]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/notify/notification.c
fsnotify: fix inotify tail drop check with path entries
[net-next-2.6.git] / fs / notify / notification.c
CommitLineData
90586523
EP
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
a2d8bc6c
EP
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
90586523
EP
34#include <linux/fs.h>
35#include <linux/init.h>
36#include <linux/kernel.h>
37#include <linux/list.h>
47882c6f 38#include <linux/module.h>
90586523
EP
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
51static struct kmem_cache *fsnotify_event_cachep;
a2d8bc6c
EP
52static 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 */
59static struct fsnotify_event q_overflow_event;
47882c6f
EP
60static 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 */
66u32 fsnotify_get_cookie(void)
67{
68 return atomic_inc_return(&fsnotify_sync_cookie);
69}
70EXPORT_SYMBOL_GPL(fsnotify_get_cookie);
a2d8bc6c
EP
71
72/* return true if the notify queue is empty, false otherwise */
73bool 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}
90586523
EP
78
79void fsnotify_get_event(struct fsnotify_event *event)
80{
81 atomic_inc(&event->refcnt);
82}
83
84void 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
e4aff117
EP
93 BUG_ON(!list_empty(&event->private_data_list));
94
62ffe5df 95 kfree(event->file_name);
90586523
EP
96 kmem_cache_free(fsnotify_event_cachep, event);
97 }
98}
99
a2d8bc6c
EP
100struct fsnotify_event_holder *fsnotify_alloc_event_holder(void)
101{
102 return kmem_cache_alloc(fsnotify_event_holder_cachep, GFP_KERNEL);
103}
104
105void fsnotify_destroy_event_holder(struct fsnotify_event_holder *holder)
106{
107 kmem_cache_free(fsnotify_event_holder_cachep, holder);
108}
109
110/*
e4aff117
EP
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 */
114struct 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.
a2d8bc6c
EP
134 */
135static 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) &&
4a148ba9
EP
139 (old->data_type == new->data_type) &&
140 (old->name_len == new->name_len)) {
a2d8bc6c
EP
141 switch (old->data_type) {
142 case (FSNOTIFY_EVENT_INODE):
4a148ba9
EP
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))
a2d8bc6c
EP
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;
c05594b6 154 break;
a2d8bc6c
EP
155 case (FSNOTIFY_EVENT_NONE):
156 return true;
157 };
158 }
159 return false;
160}
161
90586523 162/*
a2d8bc6c
EP
163 * Add an event to the group notification queue. The group can later pull this
164 * event off the queue to deal with. If the event is successfully added to the
165 * group's notification queue, a reference is taken on event.
90586523 166 */
e4aff117
EP
167int fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_event *event,
168 struct fsnotify_event_private_data *priv)
a2d8bc6c
EP
169{
170 struct fsnotify_event_holder *holder = NULL;
171 struct list_head *list = &group->notification_list;
172 struct fsnotify_event_holder *last_holder;
173 struct fsnotify_event *last_event;
174
e4aff117
EP
175 /* easy to tell if priv was attached to the event */
176 INIT_LIST_HEAD(&priv->event_list);
177
a2d8bc6c
EP
178 /*
179 * There is one fsnotify_event_holder embedded inside each fsnotify_event.
180 * Check if we expect to be able to use that holder. If not alloc a new
181 * holder.
182 * For the overflow event it's possible that something will use the in
183 * event holder before we get the lock so we may need to jump back and
184 * alloc a new holder, this can't happen for most events...
185 */
186 if (!list_empty(&event->holder.event_list)) {
187alloc_holder:
188 holder = fsnotify_alloc_event_holder();
189 if (!holder)
190 return -ENOMEM;
191 }
192
193 mutex_lock(&group->notification_mutex);
194
e4aff117 195 if (group->q_len >= group->max_events) {
a2d8bc6c 196 event = &q_overflow_event;
e4aff117
EP
197 /* sorry, no private data on the overflow event */
198 priv = NULL;
199 }
a2d8bc6c
EP
200
201 spin_lock(&event->lock);
202
203 if (list_empty(&event->holder.event_list)) {
204 if (unlikely(holder))
205 fsnotify_destroy_event_holder(holder);
206 holder = &event->holder;
207 } else if (unlikely(!holder)) {
208 /* between the time we checked above and got the lock the in
209 * event holder was used, go back and get a new one */
210 spin_unlock(&event->lock);
211 mutex_unlock(&group->notification_mutex);
212 goto alloc_holder;
213 }
214
215 if (!list_empty(list)) {
216 last_holder = list_entry(list->prev, struct fsnotify_event_holder, event_list);
217 last_event = last_holder->event;
218 if (event_compare(last_event, event)) {
219 spin_unlock(&event->lock);
220 mutex_unlock(&group->notification_mutex);
221 if (holder != &event->holder)
222 fsnotify_destroy_event_holder(holder);
e4aff117 223 return -EEXIST;
a2d8bc6c
EP
224 }
225 }
226
227 group->q_len++;
228 holder->event = event;
229
230 fsnotify_get_event(event);
231 list_add_tail(&holder->event_list, list);
e4aff117
EP
232 if (priv)
233 list_add_tail(&priv->event_list, &event->private_data_list);
a2d8bc6c
EP
234 spin_unlock(&event->lock);
235 mutex_unlock(&group->notification_mutex);
236
237 wake_up(&group->notification_waitq);
238 return 0;
239}
240
241/*
242 * Remove and return the first event from the notification list. There is a
243 * reference held on this event since it was on the list. It is the responsibility
244 * of the caller to drop this reference.
245 */
246struct fsnotify_event *fsnotify_remove_notify_event(struct fsnotify_group *group)
90586523
EP
247{
248 struct fsnotify_event *event;
a2d8bc6c 249 struct fsnotify_event_holder *holder;
90586523 250
a2d8bc6c 251 BUG_ON(!mutex_is_locked(&group->notification_mutex));
90586523 252
a2d8bc6c
EP
253 holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
254
255 event = holder->event;
256
257 spin_lock(&event->lock);
258 holder->event = NULL;
259 list_del_init(&holder->event_list);
260 spin_unlock(&event->lock);
261
262 /* event == holder means we are referenced through the in event holder */
263 if (holder != &event->holder)
264 fsnotify_destroy_event_holder(holder);
265
266 group->q_len--;
267
268 return event;
269}
270
271/*
272 * This will not remove the event, that must be done with fsnotify_remove_notify_event()
273 */
274struct fsnotify_event *fsnotify_peek_notify_event(struct fsnotify_group *group)
275{
276 struct fsnotify_event *event;
277 struct fsnotify_event_holder *holder;
278
279 BUG_ON(!mutex_is_locked(&group->notification_mutex));
280
281 holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
282 event = holder->event;
283
284 return event;
285}
286
287/*
288 * Called when a group is being torn down to clean up any outstanding
289 * event notifications.
290 */
291void fsnotify_flush_notify(struct fsnotify_group *group)
292{
293 struct fsnotify_event *event;
e4aff117 294 struct fsnotify_event_private_data *priv;
a2d8bc6c
EP
295
296 mutex_lock(&group->notification_mutex);
297 while (!fsnotify_notify_queue_is_empty(group)) {
298 event = fsnotify_remove_notify_event(group);
e4aff117
EP
299 /* if they don't implement free_event_priv they better not have attached any */
300 if (group->ops->free_event_priv) {
301 spin_lock(&event->lock);
302 priv = fsnotify_remove_priv_from_event(group, event);
303 spin_unlock(&event->lock);
304 if (priv)
305 group->ops->free_event_priv(priv);
306 }
a2d8bc6c
EP
307 fsnotify_put_event(event); /* matches fsnotify_add_notify_event */
308 }
309 mutex_unlock(&group->notification_mutex);
310}
311
312static void initialize_event(struct fsnotify_event *event)
313{
314 event->holder.event = NULL;
315 INIT_LIST_HEAD(&event->holder.event_list);
90586523
EP
316 atomic_set(&event->refcnt, 1);
317
318 spin_lock_init(&event->lock);
319
320 event->path.dentry = NULL;
321 event->path.mnt = NULL;
322 event->inode = NULL;
a2d8bc6c 323 event->data_type = FSNOTIFY_EVENT_NONE;
90586523 324
e4aff117
EP
325 INIT_LIST_HEAD(&event->private_data_list);
326
a2d8bc6c 327 event->to_tell = NULL;
62ffe5df
EP
328
329 event->file_name = NULL;
330 event->name_len = 0;
47882c6f
EP
331
332 event->sync_cookie = 0;
a2d8bc6c
EP
333}
334
335/*
336 * fsnotify_create_event - Allocate a new event which will be sent to each
337 * group's handle_event function if the group was interested in this
338 * particular event.
339 *
340 * @to_tell the inode which is supposed to receive the event (sometimes a
341 * parent of the inode to which the event happened.
342 * @mask what actually happened.
343 * @data pointer to the object which was actually affected
344 * @data_type flag indication if the data is a file, path, inode, nothing...
62ffe5df 345 * @name the filename, if available
a2d8bc6c 346 */
47882c6f
EP
347struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask, void *data,
348 int data_type, const char *name, u32 cookie)
a2d8bc6c
EP
349{
350 struct fsnotify_event *event;
351
352 event = kmem_cache_alloc(fsnotify_event_cachep, GFP_KERNEL);
353 if (!event)
354 return NULL;
355
356 initialize_event(event);
62ffe5df
EP
357
358 if (name) {
359 event->file_name = kstrdup(name, GFP_KERNEL);
360 if (!event->file_name) {
361 kmem_cache_free(fsnotify_event_cachep, event);
362 return NULL;
363 }
364 event->name_len = strlen(event->file_name);
365 }
47882c6f
EP
366
367 event->sync_cookie = cookie;
90586523
EP
368 event->to_tell = to_tell;
369
370 switch (data_type) {
371 case FSNOTIFY_EVENT_FILE: {
372 struct file *file = data;
373 struct path *path = &file->f_path;
374 event->path.dentry = path->dentry;
375 event->path.mnt = path->mnt;
376 path_get(&event->path);
377 event->data_type = FSNOTIFY_EVENT_PATH;
378 break;
379 }
380 case FSNOTIFY_EVENT_PATH: {
381 struct path *path = data;
382 event->path.dentry = path->dentry;
383 event->path.mnt = path->mnt;
384 path_get(&event->path);
385 event->data_type = FSNOTIFY_EVENT_PATH;
386 break;
387 }
388 case FSNOTIFY_EVENT_INODE:
389 event->inode = data;
390 event->data_type = FSNOTIFY_EVENT_INODE;
391 break;
392 case FSNOTIFY_EVENT_NONE:
393 event->inode = NULL;
394 event->path.dentry = NULL;
395 event->path.mnt = NULL;
396 break;
397 default:
398 BUG();
399 }
400
401 event->mask = mask;
402
403 return event;
404}
405
406__init int fsnotify_notification_init(void)
407{
408 fsnotify_event_cachep = KMEM_CACHE(fsnotify_event, SLAB_PANIC);
a2d8bc6c
EP
409 fsnotify_event_holder_cachep = KMEM_CACHE(fsnotify_event_holder, SLAB_PANIC);
410
411 initialize_event(&q_overflow_event);
412 q_overflow_event.mask = FS_Q_OVERFLOW;
90586523
EP
413
414 return 0;
415}
416subsys_initcall(fsnotify_notification_init);
417