fsnotify.c

  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 */
 20void __fsnotify_inode_delete(struct inode *inode)
 21{
 22	fsnotify_clear_marks_by_inode(inode);
 23}
 24EXPORT_SYMBOL_GPL(__fsnotify_inode_delete);
 25
 26void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
 27{
 28	fsnotify_clear_marks_by_mount(mnt);
 29}
 30
 31void __fsnotify_mntns_delete(struct mnt_namespace *mntns)
 32{
 33	fsnotify_clear_marks_by_mntns(mntns);
 34}
 35
 36/**
 37 * fsnotify_unmount_inodes - an sb is unmounting.  handle any watched inodes.
 38 * @sb: superblock being unmounted.
 39 *
 40 * Called during unmount with no locks held, so needs to be safe against
 41 * concurrent modifiers. We temporarily drop sb->s_inode_list_lock and CAN block.
 42 */
 43static void fsnotify_unmount_inodes(struct super_block *sb)
 44{
 45	struct inode *inode, *iput_inode = NULL;
 46
 47	spin_lock(&sb->s_inode_list_lock);
 48	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
 49		/*
 50		 * We cannot __iget() an inode in state I_FREEING,
 51		 * I_WILL_FREE, or I_NEW which is fine because by that point
 52		 * the inode cannot have any associated watches.
 53		 */
 54		spin_lock(&inode->i_lock);
 55		if (inode_state_read(inode) & (I_FREEING | I_WILL_FREE | I_NEW)) {
 56			spin_unlock(&inode->i_lock);
 57			continue;
 58		}
 59
 60		/*
 61		 * If i_count is zero, the inode cannot have any watches and
 62		 * doing an __iget/iput with SB_ACTIVE clear would actually
 63		 * evict all inodes with zero i_count from icache which is
 64		 * unnecessarily violent and may in fact be illegal to do.
 65		 * However, we should have been called /after/ evict_inodes
 66		 * removed all zero refcount inodes, in any case.  Test to
 67		 * be sure.
 68		 */
 69		if (!icount_read(inode)) {
 70			spin_unlock(&inode->i_lock);
 71			continue;
 72		}
 73
 74		__iget(inode);
 75		spin_unlock(&inode->i_lock);
 76		spin_unlock(&sb->s_inode_list_lock);
 77
 78		iput(iput_inode);
 79
 80		/* for each watch, send FS_UNMOUNT and then remove it */
 81		fsnotify_inode(inode, FS_UNMOUNT);
 82
 83		fsnotify_inode_delete(inode);
 84
 85		iput_inode = inode;
 86
 87		cond_resched();
 88		spin_lock(&sb->s_inode_list_lock);
 89	}
 90	spin_unlock(&sb->s_inode_list_lock);
 91
 92	iput(iput_inode);
 93}
 94
 95void fsnotify_sb_delete(struct super_block *sb)
 96{
 97	struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
 98
 99	/* Were any marks ever added to any object on this sb? */
100	if (!sbinfo)
101		return;
102
103	fsnotify_unmount_inodes(sb);
104	fsnotify_clear_marks_by_sb(sb);
105	/* Wait for outstanding object references from connectors */
106	wait_var_event(fsnotify_sb_watched_objects(sb),
107		       !atomic_long_read(fsnotify_sb_watched_objects(sb)));
108	WARN_ON(fsnotify_sb_has_priority_watchers(sb, FSNOTIFY_PRIO_CONTENT));
109	WARN_ON(fsnotify_sb_has_priority_watchers(sb,
110						  FSNOTIFY_PRIO_PRE_CONTENT));
111}
112
113void fsnotify_sb_free(struct super_block *sb)
114{
115	kfree(sb->s_fsnotify_info);
116}
117
118/*
119 * Given an inode, first check if we care what happens to our children.  Inotify
120 * and dnotify both tell their parents about events.  If we care about any event
121 * on a child we run all of our children and set a dentry flag saying that the
122 * parent cares.  Thus when an event happens on a child it can quickly tell
123 * if there is a need to find a parent and send the event to the parent.
124 */
125void fsnotify_set_children_dentry_flags(struct inode *inode)
126{
127	struct dentry *alias;
128
129	if (!S_ISDIR(inode->i_mode))
130		return;
131
132	spin_lock(&inode->i_lock);
133	/* run all of the dentries associated with this inode.  Since this is a
134	 * directory, there damn well better only be one item on this list */
135	hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
136		struct dentry *child;
137
138		/* run all of the children of the original inode and fix their
139		 * d_flags to indicate parental interest (their parent is the
140		 * original inode) */
141		spin_lock(&alias->d_lock);
142		hlist_for_each_entry(child, &alias->d_children, d_sib) {
143			if (!child->d_inode)
144				continue;
145
146			spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
147			child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
148			spin_unlock(&child->d_lock);
149		}
150		spin_unlock(&alias->d_lock);
151	}
152	spin_unlock(&inode->i_lock);
153}
154
155/*
156 * Lazily clear false positive PARENT_WATCHED flag for child whose parent had
157 * stopped watching children.
158 */
159static void fsnotify_clear_child_dentry_flag(struct inode *pinode,
160					     struct dentry *dentry)
161{
162	spin_lock(&dentry->d_lock);
163	/*
164	 * d_lock is a sufficient barrier to prevent observing a non-watched
165	 * parent state from before the fsnotify_set_children_dentry_flags()
166	 * or fsnotify_update_flags() call that had set PARENT_WATCHED.
167	 */
168	if (!fsnotify_inode_watches_children(pinode))
169		dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
170	spin_unlock(&dentry->d_lock);
171}
172
173/* Are inode/sb/mount interested in parent and name info with this event? */
174static bool fsnotify_event_needs_parent(struct inode *inode, __u32 mnt_mask,
175					__u32 mask)
176{
177	__u32 marks_mask = 0;
178
179	/* We only send parent/name to inode/sb/mount for events on non-dir */
180	if (mask & FS_ISDIR)
181		return false;
182
183	/*
184	 * All events that are possible on child can also may be reported with
185	 * parent/name info to inode/sb/mount.  Otherwise, a watching parent
186	 * could result in events reported with unexpected name info to sb/mount.
187	 */
188	BUILD_BUG_ON(FS_EVENTS_POSS_ON_CHILD & ~FS_EVENTS_POSS_TO_PARENT);
189
190	/* Did either inode/sb/mount subscribe for events with parent/name? */
191	marks_mask |= fsnotify_parent_needed_mask(
192				READ_ONCE(inode->i_fsnotify_mask));
193	marks_mask |= fsnotify_parent_needed_mask(
194				READ_ONCE(inode->i_sb->s_fsnotify_mask));
195	marks_mask |= fsnotify_parent_needed_mask(mnt_mask);
196
197	/* Did they subscribe for this event with parent/name info? */
198	return mask & marks_mask;
199}
200
201/* Are there any inode/mount/sb objects that watch for these events? */
202static inline __u32 fsnotify_object_watched(struct inode *inode, __u32 mnt_mask,
203					    __u32 mask)
204{
205	__u32 marks_mask = READ_ONCE(inode->i_fsnotify_mask) | mnt_mask |
206			   READ_ONCE(inode->i_sb->s_fsnotify_mask);
207
208	return mask & marks_mask & ALL_FSNOTIFY_EVENTS;
209}
210
211/* Report pre-content event with optional range info */
212int fsnotify_pre_content(const struct path *path, const loff_t *ppos,
213			 size_t count)
214{
215	struct file_range range;
216
217	/* Report page aligned range only when pos is known */
218	if (!ppos)
219		return fsnotify_path(path, FS_PRE_ACCESS);
220
221	range.path = path;
222	range.pos = PAGE_ALIGN_DOWN(*ppos);
223	range.count = PAGE_ALIGN(*ppos + count) - range.pos;
224
225	return fsnotify_parent(path->dentry, FS_PRE_ACCESS, &range,
226			       FSNOTIFY_EVENT_FILE_RANGE);
227}
228
229/*
230 * Notify this dentry's parent about a child's events with child name info
231 * if parent is watching or if inode/sb/mount are interested in events with
232 * parent and name info.
233 *
234 * Notify only the child without name info if parent is not watching and
235 * inode/sb/mount are not interested in events with parent and name info.
236 */
237int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
238		      int data_type)
239{
240	const struct path *path = fsnotify_data_path(data, data_type);
241	__u32 mnt_mask = path ?
242		READ_ONCE(real_mount(path->mnt)->mnt_fsnotify_mask) : 0;
243	struct inode *inode = d_inode(dentry);
244	struct dentry *parent;
245	bool parent_watched = dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED;
246	bool parent_needed, parent_interested;
247	__u32 p_mask;
248	struct inode *p_inode = NULL;
249	struct name_snapshot name;
250	struct qstr *file_name = NULL;
251	int ret = 0;
252
253	/* Optimize the likely case of nobody watching this path */
254	if (likely(!parent_watched &&
255		   !fsnotify_object_watched(inode, mnt_mask, mask)))
256		return 0;
257
258	parent = NULL;
259	parent_needed = fsnotify_event_needs_parent(inode, mnt_mask, mask);
260	if (!parent_watched && !parent_needed)
261		goto notify;
262
263	/* Does parent inode care about events on children? */
264	parent = dget_parent(dentry);
265	p_inode = parent->d_inode;
266	p_mask = fsnotify_inode_watches_children(p_inode);
267	if (unlikely(parent_watched && !p_mask))
268		fsnotify_clear_child_dentry_flag(p_inode, dentry);
269
270	/*
271	 * Include parent/name in notification either if some notification
272	 * groups require parent info or the parent is interested in this event.
273	 * The parent interest in ACCESS/MODIFY events does not apply to special
274	 * files, where read/write are not on the filesystem of the parent and
275	 * events can provide an undesirable side-channel for information
276	 * exfiltration.
277	 */
278	parent_interested = mask & p_mask & ALL_FSNOTIFY_EVENTS &&
279			    !(data_type == FSNOTIFY_EVENT_PATH &&
280			      d_is_special(dentry) &&
281			      (mask & (FS_ACCESS | FS_MODIFY)));
282	if (parent_needed || parent_interested) {
283		/* When notifying parent, child should be passed as data */
284		WARN_ON_ONCE(inode != fsnotify_data_inode(data, data_type));
285
286		/* Notify both parent and child with child name info */
287		take_dentry_name_snapshot(&name, dentry);
288		file_name = &name.name;
289		if (parent_interested)
290			mask |= FS_EVENT_ON_CHILD;
291	}
292
293notify:
294	ret = fsnotify(mask, data, data_type, p_inode, file_name, inode, 0);
295
296	if (file_name)
297		release_dentry_name_snapshot(&name);
298	dput(parent);
299
300	return ret;
301}
302EXPORT_SYMBOL_GPL(__fsnotify_parent);
303
304static int fsnotify_handle_inode_event(struct fsnotify_group *group,
305				       struct fsnotify_mark *inode_mark,
306				       u32 mask, const void *data, int data_type,
307				       struct inode *dir, const struct qstr *name,
308				       u32 cookie)
309{
310	const struct path *path = fsnotify_data_path(data, data_type);
311	struct inode *inode = fsnotify_data_inode(data, data_type);
312	const struct fsnotify_ops *ops = group->ops;
313
314	if (WARN_ON_ONCE(!ops->handle_inode_event))
315		return 0;
316
317	if (WARN_ON_ONCE(!inode && !dir))
318		return 0;
319
320	if ((inode_mark->flags & FSNOTIFY_MARK_FLAG_EXCL_UNLINK) &&
321	    path && d_unlinked(path->dentry))
322		return 0;
323
324	/* Check interest of this mark in case event was sent with two marks */
325	if (!(mask & inode_mark->mask & ALL_FSNOTIFY_EVENTS))
326		return 0;
327
328	return ops->handle_inode_event(inode_mark, mask, inode, dir, name, cookie);
329}
330
331static int fsnotify_handle_event(struct fsnotify_group *group, __u32 mask,
332				 const void *data, int data_type,
333				 struct inode *dir, const struct qstr *name,
334				 u32 cookie, struct fsnotify_iter_info *iter_info)
335{
336	struct fsnotify_mark *inode_mark = fsnotify_iter_inode_mark(iter_info);
337	struct fsnotify_mark *parent_mark = fsnotify_iter_parent_mark(iter_info);
338	int ret;
339
340	if (WARN_ON_ONCE(fsnotify_iter_sb_mark(iter_info)) ||
341	    WARN_ON_ONCE(fsnotify_iter_vfsmount_mark(iter_info)))
342		return 0;
343
344	/*
345	 * For FS_RENAME, 'dir' is old dir and 'data' is new dentry.
346	 * The only ->handle_inode_event() backend that supports FS_RENAME is
347	 * dnotify, where it means file was renamed within same parent.
348	 */
349	if (mask & FS_RENAME) {
350		struct dentry *moved = fsnotify_data_dentry(data, data_type);
351
352		if (dir != moved->d_parent->d_inode)
353			return 0;
354	}
355
356	if (parent_mark) {
357		ret = fsnotify_handle_inode_event(group, parent_mark, mask,
358						  data, data_type, dir, name, 0);
359		if (ret)
360			return ret;
361	}
362
363	if (!inode_mark)
364		return 0;
365
366	/*
367	 * Some events can be sent on both parent dir and child marks (e.g.
368	 * FS_ATTRIB).  If both parent dir and child are watching, report the
369	 * event once to parent dir with name (if interested) and once to child
370	 * without name (if interested).
371	 *
372	 * In any case regardless whether the parent is watching or not, the
373	 * child watcher is expecting an event without the FS_EVENT_ON_CHILD
374	 * flag. The file name is expected if and only if this is a directory
375	 * event.
376	 */
377	mask &= ~FS_EVENT_ON_CHILD;
378	if (!(mask & ALL_FSNOTIFY_DIRENT_EVENTS)) {
379		dir = NULL;
380		name = NULL;
381	}
382
383	return fsnotify_handle_inode_event(group, inode_mark, mask, data, data_type,
384					   dir, name, cookie);
385}
386
387static int send_to_group(__u32 mask, const void *data, int data_type,
388			 struct inode *dir, const struct qstr *file_name,
389			 u32 cookie, struct fsnotify_iter_info *iter_info)
390{
391	struct fsnotify_group *group = NULL;
392	__u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
393	__u32 marks_mask = 0;
394	__u32 marks_ignore_mask = 0;
395	bool is_dir = mask & FS_ISDIR;
396	struct fsnotify_mark *mark;
397	int type;
398
399	if (!iter_info->report_mask)
400		return 0;
401
402	/* clear ignored on inode modification */
403	if (mask & FS_MODIFY) {
404		fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
405			if (!(mark->flags &
406			      FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
407				mark->ignore_mask = 0;
408		}
409	}
410
411	/* Are any of the group marks interested in this event? */
412	fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
413		group = mark->group;
414		marks_mask |= mark->mask;
415		marks_ignore_mask |=
416			fsnotify_effective_ignore_mask(mark, is_dir, type);
417	}
418
419	pr_debug("%s: group=%p mask=%x marks_mask=%x marks_ignore_mask=%x data=%p data_type=%d dir=%p cookie=%d\n",
420		 __func__, group, mask, marks_mask, marks_ignore_mask,
421		 data, data_type, dir, cookie);
422
423	if (!(test_mask & marks_mask & ~marks_ignore_mask))
424		return 0;
425
426	if (group->ops->handle_event) {
427		return group->ops->handle_event(group, mask, data, data_type, dir,
428						file_name, cookie, iter_info);
429	}
430
431	return fsnotify_handle_event(group, mask, data, data_type, dir,
432				     file_name, cookie, iter_info);
433}
434
435static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector *const *connp)
436{
437	struct fsnotify_mark_connector *conn;
438	struct hlist_node *node = NULL;
439
440	conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
441	if (conn)
442		node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
443
444	return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
445}
446
447static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
448{
449	struct hlist_node *node = NULL;
450
451	if (mark)
452		node = srcu_dereference(mark->obj_list.next,
453					&fsnotify_mark_srcu);
454
455	return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
456}
457
458/*
459 * iter_info is a multi head priority queue of marks.
460 * Pick a subset of marks from queue heads, all with the same group
461 * and set the report_mask to a subset of the selected marks.
462 * Returns false if there are no more groups to iterate.
463 */
464static bool fsnotify_iter_select_report_types(
465		struct fsnotify_iter_info *iter_info)
466{
467	struct fsnotify_group *max_prio_group = NULL;
468	struct fsnotify_mark *mark;
469	int type;
470
471	/* Choose max prio group among groups of all queue heads */
472	fsnotify_foreach_iter_type(type) {
473		mark = iter_info->marks[type];
474		if (mark &&
475		    fsnotify_compare_groups(max_prio_group, mark->group) > 0)
476			max_prio_group = mark->group;
477	}
478
479	if (!max_prio_group)
480		return false;
481
482	/* Set the report mask for marks from same group as max prio group */
483	iter_info->current_group = max_prio_group;
484	iter_info->report_mask = 0;
485	fsnotify_foreach_iter_type(type) {
486		mark = iter_info->marks[type];
487		if (mark && mark->group == iter_info->current_group) {
488			/*
489			 * FSNOTIFY_ITER_TYPE_PARENT indicates that this inode
490			 * is watching children and interested in this event,
491			 * which is an event possible on child.
492			 * But is *this mark* watching children?
493			 */
494			if (type == FSNOTIFY_ITER_TYPE_PARENT &&
495			    !(mark->mask & FS_EVENT_ON_CHILD) &&
496			    !(fsnotify_ignore_mask(mark) & FS_EVENT_ON_CHILD))
497				continue;
498
499			fsnotify_iter_set_report_type(iter_info, type);
500		}
501	}
502
503	return true;
504}
505
506/*
507 * Pop from iter_info multi head queue, the marks that belong to the group of
508 * current iteration step.
509 */
510static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
511{
512	struct fsnotify_mark *mark;
513	int type;
514
515	/*
516	 * We cannot use fsnotify_foreach_iter_mark_type() here because we
517	 * may need to advance a mark of type X that belongs to current_group
518	 * but was not selected for reporting.
519	 */
520	fsnotify_foreach_iter_type(type) {
521		mark = iter_info->marks[type];
522		if (mark && mark->group == iter_info->current_group)
523			iter_info->marks[type] =
524				fsnotify_next_mark(iter_info->marks[type]);
525	}
526}
527
528/*
529 * fsnotify - This is the main call to fsnotify.
530 *
531 * The VFS calls into hook specific functions in linux/fsnotify.h.
532 * Those functions then in turn call here.  Here will call out to all of the
533 * registered fsnotify_group.  Those groups can then use the notification event
534 * in whatever means they feel necessary.
535 *
536 * @mask:	event type and flags
537 * @data:	object that event happened on
538 * @data_type:	type of object for fanotify_data_XXX() accessors
539 * @dir:	optional directory associated with event -
540 *		if @file_name is not NULL, this is the directory that
541 *		@file_name is relative to
542 * @file_name:	optional file name associated with event
543 * @inode:	optional inode associated with event -
544 *		If @dir and @inode are both non-NULL, event may be
545 *		reported to both.
546 * @cookie:	inotify rename cookie
547 */
548int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
549	     const struct qstr *file_name, struct inode *inode, u32 cookie)
550{
551	const struct path *path = fsnotify_data_path(data, data_type);
552	struct super_block *sb = fsnotify_data_sb(data, data_type);
553	const struct fsnotify_mnt *mnt_data = fsnotify_data_mnt(data, data_type);
554	struct fsnotify_sb_info *sbinfo = sb ? fsnotify_sb_info(sb) : NULL;
555	struct fsnotify_iter_info iter_info = {};
556	struct mount *mnt = NULL;
557	struct inode *inode2 = NULL;
558	struct dentry *moved;
559	int inode2_type;
560	int ret = 0;
561	__u32 test_mask, marks_mask = 0;
562
563	if (path)
564		mnt = real_mount(path->mnt);
565
566	if (!inode) {
567		/* Dirent event - report on TYPE_INODE to dir */
568		inode = dir;
569		/* For FS_RENAME, inode is old_dir and inode2 is new_dir */
570		if (mask & FS_RENAME) {
571			moved = fsnotify_data_dentry(data, data_type);
572			inode2 = moved->d_parent->d_inode;
573			inode2_type = FSNOTIFY_ITER_TYPE_INODE2;
574		}
575	} else if (mask & FS_EVENT_ON_CHILD) {
576		/*
577		 * Event on child - report on TYPE_PARENT to dir if it is
578		 * watching children and on TYPE_INODE to child.
579		 */
580		inode2 = dir;
581		inode2_type = FSNOTIFY_ITER_TYPE_PARENT;
582	}
583
584	/*
585	 * Optimization: srcu_read_lock() has a memory barrier which can
586	 * be expensive.  It protects walking the *_fsnotify_marks lists.
587	 * However, if we do not walk the lists, we do not have to do
588	 * SRCU because we have no references to any objects and do not
589	 * need SRCU to keep them "alive".
590	 */
591	if ((!sbinfo || !sbinfo->sb_marks) &&
592	    (!mnt || !mnt->mnt_fsnotify_marks) &&
593	    (!inode || !inode->i_fsnotify_marks) &&
594	    (!inode2 || !inode2->i_fsnotify_marks) &&
595	    (!mnt_data || !mnt_data->ns->n_fsnotify_marks))
596		return 0;
597
598	if (sb)
599		marks_mask |= READ_ONCE(sb->s_fsnotify_mask);
600	if (mnt)
601		marks_mask |= READ_ONCE(mnt->mnt_fsnotify_mask);
602	if (inode)
603		marks_mask |= READ_ONCE(inode->i_fsnotify_mask);
604	if (inode2)
605		marks_mask |= READ_ONCE(inode2->i_fsnotify_mask);
606	if (mnt_data)
607		marks_mask |= READ_ONCE(mnt_data->ns->n_fsnotify_mask);
608
609	/*
610	 * If this is a modify event we may need to clear some ignore masks.
611	 * In that case, the object with ignore masks will have the FS_MODIFY
612	 * event in its mask.
613	 * Otherwise, return if none of the marks care about this type of event.
614	 */
615	test_mask = (mask & ALL_FSNOTIFY_EVENTS);
616	if (!(test_mask & marks_mask))
617		return 0;
618
619	iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
620
621	if (sbinfo) {
622		iter_info.marks[FSNOTIFY_ITER_TYPE_SB] =
623			fsnotify_first_mark(&sbinfo->sb_marks);
624	}
625	if (mnt) {
626		iter_info.marks[FSNOTIFY_ITER_TYPE_VFSMOUNT] =
627			fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
628	}
629	if (inode) {
630		iter_info.marks[FSNOTIFY_ITER_TYPE_INODE] =
631			fsnotify_first_mark(&inode->i_fsnotify_marks);
632	}
633	if (inode2) {
634		iter_info.marks[inode2_type] =
635			fsnotify_first_mark(&inode2->i_fsnotify_marks);
636	}
637	if (mnt_data) {
638		iter_info.marks[FSNOTIFY_ITER_TYPE_MNTNS] =
639			fsnotify_first_mark(&mnt_data->ns->n_fsnotify_marks);
640	}
641
642	/*
643	 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
644	 * ignore masks are properly reflected for mount/sb mark notifications.
645	 * That's why this traversal is so complicated...
646	 */
647	while (fsnotify_iter_select_report_types(&iter_info)) {
648		ret = send_to_group(mask, data, data_type, dir, file_name,
649				    cookie, &iter_info);
650
651		if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
652			goto out;
653
654		fsnotify_iter_next(&iter_info);
655	}
656	ret = 0;
657out:
658	srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
659
660	return ret;
661}
662EXPORT_SYMBOL_GPL(fsnotify);
663
664#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
665/*
666 * At open time we check fsnotify_sb_has_priority_watchers(), call the open perm
667 * hook and set the FMODE_NONOTIFY_ mode bits accordignly.
668 * Later, fsnotify permission hooks do not check if there are permission event
669 * watches, but that there were permission event watches at open time.
670 */
671int fsnotify_open_perm_and_set_mode(struct file *file)
672{
673	struct dentry *dentry = file->f_path.dentry, *parent;
674	struct super_block *sb = dentry->d_sb;
675	__u32 mnt_mask, p_mask = 0;
676
677	/* Is it a file opened by fanotify? */
678	if (FMODE_FSNOTIFY_NONE(file->f_mode))
679		return 0;
680
681	/*
682	 * Permission events is a super set of pre-content events, so if there
683	 * are no permission event watchers, there are also no pre-content event
684	 * watchers and this is implied from the single FMODE_NONOTIFY_PERM bit.
685	 */
686	if (likely(!fsnotify_sb_has_priority_watchers(sb,
687						FSNOTIFY_PRIO_CONTENT))) {
688		file_set_fsnotify_mode(file, FMODE_NONOTIFY_PERM);
689		return 0;
690	}
691
692	/*
693	 * OK, there are some permission event watchers. Check if anybody is
694	 * watching for permission events on *this* file.
695	 */
696	mnt_mask = READ_ONCE(real_mount(file->f_path.mnt)->mnt_fsnotify_mask);
697	p_mask = fsnotify_object_watched(d_inode(dentry), mnt_mask,
698					 ALL_FSNOTIFY_PERM_EVENTS);
699	if (dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED) {
700		parent = dget_parent(dentry);
701		p_mask |= fsnotify_inode_watches_children(d_inode(parent));
702		dput(parent);
703	}
704
705	/*
706	 * Legacy FAN_ACCESS_PERM events have very high performance overhead,
707	 * so unlikely to be used in the wild. If they are used there will be
708	 * no optimizations at all.
709	 */
710	if (unlikely(p_mask & FS_ACCESS_PERM)) {
711		/* Enable all permission and pre-content events */
712		file_set_fsnotify_mode(file, 0);
713		goto open_perm;
714	}
715
716	/*
717	 * Pre-content events are only supported on regular files.
718	 * If there are pre-content event watchers and no permission access
719	 * watchers, set FMODE_NONOTIFY | FMODE_NONOTIFY_PERM to indicate that.
720	 * That is the common case with HSM service.
721	 */
722	if (d_is_reg(dentry) && (p_mask & FSNOTIFY_PRE_CONTENT_EVENTS)) {
723		file_set_fsnotify_mode(file, FMODE_NONOTIFY |
724					     FMODE_NONOTIFY_PERM);
725		goto open_perm;
726	}
727
728	/* Nobody watching permission and pre-content events on this file */
729	file_set_fsnotify_mode(file, FMODE_NONOTIFY_PERM);
730
731open_perm:
732	/*
733	 * Send open perm events depending on object masks and regardless of
734	 * FMODE_NONOTIFY_PERM.
735	 */
736	if (file->f_flags & __FMODE_EXEC && p_mask & FS_OPEN_EXEC_PERM) {
737		int ret = fsnotify_path(&file->f_path, FS_OPEN_EXEC_PERM);
738
739		if (ret)
740			return ret;
741	}
742
743	if (p_mask & FS_OPEN_PERM)
744		return fsnotify_path(&file->f_path, FS_OPEN_PERM);
745
746	return 0;
747}
748#endif
749
750void fsnotify_mnt(__u32 mask, struct mnt_namespace *ns, struct vfsmount *mnt)
751{
752	struct fsnotify_mnt data = {
753		.ns = ns,
754		.mnt_id = real_mount(mnt)->mnt_id_unique,
755	};
756
757	if (WARN_ON_ONCE(!ns))
758		return;
759
760	/*
761	 * This is an optimization as well as making sure fsnotify_init() has
762	 * been called.
763	 */
764	if (!ns->n_fsnotify_marks)
765		return;
766
767	fsnotify(mask, &data, FSNOTIFY_EVENT_MNT, NULL, NULL, NULL, 0);
768}
769
770static __init int fsnotify_init(void)
771{
772	int ret;
773
774	BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 26);
775
776	ret = init_srcu_struct(&fsnotify_mark_srcu);
777	if (ret)
778		panic("initializing fsnotify_mark_srcu");
779
780	fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
781						    SLAB_PANIC);
782
783	return 0;
784}
785core_initcall(fsnotify_init);