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.
61 if (!atomic_read(&inode->i_count)) {
62 spin_unlock(&inode->i_lock);
67 spin_unlock(&inode->i_lock);
68 spin_unlock(&sb->s_inode_list_lock);
73 /* for each watch, send FS_UNMOUNT and then remove it */
74 fsnotify(inode, FS_UNMOUNT, inode, FSNOTIFY_EVENT_INODE, NULL, 0);
76 fsnotify_inode_delete(inode);
80 spin_lock(&sb->s_inode_list_lock);
82 spin_unlock(&sb->s_inode_list_lock);
86 /* Wait for outstanding inode references from connectors */
87 wait_var_event(&sb->s_fsnotify_inode_refs,
88 !atomic_long_read(&sb->s_fsnotify_inode_refs));
91 void fsnotify_sb_delete(struct super_block *sb)
93 fsnotify_unmount_inodes(sb);
94 fsnotify_clear_marks_by_sb(sb);
98 * Given an inode, first check if we care what happens to our children. Inotify
99 * and dnotify both tell their parents about events. If we care about any event
100 * on a child we run all of our children and set a dentry flag saying that the
101 * parent cares. Thus when an event happens on a child it can quickly tell if
102 * if there is a need to find a parent and send the event to the parent.
104 void __fsnotify_update_child_dentry_flags(struct inode *inode)
106 struct dentry *alias;
109 if (!S_ISDIR(inode->i_mode))
112 /* determine if the children should tell inode about their events */
113 watched = fsnotify_inode_watches_children(inode);
115 spin_lock(&inode->i_lock);
116 /* run all of the dentries associated with this inode. Since this is a
117 * directory, there damn well better only be one item on this list */
118 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
119 struct dentry *child;
121 /* run all of the children of the original inode and fix their
122 * d_flags to indicate parental interest (their parent is the
124 spin_lock(&alias->d_lock);
125 list_for_each_entry(child, &alias->d_subdirs, d_child) {
129 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
131 child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
133 child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
134 spin_unlock(&child->d_lock);
136 spin_unlock(&alias->d_lock);
138 spin_unlock(&inode->i_lock);
141 /* Notify this dentry's parent about a child's events. */
142 int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask)
144 struct dentry *parent;
145 struct inode *p_inode;
149 dentry = path->dentry;
151 if (!(dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED))
154 parent = dget_parent(dentry);
155 p_inode = parent->d_inode;
157 if (unlikely(!fsnotify_inode_watches_children(p_inode))) {
158 __fsnotify_update_child_dentry_flags(p_inode);
159 } else if (p_inode->i_fsnotify_mask & mask & ALL_FSNOTIFY_EVENTS) {
160 struct name_snapshot name;
162 /* we are notifying a parent so come up with the new mask which
163 * specifies these are events which came from a child. */
164 mask |= FS_EVENT_ON_CHILD;
166 take_dentry_name_snapshot(&name, dentry);
168 ret = fsnotify(p_inode, mask, path, FSNOTIFY_EVENT_PATH,
171 ret = fsnotify(p_inode, mask, dentry->d_inode, FSNOTIFY_EVENT_INODE,
173 release_dentry_name_snapshot(&name);
180 EXPORT_SYMBOL_GPL(__fsnotify_parent);
182 static int send_to_group(struct inode *to_tell,
183 __u32 mask, const void *data,
184 int data_is, u32 cookie,
185 const struct qstr *file_name,
186 struct fsnotify_iter_info *iter_info)
188 struct fsnotify_group *group = NULL;
189 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
190 __u32 marks_mask = 0;
191 __u32 marks_ignored_mask = 0;
192 struct fsnotify_mark *mark;
195 if (WARN_ON(!iter_info->report_mask))
198 /* clear ignored on inode modification */
199 if (mask & FS_MODIFY) {
200 fsnotify_foreach_obj_type(type) {
201 if (!fsnotify_iter_should_report_type(iter_info, type))
203 mark = iter_info->marks[type];
205 !(mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
206 mark->ignored_mask = 0;
210 fsnotify_foreach_obj_type(type) {
211 if (!fsnotify_iter_should_report_type(iter_info, type))
213 mark = iter_info->marks[type];
214 /* does the object mark tell us to do something? */
217 marks_mask |= mark->mask;
218 marks_ignored_mask |= mark->ignored_mask;
222 pr_debug("%s: group=%p to_tell=%p mask=%x marks_mask=%x marks_ignored_mask=%x"
223 " data=%p data_is=%d cookie=%d\n",
224 __func__, group, to_tell, mask, marks_mask, marks_ignored_mask,
225 data, data_is, cookie);
227 if (!(test_mask & marks_mask & ~marks_ignored_mask))
230 return group->ops->handle_event(group, to_tell, mask, data, data_is,
231 file_name, cookie, iter_info);
234 static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
236 struct fsnotify_mark_connector *conn;
237 struct hlist_node *node = NULL;
239 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
241 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
243 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
246 static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
248 struct hlist_node *node = NULL;
251 node = srcu_dereference(mark->obj_list.next,
252 &fsnotify_mark_srcu);
254 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
258 * iter_info is a multi head priority queue of marks.
259 * Pick a subset of marks from queue heads, all with the
260 * same group and set the report_mask for selected subset.
261 * Returns the report_mask of the selected subset.
263 static unsigned int fsnotify_iter_select_report_types(
264 struct fsnotify_iter_info *iter_info)
266 struct fsnotify_group *max_prio_group = NULL;
267 struct fsnotify_mark *mark;
270 /* Choose max prio group among groups of all queue heads */
271 fsnotify_foreach_obj_type(type) {
272 mark = iter_info->marks[type];
274 fsnotify_compare_groups(max_prio_group, mark->group) > 0)
275 max_prio_group = mark->group;
281 /* Set the report mask for marks from same group as max prio group */
282 iter_info->report_mask = 0;
283 fsnotify_foreach_obj_type(type) {
284 mark = iter_info->marks[type];
286 fsnotify_compare_groups(max_prio_group, mark->group) == 0)
287 fsnotify_iter_set_report_type(iter_info, type);
290 return iter_info->report_mask;
294 * Pop from iter_info multi head queue, the marks that were iterated in the
295 * current iteration step.
297 static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
301 fsnotify_foreach_obj_type(type) {
302 if (fsnotify_iter_should_report_type(iter_info, type))
303 iter_info->marks[type] =
304 fsnotify_next_mark(iter_info->marks[type]);
309 * This is the main call to fsnotify. The VFS calls into hook specific functions
310 * in linux/fsnotify.h. Those functions then in turn call here. Here will call
311 * out to all of the registered fsnotify_group. Those groups can then use the
312 * notification event in whatever means they feel necessary.
314 int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
315 const struct qstr *file_name, u32 cookie)
317 struct fsnotify_iter_info iter_info = {};
318 struct super_block *sb = to_tell->i_sb;
319 struct mount *mnt = NULL;
320 __u32 mnt_or_sb_mask = sb->s_fsnotify_mask;
322 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
324 if (data_is == FSNOTIFY_EVENT_PATH) {
325 mnt = real_mount(((const struct path *)data)->mnt);
326 mnt_or_sb_mask |= mnt->mnt_fsnotify_mask;
328 /* An event "on child" is not intended for a mount/sb mark */
329 if (mask & FS_EVENT_ON_CHILD)
333 * Optimization: srcu_read_lock() has a memory barrier which can
334 * be expensive. It protects walking the *_fsnotify_marks lists.
335 * However, if we do not walk the lists, we do not have to do
336 * SRCU because we have no references to any objects and do not
337 * need SRCU to keep them "alive".
339 if (!to_tell->i_fsnotify_marks && !sb->s_fsnotify_marks &&
340 (!mnt || !mnt->mnt_fsnotify_marks))
343 * if this is a modify event we may need to clear the ignored masks
344 * otherwise return if neither the inode nor the vfsmount/sb care about
345 * this type of event.
347 if (!(mask & FS_MODIFY) &&
348 !(test_mask & (to_tell->i_fsnotify_mask | mnt_or_sb_mask)))
351 iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
353 iter_info.marks[FSNOTIFY_OBJ_TYPE_INODE] =
354 fsnotify_first_mark(&to_tell->i_fsnotify_marks);
355 iter_info.marks[FSNOTIFY_OBJ_TYPE_SB] =
356 fsnotify_first_mark(&sb->s_fsnotify_marks);
358 iter_info.marks[FSNOTIFY_OBJ_TYPE_VFSMOUNT] =
359 fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
363 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
364 * ignore masks are properly reflected for mount/sb mark notifications.
365 * That's why this traversal is so complicated...
367 while (fsnotify_iter_select_report_types(&iter_info)) {
368 ret = send_to_group(to_tell, mask, data, data_is, cookie,
369 file_name, &iter_info);
371 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
374 fsnotify_iter_next(&iter_info);
378 srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
382 EXPORT_SYMBOL_GPL(fsnotify);
384 extern struct kmem_cache *fsnotify_mark_connector_cachep;
386 static __init int fsnotify_init(void)
390 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 25);
392 ret = init_srcu_struct(&fsnotify_mark_srcu);
394 panic("initializing fsnotify_mark_srcu");
396 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
401 core_initcall(fsnotify_init);