projects/linux·blob master · refs

733 lines · 21.9 KiB

1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <[email protected]>
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#define CREATE_TRACE_POINTS
18#include <trace/events/fsnotify.h>
19
20/*
21 * Clear all of the marks on an inode when it is being evicted from core
22 */
23void __fsnotify_inode_delete(struct inode *inode)
24{
25 fsnotify_clear_marks_by_inode(inode);
26}
27EXPORT_SYMBOL_GPL(__fsnotify_inode_delete);
28
29void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
30{
31 fsnotify_clear_marks_by_mount(mnt);
32}
33
34void __fsnotify_mntns_delete(struct mnt_namespace *mntns)
35{
36 fsnotify_clear_marks_by_mntns(mntns);
37}
38
39void fsnotify_sb_delete(struct super_block *sb)
40{
41 struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
42
43 /* Were any marks ever added to any object on this sb? */
44 if (!sbinfo)
45 return;
46
47 fsnotify_unmount_inodes(sbinfo);
48 fsnotify_clear_marks_by_sb(sb);
49 /* Wait for outstanding object references from connectors */
50 wait_var_event(fsnotify_sb_watched_objects(sb),
51 !atomic_long_read(fsnotify_sb_watched_objects(sb)));
52 WARN_ON(fsnotify_sb_has_priority_watchers(sb, FSNOTIFY_PRIO_CONTENT));
53 WARN_ON(fsnotify_sb_has_priority_watchers(sb,
54 FSNOTIFY_PRIO_PRE_CONTENT));
55}
56
57void fsnotify_sb_free(struct super_block *sb)
58{
59 if (sb->s_fsnotify_info) {
60 WARN_ON_ONCE(!list_empty(&sb->s_fsnotify_info->inode_conn_list));
61 kfree(sb->s_fsnotify_info);
62 }
63}
64
65/*
66 * Given an inode, first check if we care what happens to our children. Inotify
67 * and dnotify both tell their parents about events. If we care about any event
68 * on a child we run all of our children and set a dentry flag saying that the
69 * parent cares. Thus when an event happens on a child it can quickly tell
70 * if there is a need to find a parent and send the event to the parent.
71 */
72void fsnotify_set_children_dentry_flags(struct inode *inode)
73{
74 struct dentry *alias;
75
76 if (!S_ISDIR(inode->i_mode))
77 return;
78
79 spin_lock(&inode->i_lock);
80 /* run all of the dentries associated with this inode. Since this is a
81 * directory, there damn well better only be one item on this list */
82 for_each_alias(alias, inode) {
83 struct dentry *child;
84
85 /* run all of the children of the original inode and fix their
86 * d_flags to indicate parental interest (their parent is the
87 * original inode) */
88 spin_lock(&alias->d_lock);
89 hlist_for_each_entry(child, &alias->d_children, d_sib) {
90 if (!child->d_inode)
91 continue;
92
93 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
94 child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
95 spin_unlock(&child->d_lock);
96 }
97 spin_unlock(&alias->d_lock);
98 }
99 spin_unlock(&inode->i_lock);
100}
101
102/*
103 * Lazily clear false positive PARENT_WATCHED flag for child whose parent had
104 * stopped watching children.
105 */
106static void fsnotify_clear_child_dentry_flag(struct inode *pinode,
107 struct dentry *dentry)
108{
109 spin_lock(&dentry->d_lock);
110 /*
111 * d_lock is a sufficient barrier to prevent observing a non-watched
112 * parent state from before the fsnotify_set_children_dentry_flags()
113 * or fsnotify_update_flags() call that had set PARENT_WATCHED.
114 */
115 if (!fsnotify_inode_watches_children(pinode))
116 dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
117 spin_unlock(&dentry->d_lock);
118}
119
120/* Are inode/sb/mount interested in parent and name info with this event? */
121static bool fsnotify_event_needs_parent(struct inode *inode, __u32 mnt_mask,
122 __u32 mask)
123{
124 __u32 marks_mask = 0;
125
126 /* We only send parent/name to inode/sb/mount for events on non-dir */
127 if (mask & FS_ISDIR)
128 return false;
129
130 /*
131 * All events that are possible on child can also may be reported with
132 * parent/name info to inode/sb/mount. Otherwise, a watching parent
133 * could result in events reported with unexpected name info to sb/mount.
134 */
135 BUILD_BUG_ON(FS_EVENTS_POSS_ON_CHILD & ~FS_EVENTS_POSS_TO_PARENT);
136
137 /* Did either inode/sb/mount subscribe for events with parent/name? */
138 marks_mask |= fsnotify_parent_needed_mask(
139 READ_ONCE(inode->i_fsnotify_mask));
140 marks_mask |= fsnotify_parent_needed_mask(
141 READ_ONCE(inode->i_sb->s_fsnotify_mask));
142 marks_mask |= fsnotify_parent_needed_mask(mnt_mask);
143
144 /* Did they subscribe for this event with parent/name info? */
145 return mask & marks_mask;
146}
147
148/* Are there any inode/mount/sb objects that watch for these events? */
149static inline __u32 fsnotify_object_watched(struct inode *inode, __u32 mnt_mask,
150 __u32 mask)
151{
152 __u32 marks_mask = READ_ONCE(inode->i_fsnotify_mask) | mnt_mask |
153 READ_ONCE(inode->i_sb->s_fsnotify_mask);
154
155 return mask & marks_mask & ALL_FSNOTIFY_EVENTS;
156}
157
158/* Report pre-content event with optional range info */
159int fsnotify_pre_content(const struct path *path, const loff_t *ppos,
160 size_t count)
161{
162 struct file_range range;
163
164 /* Report page aligned range only when pos is known */
165 if (!ppos)
166 return fsnotify_path(path, FS_PRE_ACCESS);
167
168 range.path = path;
169 range.pos = PAGE_ALIGN_DOWN(*ppos);
170 range.count = PAGE_ALIGN(*ppos + count) - range.pos;
171
172 return fsnotify_parent(path->dentry, FS_PRE_ACCESS, &range,
173 FSNOTIFY_EVENT_FILE_RANGE);
174}
175
176/*
177 * Notify this dentry's parent about a child's events with child name info
178 * if parent is watching or if inode/sb/mount are interested in events with
179 * parent and name info.
180 *
181 * Notify only the child without name info if parent is not watching and
182 * inode/sb/mount are not interested in events with parent and name info.
183 */
184int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
185 int data_type)
186{
187 const struct path *path = fsnotify_data_path(data, data_type);
188 __u32 mnt_mask = path ?
189 READ_ONCE(real_mount(path->mnt)->mnt_fsnotify_mask) : 0;
190 struct inode *inode = d_inode(dentry);
191 struct dentry *parent;
192 bool parent_watched = dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED;
193 bool parent_needed, parent_interested;
194 __u32 p_mask;
195 struct inode *p_inode = NULL;
196 struct name_snapshot name;
197 struct qstr *file_name = NULL;
198 int ret = 0;
199
200 /* Optimize the likely case of nobody watching this path */
201 if (likely(!parent_watched &&
202 !fsnotify_object_watched(inode, mnt_mask, mask)))
203 return 0;
204
205 parent = NULL;
206 parent_needed = fsnotify_event_needs_parent(inode, mnt_mask, mask);
207 if (!parent_watched && !parent_needed)
208 goto notify;
209
210 /* Does parent inode care about events on children? */
211 parent = dget_parent(dentry);
212 p_inode = parent->d_inode;
213 p_mask = fsnotify_inode_watches_children(p_inode);
214 if (unlikely(parent_watched && !p_mask))
215 fsnotify_clear_child_dentry_flag(p_inode, dentry);
216
217 /*
218 * Include parent/name in notification either if some notification
219 * groups require parent info or the parent is interested in this event.
220 * The parent interest in ACCESS/MODIFY events does not apply to special
221 * files, where read/write are not on the filesystem of the parent and
222 * events can provide an undesirable side-channel for information
223 * exfiltration.
224 */
225 parent_interested = mask & p_mask & ALL_FSNOTIFY_EVENTS &&
226 !(data_type == FSNOTIFY_EVENT_PATH &&
227 d_is_special(dentry) &&
228 (mask & (FS_ACCESS | FS_MODIFY)));
229 if (parent_needed || parent_interested) {
230 /* When notifying parent, child should be passed as data */
231 WARN_ON_ONCE(inode != fsnotify_data_inode(data, data_type));
232
233 /* Notify both parent and child with child name info */
234 take_dentry_name_snapshot(&name, dentry);
235 file_name = &name.name;
236 if (parent_interested)
237 mask |= FS_EVENT_ON_CHILD;
238 }
239
240notify:
241 ret = fsnotify(mask, data, data_type, p_inode, file_name, inode, 0);
242
243 if (file_name)
244 release_dentry_name_snapshot(&name);
245 dput(parent);
246
247 return ret;
248}
249EXPORT_SYMBOL_GPL(__fsnotify_parent);
250
251static int fsnotify_handle_inode_event(struct fsnotify_group *group,
252 struct fsnotify_mark *inode_mark,
253 u32 mask, const void *data, int data_type,
254 struct inode *dir, const struct qstr *name,
255 u32 cookie)
256{
257 const struct path *path = fsnotify_data_path(data, data_type);
258 struct inode *inode = fsnotify_data_inode(data, data_type);
259 const struct fsnotify_ops *ops = group->ops;
260
261 if (WARN_ON_ONCE(!ops->handle_inode_event))
262 return 0;
263
264 if (WARN_ON_ONCE(!inode && !dir))
265 return 0;
266
267 if ((inode_mark->flags & FSNOTIFY_MARK_FLAG_EXCL_UNLINK) &&
268 path && d_unlinked(path->dentry))
269 return 0;
270
271 /* Check interest of this mark in case event was sent with two marks */
272 if (!(mask & inode_mark->mask & ALL_FSNOTIFY_EVENTS))
273 return 0;
274
275 return ops->handle_inode_event(inode_mark, mask, inode, dir, name, cookie);
276}
277
278static int fsnotify_handle_event(struct fsnotify_group *group, __u32 mask,
279 const void *data, int data_type,
280 struct inode *dir, const struct qstr *name,
281 u32 cookie, struct fsnotify_iter_info *iter_info)
282{
283 struct fsnotify_mark *inode_mark = fsnotify_iter_inode_mark(iter_info);
284 struct fsnotify_mark *parent_mark = fsnotify_iter_parent_mark(iter_info);
285 int ret;
286
287 if (WARN_ON_ONCE(fsnotify_iter_sb_mark(iter_info)) ||
288 WARN_ON_ONCE(fsnotify_iter_vfsmount_mark(iter_info)))
289 return 0;
290
291 /*
292 * For FS_RENAME, 'dir' is old dir and 'data' is new dentry.
293 * The only ->handle_inode_event() backend that supports FS_RENAME is
294 * dnotify, where it means file was renamed within same parent.
295 */
296 if (mask & FS_RENAME) {
297 struct dentry *moved = fsnotify_data_dentry(data, data_type);
298
299 if (dir != moved->d_parent->d_inode)
300 return 0;
301 }
302
303 if (parent_mark) {
304 ret = fsnotify_handle_inode_event(group, parent_mark, mask,
305 data, data_type, dir, name, 0);
306 if (ret)
307 return ret;
308 }
309
310 if (!inode_mark)
311 return 0;
312
313 /*
314 * Some events can be sent on both parent dir and child marks (e.g.
315 * FS_ATTRIB). If both parent dir and child are watching, report the
316 * event once to parent dir with name (if interested) and once to child
317 * without name (if interested).
318 *
319 * In any case regardless whether the parent is watching or not, the
320 * child watcher is expecting an event without the FS_EVENT_ON_CHILD
321 * flag. The file name is expected if and only if this is a directory
322 * event.
323 */
324 mask &= ~FS_EVENT_ON_CHILD;
325 if (!(mask & ALL_FSNOTIFY_DIRENT_EVENTS)) {
326 dir = NULL;
327 name = NULL;
328 }
329
330 return fsnotify_handle_inode_event(group, inode_mark, mask, data, data_type,
331 dir, name, cookie);
332}
333
334static int send_to_group(__u32 mask, const void *data, int data_type,
335 struct inode *dir, const struct qstr *file_name,
336 u32 cookie, struct fsnotify_iter_info *iter_info)
337{
338 struct fsnotify_group *group = NULL;
339 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
340 __u32 marks_mask = 0;
341 __u32 marks_ignore_mask = 0;
342 bool is_dir = mask & FS_ISDIR;
343 struct fsnotify_mark *mark;
344 int type;
345
346 if (!iter_info->report_mask)
347 return 0;
348
349 /* clear ignored on inode modification */
350 if (mask & FS_MODIFY) {
351 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
352 if (!(mark->flags &
353 FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
354 mark->ignore_mask = 0;
355 }
356 }
357
358 /* Are any of the group marks interested in this event? */
359 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
360 group = mark->group;
361 marks_mask |= mark->mask;
362 marks_ignore_mask |=
363 fsnotify_effective_ignore_mask(mark, is_dir, type);
364 }
365
366 pr_debug("%s: group=%p mask=%x marks_mask=%x marks_ignore_mask=%x data=%p data_type=%d dir=%p cookie=%d\n",
367 __func__, group, mask, marks_mask, marks_ignore_mask,
368 data, data_type, dir, cookie);
369
370 if (!(test_mask & marks_mask & ~marks_ignore_mask))
371 return 0;
372
373 if (group->ops->handle_event) {
374 return group->ops->handle_event(group, mask, data, data_type, dir,
375 file_name, cookie, iter_info);
376 }
377
378 return fsnotify_handle_event(group, mask, data, data_type, dir,
379 file_name, cookie, iter_info);
380}
381
382static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector *const *connp)
383{
384 struct fsnotify_mark_connector *conn;
385 struct hlist_node *node = NULL;
386
387 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
388 if (conn)
389 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
390
391 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
392}
393
394struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
395{
396 struct hlist_node *node = NULL;
397
398 if (mark)
399 node = srcu_dereference(mark->obj_list.next,
400 &fsnotify_mark_srcu);
401
402 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
403}
404
405/*
406 * iter_info is a multi head priority queue of marks.
407 * Pick a subset of marks from queue heads, all with the same group
408 * and set the report_mask to a subset of the selected marks.
409 * Returns false if there are no more groups to iterate.
410 */
411static bool fsnotify_iter_select_report_types(
412 struct fsnotify_iter_info *iter_info)
413{
414 struct fsnotify_group *max_prio_group = NULL;
415 struct fsnotify_mark *mark;
416 int type;
417
418 /* Choose max prio group among groups of all queue heads */
419 fsnotify_foreach_iter_type(type) {
420 mark = iter_info->marks[type];
421 if (mark &&
422 fsnotify_compare_groups(max_prio_group, mark->group) > 0)
423 max_prio_group = mark->group;
424 }
425
426 if (!max_prio_group)
427 return false;
428
429 /* Set the report mask for marks from same group as max prio group */
430 iter_info->current_group = max_prio_group;
431 iter_info->report_mask = 0;
432 fsnotify_foreach_iter_type(type) {
433 mark = iter_info->marks[type];
434 if (mark && mark->group == iter_info->current_group) {
435 /*
436 * FSNOTIFY_ITER_TYPE_PARENT indicates that this inode
437 * is watching children and interested in this event,
438 * which is an event possible on child.
439 * But is *this mark* watching children?
440 */
441 if (type == FSNOTIFY_ITER_TYPE_PARENT &&
442 !(mark->mask & FS_EVENT_ON_CHILD) &&
443 !(fsnotify_ignore_mask(mark) & FS_EVENT_ON_CHILD))
444 continue;
445
446 fsnotify_iter_set_report_type(iter_info, type);
447 }
448 }
449
450 return true;
451}
452
453/*
454 * Pop from iter_info multi head queue, the marks that belong to the group of
455 * current iteration step.
456 */
457static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
458{
459 struct fsnotify_mark *mark;
460 int type;
461
462 /*
463 * We cannot use fsnotify_foreach_iter_mark_type() here because we
464 * may need to advance a mark of type X that belongs to current_group
465 * but was not selected for reporting.
466 */
467 fsnotify_foreach_iter_type(type) {
468 mark = iter_info->marks[type];
469 if (mark && mark->group == iter_info->current_group)
470 iter_info->marks[type] =
471 fsnotify_next_mark(iter_info->marks[type]);
472 }
473}
474
475/*
476 * fsnotify - This is the main call to fsnotify.
477 *
478 * The VFS calls into hook specific functions in linux/fsnotify.h.
479 * Those functions then in turn call here. Here will call out to all of the
480 * registered fsnotify_group. Those groups can then use the notification event
481 * in whatever means they feel necessary.
482 *
483 * @mask: event type and flags
484 * @data: object that event happened on
485 * @data_type: type of object for fanotify_data_XXX() accessors
486 * @dir: optional directory associated with event -
487 * if @file_name is not NULL, this is the directory that
488 * @file_name is relative to
489 * @file_name: optional file name associated with event
490 * @inode: optional inode associated with event -
491 * If @dir and @inode are both non-NULL, event may be
492 * reported to both.
493 * @cookie: inotify rename cookie
494 */
495int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
496 const struct qstr *file_name, struct inode *inode, u32 cookie)
497{
498 const struct path *path = fsnotify_data_path(data, data_type);
499 struct super_block *sb = fsnotify_data_sb(data, data_type);
500 const struct fsnotify_mnt *mnt_data = fsnotify_data_mnt(data, data_type);
501 struct fsnotify_sb_info *sbinfo = sb ? fsnotify_sb_info(sb) : NULL;
502 struct fsnotify_iter_info iter_info = {};
503 struct mount *mnt = NULL;
504 struct inode *inode2 = NULL;
505 struct dentry *moved;
506 int inode2_type;
507 int ret = 0;
508 __u32 test_mask, marks_mask = 0;
509
510 trace_fsnotify(mask, data, data_type, dir, file_name, inode, cookie);
511
512 if (path)
513 mnt = real_mount(path->mnt);
514
515 if (!inode) {
516 /* Dirent event - report on TYPE_INODE to dir */
517 inode = dir;
518 /* For FS_RENAME, inode is old_dir and inode2 is new_dir */
519 if (mask & FS_RENAME) {
520 moved = fsnotify_data_dentry(data, data_type);
521 inode2 = moved->d_parent->d_inode;
522 inode2_type = FSNOTIFY_ITER_TYPE_INODE2;
523 }
524 } else if (mask & FS_EVENT_ON_CHILD) {
525 /*
526 * Event on child - report on TYPE_PARENT to dir if it is
527 * watching children and on TYPE_INODE to child.
528 */
529 inode2 = dir;
530 inode2_type = FSNOTIFY_ITER_TYPE_PARENT;
531 }
532
533 /*
534 * Optimization: srcu_read_lock() has a memory barrier which can
535 * be expensive. It protects walking the *_fsnotify_marks lists.
536 * However, if we do not walk the lists, we do not have to do
537 * SRCU because we have no references to any objects and do not
538 * need SRCU to keep them "alive".
539 */
540 if ((!sbinfo || !sbinfo->sb_marks) &&
541 (!mnt || !mnt->mnt_fsnotify_marks) &&
542 (!inode || !inode->i_fsnotify_marks) &&
543 (!inode2 || !inode2->i_fsnotify_marks) &&
544 (!mnt_data || !mnt_data->ns->n_fsnotify_marks))
545 return 0;
546
547 if (sb)
548 marks_mask |= READ_ONCE(sb->s_fsnotify_mask);
549 if (mnt)
550 marks_mask |= READ_ONCE(mnt->mnt_fsnotify_mask);
551 if (inode)
552 marks_mask |= READ_ONCE(inode->i_fsnotify_mask);
553 if (inode2)
554 marks_mask |= READ_ONCE(inode2->i_fsnotify_mask);
555 if (mnt_data)
556 marks_mask |= READ_ONCE(mnt_data->ns->n_fsnotify_mask);
557
558 /*
559 * If this is a modify event we may need to clear some ignore masks.
560 * In that case, the object with ignore masks will have the FS_MODIFY
561 * event in its mask.
562 * Otherwise, return if none of the marks care about this type of event.
563 */
564 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
565 if (!(test_mask & marks_mask))
566 return 0;
567
568 iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
569
570 if (sbinfo) {
571 iter_info.marks[FSNOTIFY_ITER_TYPE_SB] =
572 fsnotify_first_mark(&sbinfo->sb_marks);
573 }
574 if (mnt) {
575 iter_info.marks[FSNOTIFY_ITER_TYPE_VFSMOUNT] =
576 fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
577 }
578 if (inode) {
579 iter_info.marks[FSNOTIFY_ITER_TYPE_INODE] =
580 fsnotify_first_mark(&inode->i_fsnotify_marks);
581 }
582 if (inode2) {
583 iter_info.marks[inode2_type] =
584 fsnotify_first_mark(&inode2->i_fsnotify_marks);
585 }
586 if (mnt_data) {
587 iter_info.marks[FSNOTIFY_ITER_TYPE_MNTNS] =
588 fsnotify_first_mark(&mnt_data->ns->n_fsnotify_marks);
589 }
590
591 /*
592 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
593 * ignore masks are properly reflected for mount/sb mark notifications.
594 * That's why this traversal is so complicated...
595 */
596 while (fsnotify_iter_select_report_types(&iter_info)) {
597 ret = send_to_group(mask, data, data_type, dir, file_name,
598 cookie, &iter_info);
599
600 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
601 goto out;
602
603 fsnotify_iter_next(&iter_info);
604 }
605 ret = 0;
606out:
607 srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
608
609 return ret;
610}
611EXPORT_SYMBOL_GPL(fsnotify);
612
613#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
614/*
615 * At open time we check fsnotify_sb_has_priority_watchers(), call the open perm
616 * hook and set the FMODE_NONOTIFY_ mode bits accordignly.
617 * Later, fsnotify permission hooks do not check if there are permission event
618 * watches, but that there were permission event watches at open time.
619 */
620int fsnotify_open_perm_and_set_mode(struct file *file)
621{
622 struct dentry *dentry = file->f_path.dentry, *parent;
623 struct super_block *sb = dentry->d_sb;
624 __u32 mnt_mask, p_mask = 0;
625
626 /* Is it a file opened by fanotify? */
627 if (FMODE_FSNOTIFY_NONE(file->f_mode))
628 return 0;
629
630 /*
631 * Permission events is a super set of pre-content events, so if there
632 * are no permission event watchers, there are also no pre-content event
633 * watchers and this is implied from the single FMODE_NONOTIFY_PERM bit.
634 */
635 if (likely(!fsnotify_sb_has_priority_watchers(sb,
636 FSNOTIFY_PRIO_CONTENT))) {
637 file_set_fsnotify_mode(file, FMODE_NONOTIFY_PERM);
638 return 0;
639 }
640
641 /*
642 * OK, there are some permission event watchers. Check if anybody is
643 * watching for permission events on *this* file.
644 */
645 mnt_mask = READ_ONCE(real_mount(file->f_path.mnt)->mnt_fsnotify_mask);
646 p_mask = fsnotify_object_watched(d_inode(dentry), mnt_mask,
647 ALL_FSNOTIFY_PERM_EVENTS);
648 if (dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED) {
649 parent = dget_parent(dentry);
650 p_mask |= fsnotify_inode_watches_children(d_inode(parent));
651 dput(parent);
652 }
653
654 /*
655 * Legacy FAN_ACCESS_PERM events have very high performance overhead,
656 * so unlikely to be used in the wild. If they are used there will be
657 * no optimizations at all.
658 */
659 if (unlikely(p_mask & FS_ACCESS_PERM)) {
660 /* Enable all permission and pre-content events */
661 file_set_fsnotify_mode(file, 0);
662 goto open_perm;
663 }
664
665 /*
666 * Pre-content events are only supported on regular files.
667 * If there are pre-content event watchers and no permission access
668 * watchers, set FMODE_NONOTIFY | FMODE_NONOTIFY_PERM to indicate that.
669 * That is the common case with HSM service.
670 */
671 if (d_is_reg(dentry) && (p_mask & FSNOTIFY_PRE_CONTENT_EVENTS)) {
672 file_set_fsnotify_mode(file, FMODE_NONOTIFY |
673 FMODE_NONOTIFY_PERM);
674 goto open_perm;
675 }
676
677 /* Nobody watching permission and pre-content events on this file */
678 file_set_fsnotify_mode(file, FMODE_NONOTIFY_PERM);
679
680open_perm:
681 /*
682 * Send open perm events depending on object masks and regardless of
683 * FMODE_NONOTIFY_PERM.
684 */
685 if (file->f_flags & __FMODE_EXEC && p_mask & FS_OPEN_EXEC_PERM) {
686 int ret = fsnotify_path(&file->f_path, FS_OPEN_EXEC_PERM);
687
688 if (ret)
689 return ret;
690 }
691
692 if (p_mask & FS_OPEN_PERM)
693 return fsnotify_path(&file->f_path, FS_OPEN_PERM);
694
695 return 0;
696}
697#endif
698
699void fsnotify_mnt(__u32 mask, struct mnt_namespace *ns, struct vfsmount *mnt)
700{
701 struct fsnotify_mnt data = {
702 .ns = ns,
703 .mnt_id = real_mount(mnt)->mnt_id_unique,
704 };
705
706 if (WARN_ON_ONCE(!ns))
707 return;
708
709 /*
710 * This is an optimization as well as making sure fsnotify_init() has
711 * been called.
712 */
713 if (!ns->n_fsnotify_marks)
714 return;
715
716 fsnotify(mask, &data, FSNOTIFY_EVENT_MNT, NULL, NULL, NULL, 0);
717}
718
719static __init int fsnotify_init(void)
720{
721 int ret;
722
723 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 26);
724
725 ret = init_srcu_struct(&fsnotify_mark_srcu);
726 if (ret)
727 panic("initializing fsnotify_mark_srcu");
728
729 fsnotify_init_connector_caches();
730
731 return 0;
732}
733core_initcall(fsnotify_init);