xref: /linux/fs/notify/mark.c (revision 7ff836f064e2c814a7504c91a4464eea45d475bd)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
4  */
5 
6 /*
7  * fsnotify inode mark locking/lifetime/and refcnting
8  *
9  * REFCNT:
10  * The group->recnt and mark->refcnt tell how many "things" in the kernel
11  * currently are referencing the objects. Both kind of objects typically will
12  * live inside the kernel with a refcnt of 2, one for its creation and one for
13  * the reference a group and a mark hold to each other.
14  * If you are holding the appropriate locks, you can take a reference and the
15  * object itself is guaranteed to survive until the reference is dropped.
16  *
17  * LOCKING:
18  * There are 3 locks involved with fsnotify inode marks and they MUST be taken
19  * in order as follows:
20  *
21  * group->mark_mutex
22  * mark->lock
23  * mark->connector->lock
24  *
25  * group->mark_mutex protects the marks_list anchored inside a given group and
26  * each mark is hooked via the g_list.  It also protects the groups private
27  * data (i.e group limits).
28 
29  * mark->lock protects the marks attributes like its masks and flags.
30  * Furthermore it protects the access to a reference of the group that the mark
31  * is assigned to as well as the access to a reference of the inode/vfsmount
32  * that is being watched by the mark.
33  *
34  * mark->connector->lock protects the list of marks anchored inside an
35  * inode / vfsmount and each mark is hooked via the i_list.
36  *
37  * A list of notification marks relating to inode / mnt is contained in
38  * fsnotify_mark_connector. That structure is alive as long as there are any
39  * marks in the list and is also protected by fsnotify_mark_srcu. A mark gets
40  * detached from fsnotify_mark_connector when last reference to the mark is
41  * dropped.  Thus having mark reference is enough to protect mark->connector
42  * pointer and to make sure fsnotify_mark_connector cannot disappear. Also
43  * because we remove mark from g_list before dropping mark reference associated
44  * with that, any mark found through g_list is guaranteed to have
45  * mark->connector set until we drop group->mark_mutex.
46  *
47  * LIFETIME:
48  * Inode marks survive between when they are added to an inode and when their
49  * refcnt==0. Marks are also protected by fsnotify_mark_srcu.
50  *
51  * The inode mark can be cleared for a number of different reasons including:
52  * - The inode is unlinked for the last time.  (fsnotify_inode_remove)
53  * - The inode is being evicted from cache. (fsnotify_inode_delete)
54  * - The fs the inode is on is unmounted.  (fsnotify_inode_delete/fsnotify_unmount_inodes)
55  * - Something explicitly requests that it be removed.  (fsnotify_destroy_mark)
56  * - The fsnotify_group associated with the mark is going away and all such marks
57  *   need to be cleaned up. (fsnotify_clear_marks_by_group)
58  *
59  * This has the very interesting property of being able to run concurrently with
60  * any (or all) other directions.
61  */
62 
63 #include <linux/fs.h>
64 #include <linux/init.h>
65 #include <linux/kernel.h>
66 #include <linux/kthread.h>
67 #include <linux/module.h>
68 #include <linux/mutex.h>
69 #include <linux/slab.h>
70 #include <linux/spinlock.h>
71 #include <linux/srcu.h>
72 #include <linux/ratelimit.h>
73 
74 #include <linux/atomic.h>
75 
76 #include <linux/fsnotify_backend.h>
77 #include "fsnotify.h"
78 
79 #define FSNOTIFY_REAPER_DELAY	(1)	/* 1 jiffy */
80 
81 struct srcu_struct fsnotify_mark_srcu;
82 struct kmem_cache *fsnotify_mark_connector_cachep;
83 
84 static DEFINE_SPINLOCK(destroy_lock);
85 static LIST_HEAD(destroy_list);
86 static struct fsnotify_mark_connector *connector_destroy_list;
87 
88 static void fsnotify_mark_destroy_workfn(struct work_struct *work);
89 static DECLARE_DELAYED_WORK(reaper_work, fsnotify_mark_destroy_workfn);
90 
91 static void fsnotify_connector_destroy_workfn(struct work_struct *work);
92 static DECLARE_WORK(connector_reaper_work, fsnotify_connector_destroy_workfn);
93 
94 void fsnotify_get_mark(struct fsnotify_mark *mark)
95 {
96 	WARN_ON_ONCE(!refcount_read(&mark->refcnt));
97 	refcount_inc(&mark->refcnt);
98 }
99 
100 static __u32 *fsnotify_conn_mask_p(struct fsnotify_mark_connector *conn)
101 {
102 	if (conn->type == FSNOTIFY_OBJ_TYPE_INODE)
103 		return &fsnotify_conn_inode(conn)->i_fsnotify_mask;
104 	else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT)
105 		return &fsnotify_conn_mount(conn)->mnt_fsnotify_mask;
106 	else if (conn->type == FSNOTIFY_OBJ_TYPE_SB)
107 		return &fsnotify_conn_sb(conn)->s_fsnotify_mask;
108 	return NULL;
109 }
110 
111 __u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn)
112 {
113 	if (WARN_ON(!fsnotify_valid_obj_type(conn->type)))
114 		return 0;
115 
116 	return *fsnotify_conn_mask_p(conn);
117 }
118 
119 static void __fsnotify_recalc_mask(struct fsnotify_mark_connector *conn)
120 {
121 	u32 new_mask = 0;
122 	struct fsnotify_mark *mark;
123 
124 	assert_spin_locked(&conn->lock);
125 	/* We can get detached connector here when inode is getting unlinked. */
126 	if (!fsnotify_valid_obj_type(conn->type))
127 		return;
128 	hlist_for_each_entry(mark, &conn->list, obj_list) {
129 		if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)
130 			new_mask |= mark->mask;
131 	}
132 	*fsnotify_conn_mask_p(conn) = new_mask;
133 }
134 
135 /*
136  * Calculate mask of events for a list of marks. The caller must make sure
137  * connector and connector->obj cannot disappear under us.  Callers achieve
138  * this by holding a mark->lock or mark->group->mark_mutex for a mark on this
139  * list.
140  */
141 void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn)
142 {
143 	if (!conn)
144 		return;
145 
146 	spin_lock(&conn->lock);
147 	__fsnotify_recalc_mask(conn);
148 	spin_unlock(&conn->lock);
149 	if (conn->type == FSNOTIFY_OBJ_TYPE_INODE)
150 		__fsnotify_update_child_dentry_flags(
151 					fsnotify_conn_inode(conn));
152 }
153 
154 /* Free all connectors queued for freeing once SRCU period ends */
155 static void fsnotify_connector_destroy_workfn(struct work_struct *work)
156 {
157 	struct fsnotify_mark_connector *conn, *free;
158 
159 	spin_lock(&destroy_lock);
160 	conn = connector_destroy_list;
161 	connector_destroy_list = NULL;
162 	spin_unlock(&destroy_lock);
163 
164 	synchronize_srcu(&fsnotify_mark_srcu);
165 	while (conn) {
166 		free = conn;
167 		conn = conn->destroy_next;
168 		kmem_cache_free(fsnotify_mark_connector_cachep, free);
169 	}
170 }
171 
172 static void *fsnotify_detach_connector_from_object(
173 					struct fsnotify_mark_connector *conn,
174 					unsigned int *type)
175 {
176 	struct inode *inode = NULL;
177 
178 	*type = conn->type;
179 	if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED)
180 		return NULL;
181 
182 	if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) {
183 		inode = fsnotify_conn_inode(conn);
184 		inode->i_fsnotify_mask = 0;
185 		atomic_long_inc(&inode->i_sb->s_fsnotify_inode_refs);
186 	} else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT) {
187 		fsnotify_conn_mount(conn)->mnt_fsnotify_mask = 0;
188 	} else if (conn->type == FSNOTIFY_OBJ_TYPE_SB) {
189 		fsnotify_conn_sb(conn)->s_fsnotify_mask = 0;
190 	}
191 
192 	rcu_assign_pointer(*(conn->obj), NULL);
193 	conn->obj = NULL;
194 	conn->type = FSNOTIFY_OBJ_TYPE_DETACHED;
195 
196 	return inode;
197 }
198 
199 static void fsnotify_final_mark_destroy(struct fsnotify_mark *mark)
200 {
201 	struct fsnotify_group *group = mark->group;
202 
203 	if (WARN_ON_ONCE(!group))
204 		return;
205 	group->ops->free_mark(mark);
206 	fsnotify_put_group(group);
207 }
208 
209 /* Drop object reference originally held by a connector */
210 static void fsnotify_drop_object(unsigned int type, void *objp)
211 {
212 	struct inode *inode;
213 	struct super_block *sb;
214 
215 	if (!objp)
216 		return;
217 	/* Currently only inode references are passed to be dropped */
218 	if (WARN_ON_ONCE(type != FSNOTIFY_OBJ_TYPE_INODE))
219 		return;
220 	inode = objp;
221 	sb = inode->i_sb;
222 	iput(inode);
223 	if (atomic_long_dec_and_test(&sb->s_fsnotify_inode_refs))
224 		wake_up_var(&sb->s_fsnotify_inode_refs);
225 }
226 
227 void fsnotify_put_mark(struct fsnotify_mark *mark)
228 {
229 	struct fsnotify_mark_connector *conn = READ_ONCE(mark->connector);
230 	void *objp = NULL;
231 	unsigned int type = FSNOTIFY_OBJ_TYPE_DETACHED;
232 	bool free_conn = false;
233 
234 	/* Catch marks that were actually never attached to object */
235 	if (!conn) {
236 		if (refcount_dec_and_test(&mark->refcnt))
237 			fsnotify_final_mark_destroy(mark);
238 		return;
239 	}
240 
241 	/*
242 	 * We have to be careful so that traversals of obj_list under lock can
243 	 * safely grab mark reference.
244 	 */
245 	if (!refcount_dec_and_lock(&mark->refcnt, &conn->lock))
246 		return;
247 
248 	hlist_del_init_rcu(&mark->obj_list);
249 	if (hlist_empty(&conn->list)) {
250 		objp = fsnotify_detach_connector_from_object(conn, &type);
251 		free_conn = true;
252 	} else {
253 		__fsnotify_recalc_mask(conn);
254 	}
255 	WRITE_ONCE(mark->connector, NULL);
256 	spin_unlock(&conn->lock);
257 
258 	fsnotify_drop_object(type, objp);
259 
260 	if (free_conn) {
261 		spin_lock(&destroy_lock);
262 		conn->destroy_next = connector_destroy_list;
263 		connector_destroy_list = conn;
264 		spin_unlock(&destroy_lock);
265 		queue_work(system_unbound_wq, &connector_reaper_work);
266 	}
267 	/*
268 	 * Note that we didn't update flags telling whether inode cares about
269 	 * what's happening with children. We update these flags from
270 	 * __fsnotify_parent() lazily when next event happens on one of our
271 	 * children.
272 	 */
273 	spin_lock(&destroy_lock);
274 	list_add(&mark->g_list, &destroy_list);
275 	spin_unlock(&destroy_lock);
276 	queue_delayed_work(system_unbound_wq, &reaper_work,
277 			   FSNOTIFY_REAPER_DELAY);
278 }
279 
280 /*
281  * Get mark reference when we found the mark via lockless traversal of object
282  * list. Mark can be already removed from the list by now and on its way to be
283  * destroyed once SRCU period ends.
284  *
285  * Also pin the group so it doesn't disappear under us.
286  */
287 static bool fsnotify_get_mark_safe(struct fsnotify_mark *mark)
288 {
289 	if (!mark)
290 		return true;
291 
292 	if (refcount_inc_not_zero(&mark->refcnt)) {
293 		spin_lock(&mark->lock);
294 		if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) {
295 			/* mark is attached, group is still alive then */
296 			atomic_inc(&mark->group->user_waits);
297 			spin_unlock(&mark->lock);
298 			return true;
299 		}
300 		spin_unlock(&mark->lock);
301 		fsnotify_put_mark(mark);
302 	}
303 	return false;
304 }
305 
306 /*
307  * Puts marks and wakes up group destruction if necessary.
308  *
309  * Pairs with fsnotify_get_mark_safe()
310  */
311 static void fsnotify_put_mark_wake(struct fsnotify_mark *mark)
312 {
313 	if (mark) {
314 		struct fsnotify_group *group = mark->group;
315 
316 		fsnotify_put_mark(mark);
317 		/*
318 		 * We abuse notification_waitq on group shutdown for waiting for
319 		 * all marks pinned when waiting for userspace.
320 		 */
321 		if (atomic_dec_and_test(&group->user_waits) && group->shutdown)
322 			wake_up(&group->notification_waitq);
323 	}
324 }
325 
326 bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info)
327 {
328 	int type;
329 
330 	fsnotify_foreach_obj_type(type) {
331 		/* This can fail if mark is being removed */
332 		if (!fsnotify_get_mark_safe(iter_info->marks[type]))
333 			goto fail;
334 	}
335 
336 	/*
337 	 * Now that both marks are pinned by refcount in the inode / vfsmount
338 	 * lists, we can drop SRCU lock, and safely resume the list iteration
339 	 * once userspace returns.
340 	 */
341 	srcu_read_unlock(&fsnotify_mark_srcu, iter_info->srcu_idx);
342 
343 	return true;
344 
345 fail:
346 	for (type--; type >= 0; type--)
347 		fsnotify_put_mark_wake(iter_info->marks[type]);
348 	return false;
349 }
350 
351 void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info)
352 {
353 	int type;
354 
355 	iter_info->srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
356 	fsnotify_foreach_obj_type(type)
357 		fsnotify_put_mark_wake(iter_info->marks[type]);
358 }
359 
360 /*
361  * Mark mark as detached, remove it from group list. Mark still stays in object
362  * list until its last reference is dropped. Note that we rely on mark being
363  * removed from group list before corresponding reference to it is dropped. In
364  * particular we rely on mark->connector being valid while we hold
365  * group->mark_mutex if we found the mark through g_list.
366  *
367  * Must be called with group->mark_mutex held. The caller must either hold
368  * reference to the mark or be protected by fsnotify_mark_srcu.
369  */
370 void fsnotify_detach_mark(struct fsnotify_mark *mark)
371 {
372 	struct fsnotify_group *group = mark->group;
373 
374 	WARN_ON_ONCE(!mutex_is_locked(&group->mark_mutex));
375 	WARN_ON_ONCE(!srcu_read_lock_held(&fsnotify_mark_srcu) &&
376 		     refcount_read(&mark->refcnt) < 1 +
377 			!!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED));
378 
379 	spin_lock(&mark->lock);
380 	/* something else already called this function on this mark */
381 	if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
382 		spin_unlock(&mark->lock);
383 		return;
384 	}
385 	mark->flags &= ~FSNOTIFY_MARK_FLAG_ATTACHED;
386 	list_del_init(&mark->g_list);
387 	spin_unlock(&mark->lock);
388 
389 	atomic_dec(&group->num_marks);
390 
391 	/* Drop mark reference acquired in fsnotify_add_mark_locked() */
392 	fsnotify_put_mark(mark);
393 }
394 
395 /*
396  * Free fsnotify mark. The mark is actually only marked as being freed.  The
397  * freeing is actually happening only once last reference to the mark is
398  * dropped from a workqueue which first waits for srcu period end.
399  *
400  * Caller must have a reference to the mark or be protected by
401  * fsnotify_mark_srcu.
402  */
403 void fsnotify_free_mark(struct fsnotify_mark *mark)
404 {
405 	struct fsnotify_group *group = mark->group;
406 
407 	spin_lock(&mark->lock);
408 	/* something else already called this function on this mark */
409 	if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
410 		spin_unlock(&mark->lock);
411 		return;
412 	}
413 	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
414 	spin_unlock(&mark->lock);
415 
416 	/*
417 	 * Some groups like to know that marks are being freed.  This is a
418 	 * callback to the group function to let it know that this mark
419 	 * is being freed.
420 	 */
421 	if (group->ops->freeing_mark)
422 		group->ops->freeing_mark(mark, group);
423 }
424 
425 void fsnotify_destroy_mark(struct fsnotify_mark *mark,
426 			   struct fsnotify_group *group)
427 {
428 	mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
429 	fsnotify_detach_mark(mark);
430 	mutex_unlock(&group->mark_mutex);
431 	fsnotify_free_mark(mark);
432 }
433 
434 /*
435  * Sorting function for lists of fsnotify marks.
436  *
437  * Fanotify supports different notification classes (reflected as priority of
438  * notification group). Events shall be passed to notification groups in
439  * decreasing priority order. To achieve this marks in notification lists for
440  * inodes and vfsmounts are sorted so that priorities of corresponding groups
441  * are descending.
442  *
443  * Furthermore correct handling of the ignore mask requires processing inode
444  * and vfsmount marks of each group together. Using the group address as
445  * further sort criterion provides a unique sorting order and thus we can
446  * merge inode and vfsmount lists of marks in linear time and find groups
447  * present in both lists.
448  *
449  * A return value of 1 signifies that b has priority over a.
450  * A return value of 0 signifies that the two marks have to be handled together.
451  * A return value of -1 signifies that a has priority over b.
452  */
453 int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b)
454 {
455 	if (a == b)
456 		return 0;
457 	if (!a)
458 		return 1;
459 	if (!b)
460 		return -1;
461 	if (a->priority < b->priority)
462 		return 1;
463 	if (a->priority > b->priority)
464 		return -1;
465 	if (a < b)
466 		return 1;
467 	return -1;
468 }
469 
470 static int fsnotify_attach_connector_to_object(fsnotify_connp_t *connp,
471 					       unsigned int type,
472 					       __kernel_fsid_t *fsid)
473 {
474 	struct inode *inode = NULL;
475 	struct fsnotify_mark_connector *conn;
476 
477 	conn = kmem_cache_alloc(fsnotify_mark_connector_cachep, GFP_KERNEL);
478 	if (!conn)
479 		return -ENOMEM;
480 	spin_lock_init(&conn->lock);
481 	INIT_HLIST_HEAD(&conn->list);
482 	conn->type = type;
483 	conn->obj = connp;
484 	/* Cache fsid of filesystem containing the object */
485 	if (fsid)
486 		conn->fsid = *fsid;
487 	else
488 		conn->fsid.val[0] = conn->fsid.val[1] = 0;
489 	if (conn->type == FSNOTIFY_OBJ_TYPE_INODE)
490 		inode = igrab(fsnotify_conn_inode(conn));
491 	/*
492 	 * cmpxchg() provides the barrier so that readers of *connp can see
493 	 * only initialized structure
494 	 */
495 	if (cmpxchg(connp, NULL, conn)) {
496 		/* Someone else created list structure for us */
497 		if (inode)
498 			iput(inode);
499 		kmem_cache_free(fsnotify_mark_connector_cachep, conn);
500 	}
501 
502 	return 0;
503 }
504 
505 /*
506  * Get mark connector, make sure it is alive and return with its lock held.
507  * This is for users that get connector pointer from inode or mount. Users that
508  * hold reference to a mark on the list may directly lock connector->lock as
509  * they are sure list cannot go away under them.
510  */
511 static struct fsnotify_mark_connector *fsnotify_grab_connector(
512 						fsnotify_connp_t *connp)
513 {
514 	struct fsnotify_mark_connector *conn;
515 	int idx;
516 
517 	idx = srcu_read_lock(&fsnotify_mark_srcu);
518 	conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
519 	if (!conn)
520 		goto out;
521 	spin_lock(&conn->lock);
522 	if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED) {
523 		spin_unlock(&conn->lock);
524 		srcu_read_unlock(&fsnotify_mark_srcu, idx);
525 		return NULL;
526 	}
527 out:
528 	srcu_read_unlock(&fsnotify_mark_srcu, idx);
529 	return conn;
530 }
531 
532 /*
533  * Add mark into proper place in given list of marks. These marks may be used
534  * for the fsnotify backend to determine which event types should be delivered
535  * to which group and for which inodes. These marks are ordered according to
536  * priority, highest number first, and then by the group's location in memory.
537  */
538 static int fsnotify_add_mark_list(struct fsnotify_mark *mark,
539 				  fsnotify_connp_t *connp, unsigned int type,
540 				  int allow_dups, __kernel_fsid_t *fsid)
541 {
542 	struct fsnotify_mark *lmark, *last = NULL;
543 	struct fsnotify_mark_connector *conn;
544 	int cmp;
545 	int err = 0;
546 
547 	if (WARN_ON(!fsnotify_valid_obj_type(type)))
548 		return -EINVAL;
549 
550 	/* Backend is expected to check for zero fsid (e.g. tmpfs) */
551 	if (fsid && WARN_ON_ONCE(!fsid->val[0] && !fsid->val[1]))
552 		return -ENODEV;
553 
554 restart:
555 	spin_lock(&mark->lock);
556 	conn = fsnotify_grab_connector(connp);
557 	if (!conn) {
558 		spin_unlock(&mark->lock);
559 		err = fsnotify_attach_connector_to_object(connp, type, fsid);
560 		if (err)
561 			return err;
562 		goto restart;
563 	} else if (fsid && (conn->fsid.val[0] || conn->fsid.val[1]) &&
564 		   (fsid->val[0] != conn->fsid.val[0] ||
565 		    fsid->val[1] != conn->fsid.val[1])) {
566 		/*
567 		 * Backend is expected to check for non uniform fsid
568 		 * (e.g. btrfs), but maybe we missed something?
569 		 * Only allow setting conn->fsid once to non zero fsid.
570 		 * inotify and non-fid fanotify groups do not set nor test
571 		 * conn->fsid.
572 		 */
573 		pr_warn_ratelimited("%s: fsid mismatch on object of type %u: "
574 				    "%x.%x != %x.%x\n", __func__, conn->type,
575 				    fsid->val[0], fsid->val[1],
576 				    conn->fsid.val[0], conn->fsid.val[1]);
577 		err = -EXDEV;
578 		goto out_err;
579 	}
580 
581 	/* is mark the first mark? */
582 	if (hlist_empty(&conn->list)) {
583 		hlist_add_head_rcu(&mark->obj_list, &conn->list);
584 		goto added;
585 	}
586 
587 	/* should mark be in the middle of the current list? */
588 	hlist_for_each_entry(lmark, &conn->list, obj_list) {
589 		last = lmark;
590 
591 		if ((lmark->group == mark->group) &&
592 		    (lmark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) &&
593 		    !allow_dups) {
594 			err = -EEXIST;
595 			goto out_err;
596 		}
597 
598 		cmp = fsnotify_compare_groups(lmark->group, mark->group);
599 		if (cmp >= 0) {
600 			hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list);
601 			goto added;
602 		}
603 	}
604 
605 	BUG_ON(last == NULL);
606 	/* mark should be the last entry.  last is the current last entry */
607 	hlist_add_behind_rcu(&mark->obj_list, &last->obj_list);
608 added:
609 	/*
610 	 * Since connector is attached to object using cmpxchg() we are
611 	 * guaranteed that connector initialization is fully visible by anyone
612 	 * seeing mark->connector set.
613 	 */
614 	WRITE_ONCE(mark->connector, conn);
615 out_err:
616 	spin_unlock(&conn->lock);
617 	spin_unlock(&mark->lock);
618 	return err;
619 }
620 
621 /*
622  * Attach an initialized mark to a given group and fs object.
623  * These marks may be used for the fsnotify backend to determine which
624  * event types should be delivered to which group.
625  */
626 int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
627 			     fsnotify_connp_t *connp, unsigned int type,
628 			     int allow_dups, __kernel_fsid_t *fsid)
629 {
630 	struct fsnotify_group *group = mark->group;
631 	int ret = 0;
632 
633 	BUG_ON(!mutex_is_locked(&group->mark_mutex));
634 
635 	/*
636 	 * LOCKING ORDER!!!!
637 	 * group->mark_mutex
638 	 * mark->lock
639 	 * mark->connector->lock
640 	 */
641 	spin_lock(&mark->lock);
642 	mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE | FSNOTIFY_MARK_FLAG_ATTACHED;
643 
644 	list_add(&mark->g_list, &group->marks_list);
645 	atomic_inc(&group->num_marks);
646 	fsnotify_get_mark(mark); /* for g_list */
647 	spin_unlock(&mark->lock);
648 
649 	ret = fsnotify_add_mark_list(mark, connp, type, allow_dups, fsid);
650 	if (ret)
651 		goto err;
652 
653 	if (mark->mask)
654 		fsnotify_recalc_mask(mark->connector);
655 
656 	return ret;
657 err:
658 	spin_lock(&mark->lock);
659 	mark->flags &= ~(FSNOTIFY_MARK_FLAG_ALIVE |
660 			 FSNOTIFY_MARK_FLAG_ATTACHED);
661 	list_del_init(&mark->g_list);
662 	spin_unlock(&mark->lock);
663 	atomic_dec(&group->num_marks);
664 
665 	fsnotify_put_mark(mark);
666 	return ret;
667 }
668 
669 int fsnotify_add_mark(struct fsnotify_mark *mark, fsnotify_connp_t *connp,
670 		      unsigned int type, int allow_dups, __kernel_fsid_t *fsid)
671 {
672 	int ret;
673 	struct fsnotify_group *group = mark->group;
674 
675 	mutex_lock(&group->mark_mutex);
676 	ret = fsnotify_add_mark_locked(mark, connp, type, allow_dups, fsid);
677 	mutex_unlock(&group->mark_mutex);
678 	return ret;
679 }
680 
681 /*
682  * Given a list of marks, find the mark associated with given group. If found
683  * take a reference to that mark and return it, else return NULL.
684  */
685 struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp,
686 					 struct fsnotify_group *group)
687 {
688 	struct fsnotify_mark_connector *conn;
689 	struct fsnotify_mark *mark;
690 
691 	conn = fsnotify_grab_connector(connp);
692 	if (!conn)
693 		return NULL;
694 
695 	hlist_for_each_entry(mark, &conn->list, obj_list) {
696 		if (mark->group == group &&
697 		    (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
698 			fsnotify_get_mark(mark);
699 			spin_unlock(&conn->lock);
700 			return mark;
701 		}
702 	}
703 	spin_unlock(&conn->lock);
704 	return NULL;
705 }
706 
707 /* Clear any marks in a group with given type mask */
708 void fsnotify_clear_marks_by_group(struct fsnotify_group *group,
709 				   unsigned int type_mask)
710 {
711 	struct fsnotify_mark *lmark, *mark;
712 	LIST_HEAD(to_free);
713 	struct list_head *head = &to_free;
714 
715 	/* Skip selection step if we want to clear all marks. */
716 	if (type_mask == FSNOTIFY_OBJ_ALL_TYPES_MASK) {
717 		head = &group->marks_list;
718 		goto clear;
719 	}
720 	/*
721 	 * We have to be really careful here. Anytime we drop mark_mutex, e.g.
722 	 * fsnotify_clear_marks_by_inode() can come and free marks. Even in our
723 	 * to_free list so we have to use mark_mutex even when accessing that
724 	 * list. And freeing mark requires us to drop mark_mutex. So we can
725 	 * reliably free only the first mark in the list. That's why we first
726 	 * move marks to free to to_free list in one go and then free marks in
727 	 * to_free list one by one.
728 	 */
729 	mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
730 	list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
731 		if ((1U << mark->connector->type) & type_mask)
732 			list_move(&mark->g_list, &to_free);
733 	}
734 	mutex_unlock(&group->mark_mutex);
735 
736 clear:
737 	while (1) {
738 		mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
739 		if (list_empty(head)) {
740 			mutex_unlock(&group->mark_mutex);
741 			break;
742 		}
743 		mark = list_first_entry(head, struct fsnotify_mark, g_list);
744 		fsnotify_get_mark(mark);
745 		fsnotify_detach_mark(mark);
746 		mutex_unlock(&group->mark_mutex);
747 		fsnotify_free_mark(mark);
748 		fsnotify_put_mark(mark);
749 	}
750 }
751 
752 /* Destroy all marks attached to an object via connector */
753 void fsnotify_destroy_marks(fsnotify_connp_t *connp)
754 {
755 	struct fsnotify_mark_connector *conn;
756 	struct fsnotify_mark *mark, *old_mark = NULL;
757 	void *objp;
758 	unsigned int type;
759 
760 	conn = fsnotify_grab_connector(connp);
761 	if (!conn)
762 		return;
763 	/*
764 	 * We have to be careful since we can race with e.g.
765 	 * fsnotify_clear_marks_by_group() and once we drop the conn->lock, the
766 	 * list can get modified. However we are holding mark reference and
767 	 * thus our mark cannot be removed from obj_list so we can continue
768 	 * iteration after regaining conn->lock.
769 	 */
770 	hlist_for_each_entry(mark, &conn->list, obj_list) {
771 		fsnotify_get_mark(mark);
772 		spin_unlock(&conn->lock);
773 		if (old_mark)
774 			fsnotify_put_mark(old_mark);
775 		old_mark = mark;
776 		fsnotify_destroy_mark(mark, mark->group);
777 		spin_lock(&conn->lock);
778 	}
779 	/*
780 	 * Detach list from object now so that we don't pin inode until all
781 	 * mark references get dropped. It would lead to strange results such
782 	 * as delaying inode deletion or blocking unmount.
783 	 */
784 	objp = fsnotify_detach_connector_from_object(conn, &type);
785 	spin_unlock(&conn->lock);
786 	if (old_mark)
787 		fsnotify_put_mark(old_mark);
788 	fsnotify_drop_object(type, objp);
789 }
790 
791 /*
792  * Nothing fancy, just initialize lists and locks and counters.
793  */
794 void fsnotify_init_mark(struct fsnotify_mark *mark,
795 			struct fsnotify_group *group)
796 {
797 	memset(mark, 0, sizeof(*mark));
798 	spin_lock_init(&mark->lock);
799 	refcount_set(&mark->refcnt, 1);
800 	fsnotify_get_group(group);
801 	mark->group = group;
802 	WRITE_ONCE(mark->connector, NULL);
803 }
804 
805 /*
806  * Destroy all marks in destroy_list, waits for SRCU period to finish before
807  * actually freeing marks.
808  */
809 static void fsnotify_mark_destroy_workfn(struct work_struct *work)
810 {
811 	struct fsnotify_mark *mark, *next;
812 	struct list_head private_destroy_list;
813 
814 	spin_lock(&destroy_lock);
815 	/* exchange the list head */
816 	list_replace_init(&destroy_list, &private_destroy_list);
817 	spin_unlock(&destroy_lock);
818 
819 	synchronize_srcu(&fsnotify_mark_srcu);
820 
821 	list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) {
822 		list_del_init(&mark->g_list);
823 		fsnotify_final_mark_destroy(mark);
824 	}
825 }
826 
827 /* Wait for all marks queued for destruction to be actually destroyed */
828 void fsnotify_wait_marks_destroyed(void)
829 {
830 	flush_delayed_work(&reaper_work);
831 }
832