[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/module.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;
static atomic_t fsnotify_sync_cookie = ATOMIC_INIT(0);

/**
 * fsnotify_get_cookie - return a unique cookie for use in synchronizing events.
 * Called from fsnotify_move, which is inlined into filesystem modules.
 */
u32 fsnotify_get_cookie(void)
{
	return atomic_inc_return(&fsnotify_sync_cookie);
}
EXPORT_SYMBOL_GPL(fsnotify_get_cookie);

/* 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);

		BUG_ON(!list_empty(&event->private_data_list));

		kfree(event->file_name);
		put_pid(event->tgid);
		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)
{
	if (holder)
		kmem_cache_free(fsnotify_event_holder_cachep, holder);
}

/*
 * Find the private data that the group previously attached to this event when
 * the group added the event to the notification queue (fsnotify_add_notify_event)
 */
struct fsnotify_event_private_data *fsnotify_remove_priv_from_event(struct fsnotify_group *group, struct fsnotify_event *event)
{
	struct fsnotify_event_private_data *lpriv;
	struct fsnotify_event_private_data *priv = NULL;

	assert_spin_locked(&event->lock);

	list_for_each_entry(lpriv, &event->private_data_list, event_list) {
		if (lpriv->group == group) {
			priv = lpriv;
			list_del(&priv->event_list);
			break;
		}
	}
	return priv;
}

/*
 * 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_private_data *priv,
			      int (*merge)(struct list_head *,
					   struct fsnotify_event *,
					   void **arg),
			      void **arg)
{
	struct fsnotify_event_holder *holder = NULL;
	struct list_head *list = &group->notification_list;
	int rc = 0;

	/*
	 * 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;
		rc = -EOVERFLOW;
		/* sorry, no private data on the overflow event */
		priv = NULL;
	}

	if (!list_empty(list) && merge) {
		int ret;

		ret = merge(list, event, arg);
		if (ret) {
			mutex_unlock(&group->notification_mutex);
			if (holder != &event->holder)
				fsnotify_destroy_event_holder(holder);
			return ret;
		}
	}

	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;
	}

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

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

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

/*
 * 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;
	struct fsnotify_event_private_data *priv;

	mutex_lock(&group->notification_mutex);
	while (!fsnotify_notify_queue_is_empty(group)) {
		event = fsnotify_remove_notify_event(group);
		/* if they don't implement free_event_priv they better not have attached any */
		if (group->ops->free_event_priv) {
			spin_lock(&event->lock);
			priv = fsnotify_remove_priv_from_event(group, event);
			spin_unlock(&event->lock);
			if (priv)
				group->ops->free_event_priv(priv);
		}
		fsnotify_put_event(event); /* matches fsnotify_add_notify_event */
	}
	mutex_unlock(&group->notification_mutex);
}

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

	spin_lock_init(&event->lock);

	INIT_LIST_HEAD(&event->private_data_list);
}

/*
 * Caller damn well better be holding whatever mutex is protecting the
 * old_holder->event_list and the new_event must be a clean event which
 * cannot be found anywhere else in the kernel.
 */
int fsnotify_replace_event(struct fsnotify_event_holder *old_holder,
			   struct fsnotify_event *new_event)
{
	struct fsnotify_event *old_event = old_holder->event;
	struct fsnotify_event_holder *new_holder = &new_event->holder;

	enum event_spinlock_class {
		SPINLOCK_OLD,
		SPINLOCK_NEW,
	};

	/*
	 * if the new_event's embedded holder is in use someone
	 * screwed up and didn't give us a clean new event.
	 */
	BUG_ON(!list_empty(&new_holder->event_list));

	spin_lock_nested(&old_event->lock, SPINLOCK_OLD);
	spin_lock_nested(&new_event->lock, SPINLOCK_NEW);

	new_holder->event = new_event;
	list_replace_init(&old_holder->event_list, &new_holder->event_list);

	spin_unlock(&new_event->lock);
	spin_unlock(&old_event->lock);

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

	fsnotify_get_event(new_event); /* on the list take reference */
	fsnotify_put_event(old_event); /* off the list, drop reference */

	return 0;
}

struct fsnotify_event *fsnotify_clone_event(struct fsnotify_event *old_event)
{
	struct fsnotify_event *event;

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

	memcpy(event, old_event, sizeof(*event));
	initialize_event(event);

	if (event->name_len) {
		event->file_name = kstrdup(old_event->file_name, GFP_KERNEL);
		if (!event->file_name) {
			kmem_cache_free(fsnotify_event_cachep, event);
			return NULL;
		}
	}
	event->tgid = get_pid(old_event->tgid);
	if (event->data_type == FSNOTIFY_EVENT_PATH)
		path_get(&event->path);

	return event;
}

/*
 * 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, u32 cookie,
					     gfp_t gfp)
{
	struct fsnotify_event *event;

	event = kmem_cache_zalloc(fsnotify_event_cachep, gfp);
	if (!event)
		return NULL;

	initialize_event(event);

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

	event->tgid = get_pid(task_tgid(current));
	event->sync_cookie = cookie;
	event->to_tell = to_tell;
	event->data_type = data_type;

	switch (data_type) {
	case FSNOTIFY_EVENT_PATH: {
		struct path *path = data;
		event->path.dentry = path->dentry;
		event->path.mnt = path->mnt;
		path_get(&event->path);
		break;
	}
	case FSNOTIFY_EVENT_INODE:
		event->inode = data;
		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);

	q_overflow_event = fsnotify_create_event(NULL, FS_Q_OVERFLOW, NULL,
						 FSNOTIFY_EVENT_NONE, NULL, 0,
						 GFP_KERNEL);
	if (!q_overflow_event)
		panic("unable to allocate fsnotify q_overflow_event\n");

	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>
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
	51	static struct kmem_cache *fsnotify_event_cachep;
a2d8bc6c EP	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	*/
b4277d3d	59	static struct fsnotify_event *q_overflow_event;
47882c6f EP	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);
a2d8bc6c EP	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	}
90586523 EP	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
e4aff117 EP	93	BUG_ON(!list_empty(&event->private_data_list));
e4aff117 EP	94
62ffe5df	95	kfree(event->file_name);
32c32632	96	put_pid(event->tgid);
90586523 EP	97	kmem_cache_free(fsnotify_event_cachep, event);
	98	}
	99	}
	100
a2d8bc6c EP	101	struct fsnotify_event_holder *fsnotify_alloc_event_holder(void)
	102	{
	103	return kmem_cache_alloc(fsnotify_event_holder_cachep, GFP_KERNEL);
	104	}
	105
	106	void fsnotify_destroy_event_holder(struct fsnotify_event_holder *holder)
	107	{
74766bbf EP	108	if (holder)
74766bbf EP	109	kmem_cache_free(fsnotify_event_holder_cachep, holder);
a2d8bc6c EP	110	}
	111
	112	/*
e4aff117 EP	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)
	115	*/
	116	struct fsnotify_event_private_data fsnotify_remove_priv_from_event(struct fsnotify_group group, struct fsnotify_event *event)
	117	{
	118	struct fsnotify_event_private_data *lpriv;
	119	struct fsnotify_event_private_data *priv = NULL;
	120
	121	assert_spin_locked(&event->lock);
	122
	123	list_for_each_entry(lpriv, &event->private_data_list, event_list) {
	124	if (lpriv->group == group) {
	125	priv = lpriv;
	126	list_del(&priv->event_list);
	127	break;
	128	}
	129	}
	130	return priv;
	131	}
	132
90586523	133	/*
a2d8bc6c EP	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.
90586523	137	*/
e4aff117	138	int fsnotify_add_notify_event(struct fsnotify_group group, struct fsnotify_event event,
74766bbf	139	struct fsnotify_event_private_data *priv,
6e5f77b3 EP	140	int (merge)(struct list_head ,
	141	struct fsnotify_event *,
	142	void **arg),
	143	void **arg)
a2d8bc6c EP	144	{
	145	struct fsnotify_event_holder *holder = NULL;
	146	struct list_head *list = &group->notification_list;
74766bbf	147	int rc = 0;
e4aff117	148
a2d8bc6c EP	149	/*
	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
	152	* holder.
	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...
	156	*/
	157	if (!list_empty(&event->holder.event_list)) {
	158	alloc_holder:
	159	holder = fsnotify_alloc_event_holder();
	160	if (!holder)
	161	return -ENOMEM;
	162	}
	163
	164	mutex_lock(&group->notification_mutex);
	165
e4aff117	166	if (group->q_len >= group->max_events) {
b4277d3d	167	event = q_overflow_event;
74766bbf	168	rc = -EOVERFLOW;
e4aff117 EP	169	/* sorry, no private data on the overflow event */
	170	priv = NULL;
	171	}
a2d8bc6c	172
74766bbf EP	173	if (!list_empty(list) && merge) {
	174	int ret;
	175
6e5f77b3	176	ret = merge(list, event, arg);
74766bbf EP	177	if (ret) {
	178	mutex_unlock(&group->notification_mutex);
	179	if (holder != &event->holder)
	180	fsnotify_destroy_event_holder(holder);
	181	return ret;
	182	}
	183	}
	184
a2d8bc6c EP	185	spin_lock(&event->lock);
	186
	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);
	196	goto alloc_holder;
	197	}
	198
a2d8bc6c EP	199	group->q_len++;
	200	holder->event = event;
	201
	202	fsnotify_get_event(event);
	203	list_add_tail(&holder->event_list, list);
e4aff117 EP	204	if (priv)
e4aff117 EP	205	list_add_tail(&priv->event_list, &event->private_data_list);
a2d8bc6c EP	206	spin_unlock(&event->lock);
	207	mutex_unlock(&group->notification_mutex);
	208
	209	wake_up(&group->notification_waitq);
74766bbf	210	return rc;
a2d8bc6c EP	211	}
	212
	213	/*
	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.
	217	*/
	218	struct fsnotify_event fsnotify_remove_notify_event(struct fsnotify_group group)
90586523 EP	219	{
90586523 EP	220	struct fsnotify_event *event;
a2d8bc6c	221	struct fsnotify_event_holder *holder;
90586523	222
a2d8bc6c	223	BUG_ON(!mutex_is_locked(&group->notification_mutex));
90586523	224
a2d8bc6c EP	225	holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
	226
	227	event = holder->event;
	228
	229	spin_lock(&event->lock);
	230	holder->event = NULL;
	231	list_del_init(&holder->event_list);
	232	spin_unlock(&event->lock);
	233
	234	/* event == holder means we are referenced through the in event holder */
	235	if (holder != &event->holder)
	236	fsnotify_destroy_event_holder(holder);
	237
	238	group->q_len--;
	239
	240	return event;
	241	}
	242
	243	/*
	244	* This will not remove the event, that must be done with fsnotify_remove_notify_event()
	245	*/
	246	struct fsnotify_event fsnotify_peek_notify_event(struct fsnotify_group group)
	247	{
	248	struct fsnotify_event *event;
	249	struct fsnotify_event_holder *holder;
	250
	251	BUG_ON(!mutex_is_locked(&group->notification_mutex));
	252
	253	holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
	254	event = holder->event;
	255
	256	return event;
	257	}
	258
	259	/*
	260	* Called when a group is being torn down to clean up any outstanding
	261	* event notifications.
	262	*/
	263	void fsnotify_flush_notify(struct fsnotify_group *group)
	264	{
	265	struct fsnotify_event *event;
e4aff117	266	struct fsnotify_event_private_data *priv;
a2d8bc6c EP	267
	268	mutex_lock(&group->notification_mutex);
	269	while (!fsnotify_notify_queue_is_empty(group)) {
	270	event = fsnotify_remove_notify_event(group);
e4aff117 EP	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);
	276	if (priv)
	277	group->ops->free_event_priv(priv);
	278	}
a2d8bc6c EP	279	fsnotify_put_event(event); /* matches fsnotify_add_notify_event */
	280	}
	281	mutex_unlock(&group->notification_mutex);
	282	}
	283
	284	static void initialize_event(struct fsnotify_event *event)
	285	{
a2d8bc6c	286	INIT_LIST_HEAD(&event->holder.event_list);
90586523 EP	287	atomic_set(&event->refcnt, 1);
	288
	289	spin_lock_init(&event->lock);
	290
e4aff117	291	INIT_LIST_HEAD(&event->private_data_list);
a2d8bc6c EP	292	}
a2d8bc6c EP	293
1201a536 EP	294	/*
1201a536 EP	295	* Caller damn well better be holding whatever mutex is protecting the
cac69dad EP	296	* old_holder->event_list and the new_event must be a clean event which
cac69dad EP	297	* cannot be found anywhere else in the kernel.
1201a536 EP	298	*/
	299	int fsnotify_replace_event(struct fsnotify_event_holder *old_holder,
	300	struct fsnotify_event *new_event)
	301	{
	302	struct fsnotify_event *old_event = old_holder->event;
cac69dad EP	303	struct fsnotify_event_holder *new_holder = &new_event->holder;
	304
	305	enum event_spinlock_class {
	306	SPINLOCK_OLD,
	307	SPINLOCK_NEW,
	308	};
1201a536 EP	309
1201a536 EP	310	/*
cac69dad EP	311	* if the new_event's embedded holder is in use someone
cac69dad EP	312	* screwed up and didn't give us a clean new event.
1201a536	313	*/
cac69dad	314	BUG_ON(!list_empty(&new_holder->event_list));
1201a536	315
cac69dad EP	316	spin_lock_nested(&old_event->lock, SPINLOCK_OLD);
cac69dad EP	317	spin_lock_nested(&new_event->lock, SPINLOCK_NEW);
1201a536 EP	318
	319	new_holder->event = new_event;
	320	list_replace_init(&old_holder->event_list, &new_holder->event_list);
	321
	322	spin_unlock(&new_event->lock);
	323	spin_unlock(&old_event->lock);
	324
	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);
	328
	329	fsnotify_get_event(new_event); /* on the list take reference */
	330	fsnotify_put_event(old_event); /* off the list, drop reference */
	331
	332	return 0;
	333	}
	334
b4e4e140 EP	335	struct fsnotify_event fsnotify_clone_event(struct fsnotify_event old_event)
	336	{
	337	struct fsnotify_event *event;
	338
	339	event = kmem_cache_alloc(fsnotify_event_cachep, GFP_KERNEL);
	340	if (!event)
	341	return NULL;
	342
	343	memcpy(event, old_event, sizeof(*event));
	344	initialize_event(event);
	345
	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);
	350	return NULL;
	351	}
	352	}
32c32632	353	event->tgid = get_pid(old_event->tgid);
b4e4e140 EP	354	if (event->data_type == FSNOTIFY_EVENT_PATH)
	355	path_get(&event->path);
	356
	357	return event;
	358	}
	359
a2d8bc6c EP	360	/*
	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
	363	* particular event.
	364	*
	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...
62ffe5df	370	* @name the filename, if available
a2d8bc6c	371	*/
47882c6f	372	struct fsnotify_event fsnotify_create_event(struct inode to_tell, __u32 mask, void *data,
f44aebcc EP	373	int data_type, const char *name, u32 cookie,
f44aebcc EP	374	gfp_t gfp)
a2d8bc6c EP	375	{
	376	struct fsnotify_event *event;
	377
6f3a539e	378	event = kmem_cache_zalloc(fsnotify_event_cachep, gfp);
a2d8bc6c EP	379	if (!event)
	380	return NULL;
	381
	382	initialize_event(event);
62ffe5df EP	383
62ffe5df EP	384	if (name) {
f44aebcc	385	event->file_name = kstrdup(name, gfp);
62ffe5df EP	386	if (!event->file_name) {
	387	kmem_cache_free(fsnotify_event_cachep, event);
	388	return NULL;
	389	}
	390	event->name_len = strlen(event->file_name);
	391	}
47882c6f	392
32c32632	393	event->tgid = get_pid(task_tgid(current));
47882c6f	394	event->sync_cookie = cookie;
90586523	395	event->to_tell = to_tell;
b4e4e140	396	event->data_type = data_type;
90586523 EP	397
90586523 EP	398	switch (data_type) {
90586523 EP	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);
90586523 EP	404	break;
	405	}
	406	case FSNOTIFY_EVENT_INODE:
	407	event->inode = data;
90586523 EP	408	break;
	409	case FSNOTIFY_EVENT_NONE:
	410	event->inode = NULL;
	411	event->path.dentry = NULL;
	412	event->path.mnt = NULL;
	413	break;
	414	default:
	415	BUG();
	416	}
	417
	418	event->mask = mask;
	419
	420	return event;
	421	}
	422
	423	__init int fsnotify_notification_init(void)
	424	{
	425	fsnotify_event_cachep = KMEM_CACHE(fsnotify_event, SLAB_PANIC);
a2d8bc6c EP	426	fsnotify_event_holder_cachep = KMEM_CACHE(fsnotify_event_holder, SLAB_PANIC);
a2d8bc6c EP	427
b4277d3d EP	428	q_overflow_event = fsnotify_create_event(NULL, FS_Q_OVERFLOW, NULL,
	429	FSNOTIFY_EVENT_NONE, NULL, 0,
	430	GFP_KERNEL);
	431	if (!q_overflow_event)
	432	panic("unable to allocate fsnotify q_overflow_event\n");
90586523 EP	433
	434	return 0;
	435	}
	436	subsys_initcall(fsnotify_notification_init);
	437