1644 | if (!(uval & FUTEX_OWNER_DIED)) { |
| 1645 | ret = unlock_futex_pi(uaddr, uval); |
| 1646 | if (ret == -EFAULT) |
| 1647 | goto pi_faulted; |
| 1648 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1649 | |
| 1650 | out_unlock: |
| 1651 | spin_unlock(&hb->lock); |
Peter Zijlstra | 38d47c1 | 2008-09-26 19:32:20 +0200 | [diff] [blame] | 1652 | put_futex_key(fshared, &key); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1653 | |
Darren Hart | 42d35d4 | 2008-12-29 15:49:53 -0800 | [diff] [blame] | 1654 | out: |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1655 | return ret; |
| 1656 | |
| 1657 | pi_faulted: |
| 1658 | /* |
Darren Hart | b568636 | 2008-12-18 15:06:34 -0800 | [diff] [blame] | 1659 | * We have to r/w *(int __user *)uaddr, and we have to modify it |
| 1660 | * atomically. Therefore, if we continue to fault after get_user() |
| 1661 | * below, we need to handle the fault ourselves, while still holding |
| 1662 | * the mmap_sem. This can occur if the uaddr is under contention as |
| 1663 | * we have to drop the mmap_sem in order to call get_user(). |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1664 | */ |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1665 | spin_unlock(&hb->lock); |
Darren Hart | e4dc5b7 | 2009-03-12 00:56:13 -0700 | [diff] [blame^] | 1666 | put_futex_key(fshared, &key); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1667 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1668 | ret = get_user(uval, uaddr); |
Darren Hart | b568636 | 2008-12-18 15:06:34 -0800 | [diff] [blame] | 1669 | if (!ret) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1670 | goto retry; |
| 1671 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1672 | return ret; |
| 1673 | } |
| 1674 | |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1675 | /* |
| 1676 | * Support for robust futexes: the kernel cleans up held futexes at |
| 1677 | * thread exit time. |
| 1678 | * |
| 1679 | * Implementation: user-space maintains a per-thread list of locks it |
| 1680 | * is holding. Upon do_exit(), the kernel carefully walks this list, |
| 1681 | * and marks all locks that are owned by this thread with the |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1682 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1683 | * always manipulated with the lock held, so the list is private and |
| 1684 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' |
| 1685 | * field, to allow the kernel to clean up if the thread dies after |
| 1686 | * acquiring the lock, but just before it could have added itself to |
| 1687 | * the list. There can only be one such pending lock. |
| 1688 | */ |
| 1689 | |
| 1690 | /** |
| 1691 | * sys_set_robust_list - set the robust-futex list head of a task |
| 1692 | * @head: pointer to the list-head |
| 1693 | * @len: length of the list-head, as userspace expects |
| 1694 | */ |
Heiko Carstens | 836f92a | 2009-01-14 14:14:33 +0100 | [diff] [blame] | 1695 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
| 1696 | size_t, len) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1697 | { |
Thomas Gleixner | a0c1e90 | 2008-02-23 15:23:57 -0800 | [diff] [blame] | 1698 | if (!futex_cmpxchg_enabled) |
| 1699 | return -ENOSYS; |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1700 | /* |
| 1701 | * The kernel knows only one size for now: |
| 1702 | */ |
| 1703 | if (unlikely(len != sizeof(*head))) |
| 1704 | return -EINVAL; |
| 1705 | |
| 1706 | current->robust_list = head; |
| 1707 | |
| 1708 | return 0; |
| 1709 | } |
| 1710 | |
| 1711 | /** |
| 1712 | * sys_get_robust_list - get the robust-futex list head of a task |
| 1713 | * @pid: pid of the process [zero for current task] |
| 1714 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in |
| 1715 | * @len_ptr: pointer to a length field, the kernel fills in the header size |
| 1716 | */ |
Heiko Carstens | 836f92a | 2009-01-14 14:14:33 +0100 | [diff] [blame] | 1717 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
| 1718 | struct robust_list_head __user * __user *, head_ptr, |
| 1719 | size_t __user *, len_ptr) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1720 | { |
Al Viro | ba46df9 | 2006-10-10 22:46:07 +0100 | [diff] [blame] | 1721 | struct robust_list_head __user *head; |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1722 | unsigned long ret; |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1723 | const struct cred *cred = current_cred(), *pcred; |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1724 | |
Thomas Gleixner | a0c1e90 | 2008-02-23 15:23:57 -0800 | [diff] [blame] | 1725 | if (!futex_cmpxchg_enabled) |
| 1726 | return -ENOSYS; |
| 1727 | |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1728 | if (!pid) |
| 1729 | head = current->robust_list; |
| 1730 | else { |
| 1731 | struct task_struct *p; |
| 1732 | |
| 1733 | ret = -ESRCH; |
Oleg Nesterov | aaa2a97 | 2006-09-29 02:00:55 -0700 | [diff] [blame] | 1734 | rcu_read_lock(); |
Pavel Emelyanov | 228ebcb | 2007-10-18 23:40:16 -0700 | [diff] [blame] | 1735 | p = find_task_by_vpid(pid); |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1736 | if (!p) |
| 1737 | goto err_unlock; |
| 1738 | ret = -EPERM; |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1739 | pcred = __task_cred(p); |
| 1740 | if (cred->euid != pcred->euid && |
| 1741 | cred->euid != pcred->uid && |
David Howells | 76aac0e | 2008-11-14 10:39:12 +1100 | [diff] [blame] | 1742 | !capable(CAP_SYS_PTRACE)) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1743 | goto err_unlock; |
| 1744 | head = p->robust_list; |
Oleg Nesterov | aaa2a97 | 2006-09-29 02:00:55 -0700 | [diff] [blame] | 1745 | rcu_read_unlock(); |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1746 | } |
| 1747 | |
| 1748 | if (put_user(sizeof(*head), len_ptr)) |
| 1749 | return -EFAULT; |
| 1750 | return put_user(head, head_ptr); |
| 1751 | |
| 1752 | err_unlock: |
Oleg Nesterov | aaa2a97 | 2006-09-29 02:00:55 -0700 | [diff] [blame] | 1753 | rcu_read_unlock(); |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1754 | |
| 1755 | return ret; |
| 1756 | } |
| 1757 | |
| 1758 | /* |
| 1759 | * Process a futex-list entry, check whether it's owned by the |
| 1760 | * dying task, and do notification if so: |
| 1761 | */ |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1762 | int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1763 | { |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1764 | u32 uval, nval, mval; |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1765 | |
Ingo Molnar | 8f17d3a | 2006-03-27 01:16:27 -0800 | [diff] [blame] | 1766 | retry: |
| 1767 | if (get_user(uval, uaddr)) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1768 | return -1; |
| 1769 | |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 1770 | if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1771 | /* |
| 1772 | * Ok, this dying thread is truly holding a futex |
| 1773 | * of interest. Set the OWNER_DIED bit atomically |
| 1774 | * via cmpxchg, and if the value had FUTEX_WAITERS |
| 1775 | * set, wake up a waiter (if any). (We have to do a |
| 1776 | * futex_wake() even if OWNER_DIED is already set - |
| 1777 | * to handle the rare but possible case of recursive |
| 1778 | * thread-death.) The rest of the cleanup is done in |
| 1779 | * userspace. |
| 1780 | */ |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1781 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
| 1782 | nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval); |
| 1783 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1784 | if (nval == -EFAULT) |
| 1785 | return -1; |
| 1786 | |
| 1787 | if (nval != uval) |
Ingo Molnar | 8f17d3a | 2006-03-27 01:16:27 -0800 | [diff] [blame] | 1788 | goto retry; |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1789 | |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1790 | /* |
| 1791 | * Wake robust non-PI futexes here. The wakeup of |
| 1792 | * PI futexes happens in exit_pi_state(): |
| 1793 | */ |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1794 | if (!pi && (uval & FUTEX_WAITERS)) |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 1795 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1796 | } |
| 1797 | return 0; |
| 1798 | } |
| 1799 | |
| 1800 | /* |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1801 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: |
| 1802 | */ |
| 1803 | static inline int fetch_robust_entry(struct robust_list __user **entry, |
Al Viro | ba46df9 | 2006-10-10 22:46:07 +0100 | [diff] [blame] | 1804 | struct robust_list __user * __user *head, |
| 1805 | int *pi) |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1806 | { |
| 1807 | unsigned long uentry; |
| 1808 | |
Al Viro | ba46df9 | 2006-10-10 22:46:07 +0100 | [diff] [blame] | 1809 | if (get_user(uentry, (unsigned long __user *)head)) |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1810 | return -EFAULT; |
| 1811 | |
Al Viro | ba46df9 | 2006-10-10 22:46:07 +0100 | [diff] [blame] | 1812 | *entry = (void __user *)(uentry & ~1UL); |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1813 | *pi = uentry & 1; |
| 1814 | |
| 1815 | return 0; |
| 1816 | } |
| 1817 | |
| 1818 | /* |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1819 | * Walk curr->robust_list (very carefully, it's a userspace list!) |
| 1820 | * and mark any locks found there dead, and notify any waiters. |
| 1821 | * |
| 1822 | * We silently return on any sign of list-walking problem. |
| 1823 | */ |
| 1824 | void exit_robust_list(struct task_struct *curr) |
| 1825 | { |
| 1826 | struct robust_list_head __user *head = curr->robust_list; |
Martin Schwidefsky | 9f96cb1 | 2007-10-01 01:20:13 -0700 | [diff] [blame] | 1827 | struct robust_list __user *entry, *next_entry, *pending; |
| 1828 | unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip; |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1829 | unsigned long futex_offset; |
Martin Schwidefsky | 9f96cb1 | 2007-10-01 01:20:13 -0700 | [diff] [blame] | 1830 | int rc; |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1831 | |
Thomas Gleixner | a0c1e90 | 2008-02-23 15:23:57 -0800 | [diff] [blame] | 1832 | if (!futex_cmpxchg_enabled) |
| 1833 | return; |
| 1834 | |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1835 | /* |
| 1836 | * Fetch the list head (which was registered earlier, via |
| 1837 | * sys_set_robust_list()): |
| 1838 | */ |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1839 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1840 | return; |
| 1841 | /* |
| 1842 | * Fetch the relative futex offset: |
| 1843 | */ |
| 1844 | if (get_user(futex_offset, &head->futex_offset)) |
| 1845 | return; |
| 1846 | /* |
| 1847 | * Fetch any possibly pending lock-add first, and handle it |
| 1848 | * if it exists: |
| 1849 | */ |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1850 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1851 | return; |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1852 | |
Martin Schwidefsky | 9f96cb1 | 2007-10-01 01:20:13 -0700 | [diff] [blame] | 1853 | next_entry = NULL; /* avoid warning with gcc */ |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1854 | while (entry != &head->list) { |
| 1855 | /* |
Martin Schwidefsky | 9f96cb1 | 2007-10-01 01:20:13 -0700 | [diff] [blame] | 1856 | * Fetch the next entry in the list before calling |
| 1857 | * handle_futex_death: |
| 1858 | */ |
| 1859 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); |
| 1860 | /* |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1861 | * A pending lock might already be on the list, so |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1862 | * don't process it twice: |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1863 | */ |
| 1864 | if (entry != pending) |
Al Viro | ba46df9 | 2006-10-10 22:46:07 +0100 | [diff] [blame] | 1865 | if (handle_futex_death((void __user *)entry + futex_offset, |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1866 | curr, pi)) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1867 | return; |
Martin Schwidefsky | 9f96cb1 | 2007-10-01 01:20:13 -0700 | [diff] [blame] | 1868 | if (rc) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1869 | return; |
Martin Schwidefsky | 9f96cb1 | 2007-10-01 01:20:13 -0700 | [diff] [blame] | 1870 | entry = next_entry; |
| 1871 | pi = next_pi; |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1872 | /* |
| 1873 | * Avoid excessively long or circular lists: |
| 1874 | */ |
| 1875 | if (!--limit) |
| 1876 | break; |
| 1877 | |
| 1878 | cond_resched(); |
| 1879 | } |
Martin Schwidefsky | 9f96cb1 | 2007-10-01 01:20:13 -0700 | [diff] [blame] | 1880 | |
| 1881 | if (pending) |
| 1882 | handle_futex_death((void __user *)pending + futex_offset, |
| 1883 | curr, pip); |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1884 | } |
| 1885 | |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1886 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1887 | u32 __user *uaddr2, u32 val2, u32 val3) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1888 | { |
Thomas Gleixner | 1acdac1 | 2008-11-20 10:02:53 -0800 | [diff] [blame] | 1889 | int clockrt, ret = -ENOSYS; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1890 | int cmd = op & FUTEX_CMD_MASK; |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 1891 | int fshared = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1892 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1893 | if (!(op & FUTEX_PRIVATE_FLAG)) |
Peter Zijlstra | c2f9f20 | 2008-09-26 19:32:23 +0200 | [diff] [blame] | 1894 | fshared = 1; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1895 | |
Thomas Gleixner | 1acdac1 | 2008-11-20 10:02:53 -0800 | [diff] [blame] | 1896 | clockrt = op & FUTEX_CLOCK_REALTIME; |
| 1897 | if (clockrt && cmd != FUTEX_WAIT_BITSET) |
| 1898 | return -ENOSYS; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1899 | |
| 1900 | switch (cmd) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1901 | case FUTEX_WAIT: |
Thomas Gleixner | cd68998 | 2008-02-01 17:45:14 +0100 | [diff] [blame] | 1902 | val3 = FUTEX_BITSET_MATCH_ANY; |
| 1903 | case FUTEX_WAIT_BITSET: |
Thomas Gleixner | 1acdac1 | 2008-11-20 10:02:53 -0800 | [diff] [blame] | 1904 | ret = futex_wait(uaddr, fshared, val, timeout, val3, clockrt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1905 | break; |
| 1906 | case FUTEX_WAKE: |
Thomas Gleixner | cd68998 | 2008-02-01 17:45:14 +0100 | [diff] [blame] | 1907 | val3 = FUTEX_BITSET_MATCH_ANY; |
| 1908 | case FUTEX_WAKE_BITSET: |
| 1909 | ret = futex_wake(uaddr, fshared, val, val3); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1910 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1911 | case FUTEX_REQUEUE: |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1912 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1913 | break; |
| 1914 | case FUTEX_CMP_REQUEUE: |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1915 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1916 | break; |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 1917 | case FUTEX_WAKE_OP: |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1918 | ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3); |
Jakub Jelinek | 4732efb | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 1919 | break; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1920 | case FUTEX_LOCK_PI: |
Thomas Gleixner | a0c1e90 | 2008-02-23 15:23:57 -0800 | [diff] [blame] | 1921 | if (futex_cmpxchg_enabled) |
| 1922 | ret = futex_lock_pi(uaddr, fshared, val, timeout, 0); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1923 | break; |
| 1924 | case FUTEX_UNLOCK_PI: |
Thomas Gleixner | a0c1e90 | 2008-02-23 15:23:57 -0800 | [diff] [blame] | 1925 | if (futex_cmpxchg_enabled) |
| 1926 | ret = futex_unlock_pi(uaddr, fshared); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1927 | break; |
| 1928 | case FUTEX_TRYLOCK_PI: |
Thomas Gleixner | a0c1e90 | 2008-02-23 15:23:57 -0800 | [diff] [blame] | 1929 | if (futex_cmpxchg_enabled) |
| 1930 | ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1931 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1932 | default: |
| 1933 | ret = -ENOSYS; |
| 1934 | } |
| 1935 | return ret; |
| 1936 | } |
| 1937 | |
| 1938 | |
Heiko Carstens | 17da2bd | 2009-01-14 14:14:10 +0100 | [diff] [blame] | 1939 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
| 1940 | struct timespec __user *, utime, u32 __user *, uaddr2, |
| 1941 | u32, val3) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1942 | { |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1943 | struct timespec ts; |
| 1944 | ktime_t t, *tp = NULL; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1945 | u32 val2 = 0; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1946 | int cmd = op & FUTEX_CMD_MASK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1947 | |
Thomas Gleixner | cd68998 | 2008-02-01 17:45:14 +0100 | [diff] [blame] | 1948 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || |
| 1949 | cmd == FUTEX_WAIT_BITSET)) { |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1950 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1951 | return -EFAULT; |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1952 | if (!timespec_valid(&ts)) |
Thomas Gleixner | 9741ef96 | 2006-03-31 02:31:32 -0800 | [diff] [blame] | 1953 | return -EINVAL; |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1954 | |
| 1955 | t = timespec_to_ktime(ts); |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1956 | if (cmd == FUTEX_WAIT) |
Thomas Gleixner | 5a7780e | 2008-02-13 09:20:43 +0100 | [diff] [blame] | 1957 | t = ktime_add_safe(ktime_get(), t); |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1958 | tp = &t; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1959 | } |
| 1960 | /* |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1961 | * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE. |
Andreas Schwab | f54f098 | 2007-07-31 00:38:51 -0700 | [diff] [blame] | 1962 | * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1963 | */ |
Andreas Schwab | f54f098 | 2007-07-31 00:38:51 -0700 | [diff] [blame] | 1964 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || |
| 1965 | cmd == FUTEX_WAKE_OP) |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1966 | val2 = (u32) (unsigned long) utime; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1967 | |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1968 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1969 | } |
| 1970 | |
Benjamin Herrenschmidt | f6d107f | 2008-03-27 14:52:15 +1100 | [diff] [blame] | 1971 | static int __init futex_init(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1972 | { |
Thomas Gleixner | a0c1e90 | 2008-02-23 15:23:57 -0800 | [diff] [blame] | 1973 | u32 curval; |
Thomas Gleixner | 3e4ab74 | 2008-02-23 15:23:55 -0800 | [diff] [blame] | 1974 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1975 | |
Thomas Gleixner | a0c1e90 | 2008-02-23 15:23:57 -0800 | [diff] [blame] | 1976 | /* |
| 1977 | * This will fail and we want it. Some arch implementations do |
| 1978 | * runtime detection of the futex_atomic_cmpxchg_inatomic() |
| 1979 | * functionality. We want to know that before we call in any |
| 1980 | * of the complex code paths. Also we want to prevent |
| 1981 | * registration of robust lists in that case. NULL is |
| 1982 | * guaranteed to fault and we get -EFAULT on functional |
| 1983 | * implementation, the non functional ones will return |
| 1984 | * -ENOSYS. |
| 1985 | */ |
| 1986 | curval = cmpxchg_futex_value_locked(NULL, 0, 0); |
| 1987 | if (curval == -EFAULT) |
| 1988 | futex_cmpxchg_enabled = 1; |
| 1989 | |
Thomas Gleixner | 3e4ab74 | 2008-02-23 15:23:55 -0800 | [diff] [blame] | 1990 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { |
| 1991 | plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock); |
| 1992 | spin_lock_init(&futex_queues[i].lock); |
| 1993 | } |
| 1994 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1995 | return 0; |
| 1996 | } |
Benjamin Herrenschmidt | f6d107f | 2008-03-27 14:52:15 +1100 | [diff] [blame] | 1997 | __initcall(futex_init); |