mark.c 11.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
/*
 *  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.
 */

/*
 * fsnotify inode mark locking/lifetime/and refcnting
 *
 * REFCNT:
 * The mark->refcnt tells how many "things" in the kernel currently are
 * referencing this object.  The object typically will live inside the kernel
 * with a refcnt of 2, one for each list it is on (i_list, g_list).  Any task
 * which can find this object holding the appropriete locks, can take a reference
L
Lucas De Marchi 已提交
27
 * and the object itself is guaranteed to survive until the reference is dropped.
28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87
 *
 * LOCKING:
 * There are 3 spinlocks involved with fsnotify inode marks and they MUST
 * be taken in order as follows:
 *
 * mark->lock
 * group->mark_lock
 * inode->i_lock
 *
 * mark->lock protects 2 things, mark->group and mark->inode.  You must hold
 * that lock to dereference either of these things (they could be NULL even with
 * the lock)
 *
 * group->mark_lock protects the marks_list anchored inside a given group
 * and each mark is hooked via the g_list.  It also sorta protects the
 * free_g_list, which when used is anchored by a private list on the stack of the
 * task which held the group->mark_lock.
 *
 * inode->i_lock protects the i_fsnotify_marks list anchored inside a
 * given inode and each mark is hooked via the i_list. (and sorta the
 * free_i_list)
 *
 *
 * LIFETIME:
 * Inode marks survive between when they are added to an inode and when their
 * refcnt==0.
 *
 * The inode mark can be cleared for a number of different reasons including:
 * - The inode is unlinked for the last time.  (fsnotify_inode_remove)
 * - The inode is being evicted from cache. (fsnotify_inode_delete)
 * - The fs the inode is on is unmounted.  (fsnotify_inode_delete/fsnotify_unmount_inodes)
 * - Something explicitly requests that it be removed.  (fsnotify_destroy_mark)
 * - The fsnotify_group associated with the mark is going away and all such marks
 *   need to be cleaned up. (fsnotify_clear_marks_by_group)
 *
 * Worst case we are given an inode and need to clean up all the marks on that
 * inode.  We take i_lock and walk the i_fsnotify_marks safely.  For each
 * mark on the list we take a reference (so the mark can't disappear under us).
 * We remove that mark form the inode's list of marks and we add this mark to a
 * private list anchored on the stack using i_free_list;  At this point we no
 * longer fear anything finding the mark using the inode's list of marks.
 *
 * We can safely and locklessly run the private list on the stack of everything
 * we just unattached from the original inode.  For each mark on the private list
 * we grab the mark-> and can thus dereference mark->group and mark->inode.  If
 * we see the group and inode are not NULL we take those locks.  Now holding all
 * 3 locks we can completely remove the mark from other tasks finding it in the
 * future.  Remember, 10 things might already be referencing this mark, but they
 * better be holding a ref.  We drop our reference we took before we unhooked it
 * from the inode.  When the ref hits 0 we can free the mark.
 *
 * Very similarly for freeing by group, except we use free_g_list.
 *
 * This has the very interesting property of being able to run concurrently with
 * any (or all) other directions.
 */

#include <linux/fs.h>
#include <linux/init.h>
#include <linux/kernel.h>
88
#include <linux/kthread.h>
89 90 91 92
#include <linux/module.h>
#include <linux/mutex.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
93
#include <linux/srcu.h>
94

A
Arun Sharma 已提交
95
#include <linux/atomic.h>
96 97 98 99

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

100 101 102 103 104
struct srcu_struct fsnotify_mark_srcu;
static DEFINE_SPINLOCK(destroy_lock);
static LIST_HEAD(destroy_list);
static DECLARE_WAIT_QUEUE_HEAD(destroy_waitq);

105 106 107 108 109 110 111
void fsnotify_get_mark(struct fsnotify_mark *mark)
{
	atomic_inc(&mark->refcnt);
}

void fsnotify_put_mark(struct fsnotify_mark *mark)
{
112 113 114
	if (atomic_dec_and_test(&mark->refcnt)) {
		if (mark->group)
			fsnotify_put_group(mark->group);
115
		mark->free_mark(mark);
116
	}
117 118 119 120 121 122 123
}

/*
 * Any time a mark is getting freed we end up here.
 * The caller had better be holding a reference to this mark so we don't actually
 * do the final put under the mark->lock
 */
124 125
void fsnotify_destroy_mark_locked(struct fsnotify_mark *mark,
				  struct fsnotify_group *group)
126
{
127
	struct inode *inode = NULL;
128

129 130
	BUG_ON(!mutex_is_locked(&group->mark_mutex));

131
	spin_lock(&mark->lock);
132

133 134
	/* something else already called this function on this mark */
	if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
135
		spin_unlock(&mark->lock);
136
		return;
137 138
	}

139 140
	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;

141 142
	if (mark->flags & FSNOTIFY_MARK_FLAG_INODE) {
		inode = mark->i.inode;
143
		fsnotify_destroy_inode_mark(mark);
144 145
	} else if (mark->flags & FSNOTIFY_MARK_FLAG_VFSMOUNT)
		fsnotify_destroy_vfsmount_mark(mark);
146 147 148 149 150 151
	else
		BUG();

	list_del_init(&mark->g_list);

	spin_unlock(&mark->lock);
152 153

	/* release lock temporarily */
154
	mutex_unlock(&group->mark_mutex);
155

156 157 158 159 160
	spin_lock(&destroy_lock);
	list_add(&mark->destroy_list, &destroy_list);
	spin_unlock(&destroy_lock);
	wake_up(&destroy_waitq);

161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180
	/*
	 * Some groups like to know that marks are being freed.  This is a
	 * callback to the group function to let it know that this mark
	 * is being freed.
	 */
	if (group->ops->freeing_mark)
		group->ops->freeing_mark(mark, group);

	/*
	 * __fsnotify_update_child_dentry_flags(inode);
	 *
	 * I really want to call that, but we can't, we have no idea if the inode
	 * still exists the second we drop the mark->lock.
	 *
	 * The next time an event arrive to this inode from one of it's children
	 * __fsnotify_parent will see that the inode doesn't care about it's
	 * children and will update all of these flags then.  So really this
	 * is just a lazy update (and could be a perf win...)
	 */

181
	if (inode && (mark->flags & FSNOTIFY_MARK_FLAG_OBJECT_PINNED))
182
		iput(inode);
183 184 185 186 187
	/*
	 * We don't necessarily have a ref on mark from caller so the above iput
	 * may have already destroyed it.  Don't touch from now on.
	 */

188
	atomic_dec(&group->num_marks);
189 190 191 192 193 194 195 196 197 198

	mutex_lock(&group->mark_mutex);
}

void fsnotify_destroy_mark(struct fsnotify_mark *mark,
			   struct fsnotify_group *group)
{
	mutex_lock(&group->mark_mutex);
	fsnotify_destroy_mark_locked(mark, group);
	mutex_unlock(&group->mark_mutex);
199 200
}

201 202 203 204 205 206 207 208 209 210
void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
{
	assert_spin_locked(&mark->lock);

	mark->mask = mask;

	if (mark->flags & FSNOTIFY_MARK_FLAG_INODE)
		fsnotify_set_inode_mark_mask_locked(mark, mask);
}

211 212 213 214 215 216
void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask)
{
	assert_spin_locked(&mark->lock);

	mark->ignored_mask = mask;
}
217

218 219 220 221 222
/*
 * Attach an initialized mark to a given group and fs object.
 * These marks may be used for the fsnotify backend to determine which
 * event types should be delivered to which group.
 */
223 224 225
int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
			     struct fsnotify_group *group, struct inode *inode,
			     struct vfsmount *mnt, int allow_dups)
226 227 228 229 230
{
	int ret = 0;

	BUG_ON(inode && mnt);
	BUG_ON(!inode && !mnt);
231
	BUG_ON(!mutex_is_locked(&group->mark_mutex));
232 233 234

	/*
	 * LOCKING ORDER!!!!
235
	 * group->mark_mutex
236
	 * mark->lock
237 238
	 * inode->i_lock
	 */
239
	spin_lock(&mark->lock);
240 241
	mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE;

242
	fsnotify_get_group(group);
243 244 245 246 247 248 249 250 251
	mark->group = group;
	list_add(&mark->g_list, &group->marks_list);
	atomic_inc(&group->num_marks);
	fsnotify_get_mark(mark); /* for i_list and g_list */

	if (inode) {
		ret = fsnotify_add_inode_mark(mark, group, inode, allow_dups);
		if (ret)
			goto err;
252 253 254 255
	} else if (mnt) {
		ret = fsnotify_add_vfsmount_mark(mark, group, mnt, allow_dups);
		if (ret)
			goto err;
256 257 258 259
	} else {
		BUG();
	}

260 261
	/* this will pin the object if appropriate */
	fsnotify_set_mark_mask_locked(mark, mark->mask);
262 263 264 265 266 267 268
	spin_unlock(&mark->lock);

	if (inode)
		__fsnotify_update_child_dentry_flags(inode);

	return ret;
err:
269
	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
270
	list_del_init(&mark->g_list);
271
	fsnotify_put_group(group);
272
	mark->group = NULL;
273 274 275 276
	atomic_dec(&group->num_marks);

	spin_unlock(&mark->lock);

277 278 279 280 281
	spin_lock(&destroy_lock);
	list_add(&mark->destroy_list, &destroy_list);
	spin_unlock(&destroy_lock);
	wake_up(&destroy_waitq);

282 283 284
	return ret;
}

285 286 287 288 289 290 291 292 293 294
int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
		      struct inode *inode, struct vfsmount *mnt, int allow_dups)
{
	int ret;
	mutex_lock(&group->mark_mutex);
	ret = fsnotify_add_mark_locked(mark, group, inode, mnt, allow_dups);
	mutex_unlock(&group->mark_mutex);
	return ret;
}

295
/*
E
Eric Paris 已提交
296
 * clear any marks in a group in which mark->flags & flags is true
297
 */
E
Eric Paris 已提交
298 299
void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group,
					 unsigned int flags)
300 301 302 303
{
	struct fsnotify_mark *lmark, *mark;
	LIST_HEAD(free_list);

304
	mutex_lock(&group->mark_mutex);
305
	list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
E
Eric Paris 已提交
306 307 308 309 310
		if (mark->flags & flags) {
			list_add(&mark->free_g_list, &free_list);
			list_del_init(&mark->g_list);
			fsnotify_get_mark(mark);
		}
311
	}
312
	mutex_unlock(&group->mark_mutex);
313 314

	list_for_each_entry_safe(mark, lmark, &free_list, free_g_list) {
315
		fsnotify_destroy_mark(mark, group);
316 317 318 319
		fsnotify_put_mark(mark);
	}
}

E
Eric Paris 已提交
320 321 322 323 324 325 326 327
/*
 * Given a group, destroy all of the marks associated with that group.
 */
void fsnotify_clear_marks_by_group(struct fsnotify_group *group)
{
	fsnotify_clear_marks_by_group_flags(group, (unsigned int)-1);
}

328 329 330 331 332
void fsnotify_duplicate_mark(struct fsnotify_mark *new, struct fsnotify_mark *old)
{
	assert_spin_locked(&old->lock);
	new->i.inode = old->i.inode;
	new->m.mnt = old->m.mnt;
333 334
	if (old->group)
		fsnotify_get_group(old->group);
335 336 337 338 339 340 341 342 343 344 345
	new->group = old->group;
	new->mask = old->mask;
	new->free_mark = old->free_mark;
}

/*
 * Nothing fancy, just initialize lists and locks and counters.
 */
void fsnotify_init_mark(struct fsnotify_mark *mark,
			void (*free_mark)(struct fsnotify_mark *mark))
{
346
	memset(mark, 0, sizeof(*mark));
347 348 349 350
	spin_lock_init(&mark->lock);
	atomic_set(&mark->refcnt, 1);
	mark->free_mark = free_mark;
}
351 352 353 354 355 356 357 358

static int fsnotify_mark_destroy(void *ignored)
{
	struct fsnotify_mark *mark, *next;
	LIST_HEAD(private_destroy_list);

	for (;;) {
		spin_lock(&destroy_lock);
359 360
		/* exchange the list head */
		list_replace_init(&destroy_list, &private_destroy_list);
361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387
		spin_unlock(&destroy_lock);

		synchronize_srcu(&fsnotify_mark_srcu);

		list_for_each_entry_safe(mark, next, &private_destroy_list, destroy_list) {
			list_del_init(&mark->destroy_list);
			fsnotify_put_mark(mark);
		}

		wait_event_interruptible(destroy_waitq, !list_empty(&destroy_list));
	}

	return 0;
}

static int __init fsnotify_mark_init(void)
{
	struct task_struct *thread;

	thread = kthread_run(fsnotify_mark_destroy, NULL,
			     "fsnotify_mark");
	if (IS_ERR(thread))
		panic("unable to start fsnotify mark destruction thread.");

	return 0;
}
device_initcall(fsnotify_mark_init);