xref: /linux/fs/notify/fsnotify.c (revision 69050f8d6d075dc01af7a5f2f550a8067510366f)
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 #include <linux/dcache.h>
7 #include <linux/fs.h>
8 #include <linux/gfp.h>
9 #include <linux/init.h>
10 #include <linux/module.h>
11 #include <linux/mount.h>
12 #include <linux/srcu.h>
13 
14 #include <linux/fsnotify_backend.h>
15 #include "fsnotify.h"
16 
17 /*
18  * Clear all of the marks on an inode when it is being evicted from core
19  */
20 void __fsnotify_inode_delete(struct inode *inode)
21 {
22 	fsnotify_clear_marks_by_inode(inode);
23 }
24 EXPORT_SYMBOL_GPL(__fsnotify_inode_delete);
25 
26 void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
27 {
28 	fsnotify_clear_marks_by_mount(mnt);
29 }
30 
31 void __fsnotify_mntns_delete(struct mnt_namespace *mntns)
32 {
33 	fsnotify_clear_marks_by_mntns(mntns);
34 }
35 
36 void fsnotify_sb_delete(struct super_block *sb)
37 {
38 	struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
39 
40 	/* Were any marks ever added to any object on this sb? */
41 	if (!sbinfo)
42 		return;
43 
44 	fsnotify_unmount_inodes(sbinfo);
45 	fsnotify_clear_marks_by_sb(sb);
46 	/* Wait for outstanding object references from connectors */
47 	wait_var_event(fsnotify_sb_watched_objects(sb),
48 		       !atomic_long_read(fsnotify_sb_watched_objects(sb)));
49 	WARN_ON(fsnotify_sb_has_priority_watchers(sb, FSNOTIFY_PRIO_CONTENT));
50 	WARN_ON(fsnotify_sb_has_priority_watchers(sb,
51 						  FSNOTIFY_PRIO_PRE_CONTENT));
52 }
53 
54 void fsnotify_sb_free(struct super_block *sb)
55 {
56 	if (sb->s_fsnotify_info) {
57 		WARN_ON_ONCE(!list_empty(&sb->s_fsnotify_info->inode_conn_list));
58 		kfree(sb->s_fsnotify_info);
59 	}
60 }
61 
62 /*
63  * Given an inode, first check if we care what happens to our children.  Inotify
64  * and dnotify both tell their parents about events.  If we care about any event
65  * on a child we run all of our children and set a dentry flag saying that the
66  * parent cares.  Thus when an event happens on a child it can quickly tell
67  * if there is a need to find a parent and send the event to the parent.
68  */
69 void fsnotify_set_children_dentry_flags(struct inode *inode)
70 {
71 	struct dentry *alias;
72 
73 	if (!S_ISDIR(inode->i_mode))
74 		return;
75 
76 	spin_lock(&inode->i_lock);
77 	/* run all of the dentries associated with this inode.  Since this is a
78 	 * directory, there damn well better only be one item on this list */
79 	hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
80 		struct dentry *child;
81 
82 		/* run all of the children of the original inode and fix their
83 		 * d_flags to indicate parental interest (their parent is the
84 		 * original inode) */
85 		spin_lock(&alias->d_lock);
86 		hlist_for_each_entry(child, &alias->d_children, d_sib) {
87 			if (!child->d_inode)
88 				continue;
89 
90 			spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
91 			child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
92 			spin_unlock(&child->d_lock);
93 		}
94 		spin_unlock(&alias->d_lock);
95 	}
96 	spin_unlock(&inode->i_lock);
97 }
98 
99 /*
100  * Lazily clear false positive PARENT_WATCHED flag for child whose parent had
101  * stopped watching children.
102  */
103 static void fsnotify_clear_child_dentry_flag(struct inode *pinode,
104 					     struct dentry *dentry)
105 {
106 	spin_lock(&dentry->d_lock);
107 	/*
108 	 * d_lock is a sufficient barrier to prevent observing a non-watched
109 	 * parent state from before the fsnotify_set_children_dentry_flags()
110 	 * or fsnotify_update_flags() call that had set PARENT_WATCHED.
111 	 */
112 	if (!fsnotify_inode_watches_children(pinode))
113 		dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
114 	spin_unlock(&dentry->d_lock);
115 }
116 
117 /* Are inode/sb/mount interested in parent and name info with this event? */
118 static bool fsnotify_event_needs_parent(struct inode *inode, __u32 mnt_mask,
119 					__u32 mask)
120 {
121 	__u32 marks_mask = 0;
122 
123 	/* We only send parent/name to inode/sb/mount for events on non-dir */
124 	if (mask & FS_ISDIR)
125 		return false;
126 
127 	/*
128 	 * All events that are possible on child can also may be reported with
129 	 * parent/name info to inode/sb/mount.  Otherwise, a watching parent
130 	 * could result in events reported with unexpected name info to sb/mount.
131 	 */
132 	BUILD_BUG_ON(FS_EVENTS_POSS_ON_CHILD & ~FS_EVENTS_POSS_TO_PARENT);
133 
134 	/* Did either inode/sb/mount subscribe for events with parent/name? */
135 	marks_mask |= fsnotify_parent_needed_mask(
136 				READ_ONCE(inode->i_fsnotify_mask));
137 	marks_mask |= fsnotify_parent_needed_mask(
138 				READ_ONCE(inode->i_sb->s_fsnotify_mask));
139 	marks_mask |= fsnotify_parent_needed_mask(mnt_mask);
140 
141 	/* Did they subscribe for this event with parent/name info? */
142 	return mask & marks_mask;
143 }
144 
145 /* Are there any inode/mount/sb objects that watch for these events? */
146 static inline __u32 fsnotify_object_watched(struct inode *inode, __u32 mnt_mask,
147 					    __u32 mask)
148 {
149 	__u32 marks_mask = READ_ONCE(inode->i_fsnotify_mask) | mnt_mask |
150 			   READ_ONCE(inode->i_sb->s_fsnotify_mask);
151 
152 	return mask & marks_mask & ALL_FSNOTIFY_EVENTS;
153 }
154 
155 /* Report pre-content event with optional range info */
156 int fsnotify_pre_content(const struct path *path, const loff_t *ppos,
157 			 size_t count)
158 {
159 	struct file_range range;
160 
161 	/* Report page aligned range only when pos is known */
162 	if (!ppos)
163 		return fsnotify_path(path, FS_PRE_ACCESS);
164 
165 	range.path = path;
166 	range.pos = PAGE_ALIGN_DOWN(*ppos);
167 	range.count = PAGE_ALIGN(*ppos + count) - range.pos;
168 
169 	return fsnotify_parent(path->dentry, FS_PRE_ACCESS, &range,
170 			       FSNOTIFY_EVENT_FILE_RANGE);
171 }
172 
173 /*
174  * Notify this dentry's parent about a child's events with child name info
175  * if parent is watching or if inode/sb/mount are interested in events with
176  * parent and name info.
177  *
178  * Notify only the child without name info if parent is not watching and
179  * inode/sb/mount are not interested in events with parent and name info.
180  */
181 int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
182 		      int data_type)
183 {
184 	const struct path *path = fsnotify_data_path(data, data_type);
185 	__u32 mnt_mask = path ?
186 		READ_ONCE(real_mount(path->mnt)->mnt_fsnotify_mask) : 0;
187 	struct inode *inode = d_inode(dentry);
188 	struct dentry *parent;
189 	bool parent_watched = dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED;
190 	bool parent_needed, parent_interested;
191 	__u32 p_mask;
192 	struct inode *p_inode = NULL;
193 	struct name_snapshot name;
194 	struct qstr *file_name = NULL;
195 	int ret = 0;
196 
197 	/* Optimize the likely case of nobody watching this path */
198 	if (likely(!parent_watched &&
199 		   !fsnotify_object_watched(inode, mnt_mask, mask)))
200 		return 0;
201 
202 	parent = NULL;
203 	parent_needed = fsnotify_event_needs_parent(inode, mnt_mask, mask);
204 	if (!parent_watched && !parent_needed)
205 		goto notify;
206 
207 	/* Does parent inode care about events on children? */
208 	parent = dget_parent(dentry);
209 	p_inode = parent->d_inode;
210 	p_mask = fsnotify_inode_watches_children(p_inode);
211 	if (unlikely(parent_watched && !p_mask))
212 		fsnotify_clear_child_dentry_flag(p_inode, dentry);
213 
214 	/*
215 	 * Include parent/name in notification either if some notification
216 	 * groups require parent info or the parent is interested in this event.
217 	 * The parent interest in ACCESS/MODIFY events does not apply to special
218 	 * files, where read/write are not on the filesystem of the parent and
219 	 * events can provide an undesirable side-channel for information
220 	 * exfiltration.
221 	 */
222 	parent_interested = mask & p_mask & ALL_FSNOTIFY_EVENTS &&
223 			    !(data_type == FSNOTIFY_EVENT_PATH &&
224 			      d_is_special(dentry) &&
225 			      (mask & (FS_ACCESS | FS_MODIFY)));
226 	if (parent_needed || parent_interested) {
227 		/* When notifying parent, child should be passed as data */
228 		WARN_ON_ONCE(inode != fsnotify_data_inode(data, data_type));
229 
230 		/* Notify both parent and child with child name info */
231 		take_dentry_name_snapshot(&name, dentry);
232 		file_name = &name.name;
233 		if (parent_interested)
234 			mask |= FS_EVENT_ON_CHILD;
235 	}
236 
237 notify:
238 	ret = fsnotify(mask, data, data_type, p_inode, file_name, inode, 0);
239 
240 	if (file_name)
241 		release_dentry_name_snapshot(&name);
242 	dput(parent);
243 
244 	return ret;
245 }
246 EXPORT_SYMBOL_GPL(__fsnotify_parent);
247 
248 static int fsnotify_handle_inode_event(struct fsnotify_group *group,
249 				       struct fsnotify_mark *inode_mark,
250 				       u32 mask, const void *data, int data_type,
251 				       struct inode *dir, const struct qstr *name,
252 				       u32 cookie)
253 {
254 	const struct path *path = fsnotify_data_path(data, data_type);
255 	struct inode *inode = fsnotify_data_inode(data, data_type);
256 	const struct fsnotify_ops *ops = group->ops;
257 
258 	if (WARN_ON_ONCE(!ops->handle_inode_event))
259 		return 0;
260 
261 	if (WARN_ON_ONCE(!inode && !dir))
262 		return 0;
263 
264 	if ((inode_mark->flags & FSNOTIFY_MARK_FLAG_EXCL_UNLINK) &&
265 	    path && d_unlinked(path->dentry))
266 		return 0;
267 
268 	/* Check interest of this mark in case event was sent with two marks */
269 	if (!(mask & inode_mark->mask & ALL_FSNOTIFY_EVENTS))
270 		return 0;
271 
272 	return ops->handle_inode_event(inode_mark, mask, inode, dir, name, cookie);
273 }
274 
275 static int fsnotify_handle_event(struct fsnotify_group *group, __u32 mask,
276 				 const void *data, int data_type,
277 				 struct inode *dir, const struct qstr *name,
278 				 u32 cookie, struct fsnotify_iter_info *iter_info)
279 {
280 	struct fsnotify_mark *inode_mark = fsnotify_iter_inode_mark(iter_info);
281 	struct fsnotify_mark *parent_mark = fsnotify_iter_parent_mark(iter_info);
282 	int ret;
283 
284 	if (WARN_ON_ONCE(fsnotify_iter_sb_mark(iter_info)) ||
285 	    WARN_ON_ONCE(fsnotify_iter_vfsmount_mark(iter_info)))
286 		return 0;
287 
288 	/*
289 	 * For FS_RENAME, 'dir' is old dir and 'data' is new dentry.
290 	 * The only ->handle_inode_event() backend that supports FS_RENAME is
291 	 * dnotify, where it means file was renamed within same parent.
292 	 */
293 	if (mask & FS_RENAME) {
294 		struct dentry *moved = fsnotify_data_dentry(data, data_type);
295 
296 		if (dir != moved->d_parent->d_inode)
297 			return 0;
298 	}
299 
300 	if (parent_mark) {
301 		ret = fsnotify_handle_inode_event(group, parent_mark, mask,
302 						  data, data_type, dir, name, 0);
303 		if (ret)
304 			return ret;
305 	}
306 
307 	if (!inode_mark)
308 		return 0;
309 
310 	/*
311 	 * Some events can be sent on both parent dir and child marks (e.g.
312 	 * FS_ATTRIB).  If both parent dir and child are watching, report the
313 	 * event once to parent dir with name (if interested) and once to child
314 	 * without name (if interested).
315 	 *
316 	 * In any case regardless whether the parent is watching or not, the
317 	 * child watcher is expecting an event without the FS_EVENT_ON_CHILD
318 	 * flag. The file name is expected if and only if this is a directory
319 	 * event.
320 	 */
321 	mask &= ~FS_EVENT_ON_CHILD;
322 	if (!(mask & ALL_FSNOTIFY_DIRENT_EVENTS)) {
323 		dir = NULL;
324 		name = NULL;
325 	}
326 
327 	return fsnotify_handle_inode_event(group, inode_mark, mask, data, data_type,
328 					   dir, name, cookie);
329 }
330 
331 static int send_to_group(__u32 mask, const void *data, int data_type,
332 			 struct inode *dir, const struct qstr *file_name,
333 			 u32 cookie, struct fsnotify_iter_info *iter_info)
334 {
335 	struct fsnotify_group *group = NULL;
336 	__u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
337 	__u32 marks_mask = 0;
338 	__u32 marks_ignore_mask = 0;
339 	bool is_dir = mask & FS_ISDIR;
340 	struct fsnotify_mark *mark;
341 	int type;
342 
343 	if (!iter_info->report_mask)
344 		return 0;
345 
346 	/* clear ignored on inode modification */
347 	if (mask & FS_MODIFY) {
348 		fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
349 			if (!(mark->flags &
350 			      FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
351 				mark->ignore_mask = 0;
352 		}
353 	}
354 
355 	/* Are any of the group marks interested in this event? */
356 	fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
357 		group = mark->group;
358 		marks_mask |= mark->mask;
359 		marks_ignore_mask |=
360 			fsnotify_effective_ignore_mask(mark, is_dir, type);
361 	}
362 
363 	pr_debug("%s: group=%p mask=%x marks_mask=%x marks_ignore_mask=%x data=%p data_type=%d dir=%p cookie=%d\n",
364 		 __func__, group, mask, marks_mask, marks_ignore_mask,
365 		 data, data_type, dir, cookie);
366 
367 	if (!(test_mask & marks_mask & ~marks_ignore_mask))
368 		return 0;
369 
370 	if (group->ops->handle_event) {
371 		return group->ops->handle_event(group, mask, data, data_type, dir,
372 						file_name, cookie, iter_info);
373 	}
374 
375 	return fsnotify_handle_event(group, mask, data, data_type, dir,
376 				     file_name, cookie, iter_info);
377 }
378 
379 static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector *const *connp)
380 {
381 	struct fsnotify_mark_connector *conn;
382 	struct hlist_node *node = NULL;
383 
384 	conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
385 	if (conn)
386 		node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
387 
388 	return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
389 }
390 
391 static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
392 {
393 	struct hlist_node *node = NULL;
394 
395 	if (mark)
396 		node = srcu_dereference(mark->obj_list.next,
397 					&fsnotify_mark_srcu);
398 
399 	return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
400 }
401 
402 /*
403  * iter_info is a multi head priority queue of marks.
404  * Pick a subset of marks from queue heads, all with the same group
405  * and set the report_mask to a subset of the selected marks.
406  * Returns false if there are no more groups to iterate.
407  */
408 static bool fsnotify_iter_select_report_types(
409 		struct fsnotify_iter_info *iter_info)
410 {
411 	struct fsnotify_group *max_prio_group = NULL;
412 	struct fsnotify_mark *mark;
413 	int type;
414 
415 	/* Choose max prio group among groups of all queue heads */
416 	fsnotify_foreach_iter_type(type) {
417 		mark = iter_info->marks[type];
418 		if (mark &&
419 		    fsnotify_compare_groups(max_prio_group, mark->group) > 0)
420 			max_prio_group = mark->group;
421 	}
422 
423 	if (!max_prio_group)
424 		return false;
425 
426 	/* Set the report mask for marks from same group as max prio group */
427 	iter_info->current_group = max_prio_group;
428 	iter_info->report_mask = 0;
429 	fsnotify_foreach_iter_type(type) {
430 		mark = iter_info->marks[type];
431 		if (mark && mark->group == iter_info->current_group) {
432 			/*
433 			 * FSNOTIFY_ITER_TYPE_PARENT indicates that this inode
434 			 * is watching children and interested in this event,
435 			 * which is an event possible on child.
436 			 * But is *this mark* watching children?
437 			 */
438 			if (type == FSNOTIFY_ITER_TYPE_PARENT &&
439 			    !(mark->mask & FS_EVENT_ON_CHILD) &&
440 			    !(fsnotify_ignore_mask(mark) & FS_EVENT_ON_CHILD))
441 				continue;
442 
443 			fsnotify_iter_set_report_type(iter_info, type);
444 		}
445 	}
446 
447 	return true;
448 }
449 
450 /*
451  * Pop from iter_info multi head queue, the marks that belong to the group of
452  * current iteration step.
453  */
454 static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
455 {
456 	struct fsnotify_mark *mark;
457 	int type;
458 
459 	/*
460 	 * We cannot use fsnotify_foreach_iter_mark_type() here because we
461 	 * may need to advance a mark of type X that belongs to current_group
462 	 * but was not selected for reporting.
463 	 */
464 	fsnotify_foreach_iter_type(type) {
465 		mark = iter_info->marks[type];
466 		if (mark && mark->group == iter_info->current_group)
467 			iter_info->marks[type] =
468 				fsnotify_next_mark(iter_info->marks[type]);
469 	}
470 }
471 
472 /*
473  * fsnotify - This is the main call to fsnotify.
474  *
475  * The VFS calls into hook specific functions in linux/fsnotify.h.
476  * Those functions then in turn call here.  Here will call out to all of the
477  * registered fsnotify_group.  Those groups can then use the notification event
478  * in whatever means they feel necessary.
479  *
480  * @mask:	event type and flags
481  * @data:	object that event happened on
482  * @data_type:	type of object for fanotify_data_XXX() accessors
483  * @dir:	optional directory associated with event -
484  *		if @file_name is not NULL, this is the directory that
485  *		@file_name is relative to
486  * @file_name:	optional file name associated with event
487  * @inode:	optional inode associated with event -
488  *		If @dir and @inode are both non-NULL, event may be
489  *		reported to both.
490  * @cookie:	inotify rename cookie
491  */
492 int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
493 	     const struct qstr *file_name, struct inode *inode, u32 cookie)
494 {
495 	const struct path *path = fsnotify_data_path(data, data_type);
496 	struct super_block *sb = fsnotify_data_sb(data, data_type);
497 	const struct fsnotify_mnt *mnt_data = fsnotify_data_mnt(data, data_type);
498 	struct fsnotify_sb_info *sbinfo = sb ? fsnotify_sb_info(sb) : NULL;
499 	struct fsnotify_iter_info iter_info = {};
500 	struct mount *mnt = NULL;
501 	struct inode *inode2 = NULL;
502 	struct dentry *moved;
503 	int inode2_type;
504 	int ret = 0;
505 	__u32 test_mask, marks_mask = 0;
506 
507 	if (path)
508 		mnt = real_mount(path->mnt);
509 
510 	if (!inode) {
511 		/* Dirent event - report on TYPE_INODE to dir */
512 		inode = dir;
513 		/* For FS_RENAME, inode is old_dir and inode2 is new_dir */
514 		if (mask & FS_RENAME) {
515 			moved = fsnotify_data_dentry(data, data_type);
516 			inode2 = moved->d_parent->d_inode;
517 			inode2_type = FSNOTIFY_ITER_TYPE_INODE2;
518 		}
519 	} else if (mask & FS_EVENT_ON_CHILD) {
520 		/*
521 		 * Event on child - report on TYPE_PARENT to dir if it is
522 		 * watching children and on TYPE_INODE to child.
523 		 */
524 		inode2 = dir;
525 		inode2_type = FSNOTIFY_ITER_TYPE_PARENT;
526 	}
527 
528 	/*
529 	 * Optimization: srcu_read_lock() has a memory barrier which can
530 	 * be expensive.  It protects walking the *_fsnotify_marks lists.
531 	 * However, if we do not walk the lists, we do not have to do
532 	 * SRCU because we have no references to any objects and do not
533 	 * need SRCU to keep them "alive".
534 	 */
535 	if ((!sbinfo || !sbinfo->sb_marks) &&
536 	    (!mnt || !mnt->mnt_fsnotify_marks) &&
537 	    (!inode || !inode->i_fsnotify_marks) &&
538 	    (!inode2 || !inode2->i_fsnotify_marks) &&
539 	    (!mnt_data || !mnt_data->ns->n_fsnotify_marks))
540 		return 0;
541 
542 	if (sb)
543 		marks_mask |= READ_ONCE(sb->s_fsnotify_mask);
544 	if (mnt)
545 		marks_mask |= READ_ONCE(mnt->mnt_fsnotify_mask);
546 	if (inode)
547 		marks_mask |= READ_ONCE(inode->i_fsnotify_mask);
548 	if (inode2)
549 		marks_mask |= READ_ONCE(inode2->i_fsnotify_mask);
550 	if (mnt_data)
551 		marks_mask |= READ_ONCE(mnt_data->ns->n_fsnotify_mask);
552 
553 	/*
554 	 * If this is a modify event we may need to clear some ignore masks.
555 	 * In that case, the object with ignore masks will have the FS_MODIFY
556 	 * event in its mask.
557 	 * Otherwise, return if none of the marks care about this type of event.
558 	 */
559 	test_mask = (mask & ALL_FSNOTIFY_EVENTS);
560 	if (!(test_mask & marks_mask))
561 		return 0;
562 
563 	iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
564 
565 	if (sbinfo) {
566 		iter_info.marks[FSNOTIFY_ITER_TYPE_SB] =
567 			fsnotify_first_mark(&sbinfo->sb_marks);
568 	}
569 	if (mnt) {
570 		iter_info.marks[FSNOTIFY_ITER_TYPE_VFSMOUNT] =
571 			fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
572 	}
573 	if (inode) {
574 		iter_info.marks[FSNOTIFY_ITER_TYPE_INODE] =
575 			fsnotify_first_mark(&inode->i_fsnotify_marks);
576 	}
577 	if (inode2) {
578 		iter_info.marks[inode2_type] =
579 			fsnotify_first_mark(&inode2->i_fsnotify_marks);
580 	}
581 	if (mnt_data) {
582 		iter_info.marks[FSNOTIFY_ITER_TYPE_MNTNS] =
583 			fsnotify_first_mark(&mnt_data->ns->n_fsnotify_marks);
584 	}
585 
586 	/*
587 	 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
588 	 * ignore masks are properly reflected for mount/sb mark notifications.
589 	 * That's why this traversal is so complicated...
590 	 */
591 	while (fsnotify_iter_select_report_types(&iter_info)) {
592 		ret = send_to_group(mask, data, data_type, dir, file_name,
593 				    cookie, &iter_info);
594 
595 		if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
596 			goto out;
597 
598 		fsnotify_iter_next(&iter_info);
599 	}
600 	ret = 0;
601 out:
602 	srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
603 
604 	return ret;
605 }
606 EXPORT_SYMBOL_GPL(fsnotify);
607 
608 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
609 /*
610  * At open time we check fsnotify_sb_has_priority_watchers(), call the open perm
611  * hook and set the FMODE_NONOTIFY_ mode bits accordignly.
612  * Later, fsnotify permission hooks do not check if there are permission event
613  * watches, but that there were permission event watches at open time.
614  */
615 int fsnotify_open_perm_and_set_mode(struct file *file)
616 {
617 	struct dentry *dentry = file->f_path.dentry, *parent;
618 	struct super_block *sb = dentry->d_sb;
619 	__u32 mnt_mask, p_mask = 0;
620 
621 	/* Is it a file opened by fanotify? */
622 	if (FMODE_FSNOTIFY_NONE(file->f_mode))
623 		return 0;
624 
625 	/*
626 	 * Permission events is a super set of pre-content events, so if there
627 	 * are no permission event watchers, there are also no pre-content event
628 	 * watchers and this is implied from the single FMODE_NONOTIFY_PERM bit.
629 	 */
630 	if (likely(!fsnotify_sb_has_priority_watchers(sb,
631 						FSNOTIFY_PRIO_CONTENT))) {
632 		file_set_fsnotify_mode(file, FMODE_NONOTIFY_PERM);
633 		return 0;
634 	}
635 
636 	/*
637 	 * OK, there are some permission event watchers. Check if anybody is
638 	 * watching for permission events on *this* file.
639 	 */
640 	mnt_mask = READ_ONCE(real_mount(file->f_path.mnt)->mnt_fsnotify_mask);
641 	p_mask = fsnotify_object_watched(d_inode(dentry), mnt_mask,
642 					 ALL_FSNOTIFY_PERM_EVENTS);
643 	if (dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED) {
644 		parent = dget_parent(dentry);
645 		p_mask |= fsnotify_inode_watches_children(d_inode(parent));
646 		dput(parent);
647 	}
648 
649 	/*
650 	 * Legacy FAN_ACCESS_PERM events have very high performance overhead,
651 	 * so unlikely to be used in the wild. If they are used there will be
652 	 * no optimizations at all.
653 	 */
654 	if (unlikely(p_mask & FS_ACCESS_PERM)) {
655 		/* Enable all permission and pre-content events */
656 		file_set_fsnotify_mode(file, 0);
657 		goto open_perm;
658 	}
659 
660 	/*
661 	 * Pre-content events are only supported on regular files.
662 	 * If there are pre-content event watchers and no permission access
663 	 * watchers, set FMODE_NONOTIFY | FMODE_NONOTIFY_PERM to indicate that.
664 	 * That is the common case with HSM service.
665 	 */
666 	if (d_is_reg(dentry) && (p_mask & FSNOTIFY_PRE_CONTENT_EVENTS)) {
667 		file_set_fsnotify_mode(file, FMODE_NONOTIFY |
668 					     FMODE_NONOTIFY_PERM);
669 		goto open_perm;
670 	}
671 
672 	/* Nobody watching permission and pre-content events on this file */
673 	file_set_fsnotify_mode(file, FMODE_NONOTIFY_PERM);
674 
675 open_perm:
676 	/*
677 	 * Send open perm events depending on object masks and regardless of
678 	 * FMODE_NONOTIFY_PERM.
679 	 */
680 	if (file->f_flags & __FMODE_EXEC && p_mask & FS_OPEN_EXEC_PERM) {
681 		int ret = fsnotify_path(&file->f_path, FS_OPEN_EXEC_PERM);
682 
683 		if (ret)
684 			return ret;
685 	}
686 
687 	if (p_mask & FS_OPEN_PERM)
688 		return fsnotify_path(&file->f_path, FS_OPEN_PERM);
689 
690 	return 0;
691 }
692 #endif
693 
694 void fsnotify_mnt(__u32 mask, struct mnt_namespace *ns, struct vfsmount *mnt)
695 {
696 	struct fsnotify_mnt data = {
697 		.ns = ns,
698 		.mnt_id = real_mount(mnt)->mnt_id_unique,
699 	};
700 
701 	if (WARN_ON_ONCE(!ns))
702 		return;
703 
704 	/*
705 	 * This is an optimization as well as making sure fsnotify_init() has
706 	 * been called.
707 	 */
708 	if (!ns->n_fsnotify_marks)
709 		return;
710 
711 	fsnotify(mask, &data, FSNOTIFY_EVENT_MNT, NULL, NULL, NULL, 0);
712 }
713 
714 static __init int fsnotify_init(void)
715 {
716 	int ret;
717 
718 	BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 26);
719 
720 	ret = init_srcu_struct(&fsnotify_mark_srcu);
721 	if (ret)
722 		panic("initializing fsnotify_mark_srcu");
723 
724 	fsnotify_init_connector_caches();
725 
726 	return 0;
727 }
728 core_initcall(fsnotify_init);
729