2 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2, or (at your option)
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; see the file COPYING. If not, write to
16 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
19 #include <linux/dcache.h>
21 #include <linux/gfp.h>
22 #include <linux/init.h>
23 #include <linux/module.h>
24 #include <linux/mount.h>
25 #include <linux/srcu.h>
27 #include <linux/fsnotify_backend.h>
31 * Clear all of the marks on an inode when it is being evicted from core
33 void __fsnotify_inode_delete(struct inode *inode)
35 fsnotify_clear_marks_by_inode(inode);
37 EXPORT_SYMBOL_GPL(__fsnotify_inode_delete);
39 void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
41 fsnotify_clear_marks_by_mount(mnt);
45 * fsnotify_unmount_inodes - an sb is unmounting. handle any watched inodes.
46 * @sb: superblock being unmounted.
48 * Called during unmount with no locks held, so needs to be safe against
49 * concurrent modifiers. We temporarily drop sb->s_inode_list_lock and CAN block.
51 static void fsnotify_unmount_inodes(struct super_block *sb)
53 struct inode *inode, *iput_inode = NULL;
55 spin_lock(&sb->s_inode_list_lock);
56 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
58 * We cannot __iget() an inode in state I_FREEING,
59 * I_WILL_FREE, or I_NEW which is fine because by that point
60 * the inode cannot have any associated watches.
62 spin_lock(&inode->i_lock);
63 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) {
64 spin_unlock(&inode->i_lock);
69 * If i_count is zero, the inode cannot have any watches and
70 * doing an __iget/iput with SB_ACTIVE clear would actually
71 * evict all inodes with zero i_count from icache which is
72 * unnecessarily violent and may in fact be illegal to do.
74 if (!atomic_read(&inode->i_count)) {
75 spin_unlock(&inode->i_lock);
80 spin_unlock(&inode->i_lock);
81 spin_unlock(&sb->s_inode_list_lock);
86 /* for each watch, send FS_UNMOUNT and then remove it */
87 fsnotify(inode, FS_UNMOUNT, inode, FSNOTIFY_EVENT_INODE, NULL, 0);
89 fsnotify_inode_delete(inode);
93 spin_lock(&sb->s_inode_list_lock);
95 spin_unlock(&sb->s_inode_list_lock);
99 /* Wait for outstanding inode references from connectors */
100 wait_var_event(&sb->s_fsnotify_inode_refs,
101 !atomic_long_read(&sb->s_fsnotify_inode_refs));
104 void fsnotify_sb_delete(struct super_block *sb)
106 fsnotify_unmount_inodes(sb);
107 fsnotify_clear_marks_by_sb(sb);
111 * fsnotify_nameremove - a filename was removed from a directory
113 * This is mostly called under parent vfs inode lock so name and
114 * dentry->d_parent should be stable. However there are some corner cases where
115 * inode lock is not held. So to be on the safe side and be reselient to future
116 * callers and out of tree users of d_delete(), we do not assume that d_parent
117 * and d_name are stable and we use dget_parent() and
118 * take_dentry_name_snapshot() to grab stable references.
120 void fsnotify_nameremove(struct dentry *dentry, int isdir)
122 struct dentry *parent;
123 struct name_snapshot name;
124 __u32 mask = FS_DELETE;
126 /* d_delete() of pseudo inode? (e.g. __ns_get_path() playing tricks) */
133 parent = dget_parent(dentry);
134 /* Avoid unneeded take_dentry_name_snapshot() */
135 if (!(d_inode(parent)->i_fsnotify_mask & FS_DELETE) &&
136 !(dentry->d_sb->s_fsnotify_mask & FS_DELETE))
139 take_dentry_name_snapshot(&name, dentry);
141 fsnotify(d_inode(parent), mask, d_inode(dentry), FSNOTIFY_EVENT_INODE,
144 release_dentry_name_snapshot(&name);
149 EXPORT_SYMBOL(fsnotify_nameremove);
152 * Given an inode, first check if we care what happens to our children. Inotify
153 * and dnotify both tell their parents about events. If we care about any event
154 * on a child we run all of our children and set a dentry flag saying that the
155 * parent cares. Thus when an event happens on a child it can quickly tell if
156 * if there is a need to find a parent and send the event to the parent.
158 void __fsnotify_update_child_dentry_flags(struct inode *inode)
160 struct dentry *alias;
163 if (!S_ISDIR(inode->i_mode))
166 /* determine if the children should tell inode about their events */
167 watched = fsnotify_inode_watches_children(inode);
169 spin_lock(&inode->i_lock);
170 /* run all of the dentries associated with this inode. Since this is a
171 * directory, there damn well better only be one item on this list */
172 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
173 struct dentry *child;
175 /* run all of the children of the original inode and fix their
176 * d_flags to indicate parental interest (their parent is the
178 spin_lock(&alias->d_lock);
179 list_for_each_entry(child, &alias->d_subdirs, d_child) {
183 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
185 child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
187 child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
188 spin_unlock(&child->d_lock);
190 spin_unlock(&alias->d_lock);
192 spin_unlock(&inode->i_lock);
195 /* Notify this dentry's parent about a child's events. */
196 int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask)
198 struct dentry *parent;
199 struct inode *p_inode;
203 dentry = path->dentry;
205 if (!(dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED))
208 parent = dget_parent(dentry);
209 p_inode = parent->d_inode;
211 if (unlikely(!fsnotify_inode_watches_children(p_inode))) {
212 __fsnotify_update_child_dentry_flags(p_inode);
213 } else if (p_inode->i_fsnotify_mask & mask & ALL_FSNOTIFY_EVENTS) {
214 struct name_snapshot name;
216 /* we are notifying a parent so come up with the new mask which
217 * specifies these are events which came from a child. */
218 mask |= FS_EVENT_ON_CHILD;
220 take_dentry_name_snapshot(&name, dentry);
222 ret = fsnotify(p_inode, mask, path, FSNOTIFY_EVENT_PATH,
225 ret = fsnotify(p_inode, mask, dentry->d_inode, FSNOTIFY_EVENT_INODE,
227 release_dentry_name_snapshot(&name);
234 EXPORT_SYMBOL_GPL(__fsnotify_parent);
236 static int send_to_group(struct inode *to_tell,
237 __u32 mask, const void *data,
238 int data_is, u32 cookie,
239 const struct qstr *file_name,
240 struct fsnotify_iter_info *iter_info)
242 struct fsnotify_group *group = NULL;
243 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
244 __u32 marks_mask = 0;
245 __u32 marks_ignored_mask = 0;
246 struct fsnotify_mark *mark;
249 if (WARN_ON(!iter_info->report_mask))
252 /* clear ignored on inode modification */
253 if (mask & FS_MODIFY) {
254 fsnotify_foreach_obj_type(type) {
255 if (!fsnotify_iter_should_report_type(iter_info, type))
257 mark = iter_info->marks[type];
259 !(mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
260 mark->ignored_mask = 0;
264 fsnotify_foreach_obj_type(type) {
265 if (!fsnotify_iter_should_report_type(iter_info, type))
267 mark = iter_info->marks[type];
268 /* does the object mark tell us to do something? */
271 marks_mask |= mark->mask;
272 marks_ignored_mask |= mark->ignored_mask;
276 pr_debug("%s: group=%p to_tell=%p mask=%x marks_mask=%x marks_ignored_mask=%x"
277 " data=%p data_is=%d cookie=%d\n",
278 __func__, group, to_tell, mask, marks_mask, marks_ignored_mask,
279 data, data_is, cookie);
281 if (!(test_mask & marks_mask & ~marks_ignored_mask))
284 return group->ops->handle_event(group, to_tell, mask, data, data_is,
285 file_name, cookie, iter_info);
288 static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
290 struct fsnotify_mark_connector *conn;
291 struct hlist_node *node = NULL;
293 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
295 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
297 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
300 static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
302 struct hlist_node *node = NULL;
305 node = srcu_dereference(mark->obj_list.next,
306 &fsnotify_mark_srcu);
308 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
312 * iter_info is a multi head priority queue of marks.
313 * Pick a subset of marks from queue heads, all with the
314 * same group and set the report_mask for selected subset.
315 * Returns the report_mask of the selected subset.
317 static unsigned int fsnotify_iter_select_report_types(
318 struct fsnotify_iter_info *iter_info)
320 struct fsnotify_group *max_prio_group = NULL;
321 struct fsnotify_mark *mark;
324 /* Choose max prio group among groups of all queue heads */
325 fsnotify_foreach_obj_type(type) {
326 mark = iter_info->marks[type];
328 fsnotify_compare_groups(max_prio_group, mark->group) > 0)
329 max_prio_group = mark->group;
335 /* Set the report mask for marks from same group as max prio group */
336 iter_info->report_mask = 0;
337 fsnotify_foreach_obj_type(type) {
338 mark = iter_info->marks[type];
340 fsnotify_compare_groups(max_prio_group, mark->group) == 0)
341 fsnotify_iter_set_report_type(iter_info, type);
344 return iter_info->report_mask;
348 * Pop from iter_info multi head queue, the marks that were iterated in the
349 * current iteration step.
351 static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
355 fsnotify_foreach_obj_type(type) {
356 if (fsnotify_iter_should_report_type(iter_info, type))
357 iter_info->marks[type] =
358 fsnotify_next_mark(iter_info->marks[type]);
363 * This is the main call to fsnotify. The VFS calls into hook specific functions
364 * in linux/fsnotify.h. Those functions then in turn call here. Here will call
365 * out to all of the registered fsnotify_group. Those groups can then use the
366 * notification event in whatever means they feel necessary.
368 int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
369 const struct qstr *file_name, u32 cookie)
371 struct fsnotify_iter_info iter_info = {};
372 struct super_block *sb = to_tell->i_sb;
373 struct mount *mnt = NULL;
374 __u32 mnt_or_sb_mask = sb->s_fsnotify_mask;
376 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
378 if (data_is == FSNOTIFY_EVENT_PATH) {
379 mnt = real_mount(((const struct path *)data)->mnt);
380 mnt_or_sb_mask |= mnt->mnt_fsnotify_mask;
382 /* An event "on child" is not intended for a mount/sb mark */
383 if (mask & FS_EVENT_ON_CHILD)
387 * Optimization: srcu_read_lock() has a memory barrier which can
388 * be expensive. It protects walking the *_fsnotify_marks lists.
389 * However, if we do not walk the lists, we do not have to do
390 * SRCU because we have no references to any objects and do not
391 * need SRCU to keep them "alive".
393 if (!to_tell->i_fsnotify_marks && !sb->s_fsnotify_marks &&
394 (!mnt || !mnt->mnt_fsnotify_marks))
397 * if this is a modify event we may need to clear the ignored masks
398 * otherwise return if neither the inode nor the vfsmount/sb care about
399 * this type of event.
401 if (!(mask & FS_MODIFY) &&
402 !(test_mask & (to_tell->i_fsnotify_mask | mnt_or_sb_mask)))
405 iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
407 iter_info.marks[FSNOTIFY_OBJ_TYPE_INODE] =
408 fsnotify_first_mark(&to_tell->i_fsnotify_marks);
409 iter_info.marks[FSNOTIFY_OBJ_TYPE_SB] =
410 fsnotify_first_mark(&sb->s_fsnotify_marks);
412 iter_info.marks[FSNOTIFY_OBJ_TYPE_VFSMOUNT] =
413 fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
417 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
418 * ignore masks are properly reflected for mount/sb mark notifications.
419 * That's why this traversal is so complicated...
421 while (fsnotify_iter_select_report_types(&iter_info)) {
422 ret = send_to_group(to_tell, mask, data, data_is, cookie,
423 file_name, &iter_info);
425 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
428 fsnotify_iter_next(&iter_info);
432 srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
436 EXPORT_SYMBOL_GPL(fsnotify);
438 extern struct kmem_cache *fsnotify_mark_connector_cachep;
440 static __init int fsnotify_init(void)
444 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 25);
446 ret = init_srcu_struct(&fsnotify_mark_srcu);
448 panic("initializing fsnotify_mark_srcu");
450 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
455 core_initcall(fsnotify_init);