1 /* 2 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com> 3 * 4 * This program is free software; you can redistribute it and/or modify 5 * it under the terms of the GNU General Public License as published by 6 * the Free Software Foundation; either version 2, or (at your option) 7 * any later version. 8 * 9 * This program is distributed in the hope that it will be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; see the file COPYING. If not, write to 16 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. 17 */ 18 19 /* 20 * Basic idea behind the notification queue: An fsnotify group (like inotify) 21 * sends the userspace notification about events asyncronously some time after 22 * the event happened. When inotify gets an event it will need to add that 23 * event to the group notify queue. Since a single event might need to be on 24 * multiple group's notification queues we can't add the event directly to each 25 * queue and instead add a small "event_holder" to each queue. This event_holder 26 * has a pointer back to the original event. Since the majority of events are 27 * going to end up on one, and only one, notification queue we embed one 28 * event_holder into each event. This means we have a single allocation instead 29 * of always needing two. If the embedded event_holder is already in use by 30 * another group a new event_holder (from fsnotify_event_holder_cachep) will be 31 * allocated and used. 32 */ 33 34 #include <linux/fs.h> 35 #include <linux/init.h> 36 #include <linux/kernel.h> 37 #include <linux/list.h> 38 #include <linux/module.h> 39 #include <linux/mount.h> 40 #include <linux/mutex.h> 41 #include <linux/namei.h> 42 #include <linux/path.h> 43 #include <linux/slab.h> 44 #include <linux/spinlock.h> 45 46 #include <asm/atomic.h> 47 48 #include <linux/fsnotify_backend.h> 49 #include "fsnotify.h" 50 51 static struct kmem_cache *fsnotify_event_cachep; 52 static struct kmem_cache *fsnotify_event_holder_cachep; 53 /* 54 * This is a magic event we send when the q is too full. Since it doesn't 55 * hold real event information we just keep one system wide and use it any time 56 * it is needed. It's refcnt is set 1 at kernel init time and will never 57 * get set to 0 so it will never get 'freed' 58 */ 59 static struct fsnotify_event q_overflow_event; 60 static atomic_t fsnotify_sync_cookie = ATOMIC_INIT(0); 61 62 /** 63 * fsnotify_get_cookie - return a unique cookie for use in synchronizing events. 64 * Called from fsnotify_move, which is inlined into filesystem modules. 65 */ 66 u32 fsnotify_get_cookie(void) 67 { 68 return atomic_inc_return(&fsnotify_sync_cookie); 69 } 70 EXPORT_SYMBOL_GPL(fsnotify_get_cookie); 71 72 /* return true if the notify queue is empty, false otherwise */ 73 bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group) 74 { 75 BUG_ON(!mutex_is_locked(&group->notification_mutex)); 76 return list_empty(&group->notification_list) ? true : false; 77 } 78 79 void fsnotify_get_event(struct fsnotify_event *event) 80 { 81 atomic_inc(&event->refcnt); 82 } 83 84 void fsnotify_put_event(struct fsnotify_event *event) 85 { 86 if (!event) 87 return; 88 89 if (atomic_dec_and_test(&event->refcnt)) { 90 if (event->data_type == FSNOTIFY_EVENT_PATH) 91 path_put(&event->path); 92 93 BUG_ON(!list_empty(&event->private_data_list)); 94 95 kfree(event->file_name); 96 kmem_cache_free(fsnotify_event_cachep, event); 97 } 98 } 99 100 struct fsnotify_event_holder *fsnotify_alloc_event_holder(void) 101 { 102 return kmem_cache_alloc(fsnotify_event_holder_cachep, GFP_KERNEL); 103 } 104 105 void fsnotify_destroy_event_holder(struct fsnotify_event_holder *holder) 106 { 107 kmem_cache_free(fsnotify_event_holder_cachep, holder); 108 } 109 110 /* 111 * Find the private data that the group previously attached to this event when 112 * the group added the event to the notification queue (fsnotify_add_notify_event) 113 */ 114 struct fsnotify_event_private_data *fsnotify_remove_priv_from_event(struct fsnotify_group *group, struct fsnotify_event *event) 115 { 116 struct fsnotify_event_private_data *lpriv; 117 struct fsnotify_event_private_data *priv = NULL; 118 119 assert_spin_locked(&event->lock); 120 121 list_for_each_entry(lpriv, &event->private_data_list, event_list) { 122 if (lpriv->group == group) { 123 priv = lpriv; 124 list_del(&priv->event_list); 125 break; 126 } 127 } 128 return priv; 129 } 130 131 /* 132 * Check if 2 events contain the same information. We do not compare private data 133 * but at this moment that isn't a problem for any know fsnotify listeners. 134 */ 135 static bool event_compare(struct fsnotify_event *old, struct fsnotify_event *new) 136 { 137 if ((old->mask == new->mask) && 138 (old->to_tell == new->to_tell) && 139 (old->data_type == new->data_type) && 140 (old->name_len == new->name_len)) { 141 switch (old->data_type) { 142 case (FSNOTIFY_EVENT_INODE): 143 /* remember, after old was put on the wait_q we aren't 144 * allowed to look at the inode any more, only thing 145 * left to check was if the file_name is the same */ 146 if (old->name_len && 147 !strcmp(old->file_name, new->file_name)) 148 return true; 149 break; 150 case (FSNOTIFY_EVENT_PATH): 151 if ((old->path.mnt == new->path.mnt) && 152 (old->path.dentry == new->path.dentry)) 153 return true; 154 break; 155 case (FSNOTIFY_EVENT_NONE): 156 if (old->mask & FS_Q_OVERFLOW) 157 return true; 158 else if (old->mask & FS_IN_IGNORED) 159 return false; 160 return false; 161 }; 162 } 163 return false; 164 } 165 166 /* 167 * Add an event to the group notification queue. The group can later pull this 168 * event off the queue to deal with. If the event is successfully added to the 169 * group's notification queue, a reference is taken on event. 170 */ 171 int fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_event *event, 172 struct fsnotify_event_private_data *priv) 173 { 174 struct fsnotify_event_holder *holder = NULL; 175 struct list_head *list = &group->notification_list; 176 struct fsnotify_event_holder *last_holder; 177 struct fsnotify_event *last_event; 178 int ret = 0; 179 180 /* 181 * There is one fsnotify_event_holder embedded inside each fsnotify_event. 182 * Check if we expect to be able to use that holder. If not alloc a new 183 * holder. 184 * For the overflow event it's possible that something will use the in 185 * event holder before we get the lock so we may need to jump back and 186 * alloc a new holder, this can't happen for most events... 187 */ 188 if (!list_empty(&event->holder.event_list)) { 189 alloc_holder: 190 holder = fsnotify_alloc_event_holder(); 191 if (!holder) 192 return -ENOMEM; 193 } 194 195 mutex_lock(&group->notification_mutex); 196 197 if (group->q_len >= group->max_events) { 198 event = &q_overflow_event; 199 ret = -EOVERFLOW; 200 /* sorry, no private data on the overflow event */ 201 priv = NULL; 202 } 203 204 spin_lock(&event->lock); 205 206 if (list_empty(&event->holder.event_list)) { 207 if (unlikely(holder)) 208 fsnotify_destroy_event_holder(holder); 209 holder = &event->holder; 210 } else if (unlikely(!holder)) { 211 /* between the time we checked above and got the lock the in 212 * event holder was used, go back and get a new one */ 213 spin_unlock(&event->lock); 214 mutex_unlock(&group->notification_mutex); 215 goto alloc_holder; 216 } 217 218 if (!list_empty(list)) { 219 last_holder = list_entry(list->prev, struct fsnotify_event_holder, event_list); 220 last_event = last_holder->event; 221 if (event_compare(last_event, event)) { 222 spin_unlock(&event->lock); 223 mutex_unlock(&group->notification_mutex); 224 if (holder != &event->holder) 225 fsnotify_destroy_event_holder(holder); 226 return -EEXIST; 227 } 228 } 229 230 group->q_len++; 231 holder->event = event; 232 233 fsnotify_get_event(event); 234 list_add_tail(&holder->event_list, list); 235 if (priv) 236 list_add_tail(&priv->event_list, &event->private_data_list); 237 spin_unlock(&event->lock); 238 mutex_unlock(&group->notification_mutex); 239 240 wake_up(&group->notification_waitq); 241 return ret; 242 } 243 244 /* 245 * Remove and return the first event from the notification list. There is a 246 * reference held on this event since it was on the list. It is the responsibility 247 * of the caller to drop this reference. 248 */ 249 struct fsnotify_event *fsnotify_remove_notify_event(struct fsnotify_group *group) 250 { 251 struct fsnotify_event *event; 252 struct fsnotify_event_holder *holder; 253 254 BUG_ON(!mutex_is_locked(&group->notification_mutex)); 255 256 holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list); 257 258 event = holder->event; 259 260 spin_lock(&event->lock); 261 holder->event = NULL; 262 list_del_init(&holder->event_list); 263 spin_unlock(&event->lock); 264 265 /* event == holder means we are referenced through the in event holder */ 266 if (holder != &event->holder) 267 fsnotify_destroy_event_holder(holder); 268 269 group->q_len--; 270 271 return event; 272 } 273 274 /* 275 * This will not remove the event, that must be done with fsnotify_remove_notify_event() 276 */ 277 struct fsnotify_event *fsnotify_peek_notify_event(struct fsnotify_group *group) 278 { 279 struct fsnotify_event *event; 280 struct fsnotify_event_holder *holder; 281 282 BUG_ON(!mutex_is_locked(&group->notification_mutex)); 283 284 holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list); 285 event = holder->event; 286 287 return event; 288 } 289 290 /* 291 * Called when a group is being torn down to clean up any outstanding 292 * event notifications. 293 */ 294 void fsnotify_flush_notify(struct fsnotify_group *group) 295 { 296 struct fsnotify_event *event; 297 struct fsnotify_event_private_data *priv; 298 299 mutex_lock(&group->notification_mutex); 300 while (!fsnotify_notify_queue_is_empty(group)) { 301 event = fsnotify_remove_notify_event(group); 302 /* if they don't implement free_event_priv they better not have attached any */ 303 if (group->ops->free_event_priv) { 304 spin_lock(&event->lock); 305 priv = fsnotify_remove_priv_from_event(group, event); 306 spin_unlock(&event->lock); 307 if (priv) 308 group->ops->free_event_priv(priv); 309 } 310 fsnotify_put_event(event); /* matches fsnotify_add_notify_event */ 311 } 312 mutex_unlock(&group->notification_mutex); 313 } 314 315 static void initialize_event(struct fsnotify_event *event) 316 { 317 event->holder.event = NULL; 318 INIT_LIST_HEAD(&event->holder.event_list); 319 atomic_set(&event->refcnt, 1); 320 321 spin_lock_init(&event->lock); 322 323 event->path.dentry = NULL; 324 event->path.mnt = NULL; 325 event->inode = NULL; 326 event->data_type = FSNOTIFY_EVENT_NONE; 327 328 INIT_LIST_HEAD(&event->private_data_list); 329 330 event->to_tell = NULL; 331 332 event->file_name = NULL; 333 event->name_len = 0; 334 335 event->sync_cookie = 0; 336 } 337 338 /* 339 * fsnotify_create_event - Allocate a new event which will be sent to each 340 * group's handle_event function if the group was interested in this 341 * particular event. 342 * 343 * @to_tell the inode which is supposed to receive the event (sometimes a 344 * parent of the inode to which the event happened. 345 * @mask what actually happened. 346 * @data pointer to the object which was actually affected 347 * @data_type flag indication if the data is a file, path, inode, nothing... 348 * @name the filename, if available 349 */ 350 struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask, void *data, 351 int data_type, const char *name, u32 cookie, 352 gfp_t gfp) 353 { 354 struct fsnotify_event *event; 355 356 event = kmem_cache_alloc(fsnotify_event_cachep, gfp); 357 if (!event) 358 return NULL; 359 360 initialize_event(event); 361 362 if (name) { 363 event->file_name = kstrdup(name, gfp); 364 if (!event->file_name) { 365 kmem_cache_free(fsnotify_event_cachep, event); 366 return NULL; 367 } 368 event->name_len = strlen(event->file_name); 369 } 370 371 event->sync_cookie = cookie; 372 event->to_tell = to_tell; 373 374 switch (data_type) { 375 case FSNOTIFY_EVENT_FILE: { 376 struct file *file = data; 377 struct path *path = &file->f_path; 378 event->path.dentry = path->dentry; 379 event->path.mnt = path->mnt; 380 path_get(&event->path); 381 event->data_type = FSNOTIFY_EVENT_PATH; 382 break; 383 } 384 case FSNOTIFY_EVENT_PATH: { 385 struct path *path = data; 386 event->path.dentry = path->dentry; 387 event->path.mnt = path->mnt; 388 path_get(&event->path); 389 event->data_type = FSNOTIFY_EVENT_PATH; 390 break; 391 } 392 case FSNOTIFY_EVENT_INODE: 393 event->inode = data; 394 event->data_type = FSNOTIFY_EVENT_INODE; 395 break; 396 case FSNOTIFY_EVENT_NONE: 397 event->inode = NULL; 398 event->path.dentry = NULL; 399 event->path.mnt = NULL; 400 break; 401 default: 402 BUG(); 403 } 404 405 event->mask = mask; 406 407 return event; 408 } 409 410 __init int fsnotify_notification_init(void) 411 { 412 fsnotify_event_cachep = KMEM_CACHE(fsnotify_event, SLAB_PANIC); 413 fsnotify_event_holder_cachep = KMEM_CACHE(fsnotify_event_holder, SLAB_PANIC); 414 415 initialize_event(&q_overflow_event); 416 q_overflow_event.mask = FS_Q_OVERFLOW; 417 418 return 0; 419 } 420 subsys_initcall(fsnotify_notification_init); 421 422