notification.c 13 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*
 *  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.
 */

19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
/*
 * 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.
 */

34 35 36 37
#include <linux/fs.h>
#include <linux/init.h>
#include <linux/kernel.h>
#include <linux/list.h>
38
#include <linux/module.h>
39 40 41 42 43 44 45 46 47 48 49 50 51
#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;
52 53 54 55 56 57 58
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'
 */
59
static struct fsnotify_event *q_overflow_event;
60 61 62 63 64 65 66 67 68 69 70
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);
71 72 73 74 75 76 77

/* 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;
}
78 79 80 81 82 83 84 85 86 87 88 89

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)) {
Eric Paris's avatar
Eric Paris committed
90 91
		pr_debug("%s: event=%p\n", __func__, event);

92 93 94
		if (event->data_type == FSNOTIFY_EVENT_PATH)
			path_put(&event->path);

95 96
		BUG_ON(!list_empty(&event->private_data_list));

97
		kfree(event->file_name);
98
		put_pid(event->tgid);
99 100 101 102
		kmem_cache_free(fsnotify_event_cachep, event);
	}
}

103 104 105 106 107 108 109
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)
{
110 111
	if (holder)
		kmem_cache_free(fsnotify_event_holder_cachep, holder);
112 113 114
}

/*
115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134
 * 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;
}

135
/*
136 137 138
 * 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.
139
 */
140
int fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_event *event,
141
			      struct fsnotify_event_private_data *priv,
142 143 144 145
			      int (*merge)(struct list_head *,
					   struct fsnotify_event *,
					   void **arg),
			      void **arg)
146 147 148
{
	struct fsnotify_event_holder *holder = NULL;
	struct list_head *list = &group->notification_list;
149
	int rc = 0;
150

Eric Paris's avatar
Eric Paris committed
151 152
	pr_debug("%s: group=%p event=%p priv=%p\n", __func__, group, event, priv);

153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169
	/*
	 * 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);

170
	if (group->q_len >= group->max_events) {
171
		event = q_overflow_event;
172
		rc = -EOVERFLOW;
173 174 175
		/* sorry, no private data on the overflow event */
		priv = NULL;
	}
176

177 178 179
	if (!list_empty(list) && merge) {
		int ret;

180
		ret = merge(list, event, arg);
181 182 183 184 185 186 187 188
		if (ret) {
			mutex_unlock(&group->notification_mutex);
			if (holder != &event->holder)
				fsnotify_destroy_event_holder(holder);
			return ret;
		}
	}

189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207
	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);
208 209
	if (priv)
		list_add_tail(&priv->event_list, &event->private_data_list);
210 211 212 213
	spin_unlock(&event->lock);
	mutex_unlock(&group->notification_mutex);

	wake_up(&group->notification_waitq);
214
	return rc;
215 216 217 218 219 220 221 222
}

/*
 * 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)
223 224
{
	struct fsnotify_event *event;
225
	struct fsnotify_event_holder *holder;
226

227
	BUG_ON(!mutex_is_locked(&group->notification_mutex));
228

Eric Paris's avatar
Eric Paris committed
229 230
	pr_debug("%s: group=%p\n", __func__, group);

231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271
	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;
272
	struct fsnotify_event_private_data *priv;
273 274 275 276

	mutex_lock(&group->notification_mutex);
	while (!fsnotify_notify_queue_is_empty(group)) {
		event = fsnotify_remove_notify_event(group);
277 278 279 280 281 282 283 284
		/* 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);
		}
285 286 287 288 289 290 291 292
		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);
293 294 295 296
	atomic_set(&event->refcnt, 1);

	spin_lock_init(&event->lock);

297
	INIT_LIST_HEAD(&event->private_data_list);
298 299
}

300 301
/*
 * Caller damn well better be holding whatever mutex is protecting the
302 303
 * old_holder->event_list and the new_event must be a clean event which
 * cannot be found anywhere else in the kernel.
304 305 306 307 308
 */
int fsnotify_replace_event(struct fsnotify_event_holder *old_holder,
			   struct fsnotify_event *new_event)
{
	struct fsnotify_event *old_event = old_holder->event;
309 310 311 312 313 314
	struct fsnotify_event_holder *new_holder = &new_event->holder;

	enum event_spinlock_class {
		SPINLOCK_OLD,
		SPINLOCK_NEW,
	};
315

Eric Paris's avatar
Eric Paris committed
316 317
	pr_debug("%s: old_event=%p new_event=%p\n", __func__, old_event, new_event);

318
	/*
319 320
	 * if the new_event's embedded holder is in use someone
	 * screwed up and didn't give us a clean new event.
321
	 */
322
	BUG_ON(!list_empty(&new_holder->event_list));
323

324 325
	spin_lock_nested(&old_event->lock, SPINLOCK_OLD);
	spin_lock_nested(&new_event->lock, SPINLOCK_NEW);
326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342

	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;
}

Eric Paris's avatar
Eric Paris committed
343 344 345 346 347 348 349 350
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;

Eric Paris's avatar
Eric Paris committed
351 352
	pr_debug("%s: old_event=%p new_event=%p\n", __func__, old_event, event);

Eric Paris's avatar
Eric Paris committed
353 354 355 356 357 358 359 360 361 362
	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;
		}
	}
363
	event->tgid = get_pid(old_event->tgid);
Eric Paris's avatar
Eric Paris committed
364 365 366 367 368 369
	if (event->data_type == FSNOTIFY_EVENT_PATH)
		path_get(&event->path);

	return event;
}

370 371 372 373 374 375 376 377 378 379
/*
 * 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...
380
 * @name the filename, if available
381
 */
382
struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask, void *data,
383 384
					     int data_type, const unsigned char *name,
					     u32 cookie, gfp_t gfp)
385 386 387
{
	struct fsnotify_event *event;

388
	event = kmem_cache_zalloc(fsnotify_event_cachep, gfp);
389 390 391
	if (!event)
		return NULL;

Eric Paris's avatar
Eric Paris committed
392 393 394
	pr_debug("%s: event=%p to_tell=%p mask=%x data=%p data_type=%d\n",
		 __func__, event, to_tell, mask, data, data_type);

395
	initialize_event(event);
396 397

	if (name) {
398
		event->file_name = kstrdup(name, gfp);
399 400 401 402 403 404
		if (!event->file_name) {
			kmem_cache_free(fsnotify_event_cachep, event);
			return NULL;
		}
		event->name_len = strlen(event->file_name);
	}
405

406
	event->tgid = get_pid(task_tgid(current));
407
	event->sync_cookie = cookie;
408
	event->to_tell = to_tell;
Eric Paris's avatar
Eric Paris committed
409
	event->data_type = data_type;
410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438

	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);
439 440
	fsnotify_event_holder_cachep = KMEM_CACHE(fsnotify_event_holder, SLAB_PANIC);

441 442 443 444 445
	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");
446 447 448 449 450

	return 0;
}
subsys_initcall(fsnotify_notification_init);