diff options
Diffstat (limited to 'fs/notify/inotify/inotify.c')
| -rw-r--r-- | fs/notify/inotify/inotify.c | 933 |
1 files changed, 0 insertions, 933 deletions
diff --git a/fs/notify/inotify/inotify.c b/fs/notify/inotify/inotify.c deleted file mode 100644 index 40b1cf914ccb..000000000000 --- a/fs/notify/inotify/inotify.c +++ /dev/null | |||
| @@ -1,933 +0,0 @@ | |||
| 1 | /* | ||
| 2 | * fs/inotify.c - inode-based file event notifications | ||
| 3 | * | ||
| 4 | * Authors: | ||
| 5 | * John McCutchan <ttb@tentacle.dhs.org> | ||
| 6 | * Robert Love <rml@novell.com> | ||
| 7 | * | ||
| 8 | * Kernel API added by: Amy Griffis <amy.griffis@hp.com> | ||
| 9 | * | ||
| 10 | * Copyright (C) 2005 John McCutchan | ||
| 11 | * Copyright 2006 Hewlett-Packard Development Company, L.P. | ||
| 12 | * | ||
| 13 | * This program is free software; you can redistribute it and/or modify it | ||
| 14 | * under the terms of the GNU General Public License as published by the | ||
| 15 | * Free Software Foundation; either version 2, or (at your option) any | ||
| 16 | * later version. | ||
| 17 | * | ||
| 18 | * This program is distributed in the hope that it will be useful, but | ||
| 19 | * WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 21 | * General Public License for more details. | ||
| 22 | */ | ||
| 23 | |||
| 24 | #include <linux/module.h> | ||
| 25 | #include <linux/kernel.h> | ||
| 26 | #include <linux/spinlock.h> | ||
| 27 | #include <linux/idr.h> | ||
| 28 | #include <linux/slab.h> | ||
| 29 | #include <linux/fs.h> | ||
| 30 | #include <linux/sched.h> | ||
| 31 | #include <linux/init.h> | ||
| 32 | #include <linux/list.h> | ||
| 33 | #include <linux/writeback.h> | ||
| 34 | #include <linux/inotify.h> | ||
| 35 | #include <linux/fsnotify_backend.h> | ||
| 36 | |||
| 37 | static atomic_t inotify_cookie; | ||
| 38 | |||
| 39 | /* | ||
| 40 | * Lock ordering: | ||
| 41 | * | ||
| 42 | * dentry->d_lock (used to keep d_move() away from dentry->d_parent) | ||
| 43 | * iprune_mutex (synchronize shrink_icache_memory()) | ||
| 44 | * inode_lock (protects the super_block->s_inodes list) | ||
| 45 | * inode->inotify_mutex (protects inode->inotify_watches and watches->i_list) | ||
| 46 | * inotify_handle->mutex (protects inotify_handle and watches->h_list) | ||
| 47 | * | ||
| 48 | * The inode->inotify_mutex and inotify_handle->mutex and held during execution | ||
| 49 | * of a caller's event handler. Thus, the caller must not hold any locks | ||
| 50 | * taken in their event handler while calling any of the published inotify | ||
| 51 | * interfaces. | ||
| 52 | */ | ||
| 53 | |||
| 54 | /* | ||
| 55 | * Lifetimes of the three main data structures--inotify_handle, inode, and | ||
| 56 | * inotify_watch--are managed by reference count. | ||
| 57 | * | ||
| 58 | * inotify_handle: Lifetime is from inotify_init() to inotify_destroy(). | ||
| 59 | * Additional references can bump the count via get_inotify_handle() and drop | ||
| 60 | * the count via put_inotify_handle(). | ||
| 61 | * | ||
| 62 | * inotify_watch: for inotify's purposes, lifetime is from inotify_add_watch() | ||
| 63 | * to remove_watch_no_event(). Additional references can bump the count via | ||
| 64 | * get_inotify_watch() and drop the count via put_inotify_watch(). The caller | ||
| 65 | * is reponsible for the final put after receiving IN_IGNORED, or when using | ||
| 66 | * IN_ONESHOT after receiving the first event. Inotify does the final put if | ||
| 67 | * inotify_destroy() is called. | ||
| 68 | * | ||
| 69 | * inode: Pinned so long as the inode is associated with a watch, from | ||
| 70 | * inotify_add_watch() to the final put_inotify_watch(). | ||
| 71 | */ | ||
| 72 | |||
| 73 | /* | ||
| 74 | * struct inotify_handle - represents an inotify instance | ||
| 75 | * | ||
| 76 | * This structure is protected by the mutex 'mutex'. | ||
| 77 | */ | ||
| 78 | struct inotify_handle { | ||
| 79 | struct idr idr; /* idr mapping wd -> watch */ | ||
| 80 | struct mutex mutex; /* protects this bad boy */ | ||
| 81 | struct list_head watches; /* list of watches */ | ||
| 82 | atomic_t count; /* reference count */ | ||
| 83 | u32 last_wd; /* the last wd allocated */ | ||
| 84 | const struct inotify_operations *in_ops; /* inotify caller operations */ | ||
| 85 | }; | ||
| 86 | |||
| 87 | static inline void get_inotify_handle(struct inotify_handle *ih) | ||
| 88 | { | ||
| 89 | atomic_inc(&ih->count); | ||
| 90 | } | ||
| 91 | |||
| 92 | static inline void put_inotify_handle(struct inotify_handle *ih) | ||
| 93 | { | ||
| 94 | if (atomic_dec_and_test(&ih->count)) { | ||
| 95 | idr_destroy(&ih->idr); | ||
| 96 | kfree(ih); | ||
| 97 | } | ||
| 98 | } | ||
| 99 | |||
| 100 | /** | ||
| 101 | * get_inotify_watch - grab a reference to an inotify_watch | ||
| 102 | * @watch: watch to grab | ||
| 103 | */ | ||
| 104 | void get_inotify_watch(struct inotify_watch *watch) | ||
| 105 | { | ||
| 106 | atomic_inc(&watch->count); | ||
| 107 | } | ||
| 108 | EXPORT_SYMBOL_GPL(get_inotify_watch); | ||
| 109 | |||
| 110 | int pin_inotify_watch(struct inotify_watch *watch) | ||
| 111 | { | ||
| 112 | struct super_block *sb = watch->inode->i_sb; | ||
| 113 | spin_lock(&sb_lock); | ||
| 114 | if (sb->s_count >= S_BIAS) { | ||
| 115 | atomic_inc(&sb->s_active); | ||
| 116 | spin_unlock(&sb_lock); | ||
| 117 | atomic_inc(&watch->count); | ||
| 118 | return 1; | ||
| 119 | } | ||
| 120 | spin_unlock(&sb_lock); | ||
| 121 | return 0; | ||
| 122 | } | ||
| 123 | |||
| 124 | /** | ||
| 125 | * put_inotify_watch - decrements the ref count on a given watch. cleans up | ||
| 126 | * watch references if the count reaches zero. inotify_watch is freed by | ||
| 127 | * inotify callers via the destroy_watch() op. | ||
| 128 | * @watch: watch to release | ||
| 129 | */ | ||
| 130 | void put_inotify_watch(struct inotify_watch *watch) | ||
| 131 | { | ||
| 132 | if (atomic_dec_and_test(&watch->count)) { | ||
| 133 | struct inotify_handle *ih = watch->ih; | ||
| 134 | |||
| 135 | iput(watch->inode); | ||
| 136 | ih->in_ops->destroy_watch(watch); | ||
| 137 | put_inotify_handle(ih); | ||
| 138 | } | ||
| 139 | } | ||
| 140 | EXPORT_SYMBOL_GPL(put_inotify_watch); | ||
| 141 | |||
| 142 | void unpin_inotify_watch(struct inotify_watch *watch) | ||
| 143 | { | ||
| 144 | struct super_block *sb = watch->inode->i_sb; | ||
| 145 | put_inotify_watch(watch); | ||
| 146 | deactivate_super(sb); | ||
| 147 | } | ||
| 148 | |||
| 149 | /* | ||
| 150 | * inotify_handle_get_wd - returns the next WD for use by the given handle | ||
| 151 | * | ||
| 152 | * Callers must hold ih->mutex. This function can sleep. | ||
| 153 | */ | ||
| 154 | static int inotify_handle_get_wd(struct inotify_handle *ih, | ||
| 155 | struct inotify_watch *watch) | ||
| 156 | { | ||
| 157 | int ret; | ||
| 158 | |||
| 159 | do { | ||
| 160 | if (unlikely(!idr_pre_get(&ih->idr, GFP_NOFS))) | ||
| 161 | return -ENOSPC; | ||
| 162 | ret = idr_get_new_above(&ih->idr, watch, ih->last_wd+1, &watch->wd); | ||
| 163 | } while (ret == -EAGAIN); | ||
| 164 | |||
| 165 | if (likely(!ret)) | ||
| 166 | ih->last_wd = watch->wd; | ||
| 167 | |||
| 168 | return ret; | ||
| 169 | } | ||
| 170 | |||
| 171 | /* | ||
| 172 | * inotify_inode_watched - returns nonzero if there are watches on this inode | ||
| 173 | * and zero otherwise. We call this lockless, we do not care if we race. | ||
| 174 | */ | ||
| 175 | static inline int inotify_inode_watched(struct inode *inode) | ||
| 176 | { | ||
| 177 | return !list_empty(&inode->inotify_watches); | ||
| 178 | } | ||
| 179 | |||
| 180 | /* | ||
| 181 | * Get child dentry flag into synch with parent inode. | ||
| 182 | * Flag should always be clear for negative dentrys. | ||
| 183 | */ | ||
| 184 | static void set_dentry_child_flags(struct inode *inode, int watched) | ||
| 185 | { | ||
| 186 | struct dentry *alias; | ||
| 187 | |||
| 188 | spin_lock(&dcache_lock); | ||
| 189 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { | ||
| 190 | struct dentry *child; | ||
| 191 | |||
| 192 | list_for_each_entry(child, &alias->d_subdirs, d_u.d_child) { | ||
| 193 | if (!child->d_inode) | ||
| 194 | continue; | ||
| 195 | |||
| 196 | spin_lock(&child->d_lock); | ||
| 197 | if (watched) | ||
| 198 | child->d_flags |= DCACHE_INOTIFY_PARENT_WATCHED; | ||
| 199 | else | ||
| 200 | child->d_flags &=~DCACHE_INOTIFY_PARENT_WATCHED; | ||
| 201 | spin_unlock(&child->d_lock); | ||
| 202 | } | ||
| 203 | } | ||
| 204 | spin_unlock(&dcache_lock); | ||
| 205 | } | ||
| 206 | |||
| 207 | /* | ||
| 208 | * inotify_find_handle - find the watch associated with the given inode and | ||
| 209 | * handle | ||
| 210 | * | ||
| 211 | * Callers must hold inode->inotify_mutex. | ||
| 212 | */ | ||
| 213 | static struct inotify_watch *inode_find_handle(struct inode *inode, | ||
| 214 | struct inotify_handle *ih) | ||
| 215 | { | ||
| 216 | struct inotify_watch *watch; | ||
| 217 | |||
| 218 | list_for_each_entry(watch, &inode->inotify_watches, i_list) { | ||
| 219 | if (watch->ih == ih) | ||
| 220 | return watch; | ||
| 221 | } | ||
| 222 | |||
| 223 | return NULL; | ||
| 224 | } | ||
| 225 | |||
| 226 | /* | ||
| 227 | * remove_watch_no_event - remove watch without the IN_IGNORED event. | ||
| 228 | * | ||
| 229 | * Callers must hold both inode->inotify_mutex and ih->mutex. | ||
| 230 | */ | ||
| 231 | static void remove_watch_no_event(struct inotify_watch *watch, | ||
| 232 | struct inotify_handle *ih) | ||
| 233 | { | ||
| 234 | list_del(&watch->i_list); | ||
| 235 | list_del(&watch->h_list); | ||
| 236 | |||
| 237 | if (!inotify_inode_watched(watch->inode)) | ||
| 238 | set_dentry_child_flags(watch->inode, 0); | ||
| 239 | |||
| 240 | idr_remove(&ih->idr, watch->wd); | ||
| 241 | } | ||
| 242 | |||
| 243 | /** | ||
| 244 | * inotify_remove_watch_locked - Remove a watch from both the handle and the | ||
| 245 | * inode. Sends the IN_IGNORED event signifying that the inode is no longer | ||
| 246 | * watched. May be invoked from a caller's event handler. | ||
| 247 | * @ih: inotify handle associated with watch | ||
| 248 | * @watch: watch to remove | ||
| 249 | * | ||
| 250 | * Callers must hold both inode->inotify_mutex and ih->mutex. | ||
| 251 | */ | ||
| 252 | void inotify_remove_watch_locked(struct inotify_handle *ih, | ||
| 253 | struct inotify_watch *watch) | ||
| 254 | { | ||
| 255 | remove_watch_no_event(watch, ih); | ||
| 256 | ih->in_ops->handle_event(watch, watch->wd, IN_IGNORED, 0, NULL, NULL); | ||
| 257 | } | ||
| 258 | EXPORT_SYMBOL_GPL(inotify_remove_watch_locked); | ||
| 259 | |||
| 260 | /* Kernel API for producing events */ | ||
| 261 | |||
| 262 | /* | ||
| 263 | * inotify_d_instantiate - instantiate dcache entry for inode | ||
| 264 | */ | ||
| 265 | void inotify_d_instantiate(struct dentry *entry, struct inode *inode) | ||
| 266 | { | ||
| 267 | struct dentry *parent; | ||
| 268 | |||
| 269 | if (!inode) | ||
| 270 | return; | ||
| 271 | |||
| 272 | spin_lock(&entry->d_lock); | ||
| 273 | parent = entry->d_parent; | ||
| 274 | if (parent->d_inode && inotify_inode_watched(parent->d_inode)) | ||
| 275 | entry->d_flags |= DCACHE_INOTIFY_PARENT_WATCHED; | ||
| 276 | spin_unlock(&entry->d_lock); | ||
| 277 | } | ||
| 278 | |||
| 279 | /* | ||
| 280 | * inotify_d_move - dcache entry has been moved | ||
| 281 | */ | ||
| 282 | void inotify_d_move(struct dentry *entry) | ||
| 283 | { | ||
| 284 | struct dentry *parent; | ||
| 285 | |||
| 286 | parent = entry->d_parent; | ||
| 287 | if (inotify_inode_watched(parent->d_inode)) | ||
| 288 | entry->d_flags |= DCACHE_INOTIFY_PARENT_WATCHED; | ||
| 289 | else | ||
| 290 | entry->d_flags &= ~DCACHE_INOTIFY_PARENT_WATCHED; | ||
| 291 | } | ||
| 292 | |||
| 293 | /** | ||
| 294 | * inotify_inode_queue_event - queue an event to all watches on this inode | ||
| 295 | * @inode: inode event is originating from | ||
| 296 | * @mask: event mask describing this event | ||
| 297 | * @cookie: cookie for synchronization, or zero | ||
| 298 | * @name: filename, if any | ||
| 299 | * @n_inode: inode associated with name | ||
| 300 | */ | ||
| 301 | void inotify_inode_queue_event(struct inode *inode, u32 mask, u32 cookie, | ||
| 302 | const char *name, struct inode *n_inode) | ||
| 303 | { | ||
| 304 | struct inotify_watch *watch, *next; | ||
| 305 | |||
| 306 | if (!inotify_inode_watched(inode)) | ||
| 307 | return; | ||
| 308 | |||
| 309 | mutex_lock(&inode->inotify_mutex); | ||
| 310 | list_for_each_entry_safe(watch, next, &inode->inotify_watches, i_list) { | ||
| 311 | u32 watch_mask = watch->mask; | ||
| 312 | if (watch_mask & mask) { | ||
| 313 | struct inotify_handle *ih= watch->ih; | ||
| 314 | mutex_lock(&ih->mutex); | ||
| 315 | if (watch_mask & IN_ONESHOT) | ||
| 316 | remove_watch_no_event(watch, ih); | ||
| 317 | ih->in_ops->handle_event(watch, watch->wd, mask, cookie, | ||
| 318 | name, n_inode); | ||
| 319 | mutex_unlock(&ih->mutex); | ||
| 320 | } | ||
| 321 | } | ||
| 322 | mutex_unlock(&inode->inotify_mutex); | ||
| 323 | } | ||
| 324 | EXPORT_SYMBOL_GPL(inotify_inode_queue_event); | ||
| 325 | |||
| 326 | /** | ||
| 327 | * inotify_dentry_parent_queue_event - queue an event to a dentry's parent | ||
| 328 | * @dentry: the dentry in question, we queue against this dentry's parent | ||
| 329 | * @mask: event mask describing this event | ||
| 330 | * @cookie: cookie for synchronization, or zero | ||
| 331 | * @name: filename, if any | ||
| 332 | */ | ||
| 333 | void inotify_dentry_parent_queue_event(struct dentry *dentry, u32 mask, | ||
| 334 | u32 cookie, const char *name) | ||
| 335 | { | ||
| 336 | struct dentry *parent; | ||
| 337 | struct inode *inode; | ||
| 338 | |||
| 339 | if (!(dentry->d_flags & DCACHE_INOTIFY_PARENT_WATCHED)) | ||
| 340 | return; | ||
| 341 | |||
| 342 | spin_lock(&dentry->d_lock); | ||
| 343 | parent = dentry->d_parent; | ||
| 344 | inode = parent->d_inode; | ||
| 345 | |||
| 346 | if (inotify_inode_watched(inode)) { | ||
| 347 | dget(parent); | ||
| 348 | spin_unlock(&dentry->d_lock); | ||
| 349 | inotify_inode_queue_event(inode, mask, cookie, name, | ||
| 350 | dentry->d_inode); | ||
| 351 | dput(parent); | ||
| 352 | } else | ||
| 353 | spin_unlock(&dentry->d_lock); | ||
| 354 | } | ||
| 355 | EXPORT_SYMBOL_GPL(inotify_dentry_parent_queue_event); | ||
| 356 | |||
| 357 | /** | ||
| 358 | * inotify_get_cookie - return a unique cookie for use in synchronizing events. | ||
| 359 | */ | ||
| 360 | u32 inotify_get_cookie(void) | ||
| 361 | { | ||
| 362 | return atomic_inc_return(&inotify_cookie); | ||
| 363 | } | ||
| 364 | EXPORT_SYMBOL_GPL(inotify_get_cookie); | ||
| 365 | |||
| 366 | /** | ||
| 367 | * inotify_unmount_inodes - an sb is unmounting. handle any watched inodes. | ||
| 368 | * @list: list of inodes being unmounted (sb->s_inodes) | ||
| 369 | * | ||
| 370 | * Called with inode_lock held, protecting the unmounting super block's list | ||
| 371 | * of inodes, and with iprune_mutex held, keeping shrink_icache_memory() at bay. | ||
| 372 | * We temporarily drop inode_lock, however, and CAN block. | ||
| 373 | */ | ||
| 374 | void inotify_unmount_inodes(struct list_head *list) | ||
| 375 | { | ||
| 376 | struct inode *inode, *next_i, *need_iput = NULL; | ||
| 377 | |||
| 378 | list_for_each_entry_safe(inode, next_i, list, i_sb_list) { | ||
| 379 | struct inotify_watch *watch, *next_w; | ||
| 380 | struct inode *need_iput_tmp; | ||
| 381 | struct list_head *watches; | ||
| 382 | |||
| 383 | /* | ||
| 384 | * We cannot __iget() an inode in state I_CLEAR, I_FREEING, | ||
| 385 | * I_WILL_FREE, or I_NEW which is fine because by that point | ||
| 386 | * the inode cannot have any associated watches. | ||
| 387 | */ | ||
| 388 | if (inode->i_state & (I_CLEAR|I_FREEING|I_WILL_FREE|I_NEW)) | ||
| 389 | continue; | ||
| 390 | |||
| 391 | /* | ||
| 392 | * If i_count is zero, the inode cannot have any watches and | ||
| 393 | * doing an __iget/iput with MS_ACTIVE clear would actually | ||
| 394 | * evict all inodes with zero i_count from icache which is | ||
| 395 | * unnecessarily violent and may in fact be illegal to do. | ||
| 396 | */ | ||
| 397 | if (!atomic_read(&inode->i_count)) | ||
| 398 | continue; | ||
| 399 | |||
| 400 | need_iput_tmp = need_iput; | ||
| 401 | need_iput = NULL; | ||
| 402 | /* In case inotify_remove_watch_locked() drops a reference. */ | ||
| 403 | if (inode != need_iput_tmp) | ||
| 404 | __iget(inode); | ||
| 405 | else | ||
| 406 | need_iput_tmp = NULL; | ||
| 407 | /* In case the dropping of a reference would nuke next_i. */ | ||
| 408 | if ((&next_i->i_sb_list != list) && | ||
| 409 | atomic_read(&next_i->i_count) && | ||
| 410 | !(next_i->i_state & (I_CLEAR | I_FREEING | | ||
| 411 | I_WILL_FREE))) { | ||
| 412 | __iget(next_i); | ||
| 413 | need_iput = next_i; | ||
| 414 | } | ||
| 415 | |||
| 416 | /* | ||
| 417 | * We can safely drop inode_lock here because we hold | ||
| 418 | * references on both inode and next_i. Also no new inodes | ||
| 419 | * will be added since the umount has begun. Finally, | ||
| 420 | * iprune_mutex keeps shrink_icache_memory() away. | ||
| 421 | */ | ||
| 422 | spin_unlock(&inode_lock); | ||
| 423 | |||
| 424 | if (need_iput_tmp) | ||
| 425 | iput(need_iput_tmp); | ||
| 426 | |||
| 427 | /* for each watch, send IN_UNMOUNT and then remove it */ | ||
| 428 | mutex_lock(&inode->inotify_mutex); | ||
| 429 | watches = &inode->inotify_watches; | ||
| 430 | list_for_each_entry_safe(watch, next_w, watches, i_list) { | ||
| 431 | struct inotify_handle *ih= watch->ih; | ||
| 432 | get_inotify_watch(watch); | ||
| 433 | mutex_lock(&ih->mutex); | ||
| 434 | ih->in_ops->handle_event(watch, watch->wd, IN_UNMOUNT, 0, | ||
| 435 | NULL, NULL); | ||
| 436 | inotify_remove_watch_locked(ih, watch); | ||
| 437 | mutex_unlock(&ih->mutex); | ||
| 438 | put_inotify_watch(watch); | ||
| 439 | } | ||
| 440 | mutex_unlock(&inode->inotify_mutex); | ||
| 441 | iput(inode); | ||
| 442 | |||
| 443 | spin_lock(&inode_lock); | ||
| 444 | } | ||
| 445 | } | ||
| 446 | EXPORT_SYMBOL_GPL(inotify_unmount_inodes); | ||
| 447 | |||
| 448 | /** | ||
| 449 | * inotify_inode_is_dead - an inode has been deleted, cleanup any watches | ||
| 450 | * @inode: inode that is about to be removed | ||
| 451 | */ | ||
| 452 | void inotify_inode_is_dead(struct inode *inode) | ||
| 453 | { | ||
| 454 | struct inotify_watch *watch, *next; | ||
| 455 | |||
| 456 | mutex_lock(&inode->inotify_mutex); | ||
| 457 | list_for_each_entry_safe(watch, next, &inode->inotify_watches, i_list) { | ||
| 458 | struct inotify_handle *ih = watch->ih; | ||
| 459 | mutex_lock(&ih->mutex); | ||
| 460 | inotify_remove_watch_locked(ih, watch); | ||
| 461 | mutex_unlock(&ih->mutex); | ||
| 462 | } | ||
| 463 | mutex_unlock(&inode->inotify_mutex); | ||
| 464 | } | ||
| 465 | EXPORT_SYMBOL_GPL(inotify_inode_is_dead); | ||
| 466 | |||
| 467 | /* Kernel Consumer API */ | ||
| 468 | |||
| 469 | /** | ||
| 470 | * inotify_init - allocate and initialize an inotify instance | ||
| 471 | * @ops: caller's inotify operations | ||
| 472 | */ | ||
| 473 | struct inotify_handle *inotify_init(const struct inotify_operations *ops) | ||
| 474 | { | ||
| 475 | struct inotify_handle *ih; | ||
| 476 | |||
| 477 | ih = kmalloc(sizeof(struct inotify_handle), GFP_KERNEL); | ||
| 478 | if (unlikely(!ih)) | ||
| 479 | return ERR_PTR(-ENOMEM); | ||
| 480 | |||
| 481 | idr_init(&ih->idr); | ||
| 482 | INIT_LIST_HEAD(&ih->watches); | ||
| 483 | mutex_init(&ih->mutex); | ||
| 484 | ih->last_wd = 0; | ||
| 485 | ih->in_ops = ops; | ||
| 486 | atomic_set(&ih->count, 0); | ||
| 487 | get_inotify_handle(ih); | ||
| 488 | |||
| 489 | return ih; | ||
| 490 | } | ||
| 491 | EXPORT_SYMBOL_GPL(inotify_init); | ||
| 492 | |||
| 493 | /** | ||
| 494 | * inotify_init_watch - initialize an inotify watch | ||
| 495 | * @watch: watch to initialize | ||
| 496 | */ | ||
| 497 | void inotify_init_watch(struct inotify_watch *watch) | ||
| 498 | { | ||
| 499 | INIT_LIST_HEAD(&watch->h_list); | ||
| 500 | INIT_LIST_HEAD(&watch->i_list); | ||
| 501 | atomic_set(&watch->count, 0); | ||
| 502 | get_inotify_watch(watch); /* initial get */ | ||
| 503 | } | ||
| 504 | EXPORT_SYMBOL_GPL(inotify_init_watch); | ||
| 505 | |||
| 506 | /* | ||
| 507 | * Watch removals suck violently. To kick the watch out we need (in this | ||
| 508 | * order) inode->inotify_mutex and ih->mutex. That's fine if we have | ||
| 509 | * a hold on inode; however, for all other cases we need to make damn sure | ||
| 510 | * we don't race with umount. We can *NOT* just grab a reference to a | ||
| 511 | * watch - inotify_unmount_inodes() will happily sail past it and we'll end | ||
| 512 | * with reference to inode potentially outliving its superblock. Ideally | ||
| 513 | * we just want to grab an active reference to superblock if we can; that | ||
| 514 | * will make sure we won't go into inotify_umount_inodes() until we are | ||
| 515 | * done. Cleanup is just deactivate_super(). However, that leaves a messy | ||
| 516 | * case - what if we *are* racing with umount() and active references to | ||
| 517 | * superblock can't be acquired anymore? We can bump ->s_count, grab | ||
| 518 | * ->s_umount, which will almost certainly wait until the superblock is shut | ||
| 519 | * down and the watch in question is pining for fjords. That's fine, but | ||
| 520 | * there is a problem - we might have hit the window between ->s_active | ||
| 521 | * getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock | ||
| 522 | * is past the point of no return and is heading for shutdown) and the | ||
| 523 | * moment when deactivate_super() acquires ->s_umount. We could just do | ||
| 524 | * drop_super() yield() and retry, but that's rather antisocial and this | ||
| 525 | * stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having | ||
| 526 | * found that we'd got there first (i.e. that ->s_root is non-NULL) we know | ||
| 527 | * that we won't race with inotify_umount_inodes(). So we could grab a | ||
| 528 | * reference to watch and do the rest as above, just with drop_super() instead | ||
| 529 | * of deactivate_super(), right? Wrong. We had to drop ih->mutex before we | ||
| 530 | * could grab ->s_umount. So the watch could've been gone already. | ||
| 531 | * | ||
| 532 | * That still can be dealt with - we need to save watch->wd, do idr_find() | ||
| 533 | * and compare its result with our pointer. If they match, we either have | ||
| 534 | * the damn thing still alive or we'd lost not one but two races at once, | ||
| 535 | * the watch had been killed and a new one got created with the same ->wd | ||
| 536 | * at the same address. That couldn't have happened in inotify_destroy(), | ||
| 537 | * but inotify_rm_wd() could run into that. Still, "new one got created" | ||
| 538 | * is not a problem - we have every right to kill it or leave it alone, | ||
| 539 | * whatever's more convenient. | ||
| 540 | * | ||
| 541 | * So we can use idr_find(...) == watch && watch->inode->i_sb == sb as | ||
| 542 | * "grab it and kill it" check. If it's been our original watch, we are | ||
| 543 | * fine, if it's a newcomer - nevermind, just pretend that we'd won the | ||
| 544 | * race and kill the fscker anyway; we are safe since we know that its | ||
| 545 | * superblock won't be going away. | ||
| 546 | * | ||
| 547 | * And yes, this is far beyond mere "not very pretty"; so's the entire | ||
| 548 | * concept of inotify to start with. | ||
| 549 | */ | ||
| 550 | |||
| 551 | /** | ||
| 552 | * pin_to_kill - pin the watch down for removal | ||
| 553 | * @ih: inotify handle | ||
| 554 | * @watch: watch to kill | ||
| 555 | * | ||
| 556 | * Called with ih->mutex held, drops it. Possible return values: | ||
| 557 | * 0 - nothing to do, it has died | ||
| 558 | * 1 - remove it, drop the reference and deactivate_super() | ||
| 559 | * 2 - remove it, drop the reference and drop_super(); we tried hard to avoid | ||
| 560 | * that variant, since it involved a lot of PITA, but that's the best that | ||
| 561 | * could've been done. | ||
| 562 | */ | ||
| 563 | static int pin_to_kill(struct inotify_handle *ih, struct inotify_watch *watch) | ||
| 564 | { | ||
| 565 | struct super_block *sb = watch->inode->i_sb; | ||
| 566 | s32 wd = watch->wd; | ||
| 567 | |||
| 568 | spin_lock(&sb_lock); | ||
| 569 | if (sb->s_count >= S_BIAS) { | ||
| 570 | atomic_inc(&sb->s_active); | ||
| 571 | spin_unlock(&sb_lock); | ||
| 572 | get_inotify_watch(watch); | ||
| 573 | mutex_unlock(&ih->mutex); | ||
| 574 | return 1; /* the best outcome */ | ||
| 575 | } | ||
| 576 | sb->s_count++; | ||
| 577 | spin_unlock(&sb_lock); | ||
| 578 | mutex_unlock(&ih->mutex); /* can't grab ->s_umount under it */ | ||
| 579 | down_read(&sb->s_umount); | ||
| 580 | if (likely(!sb->s_root)) { | ||
| 581 | /* fs is already shut down; the watch is dead */ | ||
| 582 | drop_super(sb); | ||
| 583 | return 0; | ||
| 584 | } | ||
| 585 | /* raced with the final deactivate_super() */ | ||
| 586 | mutex_lock(&ih->mutex); | ||
| 587 | if (idr_find(&ih->idr, wd) != watch || watch->inode->i_sb != sb) { | ||
| 588 | /* the watch is dead */ | ||
| 589 | mutex_unlock(&ih->mutex); | ||
| 590 | drop_super(sb); | ||
| 591 | return 0; | ||
| 592 | } | ||
| 593 | /* still alive or freed and reused with the same sb and wd; kill */ | ||
| 594 | get_inotify_watch(watch); | ||
| 595 | mutex_unlock(&ih->mutex); | ||
| 596 | return 2; | ||
| 597 | } | ||
| 598 | |||
| 599 | static void unpin_and_kill(struct inotify_watch *watch, int how) | ||
| 600 | { | ||
| 601 | struct super_block *sb = watch->inode->i_sb; | ||
| 602 | put_inotify_watch(watch); | ||
| 603 | switch (how) { | ||
| 604 | case 1: | ||
| 605 | deactivate_super(sb); | ||
| 606 | break; | ||
| 607 | case 2: | ||
| 608 | drop_super(sb); | ||
| 609 | } | ||
| 610 | } | ||
| 611 | |||
| 612 | /** | ||
| 613 | * inotify_destroy - clean up and destroy an inotify instance | ||
| 614 | * @ih: inotify handle | ||
| 615 | */ | ||
| 616 | void inotify_destroy(struct inotify_handle *ih) | ||
| 617 | { | ||
| 618 | /* | ||
| 619 | * Destroy all of the watches for this handle. Unfortunately, not very | ||
| 620 | * pretty. We cannot do a simple iteration over the list, because we | ||
| 621 | * do not know the inode until we iterate to the watch. But we need to | ||
| 622 | * hold inode->inotify_mutex before ih->mutex. The following works. | ||
| 623 | * | ||
| 624 | * AV: it had to become even uglier to start working ;-/ | ||
| 625 | */ | ||
| 626 | while (1) { | ||
| 627 | struct inotify_watch *watch; | ||
| 628 | struct list_head *watches; | ||
| 629 | struct super_block *sb; | ||
| 630 | struct inode *inode; | ||
| 631 | int how; | ||
| 632 | |||
| 633 | mutex_lock(&ih->mutex); | ||
| 634 | watches = &ih->watches; | ||
| 635 | if (list_empty(watches)) { | ||
| 636 | mutex_unlock(&ih->mutex); | ||
| 637 | break; | ||
| 638 | } | ||
| 639 | watch = list_first_entry(watches, struct inotify_watch, h_list); | ||
| 640 | sb = watch->inode->i_sb; | ||
| 641 | how = pin_to_kill(ih, watch); | ||
| 642 | if (!how) | ||
| 643 | continue; | ||
| 644 | |||
| 645 | inode = watch->inode; | ||
| 646 | mutex_lock(&inode->inotify_mutex); | ||
| 647 | mutex_lock(&ih->mutex); | ||
| 648 | |||
| 649 | /* make sure we didn't race with another list removal */ | ||
| 650 | if (likely(idr_find(&ih->idr, watch->wd))) { | ||
| 651 | remove_watch_no_event(watch, ih); | ||
| 652 | put_inotify_watch(watch); | ||
| 653 | } | ||
| 654 | |||
| 655 | mutex_unlock(&ih->mutex); | ||
| 656 | mutex_unlock(&inode->inotify_mutex); | ||
| 657 | unpin_and_kill(watch, how); | ||
| 658 | } | ||
| 659 | |||
| 660 | /* free this handle: the put matching the get in inotify_init() */ | ||
| 661 | put_inotify_handle(ih); | ||
| 662 | } | ||
| 663 | EXPORT_SYMBOL_GPL(inotify_destroy); | ||
| 664 | |||
| 665 | /** | ||
| 666 | * inotify_find_watch - find an existing watch for an (ih,inode) pair | ||
| 667 | * @ih: inotify handle | ||
| 668 | * @inode: inode to watch | ||
| 669 | * @watchp: pointer to existing inotify_watch | ||
| 670 | * | ||
| 671 | * Caller must pin given inode (via nameidata). | ||
| 672 | */ | ||
| 673 | s32 inotify_find_watch(struct inotify_handle *ih, struct inode *inode, | ||
| 674 | struct inotify_watch **watchp) | ||
| 675 | { | ||
| 676 | struct inotify_watch *old; | ||
| 677 | int ret = -ENOENT; | ||
| 678 | |||
| 679 | mutex_lock(&inode->inotify_mutex); | ||
| 680 | mutex_lock(&ih->mutex); | ||
| 681 | |||
| 682 | old = inode_find_handle(inode, ih); | ||
| 683 | if (unlikely(old)) { | ||
| 684 | get_inotify_watch(old); /* caller must put watch */ | ||
| 685 | *watchp = old; | ||
| 686 | ret = old->wd; | ||
| 687 | } | ||
| 688 | |||
| 689 | mutex_unlock(&ih->mutex); | ||
| 690 | mutex_unlock(&inode->inotify_mutex); | ||
| 691 | |||
| 692 | return ret; | ||
| 693 | } | ||
| 694 | EXPORT_SYMBOL_GPL(inotify_find_watch); | ||
| 695 | |||
| 696 | /** | ||
| 697 | * inotify_find_update_watch - find and update the mask of an existing watch | ||
| 698 | * @ih: inotify handle | ||
| 699 | * @inode: inode's watch to update | ||
| 700 | * @mask: mask of events to watch | ||
| 701 | * | ||
| 702 | * Caller must pin given inode (via nameidata). | ||
| 703 | */ | ||
| 704 | s32 inotify_find_update_watch(struct inotify_handle *ih, struct inode *inode, | ||
| 705 | u32 mask) | ||
| 706 | { | ||
| 707 | struct inotify_watch *old; | ||
| 708 | int mask_add = 0; | ||
| 709 | int ret; | ||
| 710 | |||
| 711 | if (mask & IN_MASK_ADD) | ||
| 712 | mask_add = 1; | ||
| 713 | |||
| 714 | /* don't allow invalid bits: we don't want flags set */ | ||
| 715 | mask &= IN_ALL_EVENTS | IN_ONESHOT; | ||
| 716 | if (unlikely(!mask)) | ||
| 717 | return -EINVAL; | ||
| 718 | |||
| 719 | mutex_lock(&inode->inotify_mutex); | ||
| 720 | mutex_lock(&ih->mutex); | ||
| 721 | |||
| 722 | /* | ||
| 723 | * Handle the case of re-adding a watch on an (inode,ih) pair that we | ||
| 724 | * are already watching. We just update the mask and return its wd. | ||
| 725 | */ | ||
| 726 | old = inode_find_handle(inode, ih); | ||
| 727 | if (unlikely(!old)) { | ||
| 728 | ret = -ENOENT; | ||
| 729 | goto out; | ||
| 730 | } | ||
| 731 | |||
| 732 | if (mask_add) | ||
| 733 | old->mask |= mask; | ||
| 734 | else | ||
| 735 | old->mask = mask; | ||
| 736 | ret = old->wd; | ||
| 737 | out: | ||
| 738 | mutex_unlock(&ih->mutex); | ||
| 739 | mutex_unlock(&inode->inotify_mutex); | ||
| 740 | return ret; | ||
| 741 | } | ||
| 742 | EXPORT_SYMBOL_GPL(inotify_find_update_watch); | ||
| 743 | |||
| 744 | /** | ||
| 745 | * inotify_add_watch - add a watch to an inotify instance | ||
| 746 | * @ih: inotify handle | ||
| 747 | * @watch: caller allocated watch structure | ||
| 748 | * @inode: inode to watch | ||
| 749 | * @mask: mask of events to watch | ||
| 750 | * | ||
| 751 | * Caller must pin given inode (via nameidata). | ||
| 752 | * Caller must ensure it only calls inotify_add_watch() once per watch. | ||
| 753 | * Calls inotify_handle_get_wd() so may sleep. | ||
| 754 | */ | ||
| 755 | s32 inotify_add_watch(struct inotify_handle *ih, struct inotify_watch *watch, | ||
| 756 | struct inode *inode, u32 mask) | ||
| 757 | { | ||
| 758 | int ret = 0; | ||
| 759 | int newly_watched; | ||
| 760 | |||
| 761 | /* don't allow invalid bits: we don't want flags set */ | ||
| 762 | mask &= IN_ALL_EVENTS | IN_ONESHOT; | ||
| 763 | if (unlikely(!mask)) | ||
| 764 | return -EINVAL; | ||
| 765 | watch->mask = mask; | ||
| 766 | |||
| 767 | mutex_lock(&inode->inotify_mutex); | ||
| 768 | mutex_lock(&ih->mutex); | ||
| 769 | |||
| 770 | /* Initialize a new watch */ | ||
| 771 | ret = inotify_handle_get_wd(ih, watch); | ||
| 772 | if (unlikely(ret)) | ||
| 773 | goto out; | ||
| 774 | ret = watch->wd; | ||
| 775 | |||
| 776 | /* save a reference to handle and bump the count to make it official */ | ||
| 777 | get_inotify_handle(ih); | ||
| 778 | watch->ih = ih; | ||
| 779 | |||
| 780 | /* | ||
| 781 | * Save a reference to the inode and bump the ref count to make it | ||
| 782 | * official. We hold a reference to nameidata, which makes this safe. | ||
| 783 | */ | ||
| 784 | watch->inode = igrab(inode); | ||
| 785 | |||
| 786 | /* Add the watch to the handle's and the inode's list */ | ||
| 787 | newly_watched = !inotify_inode_watched(inode); | ||
| 788 | list_add(&watch->h_list, &ih->watches); | ||
| 789 | list_add(&watch->i_list, &inode->inotify_watches); | ||
| 790 | /* | ||
| 791 | * Set child flags _after_ adding the watch, so there is no race | ||
| 792 | * windows where newly instantiated children could miss their parent's | ||
| 793 | * watched flag. | ||
| 794 | */ | ||
| 795 | if (newly_watched) | ||
| 796 | set_dentry_child_flags(inode, 1); | ||
| 797 | |||
| 798 | out: | ||
| 799 | mutex_unlock(&ih->mutex); | ||
| 800 | mutex_unlock(&inode->inotify_mutex); | ||
| 801 | return ret; | ||
| 802 | } | ||
| 803 | EXPORT_SYMBOL_GPL(inotify_add_watch); | ||
| 804 | |||
| 805 | /** | ||
| 806 | * inotify_clone_watch - put the watch next to existing one | ||
| 807 | * @old: already installed watch | ||
| 808 | * @new: new watch | ||
| 809 | * | ||
| 810 | * Caller must hold the inotify_mutex of inode we are dealing with; | ||
| 811 | * it is expected to remove the old watch before unlocking the inode. | ||
| 812 | */ | ||
| 813 | s32 inotify_clone_watch(struct inotify_watch *old, struct inotify_watch *new) | ||
| 814 | { | ||
| 815 | struct inotify_handle *ih = old->ih; | ||
| 816 | int ret = 0; | ||
| 817 | |||
| 818 | new->mask = old->mask; | ||
| 819 | new->ih = ih; | ||
| 820 | |||
| 821 | mutex_lock(&ih->mutex); | ||
| 822 | |||
| 823 | /* Initialize a new watch */ | ||
| 824 | ret = inotify_handle_get_wd(ih, new); | ||
| 825 | if (unlikely(ret)) | ||
| 826 | goto out; | ||
| 827 | ret = new->wd; | ||
| 828 | |||
| 829 | get_inotify_handle(ih); | ||
| 830 | |||
| 831 | new->inode = igrab(old->inode); | ||
| 832 | |||
| 833 | list_add(&new->h_list, &ih->watches); | ||
| 834 | list_add(&new->i_list, &old->inode->inotify_watches); | ||
| 835 | out: | ||
| 836 | mutex_unlock(&ih->mutex); | ||
| 837 | return ret; | ||
| 838 | } | ||
| 839 | |||
| 840 | void inotify_evict_watch(struct inotify_watch *watch) | ||
| 841 | { | ||
| 842 | get_inotify_watch(watch); | ||
| 843 | mutex_lock(&watch->ih->mutex); | ||
| 844 | inotify_remove_watch_locked(watch->ih, watch); | ||
| 845 | mutex_unlock(&watch->ih->mutex); | ||
| 846 | } | ||
| 847 | |||
| 848 | /** | ||
| 849 | * inotify_rm_wd - remove a watch from an inotify instance | ||
| 850 | * @ih: inotify handle | ||
| 851 | * @wd: watch descriptor to remove | ||
| 852 | * | ||
| 853 | * Can sleep. | ||
| 854 | */ | ||
| 855 | int inotify_rm_wd(struct inotify_handle *ih, u32 wd) | ||
| 856 | { | ||
| 857 | struct inotify_watch *watch; | ||
| 858 | struct super_block *sb; | ||
| 859 | struct inode *inode; | ||
| 860 | int how; | ||
| 861 | |||
| 862 | mutex_lock(&ih->mutex); | ||
| 863 | watch = idr_find(&ih->idr, wd); | ||
| 864 | if (unlikely(!watch)) { | ||
| 865 | mutex_unlock(&ih->mutex); | ||
| 866 | return -EINVAL; | ||
| 867 | } | ||
| 868 | sb = watch->inode->i_sb; | ||
| 869 | how = pin_to_kill(ih, watch); | ||
| 870 | if (!how) | ||
| 871 | return 0; | ||
| 872 | |||
| 873 | inode = watch->inode; | ||
| 874 | |||
| 875 | mutex_lock(&inode->inotify_mutex); | ||
| 876 | mutex_lock(&ih->mutex); | ||
| 877 | |||
| 878 | /* make sure that we did not race */ | ||
| 879 | if (likely(idr_find(&ih->idr, wd) == watch)) | ||
| 880 | inotify_remove_watch_locked(ih, watch); | ||
| 881 | |||
| 882 | mutex_unlock(&ih->mutex); | ||
| 883 | mutex_unlock(&inode->inotify_mutex); | ||
| 884 | unpin_and_kill(watch, how); | ||
| 885 | |||
| 886 | return 0; | ||
| 887 | } | ||
| 888 | EXPORT_SYMBOL_GPL(inotify_rm_wd); | ||
| 889 | |||
| 890 | /** | ||
| 891 | * inotify_rm_watch - remove a watch from an inotify instance | ||
| 892 | * @ih: inotify handle | ||
| 893 | * @watch: watch to remove | ||
| 894 | * | ||
| 895 | * Can sleep. | ||
| 896 | */ | ||
| 897 | int inotify_rm_watch(struct inotify_handle *ih, | ||
| 898 | struct inotify_watch *watch) | ||
| 899 | { | ||
| 900 | return inotify_rm_wd(ih, watch->wd); | ||
| 901 | } | ||
| 902 | EXPORT_SYMBOL_GPL(inotify_rm_watch); | ||
| 903 | |||
| 904 | /* | ||
| 905 | * inotify_setup - core initialization function | ||
| 906 | */ | ||
| 907 | static int __init inotify_setup(void) | ||
| 908 | { | ||
| 909 | BUILD_BUG_ON(IN_ACCESS != FS_ACCESS); | ||
| 910 | BUILD_BUG_ON(IN_MODIFY != FS_MODIFY); | ||
| 911 | BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB); | ||
| 912 | BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE); | ||
| 913 | BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE); | ||
| 914 | BUILD_BUG_ON(IN_OPEN != FS_OPEN); | ||
| 915 | BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM); | ||
| 916 | BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO); | ||
| 917 | BUILD_BUG_ON(IN_CREATE != FS_CREATE); | ||
| 918 | BUILD_BUG_ON(IN_DELETE != FS_DELETE); | ||
| 919 | BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF); | ||
| 920 | BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF); | ||
| 921 | BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW); | ||
| 922 | |||
| 923 | BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT); | ||
| 924 | BUILD_BUG_ON(IN_ISDIR != FS_IN_ISDIR); | ||
| 925 | BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED); | ||
| 926 | BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT); | ||
| 927 | |||
| 928 | atomic_set(&inotify_cookie, 0); | ||
| 929 | |||
| 930 | return 0; | ||
| 931 | } | ||
| 932 | |||
| 933 | module_init(inotify_setup); | ||
