xref: /linux/fs/notify/fsnotify.c (revision ef9226cd56b718c79184a3466d32984a51cb449c)
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 /**
32  * fsnotify_unmount_inodes - an sb is unmounting.  handle any watched inodes.
33  * @sb: superblock being unmounted.
34  *
35  * Called during unmount with no locks held, so needs to be safe against
36  * concurrent modifiers. We temporarily drop sb->s_inode_list_lock and CAN block.
37  */
38 static void fsnotify_unmount_inodes(struct super_block *sb)
39 {
40 	struct inode *inode, *iput_inode = NULL;
41 
42 	spin_lock(&sb->s_inode_list_lock);
43 	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
44 		/*
45 		 * We cannot __iget() an inode in state I_FREEING,
46 		 * I_WILL_FREE, or I_NEW which is fine because by that point
47 		 * the inode cannot have any associated watches.
48 		 */
49 		spin_lock(&inode->i_lock);
50 		if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) {
51 			spin_unlock(&inode->i_lock);
52 			continue;
53 		}
54 
55 		/*
56 		 * If i_count is zero, the inode cannot have any watches and
57 		 * doing an __iget/iput with SB_ACTIVE clear would actually
58 		 * evict all inodes with zero i_count from icache which is
59 		 * unnecessarily violent and may in fact be illegal to do.
60 		 * However, we should have been called /after/ evict_inodes
61 		 * removed all zero refcount inodes, in any case.  Test to
62 		 * be sure.
63 		 */
64 		if (!atomic_read(&inode->i_count)) {
65 			spin_unlock(&inode->i_lock);
66 			continue;
67 		}
68 
69 		__iget(inode);
70 		spin_unlock(&inode->i_lock);
71 		spin_unlock(&sb->s_inode_list_lock);
72 
73 		iput(iput_inode);
74 
75 		/* for each watch, send FS_UNMOUNT and then remove it */
76 		fsnotify_inode(inode, FS_UNMOUNT);
77 
78 		fsnotify_inode_delete(inode);
79 
80 		iput_inode = inode;
81 
82 		cond_resched();
83 		spin_lock(&sb->s_inode_list_lock);
84 	}
85 	spin_unlock(&sb->s_inode_list_lock);
86 
87 	iput(iput_inode);
88 }
89 
90 void fsnotify_sb_delete(struct super_block *sb)
91 {
92 	fsnotify_unmount_inodes(sb);
93 	fsnotify_clear_marks_by_sb(sb);
94 	/* Wait for outstanding object references from connectors */
95 	wait_var_event(&sb->s_fsnotify_connectors,
96 		       !atomic_long_read(&sb->s_fsnotify_connectors));
97 }
98 
99 /*
100  * Given an inode, first check if we care what happens to our children.  Inotify
101  * and dnotify both tell their parents about events.  If we care about any event
102  * on a child we run all of our children and set a dentry flag saying that the
103  * parent cares.  Thus when an event happens on a child it can quickly tell
104  * if there is a need to find a parent and send the event to the parent.
105  */
106 void __fsnotify_update_child_dentry_flags(struct inode *inode)
107 {
108 	struct dentry *alias;
109 	int watched;
110 
111 	if (!S_ISDIR(inode->i_mode))
112 		return;
113 
114 	/* determine if the children should tell inode about their events */
115 	watched = fsnotify_inode_watches_children(inode);
116 
117 	spin_lock(&inode->i_lock);
118 	/* run all of the dentries associated with this inode.  Since this is a
119 	 * directory, there damn well better only be one item on this list */
120 	hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
121 		struct dentry *child;
122 
123 		/* run all of the children of the original inode and fix their
124 		 * d_flags to indicate parental interest (their parent is the
125 		 * original inode) */
126 		spin_lock(&alias->d_lock);
127 		hlist_for_each_entry(child, &alias->d_children, d_sib) {
128 			if (!child->d_inode)
129 				continue;
130 
131 			spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
132 			if (watched)
133 				child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
134 			else
135 				child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
136 			spin_unlock(&child->d_lock);
137 		}
138 		spin_unlock(&alias->d_lock);
139 	}
140 	spin_unlock(&inode->i_lock);
141 }
142 
143 /* Are inode/sb/mount interested in parent and name info with this event? */
144 static bool fsnotify_event_needs_parent(struct inode *inode, __u32 mnt_mask,
145 					__u32 mask)
146 {
147 	__u32 marks_mask = 0;
148 
149 	/* We only send parent/name to inode/sb/mount for events on non-dir */
150 	if (mask & FS_ISDIR)
151 		return false;
152 
153 	/*
154 	 * All events that are possible on child can also may be reported with
155 	 * parent/name info to inode/sb/mount.  Otherwise, a watching parent
156 	 * could result in events reported with unexpected name info to sb/mount.
157 	 */
158 	BUILD_BUG_ON(FS_EVENTS_POSS_ON_CHILD & ~FS_EVENTS_POSS_TO_PARENT);
159 
160 	/* Did either inode/sb/mount subscribe for events with parent/name? */
161 	marks_mask |= fsnotify_parent_needed_mask(inode->i_fsnotify_mask);
162 	marks_mask |= fsnotify_parent_needed_mask(inode->i_sb->s_fsnotify_mask);
163 	marks_mask |= fsnotify_parent_needed_mask(mnt_mask);
164 
165 	/* Did they subscribe for this event with parent/name info? */
166 	return mask & marks_mask;
167 }
168 
169 /* Are there any inode/mount/sb objects that are interested in this event? */
170 static inline bool fsnotify_object_watched(struct inode *inode, __u32 mnt_mask,
171 					   __u32 mask)
172 {
173 	__u32 marks_mask = inode->i_fsnotify_mask | mnt_mask |
174 			   inode->i_sb->s_fsnotify_mask;
175 
176 	return mask & marks_mask & ALL_FSNOTIFY_EVENTS;
177 }
178 
179 /*
180  * Notify this dentry's parent about a child's events with child name info
181  * if parent is watching or if inode/sb/mount are interested in events with
182  * parent and name info.
183  *
184  * Notify only the child without name info if parent is not watching and
185  * inode/sb/mount are not interested in events with parent and name info.
186  */
187 int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
188 		      int data_type)
189 {
190 	const struct path *path = fsnotify_data_path(data, data_type);
191 	__u32 mnt_mask = path ? real_mount(path->mnt)->mnt_fsnotify_mask : 0;
192 	struct inode *inode = d_inode(dentry);
193 	struct dentry *parent;
194 	bool parent_watched = dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED;
195 	bool parent_needed, parent_interested;
196 	__u32 p_mask;
197 	struct inode *p_inode = NULL;
198 	struct name_snapshot name;
199 	struct qstr *file_name = NULL;
200 	int ret = 0;
201 
202 	/* Optimize the likely case of nobody watching this path */
203 	if (likely(!parent_watched &&
204 		   !fsnotify_object_watched(inode, mnt_mask, mask)))
205 		return 0;
206 
207 	parent = NULL;
208 	parent_needed = fsnotify_event_needs_parent(inode, mnt_mask, mask);
209 	if (!parent_watched && !parent_needed)
210 		goto notify;
211 
212 	/* Does parent inode care about events on children? */
213 	parent = dget_parent(dentry);
214 	p_inode = parent->d_inode;
215 	p_mask = fsnotify_inode_watches_children(p_inode);
216 	if (unlikely(parent_watched && !p_mask))
217 		__fsnotify_update_child_dentry_flags(p_inode);
218 
219 	/*
220 	 * Include parent/name in notification either if some notification
221 	 * groups require parent info or the parent is interested in this event.
222 	 */
223 	parent_interested = mask & p_mask & ALL_FSNOTIFY_EVENTS;
224 	if (parent_needed || parent_interested) {
225 		/* When notifying parent, child should be passed as data */
226 		WARN_ON_ONCE(inode != fsnotify_data_inode(data, data_type));
227 
228 		/* Notify both parent and child with child name info */
229 		take_dentry_name_snapshot(&name, dentry);
230 		file_name = &name.name;
231 		if (parent_interested)
232 			mask |= FS_EVENT_ON_CHILD;
233 	}
234 
235 notify:
236 	ret = fsnotify(mask, data, data_type, p_inode, file_name, inode, 0);
237 
238 	if (file_name)
239 		release_dentry_name_snapshot(&name);
240 	dput(parent);
241 
242 	return ret;
243 }
244 EXPORT_SYMBOL_GPL(__fsnotify_parent);
245 
246 static int fsnotify_handle_inode_event(struct fsnotify_group *group,
247 				       struct fsnotify_mark *inode_mark,
248 				       u32 mask, const void *data, int data_type,
249 				       struct inode *dir, const struct qstr *name,
250 				       u32 cookie)
251 {
252 	const struct path *path = fsnotify_data_path(data, data_type);
253 	struct inode *inode = fsnotify_data_inode(data, data_type);
254 	const struct fsnotify_ops *ops = group->ops;
255 
256 	if (WARN_ON_ONCE(!ops->handle_inode_event))
257 		return 0;
258 
259 	if (WARN_ON_ONCE(!inode && !dir))
260 		return 0;
261 
262 	if ((inode_mark->flags & FSNOTIFY_MARK_FLAG_EXCL_UNLINK) &&
263 	    path && d_unlinked(path->dentry))
264 		return 0;
265 
266 	/* Check interest of this mark in case event was sent with two marks */
267 	if (!(mask & inode_mark->mask & ALL_FSNOTIFY_EVENTS))
268 		return 0;
269 
270 	return ops->handle_inode_event(inode_mark, mask, inode, dir, name, cookie);
271 }
272 
273 static int fsnotify_handle_event(struct fsnotify_group *group, __u32 mask,
274 				 const void *data, int data_type,
275 				 struct inode *dir, const struct qstr *name,
276 				 u32 cookie, struct fsnotify_iter_info *iter_info)
277 {
278 	struct fsnotify_mark *inode_mark = fsnotify_iter_inode_mark(iter_info);
279 	struct fsnotify_mark *parent_mark = fsnotify_iter_parent_mark(iter_info);
280 	int ret;
281 
282 	if (WARN_ON_ONCE(fsnotify_iter_sb_mark(iter_info)) ||
283 	    WARN_ON_ONCE(fsnotify_iter_vfsmount_mark(iter_info)))
284 		return 0;
285 
286 	/*
287 	 * For FS_RENAME, 'dir' is old dir and 'data' is new dentry.
288 	 * The only ->handle_inode_event() backend that supports FS_RENAME is
289 	 * dnotify, where it means file was renamed within same parent.
290 	 */
291 	if (mask & FS_RENAME) {
292 		struct dentry *moved = fsnotify_data_dentry(data, data_type);
293 
294 		if (dir != moved->d_parent->d_inode)
295 			return 0;
296 	}
297 
298 	if (parent_mark) {
299 		ret = fsnotify_handle_inode_event(group, parent_mark, mask,
300 						  data, data_type, dir, name, 0);
301 		if (ret)
302 			return ret;
303 	}
304 
305 	if (!inode_mark)
306 		return 0;
307 
308 	if (mask & FS_EVENT_ON_CHILD) {
309 		/*
310 		 * Some events can be sent on both parent dir and child marks
311 		 * (e.g. FS_ATTRIB).  If both parent dir and child are
312 		 * watching, report the event once to parent dir with name (if
313 		 * interested) and once to child without name (if interested).
314 		 * The child watcher is expecting an event without a file name
315 		 * and without the FS_EVENT_ON_CHILD flag.
316 		 */
317 		mask &= ~FS_EVENT_ON_CHILD;
318 		dir = NULL;
319 		name = NULL;
320 	}
321 
322 	return fsnotify_handle_inode_event(group, inode_mark, mask, data, data_type,
323 					   dir, name, cookie);
324 }
325 
326 static int send_to_group(__u32 mask, const void *data, int data_type,
327 			 struct inode *dir, const struct qstr *file_name,
328 			 u32 cookie, struct fsnotify_iter_info *iter_info)
329 {
330 	struct fsnotify_group *group = NULL;
331 	__u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
332 	__u32 marks_mask = 0;
333 	__u32 marks_ignore_mask = 0;
334 	bool is_dir = mask & FS_ISDIR;
335 	struct fsnotify_mark *mark;
336 	int type;
337 
338 	if (!iter_info->report_mask)
339 		return 0;
340 
341 	/* clear ignored on inode modification */
342 	if (mask & FS_MODIFY) {
343 		fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
344 			if (!(mark->flags &
345 			      FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
346 				mark->ignore_mask = 0;
347 		}
348 	}
349 
350 	/* Are any of the group marks interested in this event? */
351 	fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
352 		group = mark->group;
353 		marks_mask |= mark->mask;
354 		marks_ignore_mask |=
355 			fsnotify_effective_ignore_mask(mark, is_dir, type);
356 	}
357 
358 	pr_debug("%s: group=%p mask=%x marks_mask=%x marks_ignore_mask=%x data=%p data_type=%d dir=%p cookie=%d\n",
359 		 __func__, group, mask, marks_mask, marks_ignore_mask,
360 		 data, data_type, dir, cookie);
361 
362 	if (!(test_mask & marks_mask & ~marks_ignore_mask))
363 		return 0;
364 
365 	if (group->ops->handle_event) {
366 		return group->ops->handle_event(group, mask, data, data_type, dir,
367 						file_name, cookie, iter_info);
368 	}
369 
370 	return fsnotify_handle_event(group, mask, data, data_type, dir,
371 				     file_name, cookie, iter_info);
372 }
373 
374 static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
375 {
376 	struct fsnotify_mark_connector *conn;
377 	struct hlist_node *node = NULL;
378 
379 	conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
380 	if (conn)
381 		node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
382 
383 	return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
384 }
385 
386 static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
387 {
388 	struct hlist_node *node = NULL;
389 
390 	if (mark)
391 		node = srcu_dereference(mark->obj_list.next,
392 					&fsnotify_mark_srcu);
393 
394 	return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
395 }
396 
397 /*
398  * iter_info is a multi head priority queue of marks.
399  * Pick a subset of marks from queue heads, all with the same group
400  * and set the report_mask to a subset of the selected marks.
401  * Returns false if there are no more groups to iterate.
402  */
403 static bool fsnotify_iter_select_report_types(
404 		struct fsnotify_iter_info *iter_info)
405 {
406 	struct fsnotify_group *max_prio_group = NULL;
407 	struct fsnotify_mark *mark;
408 	int type;
409 
410 	/* Choose max prio group among groups of all queue heads */
411 	fsnotify_foreach_iter_type(type) {
412 		mark = iter_info->marks[type];
413 		if (mark &&
414 		    fsnotify_compare_groups(max_prio_group, mark->group) > 0)
415 			max_prio_group = mark->group;
416 	}
417 
418 	if (!max_prio_group)
419 		return false;
420 
421 	/* Set the report mask for marks from same group as max prio group */
422 	iter_info->current_group = max_prio_group;
423 	iter_info->report_mask = 0;
424 	fsnotify_foreach_iter_type(type) {
425 		mark = iter_info->marks[type];
426 		if (mark && mark->group == iter_info->current_group) {
427 			/*
428 			 * FSNOTIFY_ITER_TYPE_PARENT indicates that this inode
429 			 * is watching children and interested in this event,
430 			 * which is an event possible on child.
431 			 * But is *this mark* watching children?
432 			 */
433 			if (type == FSNOTIFY_ITER_TYPE_PARENT &&
434 			    !(mark->mask & FS_EVENT_ON_CHILD) &&
435 			    !(fsnotify_ignore_mask(mark) & FS_EVENT_ON_CHILD))
436 				continue;
437 
438 			fsnotify_iter_set_report_type(iter_info, type);
439 		}
440 	}
441 
442 	return true;
443 }
444 
445 /*
446  * Pop from iter_info multi head queue, the marks that belong to the group of
447  * current iteration step.
448  */
449 static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
450 {
451 	struct fsnotify_mark *mark;
452 	int type;
453 
454 	/*
455 	 * We cannot use fsnotify_foreach_iter_mark_type() here because we
456 	 * may need to advance a mark of type X that belongs to current_group
457 	 * but was not selected for reporting.
458 	 */
459 	fsnotify_foreach_iter_type(type) {
460 		mark = iter_info->marks[type];
461 		if (mark && mark->group == iter_info->current_group)
462 			iter_info->marks[type] =
463 				fsnotify_next_mark(iter_info->marks[type]);
464 	}
465 }
466 
467 /*
468  * fsnotify - This is the main call to fsnotify.
469  *
470  * The VFS calls into hook specific functions in linux/fsnotify.h.
471  * Those functions then in turn call here.  Here will call out to all of the
472  * registered fsnotify_group.  Those groups can then use the notification event
473  * in whatever means they feel necessary.
474  *
475  * @mask:	event type and flags
476  * @data:	object that event happened on
477  * @data_type:	type of object for fanotify_data_XXX() accessors
478  * @dir:	optional directory associated with event -
479  *		if @file_name is not NULL, this is the directory that
480  *		@file_name is relative to
481  * @file_name:	optional file name associated with event
482  * @inode:	optional inode associated with event -
483  *		If @dir and @inode are both non-NULL, event may be
484  *		reported to both.
485  * @cookie:	inotify rename cookie
486  */
487 int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
488 	     const struct qstr *file_name, struct inode *inode, u32 cookie)
489 {
490 	const struct path *path = fsnotify_data_path(data, data_type);
491 	struct super_block *sb = fsnotify_data_sb(data, data_type);
492 	struct fsnotify_iter_info iter_info = {};
493 	struct mount *mnt = NULL;
494 	struct inode *inode2 = NULL;
495 	struct dentry *moved;
496 	int inode2_type;
497 	int ret = 0;
498 	__u32 test_mask, marks_mask;
499 
500 	if (path)
501 		mnt = real_mount(path->mnt);
502 
503 	if (!inode) {
504 		/* Dirent event - report on TYPE_INODE to dir */
505 		inode = dir;
506 		/* For FS_RENAME, inode is old_dir and inode2 is new_dir */
507 		if (mask & FS_RENAME) {
508 			moved = fsnotify_data_dentry(data, data_type);
509 			inode2 = moved->d_parent->d_inode;
510 			inode2_type = FSNOTIFY_ITER_TYPE_INODE2;
511 		}
512 	} else if (mask & FS_EVENT_ON_CHILD) {
513 		/*
514 		 * Event on child - report on TYPE_PARENT to dir if it is
515 		 * watching children and on TYPE_INODE to child.
516 		 */
517 		inode2 = dir;
518 		inode2_type = FSNOTIFY_ITER_TYPE_PARENT;
519 	}
520 
521 	/*
522 	 * Optimization: srcu_read_lock() has a memory barrier which can
523 	 * be expensive.  It protects walking the *_fsnotify_marks lists.
524 	 * However, if we do not walk the lists, we do not have to do
525 	 * SRCU because we have no references to any objects and do not
526 	 * need SRCU to keep them "alive".
527 	 */
528 	if (!sb->s_fsnotify_marks &&
529 	    (!mnt || !mnt->mnt_fsnotify_marks) &&
530 	    (!inode || !inode->i_fsnotify_marks) &&
531 	    (!inode2 || !inode2->i_fsnotify_marks))
532 		return 0;
533 
534 	marks_mask = sb->s_fsnotify_mask;
535 	if (mnt)
536 		marks_mask |= mnt->mnt_fsnotify_mask;
537 	if (inode)
538 		marks_mask |= inode->i_fsnotify_mask;
539 	if (inode2)
540 		marks_mask |= inode2->i_fsnotify_mask;
541 
542 
543 	/*
544 	 * If this is a modify event we may need to clear some ignore masks.
545 	 * In that case, the object with ignore masks will have the FS_MODIFY
546 	 * event in its mask.
547 	 * Otherwise, return if none of the marks care about this type of event.
548 	 */
549 	test_mask = (mask & ALL_FSNOTIFY_EVENTS);
550 	if (!(test_mask & marks_mask))
551 		return 0;
552 
553 	iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
554 
555 	iter_info.marks[FSNOTIFY_ITER_TYPE_SB] =
556 		fsnotify_first_mark(&sb->s_fsnotify_marks);
557 	if (mnt) {
558 		iter_info.marks[FSNOTIFY_ITER_TYPE_VFSMOUNT] =
559 			fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
560 	}
561 	if (inode) {
562 		iter_info.marks[FSNOTIFY_ITER_TYPE_INODE] =
563 			fsnotify_first_mark(&inode->i_fsnotify_marks);
564 	}
565 	if (inode2) {
566 		iter_info.marks[inode2_type] =
567 			fsnotify_first_mark(&inode2->i_fsnotify_marks);
568 	}
569 
570 	/*
571 	 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
572 	 * ignore masks are properly reflected for mount/sb mark notifications.
573 	 * That's why this traversal is so complicated...
574 	 */
575 	while (fsnotify_iter_select_report_types(&iter_info)) {
576 		ret = send_to_group(mask, data, data_type, dir, file_name,
577 				    cookie, &iter_info);
578 
579 		if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
580 			goto out;
581 
582 		fsnotify_iter_next(&iter_info);
583 	}
584 	ret = 0;
585 out:
586 	srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
587 
588 	return ret;
589 }
590 EXPORT_SYMBOL_GPL(fsnotify);
591 
592 static __init int fsnotify_init(void)
593 {
594 	int ret;
595 
596 	BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 23);
597 
598 	ret = init_srcu_struct(&fsnotify_mark_srcu);
599 	if (ret)
600 		panic("initializing fsnotify_mark_srcu");
601 
602 	fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
603 						    SLAB_PANIC);
604 
605 	return 0;
606 }
607 core_initcall(fsnotify_init);
608