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> |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 55 | #include <asm/futex.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 56 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 57 | #include "rtmutex_common.h" |
| 58 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8) |
| 60 | |
| 61 | /* |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 62 | * Priority Inheritance state: |
| 63 | */ |
| 64 | struct futex_pi_state { |
| 65 | /* |
| 66 | * list of 'owned' pi_state instances - these have to be |
| 67 | * cleaned up in do_exit() if the task exits prematurely: |
| 68 | */ |
| 69 | struct list_head list; |
| 70 | |
| 71 | /* |
| 72 | * The PI object: |
| 73 | */ |
| 74 | struct rt_mutex pi_mutex; |
| 75 | |
| 76 | struct task_struct *owner; |
| 77 | atomic_t refcount; |
| 78 | |
| 79 | union futex_key key; |
| 80 | }; |
| 81 | |
| 82 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 83 | * We use this hashed waitqueue instead of a normal wait_queue_t, so |
| 84 | * we can wake only the relevant ones (hashed queues may be shared). |
| 85 | * |
| 86 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 87 | * 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] | 88 | * The order of wakup is always to make the first condition true, then |
| 89 | * wake up q->waiters, then make the second condition true. |
| 90 | */ |
| 91 | struct futex_q { |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 92 | struct plist_node list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 93 | wait_queue_head_t waiters; |
| 94 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 95 | /* Which hash list lock to use: */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | spinlock_t *lock_ptr; |
| 97 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 98 | /* Key which the futex is hashed on: */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | union futex_key key; |
| 100 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 101 | /* For fd, sigio sent using these: */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 102 | int fd; |
| 103 | struct file *filp; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 104 | |
| 105 | /* Optional priority inheritance state: */ |
| 106 | struct futex_pi_state *pi_state; |
| 107 | struct task_struct *task; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 108 | }; |
| 109 | |
| 110 | /* |
| 111 | * Split the global futex_lock into every hash list lock. |
| 112 | */ |
| 113 | struct futex_hash_bucket { |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 114 | spinlock_t lock; |
| 115 | struct plist_head chain; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 116 | }; |
| 117 | |
| 118 | static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS]; |
| 119 | |
| 120 | /* Futex-fs vfsmount entry: */ |
| 121 | static struct vfsmount *futex_mnt; |
| 122 | |
| 123 | /* |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 124 | * Take mm->mmap_sem, when futex is shared |
| 125 | */ |
| 126 | static inline void futex_lock_mm(struct rw_semaphore *fshared) |
| 127 | { |
| 128 | if (fshared) |
| 129 | down_read(fshared); |
| 130 | } |
| 131 | |
| 132 | /* |
| 133 | * Release mm->mmap_sem, when the futex is shared |
| 134 | */ |
| 135 | static inline void futex_unlock_mm(struct rw_semaphore *fshared) |
| 136 | { |
| 137 | if (fshared) |
| 138 | up_read(fshared); |
| 139 | } |
| 140 | |
| 141 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | * We hash on the keys returned from get_futex_key (see below). |
| 143 | */ |
| 144 | static struct futex_hash_bucket *hash_futex(union futex_key *key) |
| 145 | { |
| 146 | u32 hash = jhash2((u32*)&key->both.word, |
| 147 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, |
| 148 | key->both.offset); |
| 149 | return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)]; |
| 150 | } |
| 151 | |
| 152 | /* |
| 153 | * Return 1 if two futex_keys are equal, 0 otherwise. |
| 154 | */ |
| 155 | static inline int match_futex(union futex_key *key1, union futex_key *key2) |
| 156 | { |
| 157 | return (key1->both.word == key2->both.word |
| 158 | && key1->both.ptr == key2->both.ptr |
| 159 | && key1->both.offset == key2->both.offset); |
| 160 | } |
| 161 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 162 | /** |
| 163 | * get_futex_key - Get parameters which are the keys for a futex. |
| 164 | * @uaddr: virtual address of the futex |
| 165 | * @shared: NULL for a PROCESS_PRIVATE futex, |
| 166 | * ¤t->mm->mmap_sem for a PROCESS_SHARED futex |
| 167 | * @key: address where result is stored. |
| 168 | * |
| 169 | * Returns a negative error code or 0 |
| 170 | * The key words are stored in *key on success. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 171 | * |
Josef "Jeff" Sipek | f3a43f3 | 2006-12-08 02:36:43 -0800 | [diff] [blame] | 172 | * 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] | 173 | * offset_within_page). For private mappings, it's (uaddr, current->mm). |
| 174 | * We can usually work out the index without swapping in the page. |
| 175 | * |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 176 | * fshared is NULL for PROCESS_PRIVATE futexes |
| 177 | * For other futexes, it points to ¤t->mm->mmap_sem and |
| 178 | * caller must have taken the reader lock. but NOT any spinlocks. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 179 | */ |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 180 | int get_futex_key(u32 __user *uaddr, struct rw_semaphore *fshared, |
| 181 | union futex_key *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 183 | unsigned long address = (unsigned long)uaddr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | struct mm_struct *mm = current->mm; |
| 185 | struct vm_area_struct *vma; |
| 186 | struct page *page; |
| 187 | int err; |
| 188 | |
| 189 | /* |
| 190 | * The futex address must be "naturally" aligned. |
| 191 | */ |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 192 | key->both.offset = address % PAGE_SIZE; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 193 | if (unlikely((address % sizeof(u32)) != 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | return -EINVAL; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 195 | address -= key->both.offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 196 | |
| 197 | /* |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 198 | * PROCESS_PRIVATE futexes are fast. |
| 199 | * As the mm cannot disappear under us and the 'key' only needs |
| 200 | * virtual address, we dont even have to find the underlying vma. |
| 201 | * Note : We do have to check 'uaddr' is a valid user address, |
| 202 | * but access_ok() should be faster than find_vma() |
| 203 | */ |
| 204 | if (!fshared) { |
| 205 | if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32)))) |
| 206 | return -EFAULT; |
| 207 | key->private.mm = mm; |
| 208 | key->private.address = address; |
| 209 | return 0; |
| 210 | } |
| 211 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 212 | * The futex is hashed differently depending on whether |
| 213 | * it's in a shared or private mapping. So check vma first. |
| 214 | */ |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 215 | vma = find_extend_vma(mm, address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 216 | if (unlikely(!vma)) |
| 217 | return -EFAULT; |
| 218 | |
| 219 | /* |
| 220 | * Permissions. |
| 221 | */ |
| 222 | if (unlikely((vma->vm_flags & (VM_IO|VM_READ)) != VM_READ)) |
| 223 | return (vma->vm_flags & VM_IO) ? -EPERM : -EACCES; |
| 224 | |
| 225 | /* |
| 226 | * Private mappings are handled in a simple way. |
| 227 | * |
| 228 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if |
| 229 | * it's a read-only handle, it's expected that futexes attach to |
| 230 | * the object not the particular process. Therefore we use |
| 231 | * VM_MAYSHARE here, not VM_SHARED which is restricted to shared |
| 232 | * mappings of _writable_ handles. |
| 233 | */ |
| 234 | if (likely(!(vma->vm_flags & VM_MAYSHARE))) { |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 235 | key->both.offset |= FUT_OFF_MMSHARED; /* reference taken on mm */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | key->private.mm = mm; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 237 | key->private.address = address; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 238 | return 0; |
| 239 | } |
| 240 | |
| 241 | /* |
| 242 | * Linear file mappings are also simple. |
| 243 | */ |
Josef "Jeff" Sipek | f3a43f3 | 2006-12-08 02:36:43 -0800 | [diff] [blame] | 244 | key->shared.inode = vma->vm_file->f_path.dentry->d_inode; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 245 | key->both.offset |= FUT_OFF_INODE; /* inode-based key. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 246 | if (likely(!(vma->vm_flags & VM_NONLINEAR))) { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 247 | key->shared.pgoff = (((address - vma->vm_start) >> PAGE_SHIFT) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 248 | + vma->vm_pgoff); |
| 249 | return 0; |
| 250 | } |
| 251 | |
| 252 | /* |
| 253 | * We could walk the page table to read the non-linear |
| 254 | * pte, and get the page index without fetching the page |
| 255 | * from swap. But that's a lot of code to duplicate here |
| 256 | * for a rare case, so we simply fetch the page. |
| 257 | */ |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 258 | err = get_user_pages(current, mm, address, 1, 0, 0, &page, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 259 | if (err >= 0) { |
| 260 | key->shared.pgoff = |
| 261 | page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 262 | put_page(page); |
| 263 | return 0; |
| 264 | } |
| 265 | return err; |
| 266 | } |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 267 | EXPORT_SYMBOL_GPL(get_futex_key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 268 | |
| 269 | /* |
| 270 | * Take a reference to the resource addressed by a key. |
| 271 | * Can be called while holding spinlocks. |
| 272 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | */ |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 274 | inline void get_futex_key_refs(union futex_key *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | { |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 276 | if (key->both.ptr == 0) |
| 277 | return; |
| 278 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { |
| 279 | case FUT_OFF_INODE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | atomic_inc(&key->shared.inode->i_count); |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 281 | break; |
| 282 | case FUT_OFF_MMSHARED: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 283 | atomic_inc(&key->private.mm->mm_count); |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 284 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | } |
| 286 | } |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 287 | EXPORT_SYMBOL_GPL(get_futex_key_refs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 288 | |
| 289 | /* |
| 290 | * Drop a reference to the resource addressed by a key. |
| 291 | * The hash bucket spinlock must not be held. |
| 292 | */ |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 293 | void drop_futex_key_refs(union futex_key *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 294 | { |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 295 | if (key->both.ptr == 0) |
| 296 | return; |
| 297 | switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { |
| 298 | case FUT_OFF_INODE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | iput(key->shared.inode); |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 300 | break; |
| 301 | case FUT_OFF_MMSHARED: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 302 | mmdrop(key->private.mm); |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 303 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 304 | } |
| 305 | } |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 306 | EXPORT_SYMBOL_GPL(drop_futex_key_refs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 307 | |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 308 | static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval) |
| 309 | { |
| 310 | u32 curval; |
| 311 | |
| 312 | pagefault_disable(); |
| 313 | curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval); |
| 314 | pagefault_enable(); |
| 315 | |
| 316 | return curval; |
| 317 | } |
| 318 | |
| 319 | static int get_futex_value_locked(u32 *dest, u32 __user *from) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 320 | { |
| 321 | int ret; |
| 322 | |
Peter Zijlstra | a866374 | 2006-12-06 20:32:20 -0800 | [diff] [blame] | 323 | pagefault_disable(); |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 324 | ret = __copy_from_user_inatomic(dest, from, sizeof(u32)); |
Peter Zijlstra | a866374 | 2006-12-06 20:32:20 -0800 | [diff] [blame] | 325 | pagefault_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 326 | |
| 327 | return ret ? -EFAULT : 0; |
| 328 | } |
| 329 | |
| 330 | /* |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 331 | * Fault handling. |
| 332 | * if fshared is non NULL, current->mm->mmap_sem is already held |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 333 | */ |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 334 | static int futex_handle_fault(unsigned long address, |
| 335 | struct rw_semaphore *fshared, int attempt) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 336 | { |
| 337 | struct vm_area_struct * vma; |
| 338 | struct mm_struct *mm = current->mm; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 339 | int ret = -EFAULT; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 340 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 341 | if (attempt > 2) |
| 342 | return ret; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 343 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 344 | if (!fshared) |
| 345 | down_read(&mm->mmap_sem); |
| 346 | vma = find_vma(mm, address); |
| 347 | if (vma && address >= vma->vm_start && |
| 348 | (vma->vm_flags & VM_WRITE)) { |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame^] | 349 | int fault; |
| 350 | fault = handle_mm_fault(mm, vma, address, 1); |
| 351 | if (unlikely((fault & VM_FAULT_ERROR))) { |
| 352 | #if 0 |
| 353 | /* XXX: let's do this when we verify it is OK */ |
| 354 | if (ret & VM_FAULT_OOM) |
| 355 | ret = -ENOMEM; |
| 356 | #endif |
| 357 | } else { |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 358 | ret = 0; |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame^] | 359 | if (fault & VM_FAULT_MAJOR) |
| 360 | current->maj_flt++; |
| 361 | else |
| 362 | current->min_flt++; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 363 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 364 | } |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 365 | if (!fshared) |
| 366 | up_read(&mm->mmap_sem); |
| 367 | return ret; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 368 | } |
| 369 | |
| 370 | /* |
| 371 | * PI code: |
| 372 | */ |
| 373 | static int refill_pi_state_cache(void) |
| 374 | { |
| 375 | struct futex_pi_state *pi_state; |
| 376 | |
| 377 | if (likely(current->pi_state_cache)) |
| 378 | return 0; |
| 379 | |
Burman Yan | 4668edc | 2006-12-06 20:38:51 -0800 | [diff] [blame] | 380 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 381 | |
| 382 | if (!pi_state) |
| 383 | return -ENOMEM; |
| 384 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 385 | INIT_LIST_HEAD(&pi_state->list); |
| 386 | /* pi_mutex gets initialized later */ |
| 387 | pi_state->owner = NULL; |
| 388 | atomic_set(&pi_state->refcount, 1); |
| 389 | |
| 390 | current->pi_state_cache = pi_state; |
| 391 | |
| 392 | return 0; |
| 393 | } |
| 394 | |
| 395 | static struct futex_pi_state * alloc_pi_state(void) |
| 396 | { |
| 397 | struct futex_pi_state *pi_state = current->pi_state_cache; |
| 398 | |
| 399 | WARN_ON(!pi_state); |
| 400 | current->pi_state_cache = NULL; |
| 401 | |
| 402 | return pi_state; |
| 403 | } |
| 404 | |
| 405 | static void free_pi_state(struct futex_pi_state *pi_state) |
| 406 | { |
| 407 | if (!atomic_dec_and_test(&pi_state->refcount)) |
| 408 | return; |
| 409 | |
| 410 | /* |
| 411 | * If pi_state->owner is NULL, the owner is most probably dying |
| 412 | * and has cleaned up the pi_state already |
| 413 | */ |
| 414 | if (pi_state->owner) { |
| 415 | spin_lock_irq(&pi_state->owner->pi_lock); |
| 416 | list_del_init(&pi_state->list); |
| 417 | spin_unlock_irq(&pi_state->owner->pi_lock); |
| 418 | |
| 419 | rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner); |
| 420 | } |
| 421 | |
| 422 | if (current->pi_state_cache) |
| 423 | kfree(pi_state); |
| 424 | else { |
| 425 | /* |
| 426 | * pi_state->list is already empty. |
| 427 | * clear pi_state->owner. |
| 428 | * refcount is at 0 - put it back to 1. |
| 429 | */ |
| 430 | pi_state->owner = NULL; |
| 431 | atomic_set(&pi_state->refcount, 1); |
| 432 | current->pi_state_cache = pi_state; |
| 433 | } |
| 434 | } |
| 435 | |
| 436 | /* |
| 437 | * Look up the task based on what TID userspace gave us. |
| 438 | * We dont trust it. |
| 439 | */ |
| 440 | static struct task_struct * futex_find_get_task(pid_t pid) |
| 441 | { |
| 442 | struct task_struct *p; |
| 443 | |
Oleg Nesterov | d359b54 | 2006-09-29 02:00:55 -0700 | [diff] [blame] | 444 | rcu_read_lock(); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 445 | p = find_task_by_pid(pid); |
Thomas Gleixner | a06381f | 2007-06-23 11:48:40 +0200 | [diff] [blame] | 446 | |
| 447 | if (!p || ((current->euid != p->euid) && (current->euid != p->uid))) |
| 448 | p = ERR_PTR(-ESRCH); |
| 449 | else |
| 450 | get_task_struct(p); |
| 451 | |
Oleg Nesterov | d359b54 | 2006-09-29 02:00:55 -0700 | [diff] [blame] | 452 | rcu_read_unlock(); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 453 | |
| 454 | return p; |
| 455 | } |
| 456 | |
| 457 | /* |
| 458 | * This task is holding PI mutexes at exit time => bad. |
| 459 | * Kernel cleans up PI-state, but userspace is likely hosed. |
| 460 | * (Robust-futex cleanup is separate and might save the day for userspace.) |
| 461 | */ |
| 462 | void exit_pi_state_list(struct task_struct *curr) |
| 463 | { |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 464 | struct list_head *next, *head = &curr->pi_state_list; |
| 465 | struct futex_pi_state *pi_state; |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 466 | struct futex_hash_bucket *hb; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 467 | union futex_key key; |
| 468 | |
| 469 | /* |
| 470 | * We are a ZOMBIE and nobody can enqueue itself on |
| 471 | * pi_state_list anymore, but we have to be careful |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 472 | * versus waiters unqueueing themselves: |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 473 | */ |
| 474 | spin_lock_irq(&curr->pi_lock); |
| 475 | while (!list_empty(head)) { |
| 476 | |
| 477 | next = head->next; |
| 478 | pi_state = list_entry(next, struct futex_pi_state, list); |
| 479 | key = pi_state->key; |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 480 | hb = hash_futex(&key); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 481 | spin_unlock_irq(&curr->pi_lock); |
| 482 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 483 | spin_lock(&hb->lock); |
| 484 | |
| 485 | spin_lock_irq(&curr->pi_lock); |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 486 | /* |
| 487 | * We dropped the pi-lock, so re-check whether this |
| 488 | * task still owns the PI-state: |
| 489 | */ |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 490 | if (head->next != next) { |
| 491 | spin_unlock(&hb->lock); |
| 492 | continue; |
| 493 | } |
| 494 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 495 | WARN_ON(pi_state->owner != curr); |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 496 | WARN_ON(list_empty(&pi_state->list)); |
| 497 | list_del_init(&pi_state->list); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 498 | pi_state->owner = NULL; |
| 499 | spin_unlock_irq(&curr->pi_lock); |
| 500 | |
| 501 | rt_mutex_unlock(&pi_state->pi_mutex); |
| 502 | |
| 503 | spin_unlock(&hb->lock); |
| 504 | |
| 505 | spin_lock_irq(&curr->pi_lock); |
| 506 | } |
| 507 | spin_unlock_irq(&curr->pi_lock); |
| 508 | } |
| 509 | |
| 510 | static int |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 511 | lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, |
| 512 | union futex_key *key, struct futex_pi_state **ps) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 513 | { |
| 514 | struct futex_pi_state *pi_state = NULL; |
| 515 | struct futex_q *this, *next; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 516 | struct plist_head *head; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 517 | struct task_struct *p; |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 518 | pid_t pid = uval & FUTEX_TID_MASK; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 519 | |
| 520 | head = &hb->chain; |
| 521 | |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 522 | plist_for_each_entry_safe(this, next, head, list) { |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 523 | if (match_futex(&this->key, key)) { |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 524 | /* |
| 525 | * Another waiter already exists - bump up |
| 526 | * the refcount and return its pi_state: |
| 527 | */ |
| 528 | pi_state = this->pi_state; |
Thomas Gleixner | 06a9ec2 | 2006-07-10 04:44:30 -0700 | [diff] [blame] | 529 | /* |
| 530 | * Userspace might have messed up non PI and PI futexes |
| 531 | */ |
| 532 | if (unlikely(!pi_state)) |
| 533 | return -EINVAL; |
| 534 | |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 535 | WARN_ON(!atomic_read(&pi_state->refcount)); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 536 | WARN_ON(pid && pi_state->owner && |
| 537 | pi_state->owner->pid != pid); |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 538 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 539 | atomic_inc(&pi_state->refcount); |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 540 | *ps = pi_state; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 541 | |
| 542 | return 0; |
| 543 | } |
| 544 | } |
| 545 | |
| 546 | /* |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 547 | * 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] | 548 | * the new pi_state to it, but bail out when TID = 0 |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 549 | */ |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 550 | if (!pid) |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 551 | return -ESRCH; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 552 | p = futex_find_get_task(pid); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 553 | if (IS_ERR(p)) |
| 554 | return PTR_ERR(p); |
| 555 | |
| 556 | /* |
| 557 | * We need to look at the task state flags to figure out, |
| 558 | * whether the task is exiting. To protect against the do_exit |
| 559 | * change of the task flags, we do this protected by |
| 560 | * p->pi_lock: |
| 561 | */ |
| 562 | spin_lock_irq(&p->pi_lock); |
| 563 | if (unlikely(p->flags & PF_EXITING)) { |
| 564 | /* |
| 565 | * The task is on the way out. When PF_EXITPIDONE is |
| 566 | * set, we know that the task has finished the |
| 567 | * cleanup: |
| 568 | */ |
| 569 | int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN; |
| 570 | |
| 571 | spin_unlock_irq(&p->pi_lock); |
| 572 | put_task_struct(p); |
| 573 | return ret; |
| 574 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 575 | |
| 576 | pi_state = alloc_pi_state(); |
| 577 | |
| 578 | /* |
| 579 | * Initialize the pi_mutex in locked state and make 'p' |
| 580 | * the owner of it: |
| 581 | */ |
| 582 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); |
| 583 | |
| 584 | /* Store the key for possible exit cleanups: */ |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 585 | pi_state->key = *key; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 586 | |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 587 | WARN_ON(!list_empty(&pi_state->list)); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 588 | list_add(&pi_state->list, &p->pi_state_list); |
| 589 | pi_state->owner = p; |
| 590 | spin_unlock_irq(&p->pi_lock); |
| 591 | |
| 592 | put_task_struct(p); |
| 593 | |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 594 | *ps = pi_state; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 595 | |
| 596 | return 0; |
| 597 | } |
| 598 | |
| 599 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | * The hash bucket lock must be held when this is called. |
| 601 | * Afterwards, the futex_q must not be accessed. |
| 602 | */ |
| 603 | static void wake_futex(struct futex_q *q) |
| 604 | { |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 605 | plist_del(&q->list, &q->list.plist); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 606 | if (q->filp) |
| 607 | send_sigio(&q->filp->f_owner, q->fd, POLL_IN); |
| 608 | /* |
| 609 | * The lock in wake_up_all() is a crucial memory barrier after the |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 610 | * plist_del() and also before assigning to q->lock_ptr. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 611 | */ |
| 612 | wake_up_all(&q->waiters); |
| 613 | /* |
| 614 | * The waiting task can free the futex_q as soon as this is written, |
| 615 | * without taking any locks. This must come last. |
Andrew Morton | 8e31108 | 2005-12-23 19:54:46 -0800 | [diff] [blame] | 616 | * |
| 617 | * A memory barrier is required here to prevent the following store |
| 618 | * to lock_ptr from getting ahead of the wakeup. Clearing the lock |
| 619 | * at the end of wake_up_all() does not prevent this store from |
| 620 | * moving. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 621 | */ |
Ralf Baechle | ccdea2f | 2006-12-06 20:40:26 -0800 | [diff] [blame] | 622 | smp_wmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 623 | q->lock_ptr = NULL; |
| 624 | } |
| 625 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 626 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) |
| 627 | { |
| 628 | struct task_struct *new_owner; |
| 629 | struct futex_pi_state *pi_state = this->pi_state; |
| 630 | u32 curval, newval; |
| 631 | |
| 632 | if (!pi_state) |
| 633 | return -EINVAL; |
| 634 | |
Ingo Molnar | 21778867 | 2007-03-16 13:38:31 -0800 | [diff] [blame] | 635 | spin_lock(&pi_state->pi_mutex.wait_lock); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 636 | new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); |
| 637 | |
| 638 | /* |
| 639 | * This happens when we have stolen the lock and the original |
| 640 | * pending owner did not enqueue itself back on the rt_mutex. |
| 641 | * Thats not a tragedy. We know that way, that a lock waiter |
| 642 | * is on the fly. We make the futex_q waiter the pending owner. |
| 643 | */ |
| 644 | if (!new_owner) |
| 645 | new_owner = this->task; |
| 646 | |
| 647 | /* |
| 648 | * We pass it to the next owner. (The WAITERS bit is always |
| 649 | * kept enabled while there is PI state around. We must also |
| 650 | * preserve the owner died bit.) |
| 651 | */ |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 652 | if (!(uval & FUTEX_OWNER_DIED)) { |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 653 | int ret = 0; |
| 654 | |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 655 | newval = FUTEX_WAITERS | new_owner->pid; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 656 | |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 657 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 658 | |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 659 | if (curval == -EFAULT) |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 660 | ret = -EFAULT; |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 661 | if (curval != uval) |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 662 | ret = -EINVAL; |
| 663 | if (ret) { |
| 664 | spin_unlock(&pi_state->pi_mutex.wait_lock); |
| 665 | return ret; |
| 666 | } |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 667 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 668 | |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 669 | spin_lock_irq(&pi_state->owner->pi_lock); |
| 670 | WARN_ON(list_empty(&pi_state->list)); |
| 671 | list_del_init(&pi_state->list); |
| 672 | spin_unlock_irq(&pi_state->owner->pi_lock); |
| 673 | |
| 674 | spin_lock_irq(&new_owner->pi_lock); |
| 675 | WARN_ON(!list_empty(&pi_state->list)); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 676 | list_add(&pi_state->list, &new_owner->pi_state_list); |
| 677 | pi_state->owner = new_owner; |
Ingo Molnar | 627371d | 2006-07-29 05:16:20 +0200 | [diff] [blame] | 678 | spin_unlock_irq(&new_owner->pi_lock); |
| 679 | |
Ingo Molnar | 21778867 | 2007-03-16 13:38:31 -0800 | [diff] [blame] | 680 | spin_unlock(&pi_state->pi_mutex.wait_lock); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 681 | rt_mutex_unlock(&pi_state->pi_mutex); |
| 682 | |
| 683 | return 0; |
| 684 | } |
| 685 | |
| 686 | static int unlock_futex_pi(u32 __user *uaddr, u32 uval) |
| 687 | { |
| 688 | u32 oldval; |
| 689 | |
| 690 | /* |
| 691 | * There is no waiter, so we unlock the futex. The owner died |
| 692 | * bit has not to be preserved here. We are the owner: |
| 693 | */ |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 694 | oldval = cmpxchg_futex_value_locked(uaddr, uval, 0); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 695 | |
| 696 | if (oldval == -EFAULT) |
| 697 | return oldval; |
| 698 | if (oldval != uval) |
| 699 | return -EAGAIN; |
| 700 | |
| 701 | return 0; |
| 702 | } |
| 703 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 704 | /* |
Ingo Molnar | 8b8f319 | 2006-07-03 00:25:05 -0700 | [diff] [blame] | 705 | * Express the locking dependencies for lockdep: |
| 706 | */ |
| 707 | static inline void |
| 708 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) |
| 709 | { |
| 710 | if (hb1 <= hb2) { |
| 711 | spin_lock(&hb1->lock); |
| 712 | if (hb1 < hb2) |
| 713 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); |
| 714 | } else { /* hb1 > hb2 */ |
| 715 | spin_lock(&hb2->lock); |
| 716 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); |
| 717 | } |
| 718 | } |
| 719 | |
| 720 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 721 | * Wake up all waiters hashed on the physical page that is mapped |
| 722 | * to this virtual address: |
| 723 | */ |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 724 | static int futex_wake(u32 __user *uaddr, struct rw_semaphore *fshared, |
| 725 | int nr_wake) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 726 | { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 727 | struct futex_hash_bucket *hb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 728 | struct futex_q *this, *next; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 729 | struct plist_head *head; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 730 | union futex_key key; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 731 | int ret; |
| 732 | |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 733 | futex_lock_mm(fshared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 734 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 735 | ret = get_futex_key(uaddr, fshared, &key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 736 | if (unlikely(ret != 0)) |
| 737 | goto out; |
| 738 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 739 | hb = hash_futex(&key); |
| 740 | spin_lock(&hb->lock); |
| 741 | head = &hb->chain; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 742 | |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 743 | plist_for_each_entry_safe(this, next, head, list) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 744 | if (match_futex (&this->key, &key)) { |
Ingo Molnar | ed6f7b1 | 2006-07-01 04:35:46 -0700 | [diff] [blame] | 745 | if (this->pi_state) { |
| 746 | ret = -EINVAL; |
| 747 | break; |
| 748 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 749 | wake_futex(this); |
| 750 | if (++ret >= nr_wake) |
| 751 | break; |
| 752 | } |
| 753 | } |
| 754 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 755 | spin_unlock(&hb->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 756 | out: |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 757 | futex_unlock_mm(fshared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 758 | return ret; |
| 759 | } |
| 760 | |
| 761 | /* |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 762 | * Wake up all waiters hashed on the physical page that is mapped |
| 763 | * to this virtual address: |
| 764 | */ |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 765 | static int |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 766 | futex_wake_op(u32 __user *uaddr1, struct rw_semaphore *fshared, |
| 767 | u32 __user *uaddr2, |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 768 | int nr_wake, int nr_wake2, int op) |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 769 | { |
| 770 | union futex_key key1, key2; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 771 | struct futex_hash_bucket *hb1, *hb2; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 772 | struct plist_head *head; |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 773 | struct futex_q *this, *next; |
| 774 | int ret, op_ret, attempt = 0; |
| 775 | |
| 776 | retryfull: |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 777 | futex_lock_mm(fshared); |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 778 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 779 | ret = get_futex_key(uaddr1, fshared, &key1); |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 780 | if (unlikely(ret != 0)) |
| 781 | goto out; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 782 | ret = get_futex_key(uaddr2, fshared, &key2); |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 783 | if (unlikely(ret != 0)) |
| 784 | goto out; |
| 785 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 786 | hb1 = hash_futex(&key1); |
| 787 | hb2 = hash_futex(&key2); |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 788 | |
| 789 | retry: |
Ingo Molnar | 8b8f319 | 2006-07-03 00:25:05 -0700 | [diff] [blame] | 790 | double_lock_hb(hb1, hb2); |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 791 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 792 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 793 | if (unlikely(op_ret < 0)) { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 794 | u32 dummy; |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 795 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 796 | spin_unlock(&hb1->lock); |
| 797 | if (hb1 != hb2) |
| 798 | spin_unlock(&hb2->lock); |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 799 | |
David Howells | 7ee1dd3 | 2006-01-06 00:11:44 -0800 | [diff] [blame] | 800 | #ifndef CONFIG_MMU |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 801 | /* |
| 802 | * we don't get EFAULT from MMU faults if we don't have an MMU, |
| 803 | * but we might get them from range checking |
| 804 | */ |
David Howells | 7ee1dd3 | 2006-01-06 00:11:44 -0800 | [diff] [blame] | 805 | ret = op_ret; |
| 806 | goto out; |
| 807 | #endif |
| 808 | |
David Gibson | 796f8d9 | 2005-11-07 00:59:33 -0800 | [diff] [blame] | 809 | if (unlikely(op_ret != -EFAULT)) { |
| 810 | ret = op_ret; |
| 811 | goto out; |
| 812 | } |
| 813 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 814 | /* |
| 815 | * futex_atomic_op_inuser needs to both read and write |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 816 | * *(int __user *)uaddr2, but we can't modify it |
| 817 | * non-atomically. Therefore, if get_user below is not |
| 818 | * enough, we need to handle the fault ourselves, while |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 819 | * still holding the mmap_sem. |
| 820 | */ |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 821 | if (attempt++) { |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 822 | ret = futex_handle_fault((unsigned long)uaddr2, |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 823 | fshared, attempt); |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 824 | if (ret) |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 825 | goto out; |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 826 | goto retry; |
| 827 | } |
| 828 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 829 | /* |
| 830 | * If we would have faulted, release mmap_sem, |
| 831 | * fault it in and start all over again. |
| 832 | */ |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 833 | futex_unlock_mm(fshared); |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 834 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 835 | ret = get_user(dummy, uaddr2); |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 836 | if (ret) |
| 837 | return ret; |
| 838 | |
| 839 | goto retryfull; |
| 840 | } |
| 841 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 842 | head = &hb1->chain; |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 843 | |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 844 | plist_for_each_entry_safe(this, next, head, list) { |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 845 | if (match_futex (&this->key, &key1)) { |
| 846 | wake_futex(this); |
| 847 | if (++ret >= nr_wake) |
| 848 | break; |
| 849 | } |
| 850 | } |
| 851 | |
| 852 | if (op_ret > 0) { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 853 | head = &hb2->chain; |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 854 | |
| 855 | op_ret = 0; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 856 | plist_for_each_entry_safe(this, next, head, list) { |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 857 | if (match_futex (&this->key, &key2)) { |
| 858 | wake_futex(this); |
| 859 | if (++op_ret >= nr_wake2) |
| 860 | break; |
| 861 | } |
| 862 | } |
| 863 | ret += op_ret; |
| 864 | } |
| 865 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 866 | spin_unlock(&hb1->lock); |
| 867 | if (hb1 != hb2) |
| 868 | spin_unlock(&hb2->lock); |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 869 | out: |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 870 | futex_unlock_mm(fshared); |
| 871 | |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 872 | return ret; |
| 873 | } |
| 874 | |
| 875 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 876 | * Requeue all waiters hashed on one physical page to another |
| 877 | * physical page. |
| 878 | */ |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 879 | static int futex_requeue(u32 __user *uaddr1, struct rw_semaphore *fshared, |
| 880 | u32 __user *uaddr2, |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 881 | int nr_wake, int nr_requeue, u32 *cmpval) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 882 | { |
| 883 | union futex_key key1, key2; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 884 | struct futex_hash_bucket *hb1, *hb2; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 885 | struct plist_head *head1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 886 | struct futex_q *this, *next; |
| 887 | int ret, drop_count = 0; |
| 888 | |
| 889 | retry: |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 890 | futex_lock_mm(fshared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 891 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 892 | ret = get_futex_key(uaddr1, fshared, &key1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 893 | if (unlikely(ret != 0)) |
| 894 | goto out; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 895 | ret = get_futex_key(uaddr2, fshared, &key2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 896 | if (unlikely(ret != 0)) |
| 897 | goto out; |
| 898 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 899 | hb1 = hash_futex(&key1); |
| 900 | hb2 = hash_futex(&key2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 901 | |
Ingo Molnar | 8b8f319 | 2006-07-03 00:25:05 -0700 | [diff] [blame] | 902 | double_lock_hb(hb1, hb2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 903 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 904 | if (likely(cmpval != NULL)) { |
| 905 | u32 curval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 906 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 907 | ret = get_futex_value_locked(&curval, uaddr1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 908 | |
| 909 | if (unlikely(ret)) { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 910 | spin_unlock(&hb1->lock); |
| 911 | if (hb1 != hb2) |
| 912 | spin_unlock(&hb2->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 | /* |
| 915 | * If we would have faulted, release mmap_sem, fault |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 916 | * it in and start all over again. |
| 917 | */ |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 918 | futex_unlock_mm(fshared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 919 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 920 | ret = get_user(curval, uaddr1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 921 | |
| 922 | if (!ret) |
| 923 | goto retry; |
| 924 | |
| 925 | return ret; |
| 926 | } |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 927 | if (curval != *cmpval) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 928 | ret = -EAGAIN; |
| 929 | goto out_unlock; |
| 930 | } |
| 931 | } |
| 932 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 933 | head1 = &hb1->chain; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 934 | plist_for_each_entry_safe(this, next, head1, list) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 935 | if (!match_futex (&this->key, &key1)) |
| 936 | continue; |
| 937 | if (++ret <= nr_wake) { |
| 938 | wake_futex(this); |
| 939 | } else { |
Sebastien Dugue | 59e0e0a | 2006-06-27 02:55:03 -0700 | [diff] [blame] | 940 | /* |
| 941 | * If key1 and key2 hash to the same bucket, no need to |
| 942 | * requeue. |
| 943 | */ |
| 944 | if (likely(head1 != &hb2->chain)) { |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 945 | plist_del(&this->list, &hb1->chain); |
| 946 | plist_add(&this->list, &hb2->chain); |
Sebastien Dugue | 59e0e0a | 2006-06-27 02:55:03 -0700 | [diff] [blame] | 947 | this->lock_ptr = &hb2->lock; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 948 | #ifdef CONFIG_DEBUG_PI_LIST |
| 949 | this->list.plist.lock = &hb2->lock; |
| 950 | #endif |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 951 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 952 | this->key = key2; |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 953 | get_futex_key_refs(&key2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 954 | drop_count++; |
| 955 | |
| 956 | if (ret - nr_wake >= nr_requeue) |
| 957 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 958 | } |
| 959 | } |
| 960 | |
| 961 | out_unlock: |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 962 | spin_unlock(&hb1->lock); |
| 963 | if (hb1 != hb2) |
| 964 | spin_unlock(&hb2->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 965 | |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 966 | /* drop_futex_key_refs() must be called outside the spinlocks. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 967 | while (--drop_count >= 0) |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 968 | drop_futex_key_refs(&key1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 969 | |
| 970 | out: |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 971 | futex_unlock_mm(fshared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 972 | return ret; |
| 973 | } |
| 974 | |
| 975 | /* The key must be already stored in q->key. */ |
| 976 | static inline struct futex_hash_bucket * |
| 977 | queue_lock(struct futex_q *q, int fd, struct file *filp) |
| 978 | { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 979 | struct futex_hash_bucket *hb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 980 | |
| 981 | q->fd = fd; |
| 982 | q->filp = filp; |
| 983 | |
| 984 | init_waitqueue_head(&q->waiters); |
| 985 | |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 986 | get_futex_key_refs(&q->key); |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 987 | hb = hash_futex(&q->key); |
| 988 | q->lock_ptr = &hb->lock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 989 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 990 | spin_lock(&hb->lock); |
| 991 | return hb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 992 | } |
| 993 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 994 | 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] | 995 | { |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 996 | int prio; |
| 997 | |
| 998 | /* |
| 999 | * The priority used to register this element is |
| 1000 | * - either the real thread-priority for the real-time threads |
| 1001 | * (i.e. threads with a priority lower than MAX_RT_PRIO) |
| 1002 | * - or MAX_RT_PRIO for non-RT threads. |
| 1003 | * Thus, all RT-threads are woken first in priority order, and |
| 1004 | * the others are woken last, in FIFO order. |
| 1005 | */ |
| 1006 | prio = min(current->normal_prio, MAX_RT_PRIO); |
| 1007 | |
| 1008 | plist_node_init(&q->list, prio); |
| 1009 | #ifdef CONFIG_DEBUG_PI_LIST |
| 1010 | q->list.plist.lock = &hb->lock; |
| 1011 | #endif |
| 1012 | plist_add(&q->list, &hb->chain); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1013 | q->task = current; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1014 | spin_unlock(&hb->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1015 | } |
| 1016 | |
| 1017 | static inline void |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1018 | queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1019 | { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1020 | spin_unlock(&hb->lock); |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 1021 | drop_futex_key_refs(&q->key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1022 | } |
| 1023 | |
| 1024 | /* |
| 1025 | * queue_me and unqueue_me must be called as a pair, each |
| 1026 | * exactly once. They are called with the hashed spinlock held. |
| 1027 | */ |
| 1028 | |
| 1029 | /* The key must be already stored in q->key. */ |
| 1030 | static void queue_me(struct futex_q *q, int fd, struct file *filp) |
| 1031 | { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1032 | struct futex_hash_bucket *hb; |
| 1033 | |
| 1034 | hb = queue_lock(q, fd, filp); |
| 1035 | __queue_me(q, hb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1036 | } |
| 1037 | |
| 1038 | /* Return 1 if we were still queued (ie. 0 means we were woken) */ |
| 1039 | static int unqueue_me(struct futex_q *q) |
| 1040 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1041 | spinlock_t *lock_ptr; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1042 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1043 | |
| 1044 | /* In the common case we don't take the spinlock, which is nice. */ |
| 1045 | retry: |
| 1046 | lock_ptr = q->lock_ptr; |
Christian Borntraeger | e91467e | 2006-08-05 12:13:52 -0700 | [diff] [blame] | 1047 | barrier(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1048 | if (lock_ptr != 0) { |
| 1049 | spin_lock(lock_ptr); |
| 1050 | /* |
| 1051 | * q->lock_ptr can change between reading it and |
| 1052 | * spin_lock(), causing us to take the wrong lock. This |
| 1053 | * corrects the race condition. |
| 1054 | * |
| 1055 | * Reasoning goes like this: if we have the wrong lock, |
| 1056 | * q->lock_ptr must have changed (maybe several times) |
| 1057 | * between reading it and the spin_lock(). It can |
| 1058 | * change again after the spin_lock() but only if it was |
| 1059 | * already changed before the spin_lock(). It cannot, |
| 1060 | * however, change back to the original value. Therefore |
| 1061 | * we can detect whether we acquired the correct lock. |
| 1062 | */ |
| 1063 | if (unlikely(lock_ptr != q->lock_ptr)) { |
| 1064 | spin_unlock(lock_ptr); |
| 1065 | goto retry; |
| 1066 | } |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 1067 | WARN_ON(plist_node_empty(&q->list)); |
| 1068 | plist_del(&q->list, &q->list.plist); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1069 | |
| 1070 | BUG_ON(q->pi_state); |
| 1071 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1072 | spin_unlock(lock_ptr); |
| 1073 | ret = 1; |
| 1074 | } |
| 1075 | |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 1076 | drop_futex_key_refs(&q->key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1077 | return ret; |
| 1078 | } |
| 1079 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1080 | /* |
| 1081 | * PI futexes can not be requeued and must remove themself from the |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1082 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry |
| 1083 | * and dropped here. |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1084 | */ |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1085 | static void unqueue_me_pi(struct futex_q *q) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1086 | { |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 1087 | WARN_ON(plist_node_empty(&q->list)); |
| 1088 | plist_del(&q->list, &q->list.plist); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1089 | |
| 1090 | BUG_ON(!q->pi_state); |
| 1091 | free_pi_state(q->pi_state); |
| 1092 | q->pi_state = NULL; |
| 1093 | |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1094 | spin_unlock(q->lock_ptr); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1095 | |
Rusty Russell | 9adef58 | 2007-05-08 00:26:42 -0700 | [diff] [blame] | 1096 | drop_futex_key_refs(&q->key); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1097 | } |
| 1098 | |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1099 | /* |
| 1100 | * Fixup the pi_state owner with current. |
| 1101 | * |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1102 | * Must be called with hash bucket lock held and mm->sem held for non |
| 1103 | * private futexes. |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1104 | */ |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1105 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1106 | struct task_struct *curr) |
| 1107 | { |
| 1108 | u32 newtid = curr->pid | FUTEX_WAITERS; |
| 1109 | struct futex_pi_state *pi_state = q->pi_state; |
| 1110 | u32 uval, curval, newval; |
| 1111 | int ret; |
| 1112 | |
| 1113 | /* Owner died? */ |
| 1114 | if (pi_state->owner != NULL) { |
| 1115 | spin_lock_irq(&pi_state->owner->pi_lock); |
| 1116 | WARN_ON(list_empty(&pi_state->list)); |
| 1117 | list_del_init(&pi_state->list); |
| 1118 | spin_unlock_irq(&pi_state->owner->pi_lock); |
| 1119 | } else |
| 1120 | newtid |= FUTEX_OWNER_DIED; |
| 1121 | |
| 1122 | pi_state->owner = curr; |
| 1123 | |
| 1124 | spin_lock_irq(&curr->pi_lock); |
| 1125 | WARN_ON(!list_empty(&pi_state->list)); |
| 1126 | list_add(&pi_state->list, &curr->pi_state_list); |
| 1127 | spin_unlock_irq(&curr->pi_lock); |
| 1128 | |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1129 | /* |
| 1130 | * We own it, so we have to replace the pending owner |
| 1131 | * TID. This must be atomic as we have preserve the |
| 1132 | * owner died bit here. |
| 1133 | */ |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1134 | ret = get_futex_value_locked(&uval, uaddr); |
| 1135 | |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1136 | while (!ret) { |
| 1137 | newval = (uval & FUTEX_OWNER_DIED) | newtid; |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1138 | |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1139 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1140 | |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1141 | if (curval == -EFAULT) |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1142 | ret = -EFAULT; |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1143 | if (curval == uval) |
| 1144 | break; |
| 1145 | uval = curval; |
| 1146 | } |
| 1147 | return ret; |
| 1148 | } |
| 1149 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1150 | /* |
| 1151 | * In case we must use restart_block to restart a futex_wait, |
| 1152 | * we encode in the 'arg3' shared capability |
| 1153 | */ |
| 1154 | #define ARG3_SHARED 1 |
| 1155 | |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1156 | static long futex_wait_restart(struct restart_block *restart); |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1157 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1158 | static int futex_wait(u32 __user *uaddr, struct rw_semaphore *fshared, |
| 1159 | u32 val, ktime_t *abs_time) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1160 | { |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1161 | struct task_struct *curr = current; |
| 1162 | DECLARE_WAITQUEUE(wait, curr); |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1163 | struct futex_hash_bucket *hb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1164 | struct futex_q q; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1165 | u32 uval; |
| 1166 | int ret; |
Thomas Gleixner | bd19723 | 2007-06-17 21:11:10 +0200 | [diff] [blame] | 1167 | struct hrtimer_sleeper t; |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1168 | int rem = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1169 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1170 | q.pi_state = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1171 | retry: |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1172 | futex_lock_mm(fshared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1173 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1174 | ret = get_futex_key(uaddr, fshared, &q.key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1175 | if (unlikely(ret != 0)) |
| 1176 | goto out_release_sem; |
| 1177 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1178 | hb = queue_lock(&q, -1, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1179 | |
| 1180 | /* |
| 1181 | * Access the page AFTER the futex is queued. |
| 1182 | * Order is important: |
| 1183 | * |
| 1184 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); |
| 1185 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } |
| 1186 | * |
| 1187 | * The basic logical guarantee of a futex is that it blocks ONLY |
| 1188 | * if cond(var) is known to be true at the time of blocking, for |
| 1189 | * any cond. If we queued after testing *uaddr, that would open |
| 1190 | * a race condition where we could block indefinitely with |
| 1191 | * cond(var) false, which would violate the guarantee. |
| 1192 | * |
| 1193 | * A consequence is that futex_wait() can return zero and absorb |
| 1194 | * a wakeup when *uaddr != val on entry to the syscall. This is |
| 1195 | * rare, but normal. |
| 1196 | * |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1197 | * for shared futexes, we hold the mmap semaphore, so the mapping |
| 1198 | * cannot have changed since we looked it up in get_futex_key. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1199 | */ |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1200 | ret = get_futex_value_locked(&uval, uaddr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1201 | |
| 1202 | if (unlikely(ret)) { |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1203 | queue_unlock(&q, hb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1204 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1205 | /* |
| 1206 | * If we would have faulted, release mmap_sem, fault it in and |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1207 | * start all over again. |
| 1208 | */ |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1209 | futex_unlock_mm(fshared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1210 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1211 | ret = get_user(uval, uaddr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1212 | |
| 1213 | if (!ret) |
| 1214 | goto retry; |
| 1215 | return ret; |
| 1216 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1217 | ret = -EWOULDBLOCK; |
| 1218 | if (uval != val) |
| 1219 | goto out_unlock_release_sem; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1220 | |
| 1221 | /* Only actually queue if *uaddr contained val. */ |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1222 | __queue_me(&q, hb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1223 | |
| 1224 | /* |
| 1225 | * Now the futex is queued and we have checked the data, we |
| 1226 | * don't want to hold mmap_sem while we sleep. |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1227 | */ |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1228 | futex_unlock_mm(fshared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1229 | |
| 1230 | /* |
| 1231 | * There might have been scheduling since the queue_me(), as we |
| 1232 | * cannot hold a spinlock across the get_user() in case it |
| 1233 | * faults, and we cannot just set TASK_INTERRUPTIBLE state when |
| 1234 | * queueing ourselves into the futex hash. This code thus has to |
| 1235 | * rely on the futex_wake() code removing us from hash when it |
| 1236 | * wakes us up. |
| 1237 | */ |
| 1238 | |
| 1239 | /* add_wait_queue is the barrier after __set_current_state. */ |
| 1240 | __set_current_state(TASK_INTERRUPTIBLE); |
| 1241 | add_wait_queue(&q.waiters, &wait); |
| 1242 | /* |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 1243 | * !plist_node_empty() is safe here without any lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1244 | * q.lock_ptr != 0 is not safe, because of ordering against wakeup. |
| 1245 | */ |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 1246 | if (likely(!plist_node_empty(&q.list))) { |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1247 | if (!abs_time) |
| 1248 | schedule(); |
| 1249 | else { |
| 1250 | hrtimer_init(&t.timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); |
| 1251 | hrtimer_init_sleeper(&t, current); |
| 1252 | t.timer.expires = *abs_time; |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1253 | |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1254 | hrtimer_start(&t.timer, t.timer.expires, HRTIMER_MODE_ABS); |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1255 | |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1256 | /* |
| 1257 | * the timer could have already expired, in which |
| 1258 | * case current would be flagged for rescheduling. |
| 1259 | * Don't bother calling schedule. |
| 1260 | */ |
| 1261 | if (likely(t.task)) |
| 1262 | schedule(); |
| 1263 | |
| 1264 | hrtimer_cancel(&t.timer); |
| 1265 | |
| 1266 | /* Flag if a timeout occured */ |
| 1267 | rem = (t.task == NULL); |
| 1268 | } |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1269 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1270 | __set_current_state(TASK_RUNNING); |
| 1271 | |
| 1272 | /* |
| 1273 | * NOTE: we don't remove ourselves from the waitqueue because |
| 1274 | * we are the only user of it. |
| 1275 | */ |
| 1276 | |
| 1277 | /* If we were woken (and unqueued), we succeeded, whatever. */ |
| 1278 | if (!unqueue_me(&q)) |
| 1279 | return 0; |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1280 | if (rem) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1281 | return -ETIMEDOUT; |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1282 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1283 | /* |
| 1284 | * We expect signal_pending(current), but another thread may |
| 1285 | * have handled it for us already. |
| 1286 | */ |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1287 | if (!abs_time) |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1288 | return -ERESTARTSYS; |
| 1289 | else { |
| 1290 | struct restart_block *restart; |
| 1291 | restart = ¤t_thread_info()->restart_block; |
| 1292 | restart->fn = futex_wait_restart; |
| 1293 | restart->arg0 = (unsigned long)uaddr; |
| 1294 | restart->arg1 = (unsigned long)val; |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1295 | restart->arg2 = (unsigned long)abs_time; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1296 | restart->arg3 = 0; |
| 1297 | if (fshared) |
| 1298 | restart->arg3 |= ARG3_SHARED; |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1299 | return -ERESTART_RESTARTBLOCK; |
| 1300 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1301 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1302 | out_unlock_release_sem: |
| 1303 | queue_unlock(&q, hb); |
| 1304 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1305 | out_release_sem: |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1306 | futex_unlock_mm(fshared); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1307 | return ret; |
| 1308 | } |
| 1309 | |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1310 | |
| 1311 | static long futex_wait_restart(struct restart_block *restart) |
| 1312 | { |
| 1313 | u32 __user *uaddr = (u32 __user *)restart->arg0; |
| 1314 | u32 val = (u32)restart->arg1; |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1315 | ktime_t *abs_time = (ktime_t *)restart->arg2; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1316 | struct rw_semaphore *fshared = NULL; |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1317 | |
| 1318 | restart->fn = do_no_restart_syscall; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1319 | if (restart->arg3 & ARG3_SHARED) |
| 1320 | fshared = ¤t->mm->mmap_sem; |
| 1321 | return (long)futex_wait(uaddr, fshared, val, abs_time); |
Nick Piggin | 72c1bbf | 2007-05-08 00:26:43 -0700 | [diff] [blame] | 1322 | } |
| 1323 | |
| 1324 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1325 | /* |
| 1326 | * Userspace tried a 0 -> TID atomic transition of the futex value |
| 1327 | * and failed. The kernel side here does the whole locking operation: |
| 1328 | * if there are waiters then it will block, it does PI, etc. (Due to |
| 1329 | * races the kernel might see a 0 value of the futex too.) |
| 1330 | */ |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1331 | static int futex_lock_pi(u32 __user *uaddr, struct rw_semaphore *fshared, |
| 1332 | int detect, ktime_t *time, int trylock) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1333 | { |
Thomas Gleixner | c5780e9 | 2006-09-08 09:47:15 -0700 | [diff] [blame] | 1334 | struct hrtimer_sleeper timeout, *to = NULL; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1335 | struct task_struct *curr = current; |
| 1336 | struct futex_hash_bucket *hb; |
| 1337 | u32 uval, newval, curval; |
| 1338 | struct futex_q q; |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1339 | int ret, lock_taken, ownerdied = 0, attempt = 0; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1340 | |
| 1341 | if (refill_pi_state_cache()) |
| 1342 | return -ENOMEM; |
| 1343 | |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1344 | if (time) { |
Thomas Gleixner | c5780e9 | 2006-09-08 09:47:15 -0700 | [diff] [blame] | 1345 | to = &timeout; |
Thomas Gleixner | c9cb2e3 | 2007-02-16 01:27:49 -0800 | [diff] [blame] | 1346 | hrtimer_init(&to->timer, CLOCK_REALTIME, HRTIMER_MODE_ABS); |
Thomas Gleixner | c5780e9 | 2006-09-08 09:47:15 -0700 | [diff] [blame] | 1347 | hrtimer_init_sleeper(to, current); |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1348 | to->timer.expires = *time; |
Thomas Gleixner | c5780e9 | 2006-09-08 09:47:15 -0700 | [diff] [blame] | 1349 | } |
| 1350 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1351 | q.pi_state = NULL; |
| 1352 | retry: |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1353 | futex_lock_mm(fshared); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1354 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1355 | ret = get_futex_key(uaddr, fshared, &q.key); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1356 | if (unlikely(ret != 0)) |
| 1357 | goto out_release_sem; |
| 1358 | |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1359 | retry_unlocked: |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1360 | hb = queue_lock(&q, -1, NULL); |
| 1361 | |
| 1362 | retry_locked: |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1363 | ret = lock_taken = 0; |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1364 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1365 | /* |
| 1366 | * To avoid races, we attempt to take the lock here again |
| 1367 | * (by doing a 0 -> TID atomic cmpxchg), while holding all |
| 1368 | * the locks. It will most likely not succeed. |
| 1369 | */ |
| 1370 | newval = current->pid; |
| 1371 | |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1372 | curval = cmpxchg_futex_value_locked(uaddr, 0, newval); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1373 | |
| 1374 | if (unlikely(curval == -EFAULT)) |
| 1375 | goto uaddr_faulted; |
| 1376 | |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1377 | /* |
| 1378 | * Detect deadlocks. In case of REQUEUE_PI this is a valid |
| 1379 | * situation and we return success to user space. |
| 1380 | */ |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1381 | if (unlikely((curval & FUTEX_TID_MASK) == current->pid)) { |
Thomas Gleixner | bd19723 | 2007-06-17 21:11:10 +0200 | [diff] [blame] | 1382 | ret = -EDEADLK; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1383 | goto out_unlock_release_sem; |
| 1384 | } |
| 1385 | |
| 1386 | /* |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1387 | * Surprise - we got the lock. Just return to userspace: |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1388 | */ |
| 1389 | if (unlikely(!curval)) |
| 1390 | goto out_unlock_release_sem; |
| 1391 | |
| 1392 | uval = curval; |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1393 | |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1394 | /* |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1395 | * Set the WAITERS flag, so the owner will know it has someone |
| 1396 | * to wake at next unlock |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1397 | */ |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1398 | newval = curval | FUTEX_WAITERS; |
| 1399 | |
| 1400 | /* |
| 1401 | * There are two cases, where a futex might have no owner (the |
Thomas Gleixner | bd19723 | 2007-06-17 21:11:10 +0200 | [diff] [blame] | 1402 | * owner TID is 0): OWNER_DIED. We take over the futex in this |
| 1403 | * case. We also do an unconditional take over, when the owner |
| 1404 | * of the futex died. |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1405 | * |
| 1406 | * This is safe as we are protected by the hash bucket lock ! |
| 1407 | */ |
| 1408 | if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { |
Thomas Gleixner | bd19723 | 2007-06-17 21:11:10 +0200 | [diff] [blame] | 1409 | /* Keep the OWNER_DIED bit */ |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1410 | newval = (curval & ~FUTEX_TID_MASK) | current->pid; |
| 1411 | ownerdied = 0; |
| 1412 | lock_taken = 1; |
| 1413 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1414 | |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1415 | curval = cmpxchg_futex_value_locked(uaddr, uval, newval); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1416 | |
| 1417 | if (unlikely(curval == -EFAULT)) |
| 1418 | goto uaddr_faulted; |
| 1419 | if (unlikely(curval != uval)) |
| 1420 | goto retry_locked; |
| 1421 | |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1422 | /* |
Thomas Gleixner | bd19723 | 2007-06-17 21:11:10 +0200 | [diff] [blame] | 1423 | * We took the lock due to owner died take over. |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1424 | */ |
Thomas Gleixner | bd19723 | 2007-06-17 21:11:10 +0200 | [diff] [blame] | 1425 | if (unlikely(lock_taken)) |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1426 | goto out_unlock_release_sem; |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1427 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1428 | /* |
| 1429 | * We dont have the lock. Look up the PI state (or create it if |
| 1430 | * we are the first waiter): |
| 1431 | */ |
Pierre Peiffer | d0aa7a7 | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1432 | ret = lookup_pi_state(uval, hb, &q.key, &q.pi_state); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1433 | |
| 1434 | if (unlikely(ret)) { |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1435 | switch (ret) { |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1436 | |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1437 | case -EAGAIN: |
| 1438 | /* |
| 1439 | * Task is exiting and we just wait for the |
| 1440 | * exit to complete. |
| 1441 | */ |
| 1442 | queue_unlock(&q, hb); |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1443 | futex_unlock_mm(fshared); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1444 | cond_resched(); |
| 1445 | goto retry; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1446 | |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1447 | case -ESRCH: |
| 1448 | /* |
| 1449 | * No owner found for this futex. Check if the |
| 1450 | * OWNER_DIED bit is set to figure out whether |
| 1451 | * this is a robust futex or not. |
| 1452 | */ |
| 1453 | if (get_futex_value_locked(&curval, uaddr)) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1454 | goto uaddr_faulted; |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1455 | |
| 1456 | /* |
| 1457 | * We simply start over in case of a robust |
| 1458 | * futex. The code above will take the futex |
| 1459 | * and return happy. |
| 1460 | */ |
| 1461 | if (curval & FUTEX_OWNER_DIED) { |
| 1462 | ownerdied = 1; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1463 | goto retry_locked; |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1464 | } |
| 1465 | default: |
| 1466 | goto out_unlock_release_sem; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1467 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1468 | } |
| 1469 | |
| 1470 | /* |
| 1471 | * Only actually queue now that the atomic ops are done: |
| 1472 | */ |
| 1473 | __queue_me(&q, hb); |
| 1474 | |
| 1475 | /* |
| 1476 | * Now the futex is queued and we have checked the data, we |
| 1477 | * don't want to hold mmap_sem while we sleep. |
| 1478 | */ |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1479 | futex_unlock_mm(fshared); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1480 | |
| 1481 | WARN_ON(!q.pi_state); |
| 1482 | /* |
| 1483 | * Block on the PI mutex: |
| 1484 | */ |
| 1485 | if (!trylock) |
| 1486 | ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1); |
| 1487 | else { |
| 1488 | ret = rt_mutex_trylock(&q.pi_state->pi_mutex); |
| 1489 | /* Fixup the trylock return value: */ |
| 1490 | ret = ret ? 0 : -EWOULDBLOCK; |
| 1491 | } |
| 1492 | |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1493 | futex_lock_mm(fshared); |
Vernon Mauery | a99e4e4 | 2006-07-01 04:35:42 -0700 | [diff] [blame] | 1494 | spin_lock(q.lock_ptr); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1495 | |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1496 | if (!ret) { |
| 1497 | /* |
| 1498 | * Got the lock. We might not be the anticipated owner |
| 1499 | * if we did a lock-steal - fix up the PI-state in |
| 1500 | * that case: |
| 1501 | */ |
| 1502 | if (q.pi_state->owner != curr) |
| 1503 | ret = fixup_pi_state_owner(uaddr, &q, curr); |
| 1504 | } else { |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1505 | /* |
| 1506 | * Catch the rare case, where the lock was released |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1507 | * when we were on the way back before we locked the |
| 1508 | * hash bucket. |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1509 | */ |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1510 | if (q.pi_state->owner == curr && |
| 1511 | rt_mutex_trylock(&q.pi_state->pi_mutex)) { |
| 1512 | ret = 0; |
| 1513 | } else { |
| 1514 | /* |
| 1515 | * Paranoia check. If we did not take the lock |
| 1516 | * in the trylock above, then we should not be |
| 1517 | * the owner of the rtmutex, neither the real |
| 1518 | * nor the pending one: |
| 1519 | */ |
| 1520 | if (rt_mutex_owner(&q.pi_state->pi_mutex) == curr) |
| 1521 | printk(KERN_ERR "futex_lock_pi: ret = %d " |
| 1522 | "pi-mutex: %p pi-state %p\n", ret, |
| 1523 | q.pi_state->pi_mutex.owner, |
| 1524 | q.pi_state->owner); |
Ingo 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 | |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1528 | /* Unqueue and drop the lock */ |
| 1529 | unqueue_me_pi(&q); |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1530 | futex_unlock_mm(fshared); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1531 | |
Thomas Gleixner | c5780e9 | 2006-09-08 09:47:15 -0700 | [diff] [blame] | 1532 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1533 | |
| 1534 | out_unlock_release_sem: |
| 1535 | queue_unlock(&q, hb); |
| 1536 | |
| 1537 | out_release_sem: |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1538 | futex_unlock_mm(fshared); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1539 | return ret; |
| 1540 | |
| 1541 | uaddr_faulted: |
| 1542 | /* |
| 1543 | * We have to r/w *(int __user *)uaddr, but we can't modify it |
| 1544 | * non-atomically. Therefore, if get_user below is not |
| 1545 | * enough, we need to handle the fault ourselves, while |
| 1546 | * still holding the mmap_sem. |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1547 | * |
| 1548 | * ... and hb->lock. :-) --ANK |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1549 | */ |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1550 | queue_unlock(&q, hb); |
| 1551 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1552 | if (attempt++) { |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1553 | ret = futex_handle_fault((unsigned long)uaddr, fshared, |
| 1554 | attempt); |
| 1555 | if (ret) |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1556 | goto out_release_sem; |
| 1557 | goto retry_unlocked; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1558 | } |
| 1559 | |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1560 | futex_unlock_mm(fshared); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1561 | |
| 1562 | ret = get_user(uval, uaddr); |
| 1563 | if (!ret && (uval != -EFAULT)) |
| 1564 | goto retry; |
| 1565 | |
| 1566 | return ret; |
| 1567 | } |
| 1568 | |
| 1569 | /* |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1570 | * Userspace attempted a TID -> 0 atomic transition, and failed. |
| 1571 | * This is the in-kernel slowpath: we look up the PI state (if any), |
| 1572 | * and do the rt-mutex unlock. |
| 1573 | */ |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1574 | static int futex_unlock_pi(u32 __user *uaddr, struct rw_semaphore *fshared) |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1575 | { |
| 1576 | struct futex_hash_bucket *hb; |
| 1577 | struct futex_q *this, *next; |
| 1578 | u32 uval; |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 1579 | struct plist_head *head; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1580 | union futex_key key; |
| 1581 | int ret, attempt = 0; |
| 1582 | |
| 1583 | retry: |
| 1584 | if (get_user(uval, uaddr)) |
| 1585 | return -EFAULT; |
| 1586 | /* |
| 1587 | * We release only a lock we actually own: |
| 1588 | */ |
| 1589 | if ((uval & FUTEX_TID_MASK) != current->pid) |
| 1590 | return -EPERM; |
| 1591 | /* |
| 1592 | * First take all the futex related locks: |
| 1593 | */ |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1594 | futex_lock_mm(fshared); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1595 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1596 | ret = get_futex_key(uaddr, fshared, &key); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1597 | if (unlikely(ret != 0)) |
| 1598 | goto out; |
| 1599 | |
| 1600 | hb = hash_futex(&key); |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1601 | retry_unlocked: |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 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)) |
| 1610 | uval = cmpxchg_futex_value_locked(uaddr, current->pid, 0); |
| 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 | */ |
| 1619 | if (unlikely(uval == current->pid)) |
| 1620 | goto out_unlock; |
| 1621 | |
| 1622 | /* |
| 1623 | * Ok, other tasks may need to be woken up - check waiters |
| 1624 | * and do the wakeup if necessary: |
| 1625 | */ |
| 1626 | head = &hb->chain; |
| 1627 | |
Pierre 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); |
| 1652 | out: |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1653 | futex_unlock_mm(fshared); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1654 | |
| 1655 | return ret; |
| 1656 | |
| 1657 | pi_faulted: |
| 1658 | /* |
| 1659 | * We have to r/w *(int __user *)uaddr, but we can't modify it |
| 1660 | * non-atomically. Therefore, if get_user below is not |
| 1661 | * enough, we need to handle the fault ourselves, while |
| 1662 | * still holding the mmap_sem. |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1663 | * |
| 1664 | * ... and hb->lock. --ANK |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1665 | */ |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1666 | spin_unlock(&hb->lock); |
| 1667 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1668 | if (attempt++) { |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1669 | ret = futex_handle_fault((unsigned long)uaddr, fshared, |
| 1670 | attempt); |
| 1671 | if (ret) |
Alexey Kuznetsov | 778e9a9 | 2007-06-08 13:47:00 -0700 | [diff] [blame] | 1672 | goto out; |
| 1673 | goto retry_unlocked; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1674 | } |
| 1675 | |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1676 | futex_unlock_mm(fshared); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1677 | |
| 1678 | ret = get_user(uval, uaddr); |
| 1679 | if (!ret && (uval != -EFAULT)) |
| 1680 | goto retry; |
| 1681 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1682 | return ret; |
| 1683 | } |
| 1684 | |
| 1685 | static int futex_close(struct inode *inode, struct file *filp) |
| 1686 | { |
| 1687 | struct futex_q *q = filp->private_data; |
| 1688 | |
| 1689 | unqueue_me(q); |
| 1690 | kfree(q); |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1691 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1692 | return 0; |
| 1693 | } |
| 1694 | |
| 1695 | /* This is one-shot: once it's gone off you need a new fd */ |
| 1696 | static unsigned int futex_poll(struct file *filp, |
| 1697 | struct poll_table_struct *wait) |
| 1698 | { |
| 1699 | struct futex_q *q = filp->private_data; |
| 1700 | int ret = 0; |
| 1701 | |
| 1702 | poll_wait(filp, &q->waiters, wait); |
| 1703 | |
| 1704 | /* |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 1705 | * plist_node_empty() is safe here without any lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1706 | * q->lock_ptr != 0 is not safe, because of ordering against wakeup. |
| 1707 | */ |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 1708 | if (plist_node_empty(&q->list)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1709 | ret = POLLIN | POLLRDNORM; |
| 1710 | |
| 1711 | return ret; |
| 1712 | } |
| 1713 | |
Helge Deller | 15ad7cd | 2006-12-06 20:40:36 -0800 | [diff] [blame] | 1714 | static const struct file_operations futex_fops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1715 | .release = futex_close, |
| 1716 | .poll = futex_poll, |
| 1717 | }; |
| 1718 | |
| 1719 | /* |
| 1720 | * Signal allows caller to avoid the race which would occur if they |
| 1721 | * set the sigio stuff up afterwards. |
| 1722 | */ |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 1723 | static int futex_fd(u32 __user *uaddr, int signal) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1724 | { |
| 1725 | struct futex_q *q; |
| 1726 | struct file *filp; |
| 1727 | int ret, err; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1728 | struct rw_semaphore *fshared; |
Andrew Morton | 19c6b6e | 2006-11-02 22:07:17 -0800 | [diff] [blame] | 1729 | static unsigned long printk_interval; |
| 1730 | |
| 1731 | if (printk_timed_ratelimit(&printk_interval, 60 * 60 * 1000)) { |
| 1732 | printk(KERN_WARNING "Process `%s' used FUTEX_FD, which " |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1733 | "will be removed from the kernel in June 2007\n", |
| 1734 | current->comm); |
Andrew Morton | 19c6b6e | 2006-11-02 22:07:17 -0800 | [diff] [blame] | 1735 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1736 | |
| 1737 | ret = -EINVAL; |
Jesper Juhl | 7ed20e1 | 2005-05-01 08:59:14 -0700 | [diff] [blame] | 1738 | if (!valid_signal(signal)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1739 | goto out; |
| 1740 | |
| 1741 | ret = get_unused_fd(); |
| 1742 | if (ret < 0) |
| 1743 | goto out; |
| 1744 | filp = get_empty_filp(); |
| 1745 | if (!filp) { |
| 1746 | put_unused_fd(ret); |
| 1747 | ret = -ENFILE; |
| 1748 | goto out; |
| 1749 | } |
| 1750 | filp->f_op = &futex_fops; |
Josef "Jeff" Sipek | f3a43f3 | 2006-12-08 02:36:43 -0800 | [diff] [blame] | 1751 | filp->f_path.mnt = mntget(futex_mnt); |
| 1752 | filp->f_path.dentry = dget(futex_mnt->mnt_root); |
| 1753 | filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1754 | |
| 1755 | if (signal) { |
Eric W. Biederman | 609d7fa | 2006-10-02 02:17:15 -0700 | [diff] [blame] | 1756 | err = __f_setown(filp, task_pid(current), PIDTYPE_PID, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1757 | if (err < 0) { |
Pekka Enberg | 39ed3fd | 2005-09-06 15:17:44 -0700 | [diff] [blame] | 1758 | goto error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1759 | } |
| 1760 | filp->f_owner.signum = signal; |
| 1761 | } |
| 1762 | |
| 1763 | q = kmalloc(sizeof(*q), GFP_KERNEL); |
| 1764 | if (!q) { |
Pekka Enberg | 39ed3fd | 2005-09-06 15:17:44 -0700 | [diff] [blame] | 1765 | err = -ENOMEM; |
| 1766 | goto error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1767 | } |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1768 | q->pi_state = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1769 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1770 | fshared = ¤t->mm->mmap_sem; |
| 1771 | down_read(fshared); |
| 1772 | err = get_futex_key(uaddr, fshared, &q->key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1773 | |
| 1774 | if (unlikely(err != 0)) { |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1775 | up_read(fshared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1776 | kfree(q); |
Pekka Enberg | 39ed3fd | 2005-09-06 15:17:44 -0700 | [diff] [blame] | 1777 | goto error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1778 | } |
| 1779 | |
| 1780 | /* |
| 1781 | * queue_me() must be called before releasing mmap_sem, because |
| 1782 | * key->shared.inode needs to be referenced while holding it. |
| 1783 | */ |
| 1784 | filp->private_data = q; |
| 1785 | |
| 1786 | queue_me(q, ret, filp); |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1787 | up_read(fshared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1788 | |
| 1789 | /* Now we map fd to filp, so userspace can access it */ |
| 1790 | fd_install(ret, filp); |
| 1791 | out: |
| 1792 | return ret; |
Pekka Enberg | 39ed3fd | 2005-09-06 15:17:44 -0700 | [diff] [blame] | 1793 | error: |
| 1794 | put_unused_fd(ret); |
| 1795 | put_filp(filp); |
| 1796 | ret = err; |
| 1797 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1798 | } |
| 1799 | |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1800 | /* |
| 1801 | * Support for robust futexes: the kernel cleans up held futexes at |
| 1802 | * thread exit time. |
| 1803 | * |
| 1804 | * Implementation: user-space maintains a per-thread list of locks it |
| 1805 | * is holding. Upon do_exit(), the kernel carefully walks this list, |
| 1806 | * and marks all locks that are owned by this thread with the |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1807 | * 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] | 1808 | * always manipulated with the lock held, so the list is private and |
| 1809 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' |
| 1810 | * field, to allow the kernel to clean up if the thread dies after |
| 1811 | * acquiring the lock, but just before it could have added itself to |
| 1812 | * the list. There can only be one such pending lock. |
| 1813 | */ |
| 1814 | |
| 1815 | /** |
| 1816 | * sys_set_robust_list - set the robust-futex list head of a task |
| 1817 | * @head: pointer to the list-head |
| 1818 | * @len: length of the list-head, as userspace expects |
| 1819 | */ |
| 1820 | asmlinkage long |
| 1821 | sys_set_robust_list(struct robust_list_head __user *head, |
| 1822 | size_t len) |
| 1823 | { |
| 1824 | /* |
| 1825 | * The kernel knows only one size for now: |
| 1826 | */ |
| 1827 | if (unlikely(len != sizeof(*head))) |
| 1828 | return -EINVAL; |
| 1829 | |
| 1830 | current->robust_list = head; |
| 1831 | |
| 1832 | return 0; |
| 1833 | } |
| 1834 | |
| 1835 | /** |
| 1836 | * sys_get_robust_list - get the robust-futex list head of a task |
| 1837 | * @pid: pid of the process [zero for current task] |
| 1838 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in |
| 1839 | * @len_ptr: pointer to a length field, the kernel fills in the header size |
| 1840 | */ |
| 1841 | asmlinkage long |
Al Viro | ba46df9 | 2006-10-10 22:46:07 +0100 | [diff] [blame] | 1842 | sys_get_robust_list(int pid, struct robust_list_head __user * __user *head_ptr, |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1843 | size_t __user *len_ptr) |
| 1844 | { |
Al Viro | ba46df9 | 2006-10-10 22:46:07 +0100 | [diff] [blame] | 1845 | struct robust_list_head __user *head; |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1846 | unsigned long ret; |
| 1847 | |
| 1848 | if (!pid) |
| 1849 | head = current->robust_list; |
| 1850 | else { |
| 1851 | struct task_struct *p; |
| 1852 | |
| 1853 | ret = -ESRCH; |
Oleg Nesterov | aaa2a97 | 2006-09-29 02:00:55 -0700 | [diff] [blame] | 1854 | rcu_read_lock(); |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1855 | p = find_task_by_pid(pid); |
| 1856 | if (!p) |
| 1857 | goto err_unlock; |
| 1858 | ret = -EPERM; |
| 1859 | if ((current->euid != p->euid) && (current->euid != p->uid) && |
| 1860 | !capable(CAP_SYS_PTRACE)) |
| 1861 | goto err_unlock; |
| 1862 | head = p->robust_list; |
Oleg Nesterov | aaa2a97 | 2006-09-29 02:00:55 -0700 | [diff] [blame] | 1863 | rcu_read_unlock(); |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1864 | } |
| 1865 | |
| 1866 | if (put_user(sizeof(*head), len_ptr)) |
| 1867 | return -EFAULT; |
| 1868 | return put_user(head, head_ptr); |
| 1869 | |
| 1870 | err_unlock: |
Oleg Nesterov | aaa2a97 | 2006-09-29 02:00:55 -0700 | [diff] [blame] | 1871 | rcu_read_unlock(); |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1872 | |
| 1873 | return ret; |
| 1874 | } |
| 1875 | |
| 1876 | /* |
| 1877 | * Process a futex-list entry, check whether it's owned by the |
| 1878 | * dying task, and do notification if so: |
| 1879 | */ |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1880 | 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] | 1881 | { |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1882 | u32 uval, nval, mval; |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1883 | |
Ingo Molnar | 8f17d3a | 2006-03-27 01:16:27 -0800 | [diff] [blame] | 1884 | retry: |
| 1885 | if (get_user(uval, uaddr)) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1886 | return -1; |
| 1887 | |
Ingo Molnar | 8f17d3a | 2006-03-27 01:16:27 -0800 | [diff] [blame] | 1888 | if ((uval & FUTEX_TID_MASK) == curr->pid) { |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1889 | /* |
| 1890 | * Ok, this dying thread is truly holding a futex |
| 1891 | * of interest. Set the OWNER_DIED bit atomically |
| 1892 | * via cmpxchg, and if the value had FUTEX_WAITERS |
| 1893 | * set, wake up a waiter (if any). (We have to do a |
| 1894 | * futex_wake() even if OWNER_DIED is already set - |
| 1895 | * to handle the rare but possible case of recursive |
| 1896 | * thread-death.) The rest of the cleanup is done in |
| 1897 | * userspace. |
| 1898 | */ |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1899 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; |
| 1900 | nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval); |
| 1901 | |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1902 | if (nval == -EFAULT) |
| 1903 | return -1; |
| 1904 | |
| 1905 | if (nval != uval) |
Ingo Molnar | 8f17d3a | 2006-03-27 01:16:27 -0800 | [diff] [blame] | 1906 | goto retry; |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1907 | |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1908 | /* |
| 1909 | * Wake robust non-PI futexes here. The wakeup of |
| 1910 | * PI futexes happens in exit_pi_state(): |
| 1911 | */ |
Thomas Gleixner | 36cf3b5 | 2007-07-15 23:41:20 -0700 | [diff] [blame] | 1912 | if (!pi && (uval & FUTEX_WAITERS)) |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1913 | futex_wake(uaddr, &curr->mm->mmap_sem, 1); |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1914 | } |
| 1915 | return 0; |
| 1916 | } |
| 1917 | |
| 1918 | /* |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1919 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: |
| 1920 | */ |
| 1921 | static inline int fetch_robust_entry(struct robust_list __user **entry, |
Al Viro | ba46df9 | 2006-10-10 22:46:07 +0100 | [diff] [blame] | 1922 | struct robust_list __user * __user *head, |
| 1923 | int *pi) |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1924 | { |
| 1925 | unsigned long uentry; |
| 1926 | |
Al Viro | ba46df9 | 2006-10-10 22:46:07 +0100 | [diff] [blame] | 1927 | if (get_user(uentry, (unsigned long __user *)head)) |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1928 | return -EFAULT; |
| 1929 | |
Al Viro | ba46df9 | 2006-10-10 22:46:07 +0100 | [diff] [blame] | 1930 | *entry = (void __user *)(uentry & ~1UL); |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1931 | *pi = uentry & 1; |
| 1932 | |
| 1933 | return 0; |
| 1934 | } |
| 1935 | |
| 1936 | /* |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1937 | * Walk curr->robust_list (very carefully, it's a userspace list!) |
| 1938 | * and mark any locks found there dead, and notify any waiters. |
| 1939 | * |
| 1940 | * We silently return on any sign of list-walking problem. |
| 1941 | */ |
| 1942 | void exit_robust_list(struct task_struct *curr) |
| 1943 | { |
| 1944 | struct robust_list_head __user *head = curr->robust_list; |
| 1945 | struct robust_list __user *entry, *pending; |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1946 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1947 | unsigned long futex_offset; |
| 1948 | |
| 1949 | /* |
| 1950 | * Fetch the list head (which was registered earlier, via |
| 1951 | * sys_set_robust_list()): |
| 1952 | */ |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1953 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1954 | return; |
| 1955 | /* |
| 1956 | * Fetch the relative futex offset: |
| 1957 | */ |
| 1958 | if (get_user(futex_offset, &head->futex_offset)) |
| 1959 | return; |
| 1960 | /* |
| 1961 | * Fetch any possibly pending lock-add first, and handle it |
| 1962 | * if it exists: |
| 1963 | */ |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1964 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1965 | return; |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1966 | |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1967 | if (pending) |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1968 | handle_futex_death((void __user *)pending + futex_offset, |
| 1969 | curr, pip); |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1970 | |
| 1971 | while (entry != &head->list) { |
| 1972 | /* |
| 1973 | * A pending lock might already be on the list, so |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 1974 | * don't process it twice: |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1975 | */ |
| 1976 | if (entry != pending) |
Al Viro | ba46df9 | 2006-10-10 22:46:07 +0100 | [diff] [blame] | 1977 | if (handle_futex_death((void __user *)entry + futex_offset, |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1978 | curr, pi)) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1979 | return; |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1980 | /* |
| 1981 | * Fetch the next entry in the list: |
| 1982 | */ |
Ingo Molnar | e3f2dde | 2006-07-29 05:17:57 +0200 | [diff] [blame] | 1983 | if (fetch_robust_entry(&entry, &entry->next, &pi)) |
Ingo Molnar | 0771dfe | 2006-03-27 01:16:22 -0800 | [diff] [blame] | 1984 | return; |
| 1985 | /* |
| 1986 | * Avoid excessively long or circular lists: |
| 1987 | */ |
| 1988 | if (!--limit) |
| 1989 | break; |
| 1990 | |
| 1991 | cond_resched(); |
| 1992 | } |
| 1993 | } |
| 1994 | |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 1995 | 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] | 1996 | u32 __user *uaddr2, u32 val2, u32 val3) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1997 | { |
| 1998 | int ret; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 1999 | int cmd = op & FUTEX_CMD_MASK; |
| 2000 | struct rw_semaphore *fshared = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2001 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 2002 | if (!(op & FUTEX_PRIVATE_FLAG)) |
| 2003 | fshared = ¤t->mm->mmap_sem; |
| 2004 | |
| 2005 | switch (cmd) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2006 | case FUTEX_WAIT: |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 2007 | ret = futex_wait(uaddr, fshared, val, timeout); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2008 | break; |
| 2009 | case FUTEX_WAKE: |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 2010 | ret = futex_wake(uaddr, fshared, val); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2011 | break; |
| 2012 | case FUTEX_FD: |
| 2013 | /* non-zero val means F_SETOWN(getpid()) & F_SETSIG(val) */ |
| 2014 | ret = futex_fd(uaddr, val); |
| 2015 | break; |
| 2016 | case FUTEX_REQUEUE: |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 2017 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2018 | break; |
| 2019 | case FUTEX_CMP_REQUEUE: |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 2020 | ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2021 | break; |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 2022 | case FUTEX_WAKE_OP: |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 2023 | ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3); |
Jakub Jelinek | 4732efbe | 2005-09-06 15:16:25 -0700 | [diff] [blame] | 2024 | break; |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 2025 | case FUTEX_LOCK_PI: |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 2026 | ret = futex_lock_pi(uaddr, fshared, val, timeout, 0); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 2027 | break; |
| 2028 | case FUTEX_UNLOCK_PI: |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 2029 | ret = futex_unlock_pi(uaddr, fshared); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 2030 | break; |
| 2031 | case FUTEX_TRYLOCK_PI: |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 2032 | ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1); |
Ingo Molnar | c87e283 | 2006-06-27 02:54:58 -0700 | [diff] [blame] | 2033 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2034 | default: |
| 2035 | ret = -ENOSYS; |
| 2036 | } |
| 2037 | return ret; |
| 2038 | } |
| 2039 | |
| 2040 | |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 2041 | asmlinkage long sys_futex(u32 __user *uaddr, int op, u32 val, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2042 | struct timespec __user *utime, u32 __user *uaddr2, |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 2043 | u32 val3) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2044 | { |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 2045 | struct timespec ts; |
| 2046 | ktime_t t, *tp = NULL; |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 2047 | u32 val2 = 0; |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 2048 | int cmd = op & FUTEX_CMD_MASK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2049 | |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 2050 | if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI)) { |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 2051 | if (copy_from_user(&ts, utime, sizeof(ts)) != 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2052 | return -EFAULT; |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 2053 | if (!timespec_valid(&ts)) |
Thomas Gleixner | 9741ef96 | 2006-03-31 02:31:32 -0800 | [diff] [blame] | 2054 | return -EINVAL; |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 2055 | |
| 2056 | t = timespec_to_ktime(ts); |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 2057 | if (cmd == FUTEX_WAIT) |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 2058 | t = ktime_add(ktime_get(), t); |
| 2059 | tp = &t; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2060 | } |
| 2061 | /* |
Eric Dumazet | 34f01cc | 2007-05-09 02:35:04 -0700 | [diff] [blame] | 2062 | * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2063 | */ |
Thomas Gleixner | bd19723 | 2007-06-17 21:11:10 +0200 | [diff] [blame] | 2064 | if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE) |
Ingo Molnar | e2970f2 | 2006-06-27 02:54:47 -0700 | [diff] [blame] | 2065 | val2 = (u32) (unsigned long) utime; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2066 | |
Pierre Peiffer | c19384b | 2007-05-09 02:35:02 -0700 | [diff] [blame] | 2067 | return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2068 | } |
| 2069 | |
David Howells | 454e239 | 2006-06-23 02:02:57 -0700 | [diff] [blame] | 2070 | static int futexfs_get_sb(struct file_system_type *fs_type, |
| 2071 | int flags, const char *dev_name, void *data, |
| 2072 | struct vfsmount *mnt) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2073 | { |
David Howells | 454e239 | 2006-06-23 02:02:57 -0700 | [diff] [blame] | 2074 | return get_sb_pseudo(fs_type, "futex", NULL, 0xBAD1DEA, mnt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2075 | } |
| 2076 | |
| 2077 | static struct file_system_type futex_fs_type = { |
| 2078 | .name = "futexfs", |
| 2079 | .get_sb = futexfs_get_sb, |
| 2080 | .kill_sb = kill_anon_super, |
| 2081 | }; |
| 2082 | |
| 2083 | static int __init init(void) |
| 2084 | { |
Akinobu Mita | 95362fa | 2006-12-06 20:39:03 -0800 | [diff] [blame] | 2085 | int i = register_filesystem(&futex_fs_type); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2086 | |
Akinobu Mita | 95362fa | 2006-12-06 20:39:03 -0800 | [diff] [blame] | 2087 | if (i) |
| 2088 | return i; |
| 2089 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2090 | futex_mnt = kern_mount(&futex_fs_type); |
Akinobu Mita | 95362fa | 2006-12-06 20:39:03 -0800 | [diff] [blame] | 2091 | if (IS_ERR(futex_mnt)) { |
| 2092 | unregister_filesystem(&futex_fs_type); |
| 2093 | return PTR_ERR(futex_mnt); |
| 2094 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2095 | |
| 2096 | for (i = 0; i < ARRAY_SIZE(futex_queues); i++) { |
Pierre Peiffer | ec92d08 | 2007-05-09 02:35:00 -0700 | [diff] [blame] | 2097 | plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2098 | spin_lock_init(&futex_queues[i].lock); |
| 2099 | } |
| 2100 | return 0; |
| 2101 | } |
| 2102 | __initcall(init); |