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 if (holder) 108 kmem_cache_free(fsnotify_event_holder_cachep, holder); 109 } 110 111 /* 112 * Find the private data that the group previously attached to this event when 113 * the group added the event to the notification queue (fsnotify_add_notify_event) 114 */ 115 struct fsnotify_event_private_data *fsnotify_remove_priv_from_event(struct fsnotify_group *group, struct fsnotify_event *event) 116 { 117 struct fsnotify_event_private_data *lpriv; 118 struct fsnotify_event_private_data *priv = NULL; 119 120 assert_spin_locked(&event->lock); 121 122 list_for_each_entry(lpriv, &event->private_data_list, event_list) { 123 if (lpriv->group == group) { 124 priv = lpriv; 125 list_del(&priv->event_list); 126 break; 127 } 128 } 129 return priv; 130 } 131 132 /* 133 * Add an event to the group notification queue. The group can later pull this 134 * event off the queue to deal with. If the event is successfully added to the 135 * group's notification queue, a reference is taken on event. 136 */ 137 int fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_event *event, 138 struct fsnotify_event_private_data *priv, 139 int (*merge)(struct list_head *, struct fsnotify_event *)) 140 { 141 struct fsnotify_event_holder *holder = NULL; 142 struct list_head *list = &group->notification_list; 143 int rc = 0; 144 145 /* 146 * There is one fsnotify_event_holder embedded inside each fsnotify_event. 147 * Check if we expect to be able to use that holder. If not alloc a new 148 * holder. 149 * For the overflow event it's possible that something will use the in 150 * event holder before we get the lock so we may need to jump back and 151 * alloc a new holder, this can't happen for most events... 152 */ 153 if (!list_empty(&event->holder.event_list)) { 154 alloc_holder: 155 holder = fsnotify_alloc_event_holder(); 156 if (!holder) 157 return -ENOMEM; 158 } 159 160 mutex_lock(&group->notification_mutex); 161 162 if (group->q_len >= group->max_events) { 163 event = q_overflow_event; 164 rc = -EOVERFLOW; 165 /* sorry, no private data on the overflow event */ 166 priv = NULL; 167 } 168 169 if (!list_empty(list) && merge) { 170 int ret; 171 172 ret = merge(list, event); 173 if (ret) { 174 mutex_unlock(&group->notification_mutex); 175 if (holder != &event->holder) 176 fsnotify_destroy_event_holder(holder); 177 return ret; 178 } 179 } 180 181 spin_lock(&event->lock); 182 183 if (list_empty(&event->holder.event_list)) { 184 if (unlikely(holder)) 185 fsnotify_destroy_event_holder(holder); 186 holder = &event->holder; 187 } else if (unlikely(!holder)) { 188 /* between the time we checked above and got the lock the in 189 * event holder was used, go back and get a new one */ 190 spin_unlock(&event->lock); 191 mutex_unlock(&group->notification_mutex); 192 goto alloc_holder; 193 } 194 195 group->q_len++; 196 holder->event = event; 197 198 fsnotify_get_event(event); 199 list_add_tail(&holder->event_list, list); 200 if (priv) 201 list_add_tail(&priv->event_list, &event->private_data_list); 202 spin_unlock(&event->lock); 203 mutex_unlock(&group->notification_mutex); 204 205 wake_up(&group->notification_waitq); 206 return rc; 207 } 208 209 /* 210 * Remove and return the first event from the notification list. There is a 211 * reference held on this event since it was on the list. It is the responsibility 212 * of the caller to drop this reference. 213 */ 214 struct fsnotify_event *fsnotify_remove_notify_event(struct fsnotify_group *group) 215 { 216 struct fsnotify_event *event; 217 struct fsnotify_event_holder *holder; 218 219 BUG_ON(!mutex_is_locked(&group->notification_mutex)); 220 221 holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list); 222 223 event = holder->event; 224 225 spin_lock(&event->lock); 226 holder->event = NULL; 227 list_del_init(&holder->event_list); 228 spin_unlock(&event->lock); 229 230 /* event == holder means we are referenced through the in event holder */ 231 if (holder != &event->holder) 232 fsnotify_destroy_event_holder(holder); 233 234 group->q_len--; 235 236 return event; 237 } 238 239 /* 240 * This will not remove the event, that must be done with fsnotify_remove_notify_event() 241 */ 242 struct fsnotify_event *fsnotify_peek_notify_event(struct fsnotify_group *group) 243 { 244 struct fsnotify_event *event; 245 struct fsnotify_event_holder *holder; 246 247 BUG_ON(!mutex_is_locked(&group->notification_mutex)); 248 249 holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list); 250 event = holder->event; 251 252 return event; 253 } 254 255 /* 256 * Called when a group is being torn down to clean up any outstanding 257 * event notifications. 258 */ 259 void fsnotify_flush_notify(struct fsnotify_group *group) 260 { 261 struct fsnotify_event *event; 262 struct fsnotify_event_private_data *priv; 263 264 mutex_lock(&group->notification_mutex); 265 while (!fsnotify_notify_queue_is_empty(group)) { 266 event = fsnotify_remove_notify_event(group); 267 /* if they don't implement free_event_priv they better not have attached any */ 268 if (group->ops->free_event_priv) { 269 spin_lock(&event->lock); 270 priv = fsnotify_remove_priv_from_event(group, event); 271 spin_unlock(&event->lock); 272 if (priv) 273 group->ops->free_event_priv(priv); 274 } 275 fsnotify_put_event(event); /* matches fsnotify_add_notify_event */ 276 } 277 mutex_unlock(&group->notification_mutex); 278 } 279 280 static void initialize_event(struct fsnotify_event *event) 281 { 282 INIT_LIST_HEAD(&event->holder.event_list); 283 atomic_set(&event->refcnt, 1); 284 285 spin_lock_init(&event->lock); 286 287 INIT_LIST_HEAD(&event->private_data_list); 288 } 289 290 /* 291 * Caller damn well better be holding whatever mutex is protecting the 292 * old_holder->event_list and the new_event must be a clean event which 293 * cannot be found anywhere else in the kernel. 294 */ 295 int fsnotify_replace_event(struct fsnotify_event_holder *old_holder, 296 struct fsnotify_event *new_event) 297 { 298 struct fsnotify_event *old_event = old_holder->event; 299 struct fsnotify_event_holder *new_holder = &new_event->holder; 300 301 enum event_spinlock_class { 302 SPINLOCK_OLD, 303 SPINLOCK_NEW, 304 }; 305 306 /* 307 * if the new_event's embedded holder is in use someone 308 * screwed up and didn't give us a clean new event. 309 */ 310 BUG_ON(!list_empty(&new_holder->event_list)); 311 312 spin_lock_nested(&old_event->lock, SPINLOCK_OLD); 313 spin_lock_nested(&new_event->lock, SPINLOCK_NEW); 314 315 new_holder->event = new_event; 316 list_replace_init(&old_holder->event_list, &new_holder->event_list); 317 318 spin_unlock(&new_event->lock); 319 spin_unlock(&old_event->lock); 320 321 /* event == holder means we are referenced through the in event holder */ 322 if (old_holder != &old_event->holder) 323 fsnotify_destroy_event_holder(old_holder); 324 325 fsnotify_get_event(new_event); /* on the list take reference */ 326 fsnotify_put_event(old_event); /* off the list, drop reference */ 327 328 return 0; 329 } 330 331 struct fsnotify_event *fsnotify_clone_event(struct fsnotify_event *old_event) 332 { 333 struct fsnotify_event *event; 334 335 event = kmem_cache_alloc(fsnotify_event_cachep, GFP_KERNEL); 336 if (!event) 337 return NULL; 338 339 memcpy(event, old_event, sizeof(*event)); 340 initialize_event(event); 341 342 if (event->name_len) { 343 event->file_name = kstrdup(old_event->file_name, GFP_KERNEL); 344 if (!event->file_name) { 345 kmem_cache_free(fsnotify_event_cachep, event); 346 return NULL; 347 } 348 } 349 if (event->data_type == FSNOTIFY_EVENT_PATH) 350 path_get(&event->path); 351 352 return event; 353 } 354 355 /* 356 * fsnotify_create_event - Allocate a new event which will be sent to each 357 * group's handle_event function if the group was interested in this 358 * particular event. 359 * 360 * @to_tell the inode which is supposed to receive the event (sometimes a 361 * parent of the inode to which the event happened. 362 * @mask what actually happened. 363 * @data pointer to the object which was actually affected 364 * @data_type flag indication if the data is a file, path, inode, nothing... 365 * @name the filename, if available 366 */ 367 struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask, void *data, 368 int data_type, const char *name, u32 cookie, 369 gfp_t gfp) 370 { 371 struct fsnotify_event *event; 372 373 event = kmem_cache_zalloc(fsnotify_event_cachep, gfp); 374 if (!event) 375 return NULL; 376 377 initialize_event(event); 378 379 if (name) { 380 event->file_name = kstrdup(name, gfp); 381 if (!event->file_name) { 382 kmem_cache_free(fsnotify_event_cachep, event); 383 return NULL; 384 } 385 event->name_len = strlen(event->file_name); 386 } 387 388 event->sync_cookie = cookie; 389 event->to_tell = to_tell; 390 event->data_type = data_type; 391 392 switch (data_type) { 393 case FSNOTIFY_EVENT_FILE: { 394 struct file *file = data; 395 struct path *path = &file->f_path; 396 event->path.dentry = path->dentry; 397 event->path.mnt = path->mnt; 398 path_get(&event->path); 399 event->data_type = FSNOTIFY_EVENT_PATH; 400 break; 401 } 402 case FSNOTIFY_EVENT_PATH: { 403 struct path *path = data; 404 event->path.dentry = path->dentry; 405 event->path.mnt = path->mnt; 406 path_get(&event->path); 407 break; 408 } 409 case FSNOTIFY_EVENT_INODE: 410 event->inode = data; 411 break; 412 case FSNOTIFY_EVENT_NONE: 413 event->inode = NULL; 414 event->path.dentry = NULL; 415 event->path.mnt = NULL; 416 break; 417 default: 418 BUG(); 419 } 420 421 event->mask = mask; 422 423 return event; 424 } 425 426 __init int fsnotify_notification_init(void) 427 { 428 fsnotify_event_cachep = KMEM_CACHE(fsnotify_event, SLAB_PANIC); 429 fsnotify_event_holder_cachep = KMEM_CACHE(fsnotify_event_holder, SLAB_PANIC); 430 431 q_overflow_event = fsnotify_create_event(NULL, FS_Q_OVERFLOW, NULL, 432 FSNOTIFY_EVENT_NONE, NULL, 0, 433 GFP_KERNEL); 434 if (!q_overflow_event) 435 panic("unable to allocate fsnotify q_overflow_event\n"); 436 437 return 0; 438 } 439 subsys_initcall(fsnotify_notification_init); 440 441