1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
4 */
5
6#include <linux/dcache.h>
7#include <linux/fs.h>
8#include <linux/gfp.h>
9#include <linux/init.h>
10#include <linux/module.h>
11#include <linux/mount.h>
12#include <linux/srcu.h>
13
14#include <linux/fsnotify_backend.h>
15#include "fsnotify.h"
16
17/*
18 * Clear all of the marks on an inode when it is being evicted from core
19 */
20void __fsnotify_inode_delete(struct inode *inode)
21{
22 fsnotify_clear_marks_by_inode(inode);
23}
24EXPORT_SYMBOL_GPL(__fsnotify_inode_delete);
25
26void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
27{
28 fsnotify_clear_marks_by_mount(mnt);
29}
30
31/**
32 * fsnotify_unmount_inodes - an sb is unmounting. handle any watched inodes.
33 * @sb: superblock being unmounted.
34 *
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.
37 */
38static void fsnotify_unmount_inodes(struct super_block *sb)
39{
40 struct inode *inode, *iput_inode = NULL;
41
42 spin_lock(lock: &sb->s_inode_list_lock);
43 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
44 /*
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.
48 */
49 spin_lock(lock: &inode->i_lock);
50 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) {
51 spin_unlock(lock: &inode->i_lock);
52 continue;
53 }
54
55 /*
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
62 * be sure.
63 */
64 if (!atomic_read(v: &inode->i_count)) {
65 spin_unlock(lock: &inode->i_lock);
66 continue;
67 }
68
69 __iget(inode);
70 spin_unlock(lock: &inode->i_lock);
71 spin_unlock(lock: &sb->s_inode_list_lock);
72
73 iput(iput_inode);
74
75 /* for each watch, send FS_UNMOUNT and then remove it */
76 fsnotify_inode(inode, FS_UNMOUNT);
77
78 fsnotify_inode_delete(inode);
79
80 iput_inode = inode;
81
82 cond_resched();
83 spin_lock(lock: &sb->s_inode_list_lock);
84 }
85 spin_unlock(lock: &sb->s_inode_list_lock);
86
87 iput(iput_inode);
88}
89
90void fsnotify_sb_delete(struct super_block *sb)
91{
92 fsnotify_unmount_inodes(sb);
93 fsnotify_clear_marks_by_sb(sb);
94 /* Wait for outstanding object references from connectors */
95 wait_var_event(&sb->s_fsnotify_connectors,
96 !atomic_long_read(&sb->s_fsnotify_connectors));
97}
98
99/*
100 * Given an inode, first check if we care what happens to our children. Inotify
101 * and dnotify both tell their parents about events. If we care about any event
102 * on a child we run all of our children and set a dentry flag saying that the
103 * parent cares. Thus when an event happens on a child it can quickly tell
104 * if there is a need to find a parent and send the event to the parent.
105 */
106void __fsnotify_update_child_dentry_flags(struct inode *inode)
107{
108 struct dentry *alias;
109 int watched;
110
111 if (!S_ISDIR(inode->i_mode))
112 return;
113
114 /* determine if the children should tell inode about their events */
115 watched = fsnotify_inode_watches_children(inode);
116
117 spin_lock(lock: &inode->i_lock);
118 /* run all of the dentries associated with this inode. Since this is a
119 * directory, there damn well better only be one item on this list */
120 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
121 struct dentry *child;
122
123 /* run all of the children of the original inode and fix their
124 * d_flags to indicate parental interest (their parent is the
125 * original inode) */
126 spin_lock(lock: &alias->d_lock);
127 list_for_each_entry(child, &alias->d_subdirs, d_child) {
128 if (!child->d_inode)
129 continue;
130
131 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
132 if (watched)
133 child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
134 else
135 child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
136 spin_unlock(lock: &child->d_lock);
137 }
138 spin_unlock(lock: &alias->d_lock);
139 }
140 spin_unlock(lock: &inode->i_lock);
141}
142
143/* Are inode/sb/mount interested in parent and name info with this event? */
144static bool fsnotify_event_needs_parent(struct inode *inode, struct mount *mnt,
145 __u32 mask)
146{
147 __u32 marks_mask = 0;
148
149 /* We only send parent/name to inode/sb/mount for events on non-dir */
150 if (mask & FS_ISDIR)
151 return false;
152
153 /*
154 * All events that are possible on child can also may be reported with
155 * parent/name info to inode/sb/mount. Otherwise, a watching parent
156 * could result in events reported with unexpected name info to sb/mount.
157 */
158 BUILD_BUG_ON(FS_EVENTS_POSS_ON_CHILD & ~FS_EVENTS_POSS_TO_PARENT);
159
160 /* Did either inode/sb/mount subscribe for events with parent/name? */
161 marks_mask |= fsnotify_parent_needed_mask(mask: inode->i_fsnotify_mask);
162 marks_mask |= fsnotify_parent_needed_mask(mask: inode->i_sb->s_fsnotify_mask);
163 if (mnt)
164 marks_mask |= fsnotify_parent_needed_mask(mask: mnt->mnt_fsnotify_mask);
165
166 /* Did they subscribe for this event with parent/name info? */
167 return mask & marks_mask;
168}
169
170/*
171 * Notify this dentry's parent about a child's events with child name info
172 * if parent is watching or if inode/sb/mount are interested in events with
173 * parent and name info.
174 *
175 * Notify only the child without name info if parent is not watching and
176 * inode/sb/mount are not interested in events with parent and name info.
177 */
178int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
179 int data_type)
180{
181 const struct path *path = fsnotify_data_path(data, data_type);
182 struct mount *mnt = path ? real_mount(mnt: path->mnt) : NULL;
183 struct inode *inode = d_inode(dentry);
184 struct dentry *parent;
185 bool parent_watched = dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED;
186 bool parent_needed, parent_interested;
187 __u32 p_mask;
188 struct inode *p_inode = NULL;
189 struct name_snapshot name;
190 struct qstr *file_name = NULL;
191 int ret = 0;
192
193 /*
194 * Do inode/sb/mount care about parent and name info on non-dir?
195 * Do they care about any event at all?
196 */
197 if (!inode->i_fsnotify_marks && !inode->i_sb->s_fsnotify_marks &&
198 (!mnt || !mnt->mnt_fsnotify_marks) && !parent_watched)
199 return 0;
200
201 parent = NULL;
202 parent_needed = fsnotify_event_needs_parent(inode, mnt, mask);
203 if (!parent_watched && !parent_needed)
204 goto notify;
205
206 /* Does parent inode care about events on children? */
207 parent = dget_parent(dentry);
208 p_inode = parent->d_inode;
209 p_mask = fsnotify_inode_watches_children(inode: p_inode);
210 if (unlikely(parent_watched && !p_mask))
211 __fsnotify_update_child_dentry_flags(inode: p_inode);
212
213 /*
214 * Include parent/name in notification either if some notification
215 * groups require parent info or the parent is interested in this event.
216 */
217 parent_interested = mask & p_mask & ALL_FSNOTIFY_EVENTS;
218 if (parent_needed || parent_interested) {
219 /* When notifying parent, child should be passed as data */
220 WARN_ON_ONCE(inode != fsnotify_data_inode(data, data_type));
221
222 /* Notify both parent and child with child name info */
223 take_dentry_name_snapshot(&name, dentry);
224 file_name = &name.name;
225 if (parent_interested)
226 mask |= FS_EVENT_ON_CHILD;
227 }
228
229notify:
230 ret = fsnotify(mask, data, data_type, dir: p_inode, name: file_name, inode, cookie: 0);
231
232 if (file_name)
233 release_dentry_name_snapshot(&name);
234 dput(parent);
235
236 return ret;
237}
238EXPORT_SYMBOL_GPL(__fsnotify_parent);
239
240static int fsnotify_handle_inode_event(struct fsnotify_group *group,
241 struct fsnotify_mark *inode_mark,
242 u32 mask, const void *data, int data_type,
243 struct inode *dir, const struct qstr *name,
244 u32 cookie)
245{
246 const struct path *path = fsnotify_data_path(data, data_type);
247 struct inode *inode = fsnotify_data_inode(data, data_type);
248 const struct fsnotify_ops *ops = group->ops;
249
250 if (WARN_ON_ONCE(!ops->handle_inode_event))
251 return 0;
252
253 if (WARN_ON_ONCE(!inode && !dir))
254 return 0;
255
256 if ((inode_mark->flags & FSNOTIFY_MARK_FLAG_EXCL_UNLINK) &&
257 path && d_unlinked(dentry: path->dentry))
258 return 0;
259
260 /* Check interest of this mark in case event was sent with two marks */
261 if (!(mask & inode_mark->mask & ALL_FSNOTIFY_EVENTS))
262 return 0;
263
264 return ops->handle_inode_event(inode_mark, mask, inode, dir, name, cookie);
265}
266
267static int fsnotify_handle_event(struct fsnotify_group *group, __u32 mask,
268 const void *data, int data_type,
269 struct inode *dir, const struct qstr *name,
270 u32 cookie, struct fsnotify_iter_info *iter_info)
271{
272 struct fsnotify_mark *inode_mark = fsnotify_iter_inode_mark(iter_info);
273 struct fsnotify_mark *parent_mark = fsnotify_iter_parent_mark(iter_info);
274 int ret;
275
276 if (WARN_ON_ONCE(fsnotify_iter_sb_mark(iter_info)) ||
277 WARN_ON_ONCE(fsnotify_iter_vfsmount_mark(iter_info)))
278 return 0;
279
280 /*
281 * For FS_RENAME, 'dir' is old dir and 'data' is new dentry.
282 * The only ->handle_inode_event() backend that supports FS_RENAME is
283 * dnotify, where it means file was renamed within same parent.
284 */
285 if (mask & FS_RENAME) {
286 struct dentry *moved = fsnotify_data_dentry(data, data_type);
287
288 if (dir != moved->d_parent->d_inode)
289 return 0;
290 }
291
292 if (parent_mark) {
293 ret = fsnotify_handle_inode_event(group, inode_mark: parent_mark, mask,
294 data, data_type, dir, name, cookie: 0);
295 if (ret)
296 return ret;
297 }
298
299 if (!inode_mark)
300 return 0;
301
302 if (mask & FS_EVENT_ON_CHILD) {
303 /*
304 * Some events can be sent on both parent dir and child marks
305 * (e.g. FS_ATTRIB). If both parent dir and child are
306 * watching, report the event once to parent dir with name (if
307 * interested) and once to child without name (if interested).
308 * The child watcher is expecting an event without a file name
309 * and without the FS_EVENT_ON_CHILD flag.
310 */
311 mask &= ~FS_EVENT_ON_CHILD;
312 dir = NULL;
313 name = NULL;
314 }
315
316 return fsnotify_handle_inode_event(group, inode_mark, mask, data, data_type,
317 dir, name, cookie);
318}
319
320static int send_to_group(__u32 mask, const void *data, int data_type,
321 struct inode *dir, const struct qstr *file_name,
322 u32 cookie, struct fsnotify_iter_info *iter_info)
323{
324 struct fsnotify_group *group = NULL;
325 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
326 __u32 marks_mask = 0;
327 __u32 marks_ignore_mask = 0;
328 bool is_dir = mask & FS_ISDIR;
329 struct fsnotify_mark *mark;
330 int type;
331
332 if (!iter_info->report_mask)
333 return 0;
334
335 /* clear ignored on inode modification */
336 if (mask & FS_MODIFY) {
337 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
338 if (!(mark->flags &
339 FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
340 mark->ignore_mask = 0;
341 }
342 }
343
344 /* Are any of the group marks interested in this event? */
345 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
346 group = mark->group;
347 marks_mask |= mark->mask;
348 marks_ignore_mask |=
349 fsnotify_effective_ignore_mask(mark, is_dir, iter_type: type);
350 }
351
352 pr_debug("%s: group=%p mask=%x marks_mask=%x marks_ignore_mask=%x data=%p data_type=%d dir=%p cookie=%d\n",
353 __func__, group, mask, marks_mask, marks_ignore_mask,
354 data, data_type, dir, cookie);
355
356 if (!(test_mask & marks_mask & ~marks_ignore_mask))
357 return 0;
358
359 if (group->ops->handle_event) {
360 return group->ops->handle_event(group, mask, data, data_type, dir,
361 file_name, cookie, iter_info);
362 }
363
364 return fsnotify_handle_event(group, mask, data, data_type, dir,
365 name: file_name, cookie, iter_info);
366}
367
368static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
369{
370 struct fsnotify_mark_connector *conn;
371 struct hlist_node *node = NULL;
372
373 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
374 if (conn)
375 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
376
377 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
378}
379
380static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
381{
382 struct hlist_node *node = NULL;
383
384 if (mark)
385 node = srcu_dereference(mark->obj_list.next,
386 &fsnotify_mark_srcu);
387
388 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
389}
390
391/*
392 * iter_info is a multi head priority queue of marks.
393 * Pick a subset of marks from queue heads, all with the same group
394 * and set the report_mask to a subset of the selected marks.
395 * Returns false if there are no more groups to iterate.
396 */
397static bool fsnotify_iter_select_report_types(
398 struct fsnotify_iter_info *iter_info)
399{
400 struct fsnotify_group *max_prio_group = NULL;
401 struct fsnotify_mark *mark;
402 int type;
403
404 /* Choose max prio group among groups of all queue heads */
405 fsnotify_foreach_iter_type(type) {
406 mark = iter_info->marks[type];
407 if (mark &&
408 fsnotify_compare_groups(a: max_prio_group, b: mark->group) > 0)
409 max_prio_group = mark->group;
410 }
411
412 if (!max_prio_group)
413 return false;
414
415 /* Set the report mask for marks from same group as max prio group */
416 iter_info->current_group = max_prio_group;
417 iter_info->report_mask = 0;
418 fsnotify_foreach_iter_type(type) {
419 mark = iter_info->marks[type];
420 if (mark && mark->group == iter_info->current_group) {
421 /*
422 * FSNOTIFY_ITER_TYPE_PARENT indicates that this inode
423 * is watching children and interested in this event,
424 * which is an event possible on child.
425 * But is *this mark* watching children?
426 */
427 if (type == FSNOTIFY_ITER_TYPE_PARENT &&
428 !(mark->mask & FS_EVENT_ON_CHILD) &&
429 !(fsnotify_ignore_mask(mark) & FS_EVENT_ON_CHILD))
430 continue;
431
432 fsnotify_iter_set_report_type(iter_info, iter_type: type);
433 }
434 }
435
436 return true;
437}
438
439/*
440 * Pop from iter_info multi head queue, the marks that belong to the group of
441 * current iteration step.
442 */
443static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
444{
445 struct fsnotify_mark *mark;
446 int type;
447
448 /*
449 * We cannot use fsnotify_foreach_iter_mark_type() here because we
450 * may need to advance a mark of type X that belongs to current_group
451 * but was not selected for reporting.
452 */
453 fsnotify_foreach_iter_type(type) {
454 mark = iter_info->marks[type];
455 if (mark && mark->group == iter_info->current_group)
456 iter_info->marks[type] =
457 fsnotify_next_mark(mark: iter_info->marks[type]);
458 }
459}
460
461/*
462 * fsnotify - This is the main call to fsnotify.
463 *
464 * The VFS calls into hook specific functions in linux/fsnotify.h.
465 * Those functions then in turn call here. Here will call out to all of the
466 * registered fsnotify_group. Those groups can then use the notification event
467 * in whatever means they feel necessary.
468 *
469 * @mask: event type and flags
470 * @data: object that event happened on
471 * @data_type: type of object for fanotify_data_XXX() accessors
472 * @dir: optional directory associated with event -
473 * if @file_name is not NULL, this is the directory that
474 * @file_name is relative to
475 * @file_name: optional file name associated with event
476 * @inode: optional inode associated with event -
477 * If @dir and @inode are both non-NULL, event may be
478 * reported to both.
479 * @cookie: inotify rename cookie
480 */
481int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
482 const struct qstr *file_name, struct inode *inode, u32 cookie)
483{
484 const struct path *path = fsnotify_data_path(data, data_type);
485 struct super_block *sb = fsnotify_data_sb(data, data_type);
486 struct fsnotify_iter_info iter_info = {};
487 struct mount *mnt = NULL;
488 struct inode *inode2 = NULL;
489 struct dentry *moved;
490 int inode2_type;
491 int ret = 0;
492 __u32 test_mask, marks_mask;
493
494 if (path)
495 mnt = real_mount(mnt: path->mnt);
496
497 if (!inode) {
498 /* Dirent event - report on TYPE_INODE to dir */
499 inode = dir;
500 /* For FS_RENAME, inode is old_dir and inode2 is new_dir */
501 if (mask & FS_RENAME) {
502 moved = fsnotify_data_dentry(data, data_type);
503 inode2 = moved->d_parent->d_inode;
504 inode2_type = FSNOTIFY_ITER_TYPE_INODE2;
505 }
506 } else if (mask & FS_EVENT_ON_CHILD) {
507 /*
508 * Event on child - report on TYPE_PARENT to dir if it is
509 * watching children and on TYPE_INODE to child.
510 */
511 inode2 = dir;
512 inode2_type = FSNOTIFY_ITER_TYPE_PARENT;
513 }
514
515 /*
516 * Optimization: srcu_read_lock() has a memory barrier which can
517 * be expensive. It protects walking the *_fsnotify_marks lists.
518 * However, if we do not walk the lists, we do not have to do
519 * SRCU because we have no references to any objects and do not
520 * need SRCU to keep them "alive".
521 */
522 if (!sb->s_fsnotify_marks &&
523 (!mnt || !mnt->mnt_fsnotify_marks) &&
524 (!inode || !inode->i_fsnotify_marks) &&
525 (!inode2 || !inode2->i_fsnotify_marks))
526 return 0;
527
528 marks_mask = sb->s_fsnotify_mask;
529 if (mnt)
530 marks_mask |= mnt->mnt_fsnotify_mask;
531 if (inode)
532 marks_mask |= inode->i_fsnotify_mask;
533 if (inode2)
534 marks_mask |= inode2->i_fsnotify_mask;
535
536
537 /*
538 * If this is a modify event we may need to clear some ignore masks.
539 * In that case, the object with ignore masks will have the FS_MODIFY
540 * event in its mask.
541 * Otherwise, return if none of the marks care about this type of event.
542 */
543 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
544 if (!(test_mask & marks_mask))
545 return 0;
546
547 iter_info.srcu_idx = srcu_read_lock(ssp: &fsnotify_mark_srcu);
548
549 iter_info.marks[FSNOTIFY_ITER_TYPE_SB] =
550 fsnotify_first_mark(connp: &sb->s_fsnotify_marks);
551 if (mnt) {
552 iter_info.marks[FSNOTIFY_ITER_TYPE_VFSMOUNT] =
553 fsnotify_first_mark(connp: &mnt->mnt_fsnotify_marks);
554 }
555 if (inode) {
556 iter_info.marks[FSNOTIFY_ITER_TYPE_INODE] =
557 fsnotify_first_mark(connp: &inode->i_fsnotify_marks);
558 }
559 if (inode2) {
560 iter_info.marks[inode2_type] =
561 fsnotify_first_mark(connp: &inode2->i_fsnotify_marks);
562 }
563
564 /*
565 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
566 * ignore masks are properly reflected for mount/sb mark notifications.
567 * That's why this traversal is so complicated...
568 */
569 while (fsnotify_iter_select_report_types(iter_info: &iter_info)) {
570 ret = send_to_group(mask, data, data_type, dir, file_name,
571 cookie, iter_info: &iter_info);
572
573 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
574 goto out;
575
576 fsnotify_iter_next(iter_info: &iter_info);
577 }
578 ret = 0;
579out:
580 srcu_read_unlock(ssp: &fsnotify_mark_srcu, idx: iter_info.srcu_idx);
581
582 return ret;
583}
584EXPORT_SYMBOL_GPL(fsnotify);
585
586static __init int fsnotify_init(void)
587{
588 int ret;
589
590 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 23);
591
592 ret = init_srcu_struct(&fsnotify_mark_srcu);
593 if (ret)
594 panic(fmt: "initializing fsnotify_mark_srcu");
595
596 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
597 SLAB_PANIC);
598
599 return 0;
600}
601core_initcall(fsnotify_init);
602

source code of linux/fs/notify/fsnotify.c