blob: 7b51b05f160c36846e47a8f1b7cb417bed9ed914 [file] [log] [blame]
Eric Paris90586522009-05-21 17:01:20 -04001/*
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
Eric Parisa2d8bc62009-05-21 17:01:37 -040019/*
20 * Basic idea behind the notification queue: An fsnotify group (like inotify)
Adam Buchbinderb3834be2012-09-19 21:48:02 -040021 * sends the userspace notification about events asynchronously some time after
Eric Parisa2d8bc62009-05-21 17:01:37 -040022 * 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
Eric Paris90586522009-05-21 17:01:20 -040034#include <linux/fs.h>
35#include <linux/init.h>
36#include <linux/kernel.h>
37#include <linux/list.h>
Eric Paris47882c62009-05-21 17:01:47 -040038#include <linux/module.h>
Eric Paris90586522009-05-21 17:01:20 -040039#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
Arun Sharma600634972011-07-26 16:09:06 -070046#include <linux/atomic.h>
Eric Paris90586522009-05-21 17:01:20 -040047
48#include <linux/fsnotify_backend.h>
49#include "fsnotify.h"
50
51static struct kmem_cache *fsnotify_event_cachep;
Eric Parisa2d8bc62009-05-21 17:01:37 -040052static 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 */
Eric Parisb4277d32009-12-17 20:12:06 -050059static struct fsnotify_event *q_overflow_event;
Eric Paris47882c62009-05-21 17:01:47 -040060static 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);
Eric Parisa2d8bc62009-05-21 17:01:37 -040071
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}
Eric Paris90586522009-05-21 17:01:20 -040078
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)) {
Eric Paris5ba08e22010-07-28 10:18:37 -040090 pr_debug("%s: event=%p\n", __func__, event);
91
Linus Torvalds20696012010-08-12 14:23:04 -070092 if (event->data_type == FSNOTIFY_EVENT_PATH)
93 path_put(&event->path);
Eric Paris90586522009-05-21 17:01:20 -040094
Eric Parise4aff112009-05-21 17:01:50 -040095 BUG_ON(!list_empty(&event->private_data_list));
96
Eric Paris62ffe5d2009-05-21 17:01:43 -040097 kfree(event->file_name);
Andreas Gruenbacher32c32632009-12-17 21:24:27 -050098 put_pid(event->tgid);
Eric Paris90586522009-05-21 17:01:20 -040099 kmem_cache_free(fsnotify_event_cachep, event);
100 }
101}
102
Eric Parisa2d8bc62009-05-21 17:01:37 -0400103struct fsnotify_event_holder *fsnotify_alloc_event_holder(void)
104{
105 return kmem_cache_alloc(fsnotify_event_holder_cachep, GFP_KERNEL);
106}
107
108void fsnotify_destroy_event_holder(struct fsnotify_event_holder *holder)
109{
Eric Paris74766bb2009-12-17 21:24:21 -0500110 if (holder)
111 kmem_cache_free(fsnotify_event_holder_cachep, holder);
Eric Parisa2d8bc62009-05-21 17:01:37 -0400112}
113
Eric Paris90586522009-05-21 17:01:20 -0400114/*
Eric Parise4aff112009-05-21 17:01:50 -0400115 * Find the private data that the group previously attached to this event when
116 * the group added the event to the notification queue (fsnotify_add_notify_event)
117 */
118struct fsnotify_event_private_data *fsnotify_remove_priv_from_event(struct fsnotify_group *group, struct fsnotify_event *event)
119{
120 struct fsnotify_event_private_data *lpriv;
121 struct fsnotify_event_private_data *priv = NULL;
122
123 assert_spin_locked(&event->lock);
124
125 list_for_each_entry(lpriv, &event->private_data_list, event_list) {
126 if (lpriv->group == group) {
127 priv = lpriv;
128 list_del(&priv->event_list);
129 break;
130 }
131 }
132 return priv;
133}
134
135/*
Eric Parisa2d8bc62009-05-21 17:01:37 -0400136 * Add an event to the group notification queue. The group can later pull this
137 * event off the queue to deal with. If the event is successfully added to the
138 * group's notification queue, a reference is taken on event.
139 */
Eric Parisf70ab542010-07-28 10:18:37 -0400140struct fsnotify_event *fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_event *event,
141 struct fsnotify_event_private_data *priv,
142 struct fsnotify_event *(*merge)(struct list_head *,
143 struct fsnotify_event *))
Eric Parisa2d8bc62009-05-21 17:01:37 -0400144{
Eric Parisf70ab542010-07-28 10:18:37 -0400145 struct fsnotify_event *return_event = NULL;
Eric Parisa2d8bc62009-05-21 17:01:37 -0400146 struct fsnotify_event_holder *holder = NULL;
147 struct list_head *list = &group->notification_list;
Eric Parise4aff112009-05-21 17:01:50 -0400148
Eric Paris5ba08e22010-07-28 10:18:37 -0400149 pr_debug("%s: group=%p event=%p priv=%p\n", __func__, group, event, priv);
150
Eric Parisa2d8bc62009-05-21 17:01:37 -0400151 /*
152 * There is one fsnotify_event_holder embedded inside each fsnotify_event.
153 * Check if we expect to be able to use that holder. If not alloc a new
154 * holder.
155 * For the overflow event it's possible that something will use the in
156 * event holder before we get the lock so we may need to jump back and
157 * alloc a new holder, this can't happen for most events...
158 */
159 if (!list_empty(&event->holder.event_list)) {
160alloc_holder:
161 holder = fsnotify_alloc_event_holder();
162 if (!holder)
Eric Parisf70ab542010-07-28 10:18:37 -0400163 return ERR_PTR(-ENOMEM);
Eric Parisa2d8bc62009-05-21 17:01:37 -0400164 }
165
166 mutex_lock(&group->notification_mutex);
167
Eric Parise4aff112009-05-21 17:01:50 -0400168 if (group->q_len >= group->max_events) {
Eric Parisb4277d32009-12-17 20:12:06 -0500169 event = q_overflow_event;
Eric Parisf70ab542010-07-28 10:18:37 -0400170
171 /*
172 * we need to return the overflow event
173 * which means we need a ref
174 */
175 fsnotify_get_event(event);
176 return_event = event;
177
Eric Parise4aff112009-05-21 17:01:50 -0400178 /* sorry, no private data on the overflow event */
179 priv = NULL;
180 }
Eric Parisa2d8bc62009-05-21 17:01:37 -0400181
Eric Paris74766bb2009-12-17 21:24:21 -0500182 if (!list_empty(list) && merge) {
Eric Parisf70ab542010-07-28 10:18:37 -0400183 struct fsnotify_event *tmp;
Eric Paris74766bb2009-12-17 21:24:21 -0500184
Eric Parisf70ab542010-07-28 10:18:37 -0400185 tmp = merge(list, event);
186 if (tmp) {
Eric Paris74766bb2009-12-17 21:24:21 -0500187 mutex_unlock(&group->notification_mutex);
Eric Parisf70ab542010-07-28 10:18:37 -0400188
189 if (return_event)
190 fsnotify_put_event(return_event);
Eric Paris74766bb2009-12-17 21:24:21 -0500191 if (holder != &event->holder)
192 fsnotify_destroy_event_holder(holder);
Eric Parisf70ab542010-07-28 10:18:37 -0400193 return tmp;
Eric Paris74766bb2009-12-17 21:24:21 -0500194 }
195 }
196
Eric Parisa2d8bc62009-05-21 17:01:37 -0400197 spin_lock(&event->lock);
198
199 if (list_empty(&event->holder.event_list)) {
200 if (unlikely(holder))
201 fsnotify_destroy_event_holder(holder);
202 holder = &event->holder;
203 } else if (unlikely(!holder)) {
204 /* between the time we checked above and got the lock the in
205 * event holder was used, go back and get a new one */
206 spin_unlock(&event->lock);
207 mutex_unlock(&group->notification_mutex);
Eric Parisf70ab542010-07-28 10:18:37 -0400208
209 if (return_event) {
210 fsnotify_put_event(return_event);
211 return_event = NULL;
212 }
213
Eric Parisa2d8bc62009-05-21 17:01:37 -0400214 goto alloc_holder;
215 }
216
Eric Parisa2d8bc62009-05-21 17:01:37 -0400217 group->q_len++;
218 holder->event = event;
219
220 fsnotify_get_event(event);
221 list_add_tail(&holder->event_list, list);
Eric Parise4aff112009-05-21 17:01:50 -0400222 if (priv)
223 list_add_tail(&priv->event_list, &event->private_data_list);
Eric Parisa2d8bc62009-05-21 17:01:37 -0400224 spin_unlock(&event->lock);
225 mutex_unlock(&group->notification_mutex);
226
227 wake_up(&group->notification_waitq);
Eric Paris0a6b6bd2011-10-14 17:43:39 -0400228 kill_fasync(&group->fsn_fa, SIGIO, POLL_IN);
Eric Parisf70ab542010-07-28 10:18:37 -0400229 return return_event;
Eric Parisa2d8bc62009-05-21 17:01:37 -0400230}
231
232/*
233 * Remove and return the first event from the notification list. There is a
234 * reference held on this event since it was on the list. It is the responsibility
235 * of the caller to drop this reference.
236 */
237struct fsnotify_event *fsnotify_remove_notify_event(struct fsnotify_group *group)
238{
239 struct fsnotify_event *event;
240 struct fsnotify_event_holder *holder;
241
242 BUG_ON(!mutex_is_locked(&group->notification_mutex));
243
Eric Paris5ba08e22010-07-28 10:18:37 -0400244 pr_debug("%s: group=%p\n", __func__, group);
245
Eric Parisa2d8bc62009-05-21 17:01:37 -0400246 holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
247
248 event = holder->event;
249
250 spin_lock(&event->lock);
251 holder->event = NULL;
252 list_del_init(&holder->event_list);
253 spin_unlock(&event->lock);
254
255 /* event == holder means we are referenced through the in event holder */
256 if (holder != &event->holder)
257 fsnotify_destroy_event_holder(holder);
258
259 group->q_len--;
260
261 return event;
262}
263
264/*
265 * This will not remove the event, that must be done with fsnotify_remove_notify_event()
266 */
267struct fsnotify_event *fsnotify_peek_notify_event(struct fsnotify_group *group)
268{
269 struct fsnotify_event *event;
270 struct fsnotify_event_holder *holder;
271
272 BUG_ON(!mutex_is_locked(&group->notification_mutex));
273
274 holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
275 event = holder->event;
276
277 return event;
278}
279
280/*
281 * Called when a group is being torn down to clean up any outstanding
282 * event notifications.
283 */
284void fsnotify_flush_notify(struct fsnotify_group *group)
285{
286 struct fsnotify_event *event;
Eric Parise4aff112009-05-21 17:01:50 -0400287 struct fsnotify_event_private_data *priv;
Eric Parisa2d8bc62009-05-21 17:01:37 -0400288
289 mutex_lock(&group->notification_mutex);
290 while (!fsnotify_notify_queue_is_empty(group)) {
291 event = fsnotify_remove_notify_event(group);
Eric Parise4aff112009-05-21 17:01:50 -0400292 /* if they don't implement free_event_priv they better not have attached any */
293 if (group->ops->free_event_priv) {
294 spin_lock(&event->lock);
295 priv = fsnotify_remove_priv_from_event(group, event);
296 spin_unlock(&event->lock);
297 if (priv)
298 group->ops->free_event_priv(priv);
299 }
Eric Parisa2d8bc62009-05-21 17:01:37 -0400300 fsnotify_put_event(event); /* matches fsnotify_add_notify_event */
301 }
302 mutex_unlock(&group->notification_mutex);
303}
304
305static void initialize_event(struct fsnotify_event *event)
306{
Eric Parisa2d8bc62009-05-21 17:01:37 -0400307 INIT_LIST_HEAD(&event->holder.event_list);
308 atomic_set(&event->refcnt, 1);
309
310 spin_lock_init(&event->lock);
311
Eric Parise4aff112009-05-21 17:01:50 -0400312 INIT_LIST_HEAD(&event->private_data_list);
Eric Parisa2d8bc62009-05-21 17:01:37 -0400313}
314
Eric Paris1201a532009-12-17 21:24:22 -0500315/*
316 * Caller damn well better be holding whatever mutex is protecting the
Eric Pariscac69da2009-12-17 21:24:22 -0500317 * old_holder->event_list and the new_event must be a clean event which
318 * cannot be found anywhere else in the kernel.
Eric Paris1201a532009-12-17 21:24:22 -0500319 */
320int fsnotify_replace_event(struct fsnotify_event_holder *old_holder,
321 struct fsnotify_event *new_event)
322{
323 struct fsnotify_event *old_event = old_holder->event;
Eric Pariscac69da2009-12-17 21:24:22 -0500324 struct fsnotify_event_holder *new_holder = &new_event->holder;
325
326 enum event_spinlock_class {
327 SPINLOCK_OLD,
328 SPINLOCK_NEW,
329 };
Eric Paris1201a532009-12-17 21:24:22 -0500330
Eric Paris5ba08e22010-07-28 10:18:37 -0400331 pr_debug("%s: old_event=%p new_event=%p\n", __func__, old_event, new_event);
332
Eric Paris1201a532009-12-17 21:24:22 -0500333 /*
Eric Pariscac69da2009-12-17 21:24:22 -0500334 * if the new_event's embedded holder is in use someone
335 * screwed up and didn't give us a clean new event.
Eric Paris1201a532009-12-17 21:24:22 -0500336 */
Eric Pariscac69da2009-12-17 21:24:22 -0500337 BUG_ON(!list_empty(&new_holder->event_list));
Eric Paris1201a532009-12-17 21:24:22 -0500338
Eric Pariscac69da2009-12-17 21:24:22 -0500339 spin_lock_nested(&old_event->lock, SPINLOCK_OLD);
340 spin_lock_nested(&new_event->lock, SPINLOCK_NEW);
Eric Paris1201a532009-12-17 21:24:22 -0500341
342 new_holder->event = new_event;
343 list_replace_init(&old_holder->event_list, &new_holder->event_list);
344
345 spin_unlock(&new_event->lock);
346 spin_unlock(&old_event->lock);
347
348 /* event == holder means we are referenced through the in event holder */
349 if (old_holder != &old_event->holder)
350 fsnotify_destroy_event_holder(old_holder);
351
352 fsnotify_get_event(new_event); /* on the list take reference */
353 fsnotify_put_event(old_event); /* off the list, drop reference */
354
355 return 0;
356}
357
Eric Parisb4e4e142009-12-17 21:24:21 -0500358struct fsnotify_event *fsnotify_clone_event(struct fsnotify_event *old_event)
359{
360 struct fsnotify_event *event;
361
362 event = kmem_cache_alloc(fsnotify_event_cachep, GFP_KERNEL);
363 if (!event)
364 return NULL;
365
Eric Paris5ba08e22010-07-28 10:18:37 -0400366 pr_debug("%s: old_event=%p new_event=%p\n", __func__, old_event, event);
367
Eric Parisb4e4e142009-12-17 21:24:21 -0500368 memcpy(event, old_event, sizeof(*event));
369 initialize_event(event);
370
371 if (event->name_len) {
372 event->file_name = kstrdup(old_event->file_name, GFP_KERNEL);
373 if (!event->file_name) {
374 kmem_cache_free(fsnotify_event_cachep, event);
375 return NULL;
376 }
377 }
Andreas Gruenbacher32c32632009-12-17 21:24:27 -0500378 event->tgid = get_pid(old_event->tgid);
Linus Torvalds20696012010-08-12 14:23:04 -0700379 if (event->data_type == FSNOTIFY_EVENT_PATH)
380 path_get(&event->path);
Eric Parisb4e4e142009-12-17 21:24:21 -0500381
382 return event;
383}
384
Eric Parisa2d8bc62009-05-21 17:01:37 -0400385/*
386 * fsnotify_create_event - Allocate a new event which will be sent to each
387 * group's handle_event function if the group was interested in this
388 * particular event.
389 *
390 * @to_tell the inode which is supposed to receive the event (sometimes a
391 * parent of the inode to which the event happened.
392 * @mask what actually happened.
393 * @data pointer to the object which was actually affected
394 * @data_type flag indication if the data is a file, path, inode, nothing...
Eric Paris62ffe5d2009-05-21 17:01:43 -0400395 * @name the filename, if available
Eric Paris90586522009-05-21 17:01:20 -0400396 */
Eric Paris47882c62009-05-21 17:01:47 -0400397struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask, void *data,
Eric Paris59b0df22010-02-08 12:53:52 -0500398 int data_type, const unsigned char *name,
399 u32 cookie, gfp_t gfp)
Eric Paris90586522009-05-21 17:01:20 -0400400{
401 struct fsnotify_event *event;
402
Eric Paris6f3a5392009-12-17 20:12:07 -0500403 event = kmem_cache_zalloc(fsnotify_event_cachep, gfp);
Eric Paris90586522009-05-21 17:01:20 -0400404 if (!event)
405 return NULL;
406
Eric Paris5ba08e22010-07-28 10:18:37 -0400407 pr_debug("%s: event=%p to_tell=%p mask=%x data=%p data_type=%d\n",
408 __func__, event, to_tell, mask, data, data_type);
409
Eric Parisa2d8bc62009-05-21 17:01:37 -0400410 initialize_event(event);
Eric Paris62ffe5d2009-05-21 17:01:43 -0400411
412 if (name) {
Eric Parisf44aebc2009-07-15 15:49:52 -0400413 event->file_name = kstrdup(name, gfp);
Eric Paris62ffe5d2009-05-21 17:01:43 -0400414 if (!event->file_name) {
415 kmem_cache_free(fsnotify_event_cachep, event);
416 return NULL;
417 }
418 event->name_len = strlen(event->file_name);
419 }
Eric Paris47882c62009-05-21 17:01:47 -0400420
Andreas Gruenbacher32c32632009-12-17 21:24:27 -0500421 event->tgid = get_pid(task_tgid(current));
Eric Paris47882c62009-05-21 17:01:47 -0400422 event->sync_cookie = cookie;
Eric Paris90586522009-05-21 17:01:20 -0400423 event->to_tell = to_tell;
Eric Parisb4e4e142009-12-17 21:24:21 -0500424 event->data_type = data_type;
Eric Paris90586522009-05-21 17:01:20 -0400425
426 switch (data_type) {
Linus Torvalds20696012010-08-12 14:23:04 -0700427 case FSNOTIFY_EVENT_PATH: {
428 struct path *path = data;
429 event->path.dentry = path->dentry;
430 event->path.mnt = path->mnt;
431 path_get(&event->path);
Eric Paris90586522009-05-21 17:01:20 -0400432 break;
433 }
434 case FSNOTIFY_EVENT_INODE:
435 event->inode = data;
Eric Paris90586522009-05-21 17:01:20 -0400436 break;
437 case FSNOTIFY_EVENT_NONE:
438 event->inode = NULL;
Linus Torvalds20696012010-08-12 14:23:04 -0700439 event->path.dentry = NULL;
440 event->path.mnt = NULL;
Eric Paris90586522009-05-21 17:01:20 -0400441 break;
442 default:
443 BUG();
444 }
445
446 event->mask = mask;
447
448 return event;
449}
450
H Hartley Sweeten9710a782012-03-23 15:01:53 -0700451static __init int fsnotify_notification_init(void)
Eric Paris90586522009-05-21 17:01:20 -0400452{
453 fsnotify_event_cachep = KMEM_CACHE(fsnotify_event, SLAB_PANIC);
Eric Parisa2d8bc62009-05-21 17:01:37 -0400454 fsnotify_event_holder_cachep = KMEM_CACHE(fsnotify_event_holder, SLAB_PANIC);
455
Eric Parisb4277d32009-12-17 20:12:06 -0500456 q_overflow_event = fsnotify_create_event(NULL, FS_Q_OVERFLOW, NULL,
457 FSNOTIFY_EVENT_NONE, NULL, 0,
458 GFP_KERNEL);
459 if (!q_overflow_event)
460 panic("unable to allocate fsnotify q_overflow_event\n");
Eric Paris90586522009-05-21 17:01:20 -0400461
462 return 0;
463}
464subsys_initcall(fsnotify_notification_init);