1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
6 #include <linux/dcache.h>
9 #include <linux/init.h>
10 #include <linux/module.h>
11 #include <linux/mount.h>
12 #include <linux/srcu.h>
14 #include <linux/fsnotify_backend.h>
18 * Clear all of the marks on an inode when it is being evicted from core
20 void __fsnotify_inode_delete(struct inode *inode)
22 fsnotify_clear_marks_by_inode(inode);
24 EXPORT_SYMBOL_GPL(__fsnotify_inode_delete);
26 void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
28 fsnotify_clear_marks_by_mount(mnt);
32 * fsnotify_unmount_inodes - an sb is unmounting. handle any watched inodes.
33 * @sb: superblock being unmounted.
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.
38 static void fsnotify_unmount_inodes(struct super_block *sb)
40 struct inode *inode, *iput_inode = NULL;
42 spin_lock(&sb->s_inode_list_lock);
43 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
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.
49 spin_lock(&inode->i_lock);
50 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) {
51 spin_unlock(&inode->i_lock);
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
64 if (!atomic_read(&inode->i_count)) {
65 spin_unlock(&inode->i_lock);
70 spin_unlock(&inode->i_lock);
71 spin_unlock(&sb->s_inode_list_lock);
75 /* for each watch, send FS_UNMOUNT and then remove it */
76 fsnotify_inode(inode, FS_UNMOUNT);
78 fsnotify_inode_delete(inode);
83 spin_lock(&sb->s_inode_list_lock);
85 spin_unlock(&sb->s_inode_list_lock);
90 void fsnotify_sb_delete(struct super_block *sb)
92 struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
94 /* Were any marks ever added to any object on this sb? */
98 fsnotify_unmount_inodes(sb);
99 fsnotify_clear_marks_by_sb(sb);
100 /* Wait for outstanding object references from connectors */
101 wait_var_event(fsnotify_sb_watched_objects(sb),
102 !atomic_long_read(fsnotify_sb_watched_objects(sb)));
103 WARN_ON(fsnotify_sb_has_priority_watchers(sb, FSNOTIFY_PRIO_CONTENT));
104 WARN_ON(fsnotify_sb_has_priority_watchers(sb,
105 FSNOTIFY_PRIO_PRE_CONTENT));
108 void fsnotify_sb_free(struct super_block *sb)
110 kfree(sb->s_fsnotify_info);
114 * Given an inode, first check if we care what happens to our children. Inotify
115 * and dnotify both tell their parents about events. If we care about any event
116 * on a child we run all of our children and set a dentry flag saying that the
117 * parent cares. Thus when an event happens on a child it can quickly tell
118 * if there is a need to find a parent and send the event to the parent.
120 void __fsnotify_update_child_dentry_flags(struct inode *inode)
122 struct dentry *alias;
125 if (!S_ISDIR(inode->i_mode))
128 /* determine if the children should tell inode about their events */
129 watched = fsnotify_inode_watches_children(inode);
131 spin_lock(&inode->i_lock);
132 /* run all of the dentries associated with this inode. Since this is a
133 * directory, there damn well better only be one item on this list */
134 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
135 struct dentry *child;
137 /* run all of the children of the original inode and fix their
138 * d_flags to indicate parental interest (their parent is the
140 spin_lock(&alias->d_lock);
141 hlist_for_each_entry(child, &alias->d_children, d_sib) {
145 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
147 child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
149 child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
150 spin_unlock(&child->d_lock);
152 spin_unlock(&alias->d_lock);
154 spin_unlock(&inode->i_lock);
157 /* Are inode/sb/mount interested in parent and name info with this event? */
158 static bool fsnotify_event_needs_parent(struct inode *inode, __u32 mnt_mask,
161 __u32 marks_mask = 0;
163 /* We only send parent/name to inode/sb/mount for events on non-dir */
168 * All events that are possible on child can also may be reported with
169 * parent/name info to inode/sb/mount. Otherwise, a watching parent
170 * could result in events reported with unexpected name info to sb/mount.
172 BUILD_BUG_ON(FS_EVENTS_POSS_ON_CHILD & ~FS_EVENTS_POSS_TO_PARENT);
174 /* Did either inode/sb/mount subscribe for events with parent/name? */
175 marks_mask |= fsnotify_parent_needed_mask(inode->i_fsnotify_mask);
176 marks_mask |= fsnotify_parent_needed_mask(inode->i_sb->s_fsnotify_mask);
177 marks_mask |= fsnotify_parent_needed_mask(mnt_mask);
179 /* Did they subscribe for this event with parent/name info? */
180 return mask & marks_mask;
183 /* Are there any inode/mount/sb objects that are interested in this event? */
184 static inline bool fsnotify_object_watched(struct inode *inode, __u32 mnt_mask,
187 __u32 marks_mask = inode->i_fsnotify_mask | mnt_mask |
188 inode->i_sb->s_fsnotify_mask;
190 return mask & marks_mask & ALL_FSNOTIFY_EVENTS;
194 * Notify this dentry's parent about a child's events with child name info
195 * if parent is watching or if inode/sb/mount are interested in events with
196 * parent and name info.
198 * Notify only the child without name info if parent is not watching and
199 * inode/sb/mount are not interested in events with parent and name info.
201 int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
204 const struct path *path = fsnotify_data_path(data, data_type);
205 __u32 mnt_mask = path ? real_mount(path->mnt)->mnt_fsnotify_mask : 0;
206 struct inode *inode = d_inode(dentry);
207 struct dentry *parent;
208 bool parent_watched = dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED;
209 bool parent_needed, parent_interested;
211 struct inode *p_inode = NULL;
212 struct name_snapshot name;
213 struct qstr *file_name = NULL;
216 /* Optimize the likely case of nobody watching this path */
217 if (likely(!parent_watched &&
218 !fsnotify_object_watched(inode, mnt_mask, mask)))
222 parent_needed = fsnotify_event_needs_parent(inode, mnt_mask, mask);
223 if (!parent_watched && !parent_needed)
226 /* Does parent inode care about events on children? */
227 parent = dget_parent(dentry);
228 p_inode = parent->d_inode;
229 p_mask = fsnotify_inode_watches_children(p_inode);
230 if (unlikely(parent_watched && !p_mask))
231 __fsnotify_update_child_dentry_flags(p_inode);
234 * Include parent/name in notification either if some notification
235 * groups require parent info or the parent is interested in this event.
237 parent_interested = mask & p_mask & ALL_FSNOTIFY_EVENTS;
238 if (parent_needed || parent_interested) {
239 /* When notifying parent, child should be passed as data */
240 WARN_ON_ONCE(inode != fsnotify_data_inode(data, data_type));
242 /* Notify both parent and child with child name info */
243 take_dentry_name_snapshot(&name, dentry);
244 file_name = &name.name;
245 if (parent_interested)
246 mask |= FS_EVENT_ON_CHILD;
250 ret = fsnotify(mask, data, data_type, p_inode, file_name, inode, 0);
253 release_dentry_name_snapshot(&name);
258 EXPORT_SYMBOL_GPL(__fsnotify_parent);
260 static int fsnotify_handle_inode_event(struct fsnotify_group *group,
261 struct fsnotify_mark *inode_mark,
262 u32 mask, const void *data, int data_type,
263 struct inode *dir, const struct qstr *name,
266 const struct path *path = fsnotify_data_path(data, data_type);
267 struct inode *inode = fsnotify_data_inode(data, data_type);
268 const struct fsnotify_ops *ops = group->ops;
270 if (WARN_ON_ONCE(!ops->handle_inode_event))
273 if (WARN_ON_ONCE(!inode && !dir))
276 if ((inode_mark->flags & FSNOTIFY_MARK_FLAG_EXCL_UNLINK) &&
277 path && d_unlinked(path->dentry))
280 /* Check interest of this mark in case event was sent with two marks */
281 if (!(mask & inode_mark->mask & ALL_FSNOTIFY_EVENTS))
284 return ops->handle_inode_event(inode_mark, mask, inode, dir, name, cookie);
287 static int fsnotify_handle_event(struct fsnotify_group *group, __u32 mask,
288 const void *data, int data_type,
289 struct inode *dir, const struct qstr *name,
290 u32 cookie, struct fsnotify_iter_info *iter_info)
292 struct fsnotify_mark *inode_mark = fsnotify_iter_inode_mark(iter_info);
293 struct fsnotify_mark *parent_mark = fsnotify_iter_parent_mark(iter_info);
296 if (WARN_ON_ONCE(fsnotify_iter_sb_mark(iter_info)) ||
297 WARN_ON_ONCE(fsnotify_iter_vfsmount_mark(iter_info)))
301 * For FS_RENAME, 'dir' is old dir and 'data' is new dentry.
302 * The only ->handle_inode_event() backend that supports FS_RENAME is
303 * dnotify, where it means file was renamed within same parent.
305 if (mask & FS_RENAME) {
306 struct dentry *moved = fsnotify_data_dentry(data, data_type);
308 if (dir != moved->d_parent->d_inode)
313 ret = fsnotify_handle_inode_event(group, parent_mark, mask,
314 data, data_type, dir, name, 0);
322 if (mask & FS_EVENT_ON_CHILD) {
324 * Some events can be sent on both parent dir and child marks
325 * (e.g. FS_ATTRIB). If both parent dir and child are
326 * watching, report the event once to parent dir with name (if
327 * interested) and once to child without name (if interested).
328 * The child watcher is expecting an event without a file name
329 * and without the FS_EVENT_ON_CHILD flag.
331 mask &= ~FS_EVENT_ON_CHILD;
336 return fsnotify_handle_inode_event(group, inode_mark, mask, data, data_type,
340 static int send_to_group(__u32 mask, const void *data, int data_type,
341 struct inode *dir, const struct qstr *file_name,
342 u32 cookie, struct fsnotify_iter_info *iter_info)
344 struct fsnotify_group *group = NULL;
345 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
346 __u32 marks_mask = 0;
347 __u32 marks_ignore_mask = 0;
348 bool is_dir = mask & FS_ISDIR;
349 struct fsnotify_mark *mark;
352 if (!iter_info->report_mask)
355 /* clear ignored on inode modification */
356 if (mask & FS_MODIFY) {
357 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
359 FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
360 mark->ignore_mask = 0;
364 /* Are any of the group marks interested in this event? */
365 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
367 marks_mask |= mark->mask;
369 fsnotify_effective_ignore_mask(mark, is_dir, type);
372 pr_debug("%s: group=%p mask=%x marks_mask=%x marks_ignore_mask=%x data=%p data_type=%d dir=%p cookie=%d\n",
373 __func__, group, mask, marks_mask, marks_ignore_mask,
374 data, data_type, dir, cookie);
376 if (!(test_mask & marks_mask & ~marks_ignore_mask))
379 if (group->ops->handle_event) {
380 return group->ops->handle_event(group, mask, data, data_type, dir,
381 file_name, cookie, iter_info);
384 return fsnotify_handle_event(group, mask, data, data_type, dir,
385 file_name, cookie, iter_info);
388 static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
390 struct fsnotify_mark_connector *conn;
391 struct hlist_node *node = NULL;
393 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
395 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
397 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
400 static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
402 struct hlist_node *node = NULL;
405 node = srcu_dereference(mark->obj_list.next,
406 &fsnotify_mark_srcu);
408 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
412 * iter_info is a multi head priority queue of marks.
413 * Pick a subset of marks from queue heads, all with the same group
414 * and set the report_mask to a subset of the selected marks.
415 * Returns false if there are no more groups to iterate.
417 static bool fsnotify_iter_select_report_types(
418 struct fsnotify_iter_info *iter_info)
420 struct fsnotify_group *max_prio_group = NULL;
421 struct fsnotify_mark *mark;
424 /* Choose max prio group among groups of all queue heads */
425 fsnotify_foreach_iter_type(type) {
426 mark = iter_info->marks[type];
428 fsnotify_compare_groups(max_prio_group, mark->group) > 0)
429 max_prio_group = mark->group;
435 /* Set the report mask for marks from same group as max prio group */
436 iter_info->current_group = max_prio_group;
437 iter_info->report_mask = 0;
438 fsnotify_foreach_iter_type(type) {
439 mark = iter_info->marks[type];
440 if (mark && mark->group == iter_info->current_group) {
442 * FSNOTIFY_ITER_TYPE_PARENT indicates that this inode
443 * is watching children and interested in this event,
444 * which is an event possible on child.
445 * But is *this mark* watching children?
447 if (type == FSNOTIFY_ITER_TYPE_PARENT &&
448 !(mark->mask & FS_EVENT_ON_CHILD) &&
449 !(fsnotify_ignore_mask(mark) & FS_EVENT_ON_CHILD))
452 fsnotify_iter_set_report_type(iter_info, type);
460 * Pop from iter_info multi head queue, the marks that belong to the group of
461 * current iteration step.
463 static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
465 struct fsnotify_mark *mark;
469 * We cannot use fsnotify_foreach_iter_mark_type() here because we
470 * may need to advance a mark of type X that belongs to current_group
471 * but was not selected for reporting.
473 fsnotify_foreach_iter_type(type) {
474 mark = iter_info->marks[type];
475 if (mark && mark->group == iter_info->current_group)
476 iter_info->marks[type] =
477 fsnotify_next_mark(iter_info->marks[type]);
482 * fsnotify - This is the main call to fsnotify.
484 * The VFS calls into hook specific functions in linux/fsnotify.h.
485 * Those functions then in turn call here. Here will call out to all of the
486 * registered fsnotify_group. Those groups can then use the notification event
487 * in whatever means they feel necessary.
489 * @mask: event type and flags
490 * @data: object that event happened on
491 * @data_type: type of object for fanotify_data_XXX() accessors
492 * @dir: optional directory associated with event -
493 * if @file_name is not NULL, this is the directory that
494 * @file_name is relative to
495 * @file_name: optional file name associated with event
496 * @inode: optional inode associated with event -
497 * If @dir and @inode are both non-NULL, event may be
499 * @cookie: inotify rename cookie
501 int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
502 const struct qstr *file_name, struct inode *inode, u32 cookie)
504 const struct path *path = fsnotify_data_path(data, data_type);
505 struct super_block *sb = fsnotify_data_sb(data, data_type);
506 struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
507 struct fsnotify_iter_info iter_info = {};
508 struct mount *mnt = NULL;
509 struct inode *inode2 = NULL;
510 struct dentry *moved;
513 __u32 test_mask, marks_mask;
516 mnt = real_mount(path->mnt);
519 /* Dirent event - report on TYPE_INODE to dir */
521 /* For FS_RENAME, inode is old_dir and inode2 is new_dir */
522 if (mask & FS_RENAME) {
523 moved = fsnotify_data_dentry(data, data_type);
524 inode2 = moved->d_parent->d_inode;
525 inode2_type = FSNOTIFY_ITER_TYPE_INODE2;
527 } else if (mask & FS_EVENT_ON_CHILD) {
529 * Event on child - report on TYPE_PARENT to dir if it is
530 * watching children and on TYPE_INODE to child.
533 inode2_type = FSNOTIFY_ITER_TYPE_PARENT;
537 * Optimization: srcu_read_lock() has a memory barrier which can
538 * be expensive. It protects walking the *_fsnotify_marks lists.
539 * However, if we do not walk the lists, we do not have to do
540 * SRCU because we have no references to any objects and do not
541 * need SRCU to keep them "alive".
543 if ((!sbinfo || !sbinfo->sb_marks) &&
544 (!mnt || !mnt->mnt_fsnotify_marks) &&
545 (!inode || !inode->i_fsnotify_marks) &&
546 (!inode2 || !inode2->i_fsnotify_marks))
549 marks_mask = sb->s_fsnotify_mask;
551 marks_mask |= mnt->mnt_fsnotify_mask;
553 marks_mask |= inode->i_fsnotify_mask;
555 marks_mask |= inode2->i_fsnotify_mask;
559 * If this is a modify event we may need to clear some ignore masks.
560 * In that case, the object with ignore masks will have the FS_MODIFY
562 * Otherwise, return if none of the marks care about this type of event.
564 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
565 if (!(test_mask & marks_mask))
568 iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
571 iter_info.marks[FSNOTIFY_ITER_TYPE_SB] =
572 fsnotify_first_mark(&sbinfo->sb_marks);
575 iter_info.marks[FSNOTIFY_ITER_TYPE_VFSMOUNT] =
576 fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
579 iter_info.marks[FSNOTIFY_ITER_TYPE_INODE] =
580 fsnotify_first_mark(&inode->i_fsnotify_marks);
583 iter_info.marks[inode2_type] =
584 fsnotify_first_mark(&inode2->i_fsnotify_marks);
588 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
589 * ignore masks are properly reflected for mount/sb mark notifications.
590 * That's why this traversal is so complicated...
592 while (fsnotify_iter_select_report_types(&iter_info)) {
593 ret = send_to_group(mask, data, data_type, dir, file_name,
596 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
599 fsnotify_iter_next(&iter_info);
603 srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
607 EXPORT_SYMBOL_GPL(fsnotify);
609 static __init int fsnotify_init(void)
613 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 23);
615 ret = init_srcu_struct(&fsnotify_mark_srcu);
617 panic("initializing fsnotify_mark_srcu");
619 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
624 core_initcall(fsnotify_init);