notification.c 13.8 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
#include <linux/file.h>
35 36 37 38
#include <linux/fs.h>
#include <linux/init.h>
#include <linux/kernel.h>
#include <linux/list.h>
39
#include <linux/module.h>
40 41 42 43 44 45 46 47 48 49 50 51 52
#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;
53 54 55 56 57 58 59
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'
 */
60
static struct fsnotify_event *q_overflow_event;
61 62 63 64 65 66 67 68 69 70 71
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);
72 73 74 75 76 77 78

/* 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;
}
79 80 81 82 83 84 85 86 87 88 89 90

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
91 92
		pr_debug("%s: event=%p\n", __func__, event);

93 94
		if (event->data_type == FSNOTIFY_EVENT_FILE)
			fput(event->file);
95

96 97
		BUG_ON(!list_empty(&event->private_data_list));

98
		kfree(event->file_name);
99
		put_pid(event->tgid);
100 101 102 103
		kmem_cache_free(fsnotify_event_cachep, event);
	}
}

104 105 106 107 108 109 110
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)
{
111 112
	if (holder)
		kmem_cache_free(fsnotify_event_holder_cachep, holder);
113 114 115
}

/*
116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
 * 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;
}

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

Eric Paris's avatar
Eric Paris committed
150 151
	pr_debug("%s: group=%p event=%p priv=%p\n", __func__, group, event, priv);

152 153 154 155 156 157 158 159 160 161 162 163
	/*
	 * 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)
164
			return ERR_PTR(-ENOMEM);
165 166 167 168
	}

	mutex_lock(&group->notification_mutex);

169
	if (group->q_len >= group->max_events) {
170
		event = q_overflow_event;
171 172 173 174 175 176 177 178

		/*
		 * we need to return the overflow event
		 * which means we need a ref
		 */
		fsnotify_get_event(event);
		return_event = event;

179 180 181
		/* sorry, no private data on the overflow event */
		priv = NULL;
	}
182

183
	if (!list_empty(list) && merge) {
184
		struct fsnotify_event *tmp;
185

186 187
		tmp = merge(list, event);
		if (tmp) {
188
			mutex_unlock(&group->notification_mutex);
189 190 191

			if (return_event)
				fsnotify_put_event(return_event);
192 193
			if (holder != &event->holder)
				fsnotify_destroy_event_holder(holder);
194
			return tmp;
195 196 197
		}
	}

198 199 200 201 202 203 204 205 206 207 208
	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);
209 210 211 212 213 214

		if (return_event) {
			fsnotify_put_event(return_event);
			return_event = NULL;
		}

215 216 217 218 219 220 221 222
		goto alloc_holder;
	}

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

	fsnotify_get_event(event);
	list_add_tail(&holder->event_list, list);
223 224
	if (priv)
		list_add_tail(&priv->event_list, &event->private_data_list);
225 226 227 228
	spin_unlock(&event->lock);
	mutex_unlock(&group->notification_mutex);

	wake_up(&group->notification_waitq);
229
	return return_event;
230 231 232 233 234 235 236 237
}

/*
 * 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)
238 239
{
	struct fsnotify_event *event;
240
	struct fsnotify_event_holder *holder;
241

242
	BUG_ON(!mutex_is_locked(&group->notification_mutex));
243

Eric Paris's avatar
Eric Paris committed
244 245
	pr_debug("%s: group=%p\n", __func__, group);

246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286
	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;
287
	struct fsnotify_event_private_data *priv;
288 289 290 291

	mutex_lock(&group->notification_mutex);
	while (!fsnotify_notify_queue_is_empty(group)) {
		event = fsnotify_remove_notify_event(group);
292 293 294 295 296 297 298 299
		/* 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);
		}
300 301 302 303 304 305 306 307
		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);
308 309 310 311
	atomic_set(&event->refcnt, 1);

	spin_lock_init(&event->lock);

312
	INIT_LIST_HEAD(&event->private_data_list);
313 314
}

315 316
/*
 * Caller damn well better be holding whatever mutex is protecting the
317 318
 * old_holder->event_list and the new_event must be a clean event which
 * cannot be found anywhere else in the kernel.
319 320 321 322 323
 */
int fsnotify_replace_event(struct fsnotify_event_holder *old_holder,
			   struct fsnotify_event *new_event)
{
	struct fsnotify_event *old_event = old_holder->event;
324 325 326 327 328 329
	struct fsnotify_event_holder *new_holder = &new_event->holder;

	enum event_spinlock_class {
		SPINLOCK_OLD,
		SPINLOCK_NEW,
	};
330

Eric Paris's avatar
Eric Paris committed
331 332
	pr_debug("%s: old_event=%p new_event=%p\n", __func__, old_event, new_event);

333
	/*
334 335
	 * if the new_event's embedded holder is in use someone
	 * screwed up and didn't give us a clean new event.
336
	 */
337
	BUG_ON(!list_empty(&new_holder->event_list));
338

339 340
	spin_lock_nested(&old_event->lock, SPINLOCK_OLD);
	spin_lock_nested(&new_event->lock, SPINLOCK_NEW);
341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357

	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
358 359 360 361 362 363 364 365
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
366 367
	pr_debug("%s: old_event=%p new_event=%p\n", __func__, old_event, event);

Eric Paris's avatar
Eric Paris committed
368 369 370 371 372 373 374 375 376 377
	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;
		}
	}
378
	event->tgid = get_pid(old_event->tgid);
379 380
	if (event->data_type == FSNOTIFY_EVENT_FILE)
		get_file(event->file);
Eric Paris's avatar
Eric Paris committed
381 382 383 384

	return event;
}

385 386 387 388 389 390 391 392 393 394
/*
 * 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...
395
 * @name the filename, if available
396
 */
397
struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask, void *data,
398 399
					     int data_type, const unsigned char *name,
					     u32 cookie, gfp_t gfp)
400 401 402
{
	struct fsnotify_event *event;

403
	event = kmem_cache_zalloc(fsnotify_event_cachep, gfp);
404 405 406
	if (!event)
		return NULL;

Eric Paris's avatar
Eric Paris committed
407 408 409
	pr_debug("%s: event=%p to_tell=%p mask=%x data=%p data_type=%d\n",
		 __func__, event, to_tell, mask, data, data_type);

410
	initialize_event(event);
411 412

	if (name) {
413
		event->file_name = kstrdup(name, gfp);
414 415 416 417 418 419
		if (!event->file_name) {
			kmem_cache_free(fsnotify_event_cachep, event);
			return NULL;
		}
		event->name_len = strlen(event->file_name);
	}
420

421
	event->tgid = get_pid(task_tgid(current));
422
	event->sync_cookie = cookie;
423
	event->to_tell = to_tell;
Eric Paris's avatar
Eric Paris committed
424
	event->data_type = data_type;
425 426

	switch (data_type) {
427 428
	case FSNOTIFY_EVENT_FILE: {
		event->file = data;
429 430 431 432 433 434 435 436 437 438 439 440 441
		/*
		 * if this file is about to disappear hold an extra reference
		 * until we return to __fput so we don't have to worry about
		 * future get/put destroying the file under us or generating
		 * additional events.  Notice that we change f_mode without
		 * holding f_lock.  This is safe since this is the only possible
		 * reference to this object in the kernel (it was about to be
		 * freed, remember?)
		 */
		if (!atomic_long_read(&event->file->f_count)) {
			event->file->f_mode |= FMODE_NONOTIFY;
			get_file(event->file);
		}
442
		get_file(event->file);
443 444 445 446 447 448 449
		break;
	}
	case FSNOTIFY_EVENT_INODE:
		event->inode = data;
		break;
	case FSNOTIFY_EVENT_NONE:
		event->inode = NULL;
450
		event->file = NULL;
451 452 453 454 455 456 457 458 459 460 461 462 463
		break;
	default:
		BUG();
	}

	event->mask = mask;

	return event;
}

__init int fsnotify_notification_init(void)
{
	fsnotify_event_cachep = KMEM_CACHE(fsnotify_event, SLAB_PANIC);
464 465
	fsnotify_event_holder_cachep = KMEM_CACHE(fsnotify_event_holder, SLAB_PANIC);

466 467 468 469 470
	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");
471 472 473 474 475

	return 0;
}
subsys_initcall(fsnotify_notification_init);