[net-next-2.6.git] / fs / notify / notification.c

/*
 *  Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 2, or (at your option)
 *  any later version.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program; see the file COPYING.  If not, write to
 *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
 */

/*
 * Basic idea behind the notification queue: An fsnotify group (like inotify)
 * sends the userspace notification about events asyncronously some time after
 * the event happened.  When inotify gets an event it will need to add that
 * event to the group notify queue.  Since a single event might need to be on
 * multiple group's notification queues we can't add the event directly to each
 * queue and instead add a small "event_holder" to each queue.  This event_holder
 * has a pointer back to the original event.  Since the majority of events are
 * going to end up on one, and only one, notification queue we embed one
 * event_holder into each event.  This means we have a single allocation instead
 * of always needing two.  If the embedded event_holder is already in use by
 * another group a new event_holder (from fsnotify_event_holder_cachep) will be
 * allocated and used.
 */

#include <linux/fs.h>
#include <linux/init.h>
#include <linux/kernel.h>
#include <linux/list.h>
#include <linux/mount.h>
#include <linux/mutex.h>
#include <linux/namei.h>
#include <linux/path.h>
#include <linux/slab.h>
#include <linux/spinlock.h>

#include <asm/atomic.h>

#include <linux/fsnotify_backend.h>
#include "fsnotify.h"

static struct kmem_cache *fsnotify_event_cachep;
static struct kmem_cache *fsnotify_event_holder_cachep;
/*
 * This is a magic event we send when the q is too full.  Since it doesn't
 * hold real event information we just keep one system wide and use it any time
 * it is needed.  It's refcnt is set 1 at kernel init time and will never
 * get set to 0 so it will never get 'freed'
 */
static struct fsnotify_event q_overflow_event;

/* return true if the notify queue is empty, false otherwise */
bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group)
{
	BUG_ON(!mutex_is_locked(&group->notification_mutex));
	return list_empty(&group->notification_list) ? true : false;
}

void fsnotify_get_event(struct fsnotify_event *event)
{
	atomic_inc(&event->refcnt);
}

void fsnotify_put_event(struct fsnotify_event *event)
{
	if (!event)
		return;

	if (atomic_dec_and_test(&event->refcnt)) {
		if (event->data_type == FSNOTIFY_EVENT_PATH)
			path_put(&event->path);

		kfree(event->file_name);
		kmem_cache_free(fsnotify_event_cachep, event);
	}
}

struct fsnotify_event_holder *fsnotify_alloc_event_holder(void)
{
	return kmem_cache_alloc(fsnotify_event_holder_cachep, GFP_KERNEL);
}

void fsnotify_destroy_event_holder(struct fsnotify_event_holder *holder)
{
	kmem_cache_free(fsnotify_event_holder_cachep, holder);
}

/*
 * check if 2 events contain the same information.
 */
static bool event_compare(struct fsnotify_event *old, struct fsnotify_event *new)
{
	if ((old->mask == new->mask) &&
	    (old->to_tell == new->to_tell) &&
	    (old->data_type == new->data_type)) {
		switch (old->data_type) {
		case (FSNOTIFY_EVENT_INODE):
			if (old->inode == new->inode)
				return true;
			break;
		case (FSNOTIFY_EVENT_PATH):
			if ((old->path.mnt == new->path.mnt) &&
			    (old->path.dentry == new->path.dentry))
				return true;
		case (FSNOTIFY_EVENT_NONE):
			return true;
		};
	}
	return false;
}

/*
 * Add an event to the group notification queue.  The group can later pull this
 * event off the queue to deal with.  If the event is successfully added to the
 * group's notification queue, a reference is taken on event.
 */
int fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_event *event)
{
	struct fsnotify_event_holder *holder = NULL;
	struct list_head *list = &group->notification_list;
	struct fsnotify_event_holder *last_holder;
	struct fsnotify_event *last_event;

	/*
	 * There is one fsnotify_event_holder embedded inside each fsnotify_event.
	 * Check if we expect to be able to use that holder.  If not alloc a new
	 * holder.
	 * For the overflow event it's possible that something will use the in
	 * event holder before we get the lock so we may need to jump back and
	 * alloc a new holder, this can't happen for most events...
	 */
	if (!list_empty(&event->holder.event_list)) {
alloc_holder:
		holder = fsnotify_alloc_event_holder();
		if (!holder)
			return -ENOMEM;
	}

	mutex_lock(&group->notification_mutex);

	if (group->q_len >= group->max_events)
		event = &q_overflow_event;

	spin_lock(&event->lock);

	if (list_empty(&event->holder.event_list)) {
		if (unlikely(holder))
			fsnotify_destroy_event_holder(holder);
		holder = &event->holder;
	} else if (unlikely(!holder)) {
		/* between the time we checked above and got the lock the in
		 * event holder was used, go back and get a new one */
		spin_unlock(&event->lock);
		mutex_unlock(&group->notification_mutex);
		goto alloc_holder;
	}

	if (!list_empty(list)) {
		last_holder = list_entry(list->prev, struct fsnotify_event_holder, event_list);
		last_event = last_holder->event;
		if (event_compare(last_event, event)) {
			spin_unlock(&event->lock);
			mutex_unlock(&group->notification_mutex);
			if (holder != &event->holder)
				fsnotify_destroy_event_holder(holder);
			return 0;
		}
	}

	group->q_len++;
	holder->event = event;

	fsnotify_get_event(event);
	list_add_tail(&holder->event_list, list);
	spin_unlock(&event->lock);
	mutex_unlock(&group->notification_mutex);

	wake_up(&group->notification_waitq);
	return 0;
}

/*
 * Remove and return the first event from the notification list.  There is a
 * reference held on this event since it was on the list.  It is the responsibility
 * of the caller to drop this reference.
 */
struct fsnotify_event *fsnotify_remove_notify_event(struct fsnotify_group *group)
{
	struct fsnotify_event *event;
	struct fsnotify_event_holder *holder;

	BUG_ON(!mutex_is_locked(&group->notification_mutex));

	holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);

	event = holder->event;

	spin_lock(&event->lock);
	holder->event = NULL;
	list_del_init(&holder->event_list);
	spin_unlock(&event->lock);

	/* event == holder means we are referenced through the in event holder */
	if (holder != &event->holder)
		fsnotify_destroy_event_holder(holder);

	group->q_len--;

	return event;
}

/*
 * This will not remove the event, that must be done with fsnotify_remove_notify_event()
 */
struct fsnotify_event *fsnotify_peek_notify_event(struct fsnotify_group *group)
{
	struct fsnotify_event *event;
	struct fsnotify_event_holder *holder;

	BUG_ON(!mutex_is_locked(&group->notification_mutex));

	holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
	event = holder->event;

	return event;
}

/*
 * Called when a group is being torn down to clean up any outstanding
 * event notifications.
 */
void fsnotify_flush_notify(struct fsnotify_group *group)
{
	struct fsnotify_event *event;

	mutex_lock(&group->notification_mutex);
	while (!fsnotify_notify_queue_is_empty(group)) {
		event = fsnotify_remove_notify_event(group);
		fsnotify_put_event(event); /* matches fsnotify_add_notify_event */
	}
	mutex_unlock(&group->notification_mutex);
}

static void initialize_event(struct fsnotify_event *event)
{
	event->holder.event = NULL;
	INIT_LIST_HEAD(&event->holder.event_list);
	atomic_set(&event->refcnt, 1);

	spin_lock_init(&event->lock);

	event->path.dentry = NULL;
	event->path.mnt = NULL;
	event->inode = NULL;
	event->data_type = FSNOTIFY_EVENT_NONE;

	event->to_tell = NULL;

	event->file_name = NULL;
	event->name_len = 0;
}

/*
 * fsnotify_create_event - Allocate a new event which will be sent to each
 * group's handle_event function if the group was interested in this
 * particular event.
 *
 * @to_tell the inode which is supposed to receive the event (sometimes a
 *	parent of the inode to which the event happened.
 * @mask what actually happened.
 * @data pointer to the object which was actually affected
 * @data_type flag indication if the data is a file, path, inode, nothing...
 * @name the filename, if available
 */
struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask,
					     void *data, int data_type, const char *name)
{
	struct fsnotify_event *event;

	event = kmem_cache_alloc(fsnotify_event_cachep, GFP_KERNEL);
	if (!event)
		return NULL;

	initialize_event(event);

	if (name) {
		event->file_name = kstrdup(name, GFP_KERNEL);
		if (!event->file_name) {
			kmem_cache_free(fsnotify_event_cachep, event);
			return NULL;
		}
		event->name_len = strlen(event->file_name);
	}
	event->to_tell = to_tell;

	switch (data_type) {
	case FSNOTIFY_EVENT_FILE: {
		struct file *file = data;
		struct path *path = &file->f_path;
		event->path.dentry = path->dentry;
		event->path.mnt = path->mnt;
		path_get(&event->path);
		event->data_type = FSNOTIFY_EVENT_PATH;
		break;
	}
	case FSNOTIFY_EVENT_PATH: {
		struct path *path = data;
		event->path.dentry = path->dentry;
		event->path.mnt = path->mnt;
		path_get(&event->path);
		event->data_type = FSNOTIFY_EVENT_PATH;
		break;
	}
	case FSNOTIFY_EVENT_INODE:
		event->inode = data;
		event->data_type = FSNOTIFY_EVENT_INODE;
		break;
	case FSNOTIFY_EVENT_NONE:
		event->inode = NULL;
		event->path.dentry = NULL;
		event->path.mnt = NULL;
		break;
	default:
		BUG();
	}

	event->mask = mask;

	return event;
}

__init int fsnotify_notification_init(void)
{
	fsnotify_event_cachep = KMEM_CACHE(fsnotify_event, SLAB_PANIC);
	fsnotify_event_holder_cachep = KMEM_CACHE(fsnotify_event_holder, SLAB_PANIC);

	initialize_event(&q_overflow_event);
	q_overflow_event.mask = FS_Q_OVERFLOW;

	return 0;
}
subsys_initcall(fsnotify_notification_init);
Commit	Line	Data
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>
	38	#include <linux/mount.h>
	39	#include <linux/mutex.h>
	40	#include <linux/namei.h>
	41	#include <linux/path.h>
	42	#include <linux/slab.h>
	43	#include <linux/spinlock.h>
	44
	45	#include <asm/atomic.h>
	46
	47	#include <linux/fsnotify_backend.h>
	48	#include "fsnotify.h"
	49
	50	static struct kmem_cache *fsnotify_event_cachep;
a2d8bc6c EP	51	static struct kmem_cache *fsnotify_event_holder_cachep;
	52	/*
	53	* This is a magic event we send when the q is too full. Since it doesn't
	54	* hold real event information we just keep one system wide and use it any time
	55	* it is needed. It's refcnt is set 1 at kernel init time and will never
	56	* get set to 0 so it will never get 'freed'
	57	*/
	58	static struct fsnotify_event q_overflow_event;
	59
	60	/* return true if the notify queue is empty, false otherwise */
	61	bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group)
	62	{
	63	BUG_ON(!mutex_is_locked(&group->notification_mutex));
	64	return list_empty(&group->notification_list) ? true : false;
	65	}
90586523 EP	66
	67	void fsnotify_get_event(struct fsnotify_event *event)
	68	{
	69	atomic_inc(&event->refcnt);
	70	}
	71
	72	void fsnotify_put_event(struct fsnotify_event *event)
	73	{
	74	if (!event)
	75	return;
	76
	77	if (atomic_dec_and_test(&event->refcnt)) {
	78	if (event->data_type == FSNOTIFY_EVENT_PATH)
	79	path_put(&event->path);
	80
62ffe5df	81	kfree(event->file_name);
90586523 EP	82	kmem_cache_free(fsnotify_event_cachep, event);
	83	}
	84	}
	85
a2d8bc6c EP	86	struct fsnotify_event_holder *fsnotify_alloc_event_holder(void)
	87	{
	88	return kmem_cache_alloc(fsnotify_event_holder_cachep, GFP_KERNEL);
	89	}
	90
	91	void fsnotify_destroy_event_holder(struct fsnotify_event_holder *holder)
	92	{
	93	kmem_cache_free(fsnotify_event_holder_cachep, holder);
	94	}
	95
	96	/*
	97	* check if 2 events contain the same information.
	98	*/
	99	static bool event_compare(struct fsnotify_event old, struct fsnotify_event new)
	100	{
	101	if ((old->mask == new->mask) &&
	102	(old->to_tell == new->to_tell) &&
	103	(old->data_type == new->data_type)) {
	104	switch (old->data_type) {
	105	case (FSNOTIFY_EVENT_INODE):
	106	if (old->inode == new->inode)
	107	return true;
	108	break;
	109	case (FSNOTIFY_EVENT_PATH):
	110	if ((old->path.mnt == new->path.mnt) &&
	111	(old->path.dentry == new->path.dentry))
	112	return true;
	113	case (FSNOTIFY_EVENT_NONE):
	114	return true;
	115	};
	116	}
	117	return false;
	118	}
	119
90586523	120	/*
a2d8bc6c EP	121	* Add an event to the group notification queue. The group can later pull this
	122	* event off the queue to deal with. If the event is successfully added to the
	123	* group's notification queue, a reference is taken on event.
90586523	124	*/
a2d8bc6c EP	125	int fsnotify_add_notify_event(struct fsnotify_group group, struct fsnotify_event event)
	126	{
	127	struct fsnotify_event_holder *holder = NULL;
	128	struct list_head *list = &group->notification_list;
	129	struct fsnotify_event_holder *last_holder;
	130	struct fsnotify_event *last_event;
	131
	132	/*
	133	* There is one fsnotify_event_holder embedded inside each fsnotify_event.
	134	* Check if we expect to be able to use that holder. If not alloc a new
	135	* holder.
	136	* For the overflow event it's possible that something will use the in
	137	* event holder before we get the lock so we may need to jump back and
	138	* alloc a new holder, this can't happen for most events...
	139	*/
	140	if (!list_empty(&event->holder.event_list)) {
	141	alloc_holder:
	142	holder = fsnotify_alloc_event_holder();
	143	if (!holder)
	144	return -ENOMEM;
	145	}
	146
	147	mutex_lock(&group->notification_mutex);
	148
	149	if (group->q_len >= group->max_events)
	150	event = &q_overflow_event;
	151
	152	spin_lock(&event->lock);
	153
	154	if (list_empty(&event->holder.event_list)) {
	155	if (unlikely(holder))
	156	fsnotify_destroy_event_holder(holder);
	157	holder = &event->holder;
	158	} else if (unlikely(!holder)) {
	159	/* between the time we checked above and got the lock the in
	160	* event holder was used, go back and get a new one */
	161	spin_unlock(&event->lock);
	162	mutex_unlock(&group->notification_mutex);
	163	goto alloc_holder;
	164	}
	165
	166	if (!list_empty(list)) {
	167	last_holder = list_entry(list->prev, struct fsnotify_event_holder, event_list);
	168	last_event = last_holder->event;
	169	if (event_compare(last_event, event)) {
	170	spin_unlock(&event->lock);
	171	mutex_unlock(&group->notification_mutex);
	172	if (holder != &event->holder)
	173	fsnotify_destroy_event_holder(holder);
	174	return 0;
	175	}
	176	}
	177
	178	group->q_len++;
	179	holder->event = event;
	180
	181	fsnotify_get_event(event);
	182	list_add_tail(&holder->event_list, list);
	183	spin_unlock(&event->lock);
	184	mutex_unlock(&group->notification_mutex);
	185
	186	wake_up(&group->notification_waitq);
	187	return 0;
	188	}
189
190	/*
191	* Remove and return the first event from the notification list. There is a
192	* reference held on this event since it was on the list. It is the responsibility
193	* of the caller to drop this reference.
194	*/
195	struct fsnotify_event fsnotify_remove_notify_event(struct fsnotify_group group)
90586523 EP	196	{
90586523 EP	197	struct fsnotify_event *event;
a2d8bc6c	198	struct fsnotify_event_holder *holder;
90586523	199
a2d8bc6c	200	BUG_ON(!mutex_is_locked(&group->notification_mutex));
90586523	201
a2d8bc6c EP	202	holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
	203
	204	event = holder->event;
	205
	206	spin_lock(&event->lock);
	207	holder->event = NULL;
	208	list_del_init(&holder->event_list);
	209	spin_unlock(&event->lock);
	210
	211	/* event == holder means we are referenced through the in event holder */
	212	if (holder != &event->holder)
	213	fsnotify_destroy_event_holder(holder);
	214
	215	group->q_len--;
	216
	217	return event;
	218	}
	219
	220	/*
	221	* This will not remove the event, that must be done with fsnotify_remove_notify_event()
	222	*/
	223	struct fsnotify_event fsnotify_peek_notify_event(struct fsnotify_group group)
	224	{
	225	struct fsnotify_event *event;
	226	struct fsnotify_event_holder *holder;
	227
	228	BUG_ON(!mutex_is_locked(&group->notification_mutex));
	229
	230	holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
	231	event = holder->event;
	232
	233	return event;
	234	}
	235
	236	/*
	237	* Called when a group is being torn down to clean up any outstanding
	238	* event notifications.
	239	*/
	240	void fsnotify_flush_notify(struct fsnotify_group *group)
	241	{
	242	struct fsnotify_event *event;
	243
	244	mutex_lock(&group->notification_mutex);
	245	while (!fsnotify_notify_queue_is_empty(group)) {
	246	event = fsnotify_remove_notify_event(group);
	247	fsnotify_put_event(event); /* matches fsnotify_add_notify_event */
	248	}
	249	mutex_unlock(&group->notification_mutex);
	250	}
	251
	252	static void initialize_event(struct fsnotify_event *event)
	253	{
	254	event->holder.event = NULL;
	255	INIT_LIST_HEAD(&event->holder.event_list);
90586523 EP	256	atomic_set(&event->refcnt, 1);
	257
	258	spin_lock_init(&event->lock);
	259
	260	event->path.dentry = NULL;
	261	event->path.mnt = NULL;
	262	event->inode = NULL;
a2d8bc6c	263	event->data_type = FSNOTIFY_EVENT_NONE;
90586523	264
a2d8bc6c	265	event->to_tell = NULL;
62ffe5df EP	266
	267	event->file_name = NULL;
	268	event->name_len = 0;
a2d8bc6c EP	269	}
	270
	271	/*
	272	* fsnotify_create_event - Allocate a new event which will be sent to each
	273	* group's handle_event function if the group was interested in this
	274	* particular event.
	275	*
	276	* @to_tell the inode which is supposed to receive the event (sometimes a
	277	* parent of the inode to which the event happened.
	278	* @mask what actually happened.
	279	* @data pointer to the object which was actually affected
	280	* @data_type flag indication if the data is a file, path, inode, nothing...
62ffe5df	281	* @name the filename, if available
a2d8bc6c EP	282	*/
a2d8bc6c EP	283	struct fsnotify_event fsnotify_create_event(struct inode to_tell, __u32 mask,
62ffe5df	284	void data, int data_type, const char name)
a2d8bc6c EP	285	{
	286	struct fsnotify_event *event;
	287
	288	event = kmem_cache_alloc(fsnotify_event_cachep, GFP_KERNEL);
	289	if (!event)
	290	return NULL;
	291
	292	initialize_event(event);
62ffe5df EP	293
	294	if (name) {
	295	event->file_name = kstrdup(name, GFP_KERNEL);
	296	if (!event->file_name) {
	297	kmem_cache_free(fsnotify_event_cachep, event);
	298	return NULL;
	299	}
	300	event->name_len = strlen(event->file_name);
	301	}
90586523 EP	302	event->to_tell = to_tell;
	303
	304	switch (data_type) {
	305	case FSNOTIFY_EVENT_FILE: {
	306	struct file *file = data;
	307	struct path *path = &file->f_path;
	308	event->path.dentry = path->dentry;
	309	event->path.mnt = path->mnt;
	310	path_get(&event->path);
	311	event->data_type = FSNOTIFY_EVENT_PATH;
	312	break;
	313	}
	314	case FSNOTIFY_EVENT_PATH: {
	315	struct path *path = data;
	316	event->path.dentry = path->dentry;
	317	event->path.mnt = path->mnt;
	318	path_get(&event->path);
	319	event->data_type = FSNOTIFY_EVENT_PATH;
	320	break;
	321	}
	322	case FSNOTIFY_EVENT_INODE:
	323	event->inode = data;
	324	event->data_type = FSNOTIFY_EVENT_INODE;
	325	break;
	326	case FSNOTIFY_EVENT_NONE:
	327	event->inode = NULL;
	328	event->path.dentry = NULL;
	329	event->path.mnt = NULL;
	330	break;
	331	default:
	332	BUG();
	333	}
	334
	335	event->mask = mask;
	336
	337	return event;
	338	}
	339
	340	__init int fsnotify_notification_init(void)
	341	{
	342	fsnotify_event_cachep = KMEM_CACHE(fsnotify_event, SLAB_PANIC);
a2d8bc6c EP	343	fsnotify_event_holder_cachep = KMEM_CACHE(fsnotify_event_holder, SLAB_PANIC);
	344
	345	initialize_event(&q_overflow_event);
	346	q_overflow_event.mask = FS_Q_OVERFLOW;
90586523 EP	347
	348	return 0;
	349	}
	350	subsys_initcall(fsnotify_notification_init);
	351