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);
76 /* for each watch, send FS_UNMOUNT and then remove it */
77 fsnotify_inode(inode, FS_UNMOUNT);
79 fsnotify_inode_delete(inode);
84 spin_lock(&sb->s_inode_list_lock);
86 spin_unlock(&sb->s_inode_list_lock);
90 /* Wait for outstanding inode references from connectors */
91 wait_var_event(&sb->s_fsnotify_inode_refs,
92 !atomic_long_read(&sb->s_fsnotify_inode_refs));
95 void fsnotify_sb_delete(struct super_block *sb)
97 fsnotify_unmount_inodes(sb);
98 fsnotify_clear_marks_by_sb(sb);
102 * Given an inode, first check if we care what happens to our children. Inotify
103 * and dnotify both tell their parents about events. If we care about any event
104 * on a child we run all of our children and set a dentry flag saying that the
105 * parent cares. Thus when an event happens on a child it can quickly tell if
106 * if there is a need to find a parent and send the event to the parent.
108 void __fsnotify_update_child_dentry_flags(struct inode *inode)
110 struct dentry *alias;
113 if (!S_ISDIR(inode->i_mode))
116 /* determine if the children should tell inode about their events */
117 watched = fsnotify_inode_watches_children(inode);
119 spin_lock(&inode->i_lock);
120 /* run all of the dentries associated with this inode. Since this is a
121 * directory, there damn well better only be one item on this list */
122 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
123 struct dentry *child;
125 /* run all of the children of the original inode and fix their
126 * d_flags to indicate parental interest (their parent is the
128 spin_lock(&alias->d_lock);
129 list_for_each_entry(child, &alias->d_subdirs, d_child) {
133 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
135 child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
137 child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
138 spin_unlock(&child->d_lock);
140 spin_unlock(&alias->d_lock);
142 spin_unlock(&inode->i_lock);
145 /* Are inode/sb/mount interested in parent and name info with this event? */
146 static bool fsnotify_event_needs_parent(struct inode *inode, struct mount *mnt,
149 __u32 marks_mask = 0;
151 /* We only send parent/name to inode/sb/mount for events on non-dir */
155 /* Did either inode/sb/mount subscribe for events with parent/name? */
156 marks_mask |= fsnotify_parent_needed_mask(inode->i_fsnotify_mask);
157 marks_mask |= fsnotify_parent_needed_mask(inode->i_sb->s_fsnotify_mask);
159 marks_mask |= fsnotify_parent_needed_mask(mnt->mnt_fsnotify_mask);
161 /* Did they subscribe for this event with parent/name info? */
162 return mask & marks_mask;
166 * Notify this dentry's parent about a child's events with child name info
167 * if parent is watching or if inode/sb/mount are interested in events with
168 * parent and name info.
170 * Notify only the child without name info if parent is not watching and
171 * inode/sb/mount are not interested in events with parent and name info.
173 int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
176 const struct path *path = fsnotify_data_path(data, data_type);
177 struct mount *mnt = path ? real_mount(path->mnt) : NULL;
178 struct inode *inode = d_inode(dentry);
179 struct dentry *parent;
180 bool parent_watched = dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED;
181 bool parent_needed, parent_interested;
183 struct inode *p_inode = NULL;
184 struct name_snapshot name;
185 struct qstr *file_name = NULL;
189 * Do inode/sb/mount care about parent and name info on non-dir?
190 * Do they care about any event at all?
192 if (!inode->i_fsnotify_marks && !inode->i_sb->s_fsnotify_marks &&
193 (!mnt || !mnt->mnt_fsnotify_marks) && !parent_watched)
197 parent_needed = fsnotify_event_needs_parent(inode, mnt, mask);
198 if (!parent_watched && !parent_needed)
201 /* Does parent inode care about events on children? */
202 parent = dget_parent(dentry);
203 p_inode = parent->d_inode;
204 p_mask = fsnotify_inode_watches_children(p_inode);
205 if (unlikely(parent_watched && !p_mask))
206 __fsnotify_update_child_dentry_flags(p_inode);
209 * Include parent/name in notification either if some notification
210 * groups require parent info or the parent is interested in this event.
212 parent_interested = mask & p_mask & ALL_FSNOTIFY_EVENTS;
213 if (parent_needed || parent_interested) {
214 /* When notifying parent, child should be passed as data */
215 WARN_ON_ONCE(inode != fsnotify_data_inode(data, data_type));
217 /* Notify both parent and child with child name info */
218 take_dentry_name_snapshot(&name, dentry);
219 file_name = &name.name;
220 if (parent_interested)
221 mask |= FS_EVENT_ON_CHILD;
225 ret = fsnotify(mask, data, data_type, p_inode, file_name, inode, 0);
228 release_dentry_name_snapshot(&name);
233 EXPORT_SYMBOL_GPL(__fsnotify_parent);
235 static int fsnotify_handle_event(struct fsnotify_group *group, __u32 mask,
236 const void *data, int data_type,
237 struct inode *dir, const struct qstr *name,
238 u32 cookie, struct fsnotify_iter_info *iter_info)
240 struct fsnotify_mark *inode_mark = fsnotify_iter_inode_mark(iter_info);
241 struct fsnotify_mark *child_mark = fsnotify_iter_child_mark(iter_info);
242 struct inode *inode = fsnotify_data_inode(data, data_type);
243 const struct fsnotify_ops *ops = group->ops;
246 if (WARN_ON_ONCE(!ops->handle_inode_event))
249 if (WARN_ON_ONCE(fsnotify_iter_sb_mark(iter_info)) ||
250 WARN_ON_ONCE(fsnotify_iter_vfsmount_mark(iter_info)))
254 * An event can be sent on child mark iterator instead of inode mark
255 * iterator because of other groups that have interest of this inode
256 * and have marks on both parent and child. We can simplify this case.
259 inode_mark = child_mark;
265 ret = ops->handle_inode_event(inode_mark, mask, inode, dir, name);
266 if (ret || !child_mark)
270 * Some events can be sent on both parent dir and child marks
271 * (e.g. FS_ATTRIB). If both parent dir and child are watching,
272 * report the event once to parent dir with name and once to child
275 return ops->handle_inode_event(child_mark, mask, inode, NULL, NULL);
278 static int send_to_group(__u32 mask, const void *data, int data_type,
279 struct inode *dir, const struct qstr *file_name,
280 u32 cookie, struct fsnotify_iter_info *iter_info)
282 struct fsnotify_group *group = NULL;
283 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
284 __u32 marks_mask = 0;
285 __u32 marks_ignored_mask = 0;
286 struct fsnotify_mark *mark;
289 if (WARN_ON(!iter_info->report_mask))
292 /* clear ignored on inode modification */
293 if (mask & FS_MODIFY) {
294 fsnotify_foreach_obj_type(type) {
295 if (!fsnotify_iter_should_report_type(iter_info, type))
297 mark = iter_info->marks[type];
299 !(mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
300 mark->ignored_mask = 0;
304 fsnotify_foreach_obj_type(type) {
305 if (!fsnotify_iter_should_report_type(iter_info, type))
307 mark = iter_info->marks[type];
308 /* does the object mark tell us to do something? */
311 marks_mask |= mark->mask;
312 marks_ignored_mask |= mark->ignored_mask;
316 pr_debug("%s: group=%p mask=%x marks_mask=%x marks_ignored_mask=%x data=%p data_type=%d dir=%p cookie=%d\n",
317 __func__, group, mask, marks_mask, marks_ignored_mask,
318 data, data_type, dir, cookie);
320 if (!(test_mask & marks_mask & ~marks_ignored_mask))
323 if (group->ops->handle_event) {
324 return group->ops->handle_event(group, mask, data, data_type, dir,
325 file_name, cookie, iter_info);
328 return fsnotify_handle_event(group, mask, data, data_type, dir,
329 file_name, cookie, iter_info);
332 static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
334 struct fsnotify_mark_connector *conn;
335 struct hlist_node *node = NULL;
337 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
339 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
341 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
344 static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
346 struct hlist_node *node = NULL;
349 node = srcu_dereference(mark->obj_list.next,
350 &fsnotify_mark_srcu);
352 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
356 * iter_info is a multi head priority queue of marks.
357 * Pick a subset of marks from queue heads, all with the
358 * same group and set the report_mask for selected subset.
359 * Returns the report_mask of the selected subset.
361 static unsigned int fsnotify_iter_select_report_types(
362 struct fsnotify_iter_info *iter_info)
364 struct fsnotify_group *max_prio_group = NULL;
365 struct fsnotify_mark *mark;
368 /* Choose max prio group among groups of all queue heads */
369 fsnotify_foreach_obj_type(type) {
370 mark = iter_info->marks[type];
372 fsnotify_compare_groups(max_prio_group, mark->group) > 0)
373 max_prio_group = mark->group;
379 /* Set the report mask for marks from same group as max prio group */
380 iter_info->report_mask = 0;
381 fsnotify_foreach_obj_type(type) {
382 mark = iter_info->marks[type];
384 fsnotify_compare_groups(max_prio_group, mark->group) == 0)
385 fsnotify_iter_set_report_type(iter_info, type);
388 return iter_info->report_mask;
392 * Pop from iter_info multi head queue, the marks that were iterated in the
393 * current iteration step.
395 static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
399 fsnotify_foreach_obj_type(type) {
400 if (fsnotify_iter_should_report_type(iter_info, type))
401 iter_info->marks[type] =
402 fsnotify_next_mark(iter_info->marks[type]);
407 * fsnotify - This is the main call to fsnotify.
409 * The VFS calls into hook specific functions in linux/fsnotify.h.
410 * Those functions then in turn call here. Here will call out to all of the
411 * registered fsnotify_group. Those groups can then use the notification event
412 * in whatever means they feel necessary.
414 * @mask: event type and flags
415 * @data: object that event happened on
416 * @data_type: type of object for fanotify_data_XXX() accessors
417 * @dir: optional directory associated with event -
418 * if @file_name is not NULL, this is the directory that
419 * @file_name is relative to
420 * @file_name: optional file name associated with event
421 * @inode: optional inode associated with event -
422 * either @dir or @inode must be non-NULL.
423 * if both are non-NULL event may be reported to both.
424 * @cookie: inotify rename cookie
426 int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
427 const struct qstr *file_name, struct inode *inode, u32 cookie)
429 const struct path *path = fsnotify_data_path(data, data_type);
430 struct fsnotify_iter_info iter_info = {};
431 struct super_block *sb;
432 struct mount *mnt = NULL;
433 struct inode *child = NULL;
435 __u32 test_mask, marks_mask;
438 mnt = real_mount(path->mnt);
441 /* Dirent event - report on TYPE_INODE to dir */
443 } else if (mask & FS_EVENT_ON_CHILD) {
445 * Event on child - report on TYPE_INODE to dir if it is
446 * watching children and on TYPE_CHILD to child.
454 * Optimization: srcu_read_lock() has a memory barrier which can
455 * be expensive. It protects walking the *_fsnotify_marks lists.
456 * However, if we do not walk the lists, we do not have to do
457 * SRCU because we have no references to any objects and do not
458 * need SRCU to keep them "alive".
460 if (!sb->s_fsnotify_marks &&
461 (!mnt || !mnt->mnt_fsnotify_marks) &&
462 (!inode || !inode->i_fsnotify_marks) &&
463 (!child || !child->i_fsnotify_marks))
466 marks_mask = sb->s_fsnotify_mask;
468 marks_mask |= mnt->mnt_fsnotify_mask;
470 marks_mask |= inode->i_fsnotify_mask;
472 marks_mask |= child->i_fsnotify_mask;
476 * if this is a modify event we may need to clear the ignored masks
477 * otherwise return if none of the marks care about this type of event.
479 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
480 if (!(mask & FS_MODIFY) && !(test_mask & marks_mask))
483 iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
485 iter_info.marks[FSNOTIFY_OBJ_TYPE_SB] =
486 fsnotify_first_mark(&sb->s_fsnotify_marks);
488 iter_info.marks[FSNOTIFY_OBJ_TYPE_VFSMOUNT] =
489 fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
492 iter_info.marks[FSNOTIFY_OBJ_TYPE_INODE] =
493 fsnotify_first_mark(&inode->i_fsnotify_marks);
496 iter_info.marks[FSNOTIFY_OBJ_TYPE_CHILD] =
497 fsnotify_first_mark(&child->i_fsnotify_marks);
501 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
502 * ignore masks are properly reflected for mount/sb mark notifications.
503 * That's why this traversal is so complicated...
505 while (fsnotify_iter_select_report_types(&iter_info)) {
506 ret = send_to_group(mask, data, data_type, dir, file_name,
509 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
512 fsnotify_iter_next(&iter_info);
516 srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
520 EXPORT_SYMBOL_GPL(fsnotify);
522 static __init int fsnotify_init(void)
526 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 25);
528 ret = init_srcu_struct(&fsnotify_mark_srcu);
530 panic("initializing fsnotify_mark_srcu");
532 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
537 core_initcall(fsnotify_init);