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
31void __fsnotify_mntns_delete(struct mnt_namespace *mntns)
32{
33 fsnotify_clear_marks_by_mntns(mntns);
34}
35
36/**
37 * fsnotify_unmount_inodes - an sb is unmounting. handle any watched inodes.
38 * @sb: superblock being unmounted.
39 *
40 * Called during unmount with no locks held, so needs to be safe against
41 * concurrent modifiers. We temporarily drop sb->s_inode_list_lock and CAN block.
42 */
43static void fsnotify_unmount_inodes(struct super_block *sb)
44{
45 struct inode *inode, *iput_inode = NULL;
46
47 spin_lock(lock: &sb->s_inode_list_lock);
48 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
49 /*
50 * We cannot __iget() an inode in state I_FREEING,
51 * I_WILL_FREE, or I_NEW which is fine because by that point
52 * the inode cannot have any associated watches.
53 */
54 spin_lock(lock: &inode->i_lock);
55 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) {
56 spin_unlock(lock: &inode->i_lock);
57 continue;
58 }
59
60 /*
61 * If i_count is zero, the inode cannot have any watches and
62 * doing an __iget/iput with SB_ACTIVE clear would actually
63 * evict all inodes with zero i_count from icache which is
64 * unnecessarily violent and may in fact be illegal to do.
65 * However, we should have been called /after/ evict_inodes
66 * removed all zero refcount inodes, in any case. Test to
67 * be sure.
68 */
69 if (!icount_read(inode)) {
70 spin_unlock(lock: &inode->i_lock);
71 continue;
72 }
73
74 __iget(inode);
75 spin_unlock(lock: &inode->i_lock);
76 spin_unlock(lock: &sb->s_inode_list_lock);
77
78 iput(iput_inode);
79
80 /* for each watch, send FS_UNMOUNT and then remove it */
81 fsnotify_inode(inode, FS_UNMOUNT);
82
83 fsnotify_inode_delete(inode);
84
85 iput_inode = inode;
86
87 cond_resched();
88 spin_lock(lock: &sb->s_inode_list_lock);
89 }
90 spin_unlock(lock: &sb->s_inode_list_lock);
91
92 iput(iput_inode);
93}
94
95void fsnotify_sb_delete(struct super_block *sb)
96{
97 struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
98
99 /* Were any marks ever added to any object on this sb? */
100 if (!sbinfo)
101 return;
102
103 fsnotify_unmount_inodes(sb);
104 fsnotify_clear_marks_by_sb(sb);
105 /* Wait for outstanding object references from connectors */
106 wait_var_event(fsnotify_sb_watched_objects(sb),
107 !atomic_long_read(fsnotify_sb_watched_objects(sb)));
108 WARN_ON(fsnotify_sb_has_priority_watchers(sb, FSNOTIFY_PRIO_CONTENT));
109 WARN_ON(fsnotify_sb_has_priority_watchers(sb,
110 FSNOTIFY_PRIO_PRE_CONTENT));
111}
112
113void fsnotify_sb_free(struct super_block *sb)
114{
115 kfree(objp: sb->s_fsnotify_info);
116}
117
118/*
119 * Given an inode, first check if we care what happens to our children. Inotify
120 * and dnotify both tell their parents about events. If we care about any event
121 * on a child we run all of our children and set a dentry flag saying that the
122 * parent cares. Thus when an event happens on a child it can quickly tell
123 * if there is a need to find a parent and send the event to the parent.
124 */
125void fsnotify_set_children_dentry_flags(struct inode *inode)
126{
127 struct dentry *alias;
128
129 if (!S_ISDIR(inode->i_mode))
130 return;
131
132 spin_lock(lock: &inode->i_lock);
133 /* run all of the dentries associated with this inode. Since this is a
134 * directory, there damn well better only be one item on this list */
135 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
136 struct dentry *child;
137
138 /* run all of the children of the original inode and fix their
139 * d_flags to indicate parental interest (their parent is the
140 * original inode) */
141 spin_lock(lock: &alias->d_lock);
142 hlist_for_each_entry(child, &alias->d_children, d_sib) {
143 if (!child->d_inode)
144 continue;
145
146 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
147 child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
148 spin_unlock(lock: &child->d_lock);
149 }
150 spin_unlock(lock: &alias->d_lock);
151 }
152 spin_unlock(lock: &inode->i_lock);
153}
154
155/*
156 * Lazily clear false positive PARENT_WATCHED flag for child whose parent had
157 * stopped watching children.
158 */
159static void fsnotify_clear_child_dentry_flag(struct inode *pinode,
160 struct dentry *dentry)
161{
162 spin_lock(lock: &dentry->d_lock);
163 /*
164 * d_lock is a sufficient barrier to prevent observing a non-watched
165 * parent state from before the fsnotify_set_children_dentry_flags()
166 * or fsnotify_update_flags() call that had set PARENT_WATCHED.
167 */
168 if (!fsnotify_inode_watches_children(inode: pinode))
169 dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
170 spin_unlock(lock: &dentry->d_lock);
171}
172
173/* Are inode/sb/mount interested in parent and name info with this event? */
174static bool fsnotify_event_needs_parent(struct inode *inode, __u32 mnt_mask,
175 __u32 mask)
176{
177 __u32 marks_mask = 0;
178
179 /* We only send parent/name to inode/sb/mount for events on non-dir */
180 if (mask & FS_ISDIR)
181 return false;
182
183 /*
184 * All events that are possible on child can also may be reported with
185 * parent/name info to inode/sb/mount. Otherwise, a watching parent
186 * could result in events reported with unexpected name info to sb/mount.
187 */
188 BUILD_BUG_ON(FS_EVENTS_POSS_ON_CHILD & ~FS_EVENTS_POSS_TO_PARENT);
189
190 /* Did either inode/sb/mount subscribe for events with parent/name? */
191 marks_mask |= fsnotify_parent_needed_mask(
192 READ_ONCE(inode->i_fsnotify_mask));
193 marks_mask |= fsnotify_parent_needed_mask(
194 READ_ONCE(inode->i_sb->s_fsnotify_mask));
195 marks_mask |= fsnotify_parent_needed_mask(mask: mnt_mask);
196
197 /* Did they subscribe for this event with parent/name info? */
198 return mask & marks_mask;
199}
200
201/* Are there any inode/mount/sb objects that watch for these events? */
202static inline __u32 fsnotify_object_watched(struct inode *inode, __u32 mnt_mask,
203 __u32 mask)
204{
205 __u32 marks_mask = READ_ONCE(inode->i_fsnotify_mask) | mnt_mask |
206 READ_ONCE(inode->i_sb->s_fsnotify_mask);
207
208 return mask & marks_mask & ALL_FSNOTIFY_EVENTS;
209}
210
211/* Report pre-content event with optional range info */
212int fsnotify_pre_content(const struct path *path, const loff_t *ppos,
213 size_t count)
214{
215 struct file_range range;
216
217 /* Report page aligned range only when pos is known */
218 if (!ppos)
219 return fsnotify_path(path, FS_PRE_ACCESS);
220
221 range.path = path;
222 range.pos = PAGE_ALIGN_DOWN(*ppos);
223 range.count = PAGE_ALIGN(*ppos + count) - range.pos;
224
225 return fsnotify_parent(dentry: path->dentry, FS_PRE_ACCESS, data: &range,
226 data_type: FSNOTIFY_EVENT_FILE_RANGE);
227}
228
229/*
230 * Notify this dentry's parent about a child's events with child name info
231 * if parent is watching or if inode/sb/mount are interested in events with
232 * parent and name info.
233 *
234 * Notify only the child without name info if parent is not watching and
235 * inode/sb/mount are not interested in events with parent and name info.
236 */
237int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
238 int data_type)
239{
240 const struct path *path = fsnotify_data_path(data, data_type);
241 __u32 mnt_mask = path ?
242 READ_ONCE(real_mount(path->mnt)->mnt_fsnotify_mask) : 0;
243 struct inode *inode = d_inode(dentry);
244 struct dentry *parent;
245 bool parent_watched = dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED;
246 bool parent_needed, parent_interested;
247 __u32 p_mask;
248 struct inode *p_inode = NULL;
249 struct name_snapshot name;
250 struct qstr *file_name = NULL;
251 int ret = 0;
252
253 /* Optimize the likely case of nobody watching this path */
254 if (likely(!parent_watched &&
255 !fsnotify_object_watched(inode, mnt_mask, mask)))
256 return 0;
257
258 parent = NULL;
259 parent_needed = fsnotify_event_needs_parent(inode, mnt_mask, mask);
260 if (!parent_watched && !parent_needed)
261 goto notify;
262
263 /* Does parent inode care about events on children? */
264 parent = dget_parent(dentry);
265 p_inode = parent->d_inode;
266 p_mask = fsnotify_inode_watches_children(inode: p_inode);
267 if (unlikely(parent_watched && !p_mask))
268 fsnotify_clear_child_dentry_flag(pinode: p_inode, dentry);
269
270 /*
271 * Include parent/name in notification either if some notification
272 * groups require parent info or the parent is interested in this event.
273 */
274 parent_interested = mask & p_mask & ALL_FSNOTIFY_EVENTS;
275 if (parent_needed || parent_interested) {
276 /* When notifying parent, child should be passed as data */
277 WARN_ON_ONCE(inode != fsnotify_data_inode(data, data_type));
278
279 /* Notify both parent and child with child name info */
280 take_dentry_name_snapshot(&name, dentry);
281 file_name = &name.name;
282 if (parent_interested)
283 mask |= FS_EVENT_ON_CHILD;
284 }
285
286notify:
287 ret = fsnotify(mask, data, data_type, dir: p_inode, name: file_name, inode, cookie: 0);
288
289 if (file_name)
290 release_dentry_name_snapshot(&name);
291 dput(parent);
292
293 return ret;
294}
295EXPORT_SYMBOL_GPL(__fsnotify_parent);
296
297static int fsnotify_handle_inode_event(struct fsnotify_group *group,
298 struct fsnotify_mark *inode_mark,
299 u32 mask, const void *data, int data_type,
300 struct inode *dir, const struct qstr *name,
301 u32 cookie)
302{
303 const struct path *path = fsnotify_data_path(data, data_type);
304 struct inode *inode = fsnotify_data_inode(data, data_type);
305 const struct fsnotify_ops *ops = group->ops;
306
307 if (WARN_ON_ONCE(!ops->handle_inode_event))
308 return 0;
309
310 if (WARN_ON_ONCE(!inode && !dir))
311 return 0;
312
313 if ((inode_mark->flags & FSNOTIFY_MARK_FLAG_EXCL_UNLINK) &&
314 path && d_unlinked(dentry: path->dentry))
315 return 0;
316
317 /* Check interest of this mark in case event was sent with two marks */
318 if (!(mask & inode_mark->mask & ALL_FSNOTIFY_EVENTS))
319 return 0;
320
321 return ops->handle_inode_event(inode_mark, mask, inode, dir, name, cookie);
322}
323
324static int fsnotify_handle_event(struct fsnotify_group *group, __u32 mask,
325 const void *data, int data_type,
326 struct inode *dir, const struct qstr *name,
327 u32 cookie, struct fsnotify_iter_info *iter_info)
328{
329 struct fsnotify_mark *inode_mark = fsnotify_iter_inode_mark(iter_info);
330 struct fsnotify_mark *parent_mark = fsnotify_iter_parent_mark(iter_info);
331 int ret;
332
333 if (WARN_ON_ONCE(fsnotify_iter_sb_mark(iter_info)) ||
334 WARN_ON_ONCE(fsnotify_iter_vfsmount_mark(iter_info)))
335 return 0;
336
337 /*
338 * For FS_RENAME, 'dir' is old dir and 'data' is new dentry.
339 * The only ->handle_inode_event() backend that supports FS_RENAME is
340 * dnotify, where it means file was renamed within same parent.
341 */
342 if (mask & FS_RENAME) {
343 struct dentry *moved = fsnotify_data_dentry(data, data_type);
344
345 if (dir != moved->d_parent->d_inode)
346 return 0;
347 }
348
349 if (parent_mark) {
350 ret = fsnotify_handle_inode_event(group, inode_mark: parent_mark, mask,
351 data, data_type, dir, name, cookie: 0);
352 if (ret)
353 return ret;
354 }
355
356 if (!inode_mark)
357 return 0;
358
359 /*
360 * Some events can be sent on both parent dir and child marks (e.g.
361 * FS_ATTRIB). If both parent dir and child are watching, report the
362 * event once to parent dir with name (if interested) and once to child
363 * without name (if interested).
364 *
365 * In any case regardless whether the parent is watching or not, the
366 * child watcher is expecting an event without the FS_EVENT_ON_CHILD
367 * flag. The file name is expected if and only if this is a directory
368 * event.
369 */
370 mask &= ~FS_EVENT_ON_CHILD;
371 if (!(mask & ALL_FSNOTIFY_DIRENT_EVENTS)) {
372 dir = NULL;
373 name = NULL;
374 }
375
376 return fsnotify_handle_inode_event(group, inode_mark, mask, data, data_type,
377 dir, name, cookie);
378}
379
380static int send_to_group(__u32 mask, const void *data, int data_type,
381 struct inode *dir, const struct qstr *file_name,
382 u32 cookie, struct fsnotify_iter_info *iter_info)
383{
384 struct fsnotify_group *group = NULL;
385 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
386 __u32 marks_mask = 0;
387 __u32 marks_ignore_mask = 0;
388 bool is_dir = mask & FS_ISDIR;
389 struct fsnotify_mark *mark;
390 int type;
391
392 if (!iter_info->report_mask)
393 return 0;
394
395 /* clear ignored on inode modification */
396 if (mask & FS_MODIFY) {
397 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
398 if (!(mark->flags &
399 FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
400 mark->ignore_mask = 0;
401 }
402 }
403
404 /* Are any of the group marks interested in this event? */
405 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
406 group = mark->group;
407 marks_mask |= mark->mask;
408 marks_ignore_mask |=
409 fsnotify_effective_ignore_mask(mark, is_dir, iter_type: type);
410 }
411
412 pr_debug("%s: group=%p mask=%x marks_mask=%x marks_ignore_mask=%x data=%p data_type=%d dir=%p cookie=%d\n",
413 __func__, group, mask, marks_mask, marks_ignore_mask,
414 data, data_type, dir, cookie);
415
416 if (!(test_mask & marks_mask & ~marks_ignore_mask))
417 return 0;
418
419 if (group->ops->handle_event) {
420 return group->ops->handle_event(group, mask, data, data_type, dir,
421 file_name, cookie, iter_info);
422 }
423
424 return fsnotify_handle_event(group, mask, data, data_type, dir,
425 name: file_name, cookie, iter_info);
426}
427
428static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector *const *connp)
429{
430 struct fsnotify_mark_connector *conn;
431 struct hlist_node *node = NULL;
432
433 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
434 if (conn)
435 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
436
437 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
438}
439
440static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
441{
442 struct hlist_node *node = NULL;
443
444 if (mark)
445 node = srcu_dereference(mark->obj_list.next,
446 &fsnotify_mark_srcu);
447
448 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
449}
450
451/*
452 * iter_info is a multi head priority queue of marks.
453 * Pick a subset of marks from queue heads, all with the same group
454 * and set the report_mask to a subset of the selected marks.
455 * Returns false if there are no more groups to iterate.
456 */
457static bool fsnotify_iter_select_report_types(
458 struct fsnotify_iter_info *iter_info)
459{
460 struct fsnotify_group *max_prio_group = NULL;
461 struct fsnotify_mark *mark;
462 int type;
463
464 /* Choose max prio group among groups of all queue heads */
465 fsnotify_foreach_iter_type(type) {
466 mark = iter_info->marks[type];
467 if (mark &&
468 fsnotify_compare_groups(a: max_prio_group, b: mark->group) > 0)
469 max_prio_group = mark->group;
470 }
471
472 if (!max_prio_group)
473 return false;
474
475 /* Set the report mask for marks from same group as max prio group */
476 iter_info->current_group = max_prio_group;
477 iter_info->report_mask = 0;
478 fsnotify_foreach_iter_type(type) {
479 mark = iter_info->marks[type];
480 if (mark && mark->group == iter_info->current_group) {
481 /*
482 * FSNOTIFY_ITER_TYPE_PARENT indicates that this inode
483 * is watching children and interested in this event,
484 * which is an event possible on child.
485 * But is *this mark* watching children?
486 */
487 if (type == FSNOTIFY_ITER_TYPE_PARENT &&
488 !(mark->mask & FS_EVENT_ON_CHILD) &&
489 !(fsnotify_ignore_mask(mark) & FS_EVENT_ON_CHILD))
490 continue;
491
492 fsnotify_iter_set_report_type(iter_info, iter_type: type);
493 }
494 }
495
496 return true;
497}
498
499/*
500 * Pop from iter_info multi head queue, the marks that belong to the group of
501 * current iteration step.
502 */
503static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
504{
505 struct fsnotify_mark *mark;
506 int type;
507
508 /*
509 * We cannot use fsnotify_foreach_iter_mark_type() here because we
510 * may need to advance a mark of type X that belongs to current_group
511 * but was not selected for reporting.
512 */
513 fsnotify_foreach_iter_type(type) {
514 mark = iter_info->marks[type];
515 if (mark && mark->group == iter_info->current_group)
516 iter_info->marks[type] =
517 fsnotify_next_mark(mark: iter_info->marks[type]);
518 }
519}
520
521/*
522 * fsnotify - This is the main call to fsnotify.
523 *
524 * The VFS calls into hook specific functions in linux/fsnotify.h.
525 * Those functions then in turn call here. Here will call out to all of the
526 * registered fsnotify_group. Those groups can then use the notification event
527 * in whatever means they feel necessary.
528 *
529 * @mask: event type and flags
530 * @data: object that event happened on
531 * @data_type: type of object for fanotify_data_XXX() accessors
532 * @dir: optional directory associated with event -
533 * if @file_name is not NULL, this is the directory that
534 * @file_name is relative to
535 * @file_name: optional file name associated with event
536 * @inode: optional inode associated with event -
537 * If @dir and @inode are both non-NULL, event may be
538 * reported to both.
539 * @cookie: inotify rename cookie
540 */
541int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
542 const struct qstr *file_name, struct inode *inode, u32 cookie)
543{
544 const struct path *path = fsnotify_data_path(data, data_type);
545 struct super_block *sb = fsnotify_data_sb(data, data_type);
546 const struct fsnotify_mnt *mnt_data = fsnotify_data_mnt(data, data_type);
547 struct fsnotify_sb_info *sbinfo = sb ? fsnotify_sb_info(sb) : NULL;
548 struct fsnotify_iter_info iter_info = {};
549 struct mount *mnt = NULL;
550 struct inode *inode2 = NULL;
551 struct dentry *moved;
552 int inode2_type;
553 int ret = 0;
554 __u32 test_mask, marks_mask = 0;
555
556 if (path)
557 mnt = real_mount(mnt: path->mnt);
558
559 if (!inode) {
560 /* Dirent event - report on TYPE_INODE to dir */
561 inode = dir;
562 /* For FS_RENAME, inode is old_dir and inode2 is new_dir */
563 if (mask & FS_RENAME) {
564 moved = fsnotify_data_dentry(data, data_type);
565 inode2 = moved->d_parent->d_inode;
566 inode2_type = FSNOTIFY_ITER_TYPE_INODE2;
567 }
568 } else if (mask & FS_EVENT_ON_CHILD) {
569 /*
570 * Event on child - report on TYPE_PARENT to dir if it is
571 * watching children and on TYPE_INODE to child.
572 */
573 inode2 = dir;
574 inode2_type = FSNOTIFY_ITER_TYPE_PARENT;
575 }
576
577 /*
578 * Optimization: srcu_read_lock() has a memory barrier which can
579 * be expensive. It protects walking the *_fsnotify_marks lists.
580 * However, if we do not walk the lists, we do not have to do
581 * SRCU because we have no references to any objects and do not
582 * need SRCU to keep them "alive".
583 */
584 if ((!sbinfo || !sbinfo->sb_marks) &&
585 (!mnt || !mnt->mnt_fsnotify_marks) &&
586 (!inode || !inode->i_fsnotify_marks) &&
587 (!inode2 || !inode2->i_fsnotify_marks) &&
588 (!mnt_data || !mnt_data->ns->n_fsnotify_marks))
589 return 0;
590
591 if (sb)
592 marks_mask |= READ_ONCE(sb->s_fsnotify_mask);
593 if (mnt)
594 marks_mask |= READ_ONCE(mnt->mnt_fsnotify_mask);
595 if (inode)
596 marks_mask |= READ_ONCE(inode->i_fsnotify_mask);
597 if (inode2)
598 marks_mask |= READ_ONCE(inode2->i_fsnotify_mask);
599 if (mnt_data)
600 marks_mask |= READ_ONCE(mnt_data->ns->n_fsnotify_mask);
601
602 /*
603 * If this is a modify event we may need to clear some ignore masks.
604 * In that case, the object with ignore masks will have the FS_MODIFY
605 * event in its mask.
606 * Otherwise, return if none of the marks care about this type of event.
607 */
608 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
609 if (!(test_mask & marks_mask))
610 return 0;
611
612 iter_info.srcu_idx = srcu_read_lock(ssp: &fsnotify_mark_srcu);
613
614 if (sbinfo) {
615 iter_info.marks[FSNOTIFY_ITER_TYPE_SB] =
616 fsnotify_first_mark(connp: &sbinfo->sb_marks);
617 }
618 if (mnt) {
619 iter_info.marks[FSNOTIFY_ITER_TYPE_VFSMOUNT] =
620 fsnotify_first_mark(connp: &mnt->mnt_fsnotify_marks);
621 }
622 if (inode) {
623 iter_info.marks[FSNOTIFY_ITER_TYPE_INODE] =
624 fsnotify_first_mark(connp: &inode->i_fsnotify_marks);
625 }
626 if (inode2) {
627 iter_info.marks[inode2_type] =
628 fsnotify_first_mark(connp: &inode2->i_fsnotify_marks);
629 }
630 if (mnt_data) {
631 iter_info.marks[FSNOTIFY_ITER_TYPE_MNTNS] =
632 fsnotify_first_mark(connp: &mnt_data->ns->n_fsnotify_marks);
633 }
634
635 /*
636 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
637 * ignore masks are properly reflected for mount/sb mark notifications.
638 * That's why this traversal is so complicated...
639 */
640 while (fsnotify_iter_select_report_types(iter_info: &iter_info)) {
641 ret = send_to_group(mask, data, data_type, dir, file_name,
642 cookie, iter_info: &iter_info);
643
644 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
645 goto out;
646
647 fsnotify_iter_next(iter_info: &iter_info);
648 }
649 ret = 0;
650out:
651 srcu_read_unlock(ssp: &fsnotify_mark_srcu, idx: iter_info.srcu_idx);
652
653 return ret;
654}
655EXPORT_SYMBOL_GPL(fsnotify);
656
657#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
658/*
659 * At open time we check fsnotify_sb_has_priority_watchers(), call the open perm
660 * hook and set the FMODE_NONOTIFY_ mode bits accordignly.
661 * Later, fsnotify permission hooks do not check if there are permission event
662 * watches, but that there were permission event watches at open time.
663 */
664int fsnotify_open_perm_and_set_mode(struct file *file)
665{
666 struct dentry *dentry = file->f_path.dentry, *parent;
667 struct super_block *sb = dentry->d_sb;
668 __u32 mnt_mask, p_mask = 0;
669
670 /* Is it a file opened by fanotify? */
671 if (FMODE_FSNOTIFY_NONE(file->f_mode))
672 return 0;
673
674 /*
675 * Permission events is a super set of pre-content events, so if there
676 * are no permission event watchers, there are also no pre-content event
677 * watchers and this is implied from the single FMODE_NONOTIFY_PERM bit.
678 */
679 if (likely(!fsnotify_sb_has_priority_watchers(sb,
680 FSNOTIFY_PRIO_CONTENT))) {
681 file_set_fsnotify_mode(file, FMODE_NONOTIFY_PERM);
682 return 0;
683 }
684
685 /*
686 * OK, there are some permission event watchers. Check if anybody is
687 * watching for permission events on *this* file.
688 */
689 mnt_mask = READ_ONCE(real_mount(file->f_path.mnt)->mnt_fsnotify_mask);
690 p_mask = fsnotify_object_watched(d_inode(dentry), mnt_mask,
691 ALL_FSNOTIFY_PERM_EVENTS);
692 if (dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED) {
693 parent = dget_parent(dentry);
694 p_mask |= fsnotify_inode_watches_children(d_inode(parent));
695 dput(parent);
696 }
697
698 /*
699 * Legacy FAN_ACCESS_PERM events have very high performance overhead,
700 * so unlikely to be used in the wild. If they are used there will be
701 * no optimizations at all.
702 */
703 if (unlikely(p_mask & FS_ACCESS_PERM)) {
704 /* Enable all permission and pre-content events */
705 file_set_fsnotify_mode(file, 0);
706 goto open_perm;
707 }
708
709 /*
710 * Pre-content events are only supported on regular files.
711 * If there are pre-content event watchers and no permission access
712 * watchers, set FMODE_NONOTIFY | FMODE_NONOTIFY_PERM to indicate that.
713 * That is the common case with HSM service.
714 */
715 if (d_is_reg(dentry) && (p_mask & FSNOTIFY_PRE_CONTENT_EVENTS)) {
716 file_set_fsnotify_mode(file, FMODE_NONOTIFY |
717 FMODE_NONOTIFY_PERM);
718 goto open_perm;
719 }
720
721 /* Nobody watching permission and pre-content events on this file */
722 file_set_fsnotify_mode(file, FMODE_NONOTIFY_PERM);
723
724open_perm:
725 /*
726 * Send open perm events depending on object masks and regardless of
727 * FMODE_NONOTIFY_PERM.
728 */
729 if (file->f_flags & __FMODE_EXEC && p_mask & FS_OPEN_EXEC_PERM) {
730 int ret = fsnotify_path(&file->f_path, FS_OPEN_EXEC_PERM);
731
732 if (ret)
733 return ret;
734 }
735
736 if (p_mask & FS_OPEN_PERM)
737 return fsnotify_path(&file->f_path, FS_OPEN_PERM);
738
739 return 0;
740}
741#endif
742
743void fsnotify_mnt(__u32 mask, struct mnt_namespace *ns, struct vfsmount *mnt)
744{
745 struct fsnotify_mnt data = {
746 .ns = ns,
747 .mnt_id = real_mount(mnt)->mnt_id_unique,
748 };
749
750 if (WARN_ON_ONCE(!ns))
751 return;
752
753 /*
754 * This is an optimization as well as making sure fsnotify_init() has
755 * been called.
756 */
757 if (!ns->n_fsnotify_marks)
758 return;
759
760 fsnotify(mask, &data, FSNOTIFY_EVENT_MNT, NULL, NULL, NULL, 0);
761}
762
763static __init int fsnotify_init(void)
764{
765 int ret;
766
767 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 26);
768
769 ret = init_srcu_struct(ssp: &fsnotify_mark_srcu);
770 if (ret)
771 panic(fmt: "initializing fsnotify_mark_srcu");
772
773 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
774 SLAB_PANIC);
775
776 return 0;
777}
778core_initcall(fsnotify_init);
779