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, FS_UNMOUNT, inode, FSNOTIFY_EVENT_INODE, NULL, 0);
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);
146 * Notify this dentry's parent about a child's events with child name info
147 * if parent is watching.
148 * Notify also the child without name info if child inode is watching.
150 int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
153 struct dentry *parent;
154 struct inode *p_inode;
157 if (!(dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED))
160 parent = dget_parent(dentry);
161 p_inode = parent->d_inode;
163 if (unlikely(!fsnotify_inode_watches_children(p_inode))) {
164 __fsnotify_update_child_dentry_flags(p_inode);
165 } else if (p_inode->i_fsnotify_mask & mask & ALL_FSNOTIFY_EVENTS) {
166 struct name_snapshot name;
169 * We are notifying a parent, so set a flag in mask to inform
170 * backend that event has information about a child entry.
172 take_dentry_name_snapshot(&name, dentry);
173 ret = fsnotify(p_inode, mask | FS_EVENT_ON_CHILD, data,
174 data_type, &name.name, 0);
175 release_dentry_name_snapshot(&name);
184 return fsnotify(d_inode(dentry), mask, data, data_type, NULL, 0);
186 EXPORT_SYMBOL_GPL(__fsnotify_parent);
188 static int send_to_group(__u32 mask, const void *data, int data_type,
189 struct inode *dir, const struct qstr *file_name,
190 u32 cookie, struct fsnotify_iter_info *iter_info)
192 struct fsnotify_group *group = NULL;
193 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
194 __u32 marks_mask = 0;
195 __u32 marks_ignored_mask = 0;
196 struct fsnotify_mark *mark;
199 if (WARN_ON(!iter_info->report_mask))
202 /* clear ignored on inode modification */
203 if (mask & FS_MODIFY) {
204 fsnotify_foreach_obj_type(type) {
205 if (!fsnotify_iter_should_report_type(iter_info, type))
207 mark = iter_info->marks[type];
209 !(mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
210 mark->ignored_mask = 0;
214 fsnotify_foreach_obj_type(type) {
215 if (!fsnotify_iter_should_report_type(iter_info, type))
217 mark = iter_info->marks[type];
218 /* does the object mark tell us to do something? */
221 marks_mask |= mark->mask;
222 marks_ignored_mask |= mark->ignored_mask;
226 pr_debug("%s: group=%p mask=%x marks_mask=%x marks_ignored_mask=%x data=%p data_type=%d dir=%p cookie=%d\n",
227 __func__, group, mask, marks_mask, marks_ignored_mask,
228 data, data_type, dir, cookie);
230 if (!(test_mask & marks_mask & ~marks_ignored_mask))
233 return group->ops->handle_event(group, mask, data, data_type, dir,
234 file_name, cookie, iter_info);
237 static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
239 struct fsnotify_mark_connector *conn;
240 struct hlist_node *node = NULL;
242 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
244 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
246 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
249 static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
251 struct hlist_node *node = NULL;
254 node = srcu_dereference(mark->obj_list.next,
255 &fsnotify_mark_srcu);
257 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
261 * iter_info is a multi head priority queue of marks.
262 * Pick a subset of marks from queue heads, all with the
263 * same group and set the report_mask for selected subset.
264 * Returns the report_mask of the selected subset.
266 static unsigned int fsnotify_iter_select_report_types(
267 struct fsnotify_iter_info *iter_info)
269 struct fsnotify_group *max_prio_group = NULL;
270 struct fsnotify_mark *mark;
273 /* Choose max prio group among groups of all queue heads */
274 fsnotify_foreach_obj_type(type) {
275 mark = iter_info->marks[type];
277 fsnotify_compare_groups(max_prio_group, mark->group) > 0)
278 max_prio_group = mark->group;
284 /* Set the report mask for marks from same group as max prio group */
285 iter_info->report_mask = 0;
286 fsnotify_foreach_obj_type(type) {
287 mark = iter_info->marks[type];
289 fsnotify_compare_groups(max_prio_group, mark->group) == 0)
290 fsnotify_iter_set_report_type(iter_info, type);
293 return iter_info->report_mask;
297 * Pop from iter_info multi head queue, the marks that were iterated in the
298 * current iteration step.
300 static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
304 fsnotify_foreach_obj_type(type) {
305 if (fsnotify_iter_should_report_type(iter_info, type))
306 iter_info->marks[type] =
307 fsnotify_next_mark(iter_info->marks[type]);
312 * This is the main call to fsnotify. The VFS calls into hook specific functions
313 * in linux/fsnotify.h. Those functions then in turn call here. Here will call
314 * out to all of the registered fsnotify_group. Those groups can then use the
315 * notification event in whatever means they feel necessary.
317 int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_type,
318 const struct qstr *file_name, u32 cookie)
320 const struct path *path = fsnotify_data_path(data, data_type);
321 struct fsnotify_iter_info iter_info = {};
322 struct super_block *sb = to_tell->i_sb;
323 struct inode *dir = file_name ? to_tell : NULL;
324 struct mount *mnt = NULL;
326 __u32 test_mask, marks_mask;
329 mnt = real_mount(path->mnt);
332 * Optimization: srcu_read_lock() has a memory barrier which can
333 * be expensive. It protects walking the *_fsnotify_marks lists.
334 * However, if we do not walk the lists, we do not have to do
335 * SRCU because we have no references to any objects and do not
336 * need SRCU to keep them "alive".
338 if (!to_tell->i_fsnotify_marks && !sb->s_fsnotify_marks &&
339 (!mnt || !mnt->mnt_fsnotify_marks))
342 /* An event "on child" is not intended for a mount/sb mark */
343 marks_mask = to_tell->i_fsnotify_mask;
344 if (!(mask & FS_EVENT_ON_CHILD)) {
345 marks_mask |= sb->s_fsnotify_mask;
347 marks_mask |= mnt->mnt_fsnotify_mask;
351 * if this is a modify event we may need to clear the ignored masks
352 * otherwise return if neither the inode nor the vfsmount/sb care about
353 * this type of event.
355 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
356 if (!(mask & FS_MODIFY) && !(test_mask & marks_mask))
359 iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
361 iter_info.marks[FSNOTIFY_OBJ_TYPE_INODE] =
362 fsnotify_first_mark(&to_tell->i_fsnotify_marks);
363 iter_info.marks[FSNOTIFY_OBJ_TYPE_SB] =
364 fsnotify_first_mark(&sb->s_fsnotify_marks);
366 iter_info.marks[FSNOTIFY_OBJ_TYPE_VFSMOUNT] =
367 fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
371 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
372 * ignore masks are properly reflected for mount/sb mark notifications.
373 * That's why this traversal is so complicated...
375 while (fsnotify_iter_select_report_types(&iter_info)) {
376 ret = send_to_group(mask, data, data_type, dir, file_name,
379 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
382 fsnotify_iter_next(&iter_info);
386 srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
390 EXPORT_SYMBOL_GPL(fsnotify);
392 static __init int fsnotify_init(void)
396 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 25);
398 ret = init_srcu_struct(&fsnotify_mark_srcu);
400 panic("initializing fsnotify_mark_srcu");
402 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
407 core_initcall(fsnotify_init);