2 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
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)
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.
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.
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
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>
46 #include <asm/atomic.h>
48 #include <linux/fsnotify_backend.h>
51 static struct kmem_cache *fsnotify_event_cachep;
52 static struct kmem_cache *fsnotify_event_holder_cachep;
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'
59 static struct fsnotify_event *q_overflow_event;
60 static atomic_t fsnotify_sync_cookie = ATOMIC_INIT(0);
63 * fsnotify_get_cookie - return a unique cookie for use in synchronizing events.
64 * Called from fsnotify_move, which is inlined into filesystem modules.
66 u32 fsnotify_get_cookie(void)
68 return atomic_inc_return(&fsnotify_sync_cookie);
70 EXPORT_SYMBOL_GPL(fsnotify_get_cookie);
72 /* return true if the notify queue is empty, false otherwise */
73 bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group)
75 BUG_ON(!mutex_is_locked(&group->notification_mutex));
76 return list_empty(&group->notification_list) ? true : false;
79 void fsnotify_get_event(struct fsnotify_event *event)
81 atomic_inc(&event->refcnt);
84 void fsnotify_put_event(struct fsnotify_event *event)
89 if (atomic_dec_and_test(&event->refcnt)) {
90 if (event->data_type == FSNOTIFY_EVENT_PATH)
91 path_put(&event->path);
93 BUG_ON(!list_empty(&event->private_data_list));
95 kfree(event->file_name);
97 kmem_cache_free(fsnotify_event_cachep, event);
101 struct fsnotify_event_holder *fsnotify_alloc_event_holder(void)
103 return kmem_cache_alloc(fsnotify_event_holder_cachep, GFP_KERNEL);
106 void fsnotify_destroy_event_holder(struct fsnotify_event_holder *holder)
109 kmem_cache_free(fsnotify_event_holder_cachep, holder);
113 * Find the private data that the group previously attached to this event when
114 * the group added the event to the notification queue (fsnotify_add_notify_event)
116 struct fsnotify_event_private_data *fsnotify_remove_priv_from_event(struct fsnotify_group *group, struct fsnotify_event *event)
118 struct fsnotify_event_private_data *lpriv;
119 struct fsnotify_event_private_data *priv = NULL;
121 assert_spin_locked(&event->lock);
123 list_for_each_entry(lpriv, &event->private_data_list, event_list) {
124 if (lpriv->group == group) {
126 list_del(&priv->event_list);
134 * Add an event to the group notification queue. The group can later pull this
135 * event off the queue to deal with. If the event is successfully added to the
136 * group's notification queue, a reference is taken on event.
138 int fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_event *event,
139 struct fsnotify_event_private_data *priv,
140 int (*merge)(struct list_head *,
141 struct fsnotify_event *,
145 struct fsnotify_event_holder *holder = NULL;
146 struct list_head *list = &group->notification_list;
150 * There is one fsnotify_event_holder embedded inside each fsnotify_event.
151 * Check if we expect to be able to use that holder. If not alloc a new
153 * For the overflow event it's possible that something will use the in
154 * event holder before we get the lock so we may need to jump back and
155 * alloc a new holder, this can't happen for most events...
157 if (!list_empty(&event->holder.event_list)) {
159 holder = fsnotify_alloc_event_holder();
164 mutex_lock(&group->notification_mutex);
166 if (group->q_len >= group->max_events) {
167 event = q_overflow_event;
169 /* sorry, no private data on the overflow event */
173 if (!list_empty(list) && merge) {
176 ret = merge(list, event, arg);
178 mutex_unlock(&group->notification_mutex);
179 if (holder != &event->holder)
180 fsnotify_destroy_event_holder(holder);
185 spin_lock(&event->lock);
187 if (list_empty(&event->holder.event_list)) {
188 if (unlikely(holder))
189 fsnotify_destroy_event_holder(holder);
190 holder = &event->holder;
191 } else if (unlikely(!holder)) {
192 /* between the time we checked above and got the lock the in
193 * event holder was used, go back and get a new one */
194 spin_unlock(&event->lock);
195 mutex_unlock(&group->notification_mutex);
200 holder->event = event;
202 fsnotify_get_event(event);
203 list_add_tail(&holder->event_list, list);
205 list_add_tail(&priv->event_list, &event->private_data_list);
206 spin_unlock(&event->lock);
207 mutex_unlock(&group->notification_mutex);
209 wake_up(&group->notification_waitq);
214 * Remove and return the first event from the notification list. There is a
215 * reference held on this event since it was on the list. It is the responsibility
216 * of the caller to drop this reference.
218 struct fsnotify_event *fsnotify_remove_notify_event(struct fsnotify_group *group)
220 struct fsnotify_event *event;
221 struct fsnotify_event_holder *holder;
223 BUG_ON(!mutex_is_locked(&group->notification_mutex));
225 holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
227 event = holder->event;
229 spin_lock(&event->lock);
230 holder->event = NULL;
231 list_del_init(&holder->event_list);
232 spin_unlock(&event->lock);
234 /* event == holder means we are referenced through the in event holder */
235 if (holder != &event->holder)
236 fsnotify_destroy_event_holder(holder);
244 * This will not remove the event, that must be done with fsnotify_remove_notify_event()
246 struct fsnotify_event *fsnotify_peek_notify_event(struct fsnotify_group *group)
248 struct fsnotify_event *event;
249 struct fsnotify_event_holder *holder;
251 BUG_ON(!mutex_is_locked(&group->notification_mutex));
253 holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
254 event = holder->event;
260 * Called when a group is being torn down to clean up any outstanding
261 * event notifications.
263 void fsnotify_flush_notify(struct fsnotify_group *group)
265 struct fsnotify_event *event;
266 struct fsnotify_event_private_data *priv;
268 mutex_lock(&group->notification_mutex);
269 while (!fsnotify_notify_queue_is_empty(group)) {
270 event = fsnotify_remove_notify_event(group);
271 /* if they don't implement free_event_priv they better not have attached any */
272 if (group->ops->free_event_priv) {
273 spin_lock(&event->lock);
274 priv = fsnotify_remove_priv_from_event(group, event);
275 spin_unlock(&event->lock);
277 group->ops->free_event_priv(priv);
279 fsnotify_put_event(event); /* matches fsnotify_add_notify_event */
281 mutex_unlock(&group->notification_mutex);
284 static void initialize_event(struct fsnotify_event *event)
286 INIT_LIST_HEAD(&event->holder.event_list);
287 atomic_set(&event->refcnt, 1);
289 spin_lock_init(&event->lock);
291 INIT_LIST_HEAD(&event->private_data_list);
295 * Caller damn well better be holding whatever mutex is protecting the
296 * old_holder->event_list and the new_event must be a clean event which
297 * cannot be found anywhere else in the kernel.
299 int fsnotify_replace_event(struct fsnotify_event_holder *old_holder,
300 struct fsnotify_event *new_event)
302 struct fsnotify_event *old_event = old_holder->event;
303 struct fsnotify_event_holder *new_holder = &new_event->holder;
305 enum event_spinlock_class {
311 * if the new_event's embedded holder is in use someone
312 * screwed up and didn't give us a clean new event.
314 BUG_ON(!list_empty(&new_holder->event_list));
316 spin_lock_nested(&old_event->lock, SPINLOCK_OLD);
317 spin_lock_nested(&new_event->lock, SPINLOCK_NEW);
319 new_holder->event = new_event;
320 list_replace_init(&old_holder->event_list, &new_holder->event_list);
322 spin_unlock(&new_event->lock);
323 spin_unlock(&old_event->lock);
325 /* event == holder means we are referenced through the in event holder */
326 if (old_holder != &old_event->holder)
327 fsnotify_destroy_event_holder(old_holder);
329 fsnotify_get_event(new_event); /* on the list take reference */
330 fsnotify_put_event(old_event); /* off the list, drop reference */
335 struct fsnotify_event *fsnotify_clone_event(struct fsnotify_event *old_event)
337 struct fsnotify_event *event;
339 event = kmem_cache_alloc(fsnotify_event_cachep, GFP_KERNEL);
343 memcpy(event, old_event, sizeof(*event));
344 initialize_event(event);
346 if (event->name_len) {
347 event->file_name = kstrdup(old_event->file_name, GFP_KERNEL);
348 if (!event->file_name) {
349 kmem_cache_free(fsnotify_event_cachep, event);
353 event->tgid = get_pid(old_event->tgid);
354 if (event->data_type == FSNOTIFY_EVENT_PATH)
355 path_get(&event->path);
361 * fsnotify_create_event - Allocate a new event which will be sent to each
362 * group's handle_event function if the group was interested in this
365 * @to_tell the inode which is supposed to receive the event (sometimes a
366 * parent of the inode to which the event happened.
367 * @mask what actually happened.
368 * @data pointer to the object which was actually affected
369 * @data_type flag indication if the data is a file, path, inode, nothing...
370 * @name the filename, if available
372 struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask, void *data,
373 int data_type, const char *name, u32 cookie,
376 struct fsnotify_event *event;
378 event = kmem_cache_zalloc(fsnotify_event_cachep, gfp);
382 initialize_event(event);
385 event->file_name = kstrdup(name, gfp);
386 if (!event->file_name) {
387 kmem_cache_free(fsnotify_event_cachep, event);
390 event->name_len = strlen(event->file_name);
393 event->tgid = get_pid(task_tgid(current));
394 event->sync_cookie = cookie;
395 event->to_tell = to_tell;
396 event->data_type = data_type;
399 case FSNOTIFY_EVENT_PATH: {
400 struct path *path = data;
401 event->path.dentry = path->dentry;
402 event->path.mnt = path->mnt;
403 path_get(&event->path);
406 case FSNOTIFY_EVENT_INODE:
409 case FSNOTIFY_EVENT_NONE:
411 event->path.dentry = NULL;
412 event->path.mnt = NULL;
423 __init int fsnotify_notification_init(void)
425 fsnotify_event_cachep = KMEM_CACHE(fsnotify_event, SLAB_PANIC);
426 fsnotify_event_holder_cachep = KMEM_CACHE(fsnotify_event_holder, SLAB_PANIC);
428 q_overflow_event = fsnotify_create_event(NULL, FS_Q_OVERFLOW, NULL,
429 FSNOTIFY_EVENT_NONE, NULL, 0,
431 if (!q_overflow_event)
432 panic("unable to allocate fsnotify q_overflow_event\n");
436 subsys_initcall(fsnotify_notification_init);