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 */
20 void __fsnotify_inode_delete(struct inode *inode)
21 {
22 fsnotify_clear_marks_by_inode(inode);
23 }
24 EXPORT_SYMBOL_GPL(__fsnotify_inode_delete);
25
26 void __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 */
38 static void fsnotify_unmount_inodes(struct super_block *sb)
39 {
40 struct inode *inode, *iput_inode = NULL;
41
42 spin_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(&inode->i_lock);
50 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) {
51 spin_unlock(&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(&inode->i_count)) {
65 spin_unlock(&inode->i_lock);
66 continue;
67 }
68
69 __iget(inode);
70 spin_unlock(&inode->i_lock);
71 spin_unlock(&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(&sb->s_inode_list_lock);
84 }
85 spin_unlock(&sb->s_inode_list_lock);
86
87 iput(iput_inode);
88 }
89
90 void fsnotify_sb_delete(struct super_block *sb)
91 {
92 struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
93
94 /* Were any marks ever added to any object on this sb? */
95 if (!sbinfo)
96 return;
97
98 fsnotify_unmount_inodes(sb);
99 fsnotify_clear_marks_by_sb(sb);
100 /* Wait for outstanding object references from connectors */
101 wait_var_event(fsnotify_sb_watched_objects(sb),
102 !atomic_long_read(fsnotify_sb_watched_objects(sb)));
103 WARN_ON(fsnotify_sb_has_priority_watchers(sb, FSNOTIFY_PRIO_CONTENT));
104 WARN_ON(fsnotify_sb_has_priority_watchers(sb,
105 FSNOTIFY_PRIO_PRE_CONTENT));
106 }
107
108 void fsnotify_sb_free(struct super_block *sb)
109 {
110 kfree(sb->s_fsnotify_info);
111 }
112
113 /*
114 * Given an inode, first check if we care what happens to our children. Inotify
115 * and dnotify both tell their parents about events. If we care about any event
116 * on a child we run all of our children and set a dentry flag saying that the
117 * parent cares. Thus when an event happens on a child it can quickly tell
118 * if there is a need to find a parent and send the event to the parent.
119 */
120 void fsnotify_set_children_dentry_flags(struct inode *inode)
121 {
122 struct dentry *alias;
123
124 if (!S_ISDIR(inode->i_mode))
125 return;
126
127 spin_lock(&inode->i_lock);
128 /* run all of the dentries associated with this inode. Since this is a
129 * directory, there damn well better only be one item on this list */
130 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
131 struct dentry *child;
132
133 /* run all of the children of the original inode and fix their
134 * d_flags to indicate parental interest (their parent is the
135 * original inode) */
136 spin_lock(&alias->d_lock);
137 hlist_for_each_entry(child, &alias->d_children, d_sib) {
138 if (!child->d_inode)
139 continue;
140
141 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
142 child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
143 spin_unlock(&child->d_lock);
144 }
145 spin_unlock(&alias->d_lock);
146 }
147 spin_unlock(&inode->i_lock);
148 }
149
150 /*
151 * Lazily clear false positive PARENT_WATCHED flag for child whose parent had
152 * stopped watching children.
153 */
154 static void fsnotify_clear_child_dentry_flag(struct inode *pinode,
155 struct dentry *dentry)
156 {
157 spin_lock(&dentry->d_lock);
158 /*
159 * d_lock is a sufficient barrier to prevent observing a non-watched
160 * parent state from before the fsnotify_set_children_dentry_flags()
161 * or fsnotify_update_flags() call that had set PARENT_WATCHED.
162 */
163 if (!fsnotify_inode_watches_children(pinode))
164 dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
165 spin_unlock(&dentry->d_lock);
166 }
167
168 /* Are inode/sb/mount interested in parent and name info with this event? */
169 static bool fsnotify_event_needs_parent(struct inode *inode, __u32 mnt_mask,
170 __u32 mask)
171 {
172 __u32 marks_mask = 0;
173
174 /* We only send parent/name to inode/sb/mount for events on non-dir */
175 if (mask & FS_ISDIR)
176 return false;
177
178 /*
179 * All events that are possible on child can also may be reported with
180 * parent/name info to inode/sb/mount. Otherwise, a watching parent
181 * could result in events reported with unexpected name info to sb/mount.
182 */
183 BUILD_BUG_ON(FS_EVENTS_POSS_ON_CHILD & ~FS_EVENTS_POSS_TO_PARENT);
184
185 /* Did either inode/sb/mount subscribe for events with parent/name? */
186 marks_mask |= fsnotify_parent_needed_mask(inode->i_fsnotify_mask);
187 marks_mask |= fsnotify_parent_needed_mask(inode->i_sb->s_fsnotify_mask);
188 marks_mask |= fsnotify_parent_needed_mask(mnt_mask);
189
190 /* Did they subscribe for this event with parent/name info? */
191 return mask & marks_mask;
192 }
193
194 /* Are there any inode/mount/sb objects that are interested in this event? */
195 static inline bool fsnotify_object_watched(struct inode *inode, __u32 mnt_mask,
196 __u32 mask)
197 {
198 __u32 marks_mask = inode->i_fsnotify_mask | mnt_mask |
199 inode->i_sb->s_fsnotify_mask;
200
201 return mask & marks_mask & ALL_FSNOTIFY_EVENTS;
202 }
203
204 /*
205 * Notify this dentry's parent about a child's events with child name info
206 * if parent is watching or if inode/sb/mount are interested in events with
207 * parent and name info.
208 *
209 * Notify only the child without name info if parent is not watching and
210 * inode/sb/mount are not interested in events with parent and name info.
211 */
212 int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
213 int data_type)
214 {
215 const struct path *path = fsnotify_data_path(data, data_type);
216 __u32 mnt_mask = path ? real_mount(path->mnt)->mnt_fsnotify_mask : 0;
217 struct inode *inode = d_inode(dentry);
218 struct dentry *parent;
219 bool parent_watched = dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED;
220 bool parent_needed, parent_interested;
221 __u32 p_mask;
222 struct inode *p_inode = NULL;
223 struct name_snapshot name;
224 struct qstr *file_name = NULL;
225 int ret = 0;
226
227 /* Optimize the likely case of nobody watching this path */
228 if (likely(!parent_watched &&
229 !fsnotify_object_watched(inode, mnt_mask, mask)))
230 return 0;
231
232 parent = NULL;
233 parent_needed = fsnotify_event_needs_parent(inode, mnt_mask, mask);
234 if (!parent_watched && !parent_needed)
235 goto notify;
236
237 /* Does parent inode care about events on children? */
238 parent = dget_parent(dentry);
239 p_inode = parent->d_inode;
240 p_mask = fsnotify_inode_watches_children(p_inode);
241 if (unlikely(parent_watched && !p_mask))
242 fsnotify_clear_child_dentry_flag(p_inode, dentry);
243
244 /*
245 * Include parent/name in notification either if some notification
246 * groups require parent info or the parent is interested in this event.
247 */
248 parent_interested = mask & p_mask & ALL_FSNOTIFY_EVENTS;
249 if (parent_needed || parent_interested) {
250 /* When notifying parent, child should be passed as data */
251 WARN_ON_ONCE(inode != fsnotify_data_inode(data, data_type));
252
253 /* Notify both parent and child with child name info */
254 take_dentry_name_snapshot(&name, dentry);
255 file_name = &name.name;
256 if (parent_interested)
257 mask |= FS_EVENT_ON_CHILD;
258 }
259
260 notify:
261 ret = fsnotify(mask, data, data_type, p_inode, file_name, inode, 0);
262
263 if (file_name)
264 release_dentry_name_snapshot(&name);
265 dput(parent);
266
267 return ret;
268 }
269 EXPORT_SYMBOL_GPL(__fsnotify_parent);
270
271 static int fsnotify_handle_inode_event(struct fsnotify_group *group,
272 struct fsnotify_mark *inode_mark,
273 u32 mask, const void *data, int data_type,
274 struct inode *dir, const struct qstr *name,
275 u32 cookie)
276 {
277 const struct path *path = fsnotify_data_path(data, data_type);
278 struct inode *inode = fsnotify_data_inode(data, data_type);
279 const struct fsnotify_ops *ops = group->ops;
280
281 if (WARN_ON_ONCE(!ops->handle_inode_event))
282 return 0;
283
284 if (WARN_ON_ONCE(!inode && !dir))
285 return 0;
286
287 if ((inode_mark->flags & FSNOTIFY_MARK_FLAG_EXCL_UNLINK) &&
288 path && d_unlinked(path->dentry))
289 return 0;
290
291 /* Check interest of this mark in case event was sent with two marks */
292 if (!(mask & inode_mark->mask & ALL_FSNOTIFY_EVENTS))
293 return 0;
294
295 return ops->handle_inode_event(inode_mark, mask, inode, dir, name, cookie);
296 }
297
298 static int fsnotify_handle_event(struct fsnotify_group *group, __u32 mask,
299 const void *data, int data_type,
300 struct inode *dir, const struct qstr *name,
301 u32 cookie, struct fsnotify_iter_info *iter_info)
302 {
303 struct fsnotify_mark *inode_mark = fsnotify_iter_inode_mark(iter_info);
304 struct fsnotify_mark *parent_mark = fsnotify_iter_parent_mark(iter_info);
305 int ret;
306
307 if (WARN_ON_ONCE(fsnotify_iter_sb_mark(iter_info)) ||
308 WARN_ON_ONCE(fsnotify_iter_vfsmount_mark(iter_info)))
309 return 0;
310
311 /*
312 * For FS_RENAME, 'dir' is old dir and 'data' is new dentry.
313 * The only ->handle_inode_event() backend that supports FS_RENAME is
314 * dnotify, where it means file was renamed within same parent.
315 */
316 if (mask & FS_RENAME) {
317 struct dentry *moved = fsnotify_data_dentry(data, data_type);
318
319 if (dir != moved->d_parent->d_inode)
320 return 0;
321 }
322
323 if (parent_mark) {
324 ret = fsnotify_handle_inode_event(group, parent_mark, mask,
325 data, data_type, dir, name, 0);
326 if (ret)
327 return ret;
328 }
329
330 if (!inode_mark)
331 return 0;
332
333 if (mask & FS_EVENT_ON_CHILD) {
334 /*
335 * Some events can be sent on both parent dir and child marks
336 * (e.g. FS_ATTRIB). If both parent dir and child are
337 * watching, report the event once to parent dir with name (if
338 * interested) and once to child without name (if interested).
339 * The child watcher is expecting an event without a file name
340 * and without the FS_EVENT_ON_CHILD flag.
341 */
342 mask &= ~FS_EVENT_ON_CHILD;
343 dir = NULL;
344 name = NULL;
345 }
346
347 return fsnotify_handle_inode_event(group, inode_mark, mask, data, data_type,
348 dir, name, cookie);
349 }
350
351 static int send_to_group(__u32 mask, const void *data, int data_type,
352 struct inode *dir, const struct qstr *file_name,
353 u32 cookie, struct fsnotify_iter_info *iter_info)
354 {
355 struct fsnotify_group *group = NULL;
356 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
357 __u32 marks_mask = 0;
358 __u32 marks_ignore_mask = 0;
359 bool is_dir = mask & FS_ISDIR;
360 struct fsnotify_mark *mark;
361 int type;
362
363 if (!iter_info->report_mask)
364 return 0;
365
366 /* clear ignored on inode modification */
367 if (mask & FS_MODIFY) {
368 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
369 if (!(mark->flags &
370 FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
371 mark->ignore_mask = 0;
372 }
373 }
374
375 /* Are any of the group marks interested in this event? */
376 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
377 group = mark->group;
378 marks_mask |= mark->mask;
379 marks_ignore_mask |=
380 fsnotify_effective_ignore_mask(mark, is_dir, type);
381 }
382
383 pr_debug("%s: group=%p mask=%x marks_mask=%x marks_ignore_mask=%x data=%p data_type=%d dir=%p cookie=%d\n",
384 __func__, group, mask, marks_mask, marks_ignore_mask,
385 data, data_type, dir, cookie);
386
387 if (!(test_mask & marks_mask & ~marks_ignore_mask))
388 return 0;
389
390 if (group->ops->handle_event) {
391 return group->ops->handle_event(group, mask, data, data_type, dir,
392 file_name, cookie, iter_info);
393 }
394
395 return fsnotify_handle_event(group, mask, data, data_type, dir,
396 file_name, cookie, iter_info);
397 }
398
399 static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
400 {
401 struct fsnotify_mark_connector *conn;
402 struct hlist_node *node = NULL;
403
404 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
405 if (conn)
406 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
407
408 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
409 }
410
411 static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
412 {
413 struct hlist_node *node = NULL;
414
415 if (mark)
416 node = srcu_dereference(mark->obj_list.next,
417 &fsnotify_mark_srcu);
418
419 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
420 }
421
422 /*
423 * iter_info is a multi head priority queue of marks.
424 * Pick a subset of marks from queue heads, all with the same group
425 * and set the report_mask to a subset of the selected marks.
426 * Returns false if there are no more groups to iterate.
427 */
428 static bool fsnotify_iter_select_report_types(
429 struct fsnotify_iter_info *iter_info)
430 {
431 struct fsnotify_group *max_prio_group = NULL;
432 struct fsnotify_mark *mark;
433 int type;
434
435 /* Choose max prio group among groups of all queue heads */
436 fsnotify_foreach_iter_type(type) {
437 mark = iter_info->marks[type];
438 if (mark &&
439 fsnotify_compare_groups(max_prio_group, mark->group) > 0)
440 max_prio_group = mark->group;
441 }
442
443 if (!max_prio_group)
444 return false;
445
446 /* Set the report mask for marks from same group as max prio group */
447 iter_info->current_group = max_prio_group;
448 iter_info->report_mask = 0;
449 fsnotify_foreach_iter_type(type) {
450 mark = iter_info->marks[type];
451 if (mark && mark->group == iter_info->current_group) {
452 /*
453 * FSNOTIFY_ITER_TYPE_PARENT indicates that this inode
454 * is watching children and interested in this event,
455 * which is an event possible on child.
456 * But is *this mark* watching children?
457 */
458 if (type == FSNOTIFY_ITER_TYPE_PARENT &&
459 !(mark->mask & FS_EVENT_ON_CHILD) &&
460 !(fsnotify_ignore_mask(mark) & FS_EVENT_ON_CHILD))
461 continue;
462
463 fsnotify_iter_set_report_type(iter_info, type);
464 }
465 }
466
467 return true;
468 }
469
470 /*
471 * Pop from iter_info multi head queue, the marks that belong to the group of
472 * current iteration step.
473 */
474 static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
475 {
476 struct fsnotify_mark *mark;
477 int type;
478
479 /*
480 * We cannot use fsnotify_foreach_iter_mark_type() here because we
481 * may need to advance a mark of type X that belongs to current_group
482 * but was not selected for reporting.
483 */
484 fsnotify_foreach_iter_type(type) {
485 mark = iter_info->marks[type];
486 if (mark && mark->group == iter_info->current_group)
487 iter_info->marks[type] =
488 fsnotify_next_mark(iter_info->marks[type]);
489 }
490 }
491
492 /*
493 * fsnotify - This is the main call to fsnotify.
494 *
495 * The VFS calls into hook specific functions in linux/fsnotify.h.
496 * Those functions then in turn call here. Here will call out to all of the
497 * registered fsnotify_group. Those groups can then use the notification event
498 * in whatever means they feel necessary.
499 *
500 * @mask: event type and flags
501 * @data: object that event happened on
502 * @data_type: type of object for fanotify_data_XXX() accessors
503 * @dir: optional directory associated with event -
504 * if @file_name is not NULL, this is the directory that
505 * @file_name is relative to
506 * @file_name: optional file name associated with event
507 * @inode: optional inode associated with event -
508 * If @dir and @inode are both non-NULL, event may be
509 * reported to both.
510 * @cookie: inotify rename cookie
511 */
512 int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
513 const struct qstr *file_name, struct inode *inode, u32 cookie)
514 {
515 const struct path *path = fsnotify_data_path(data, data_type);
516 struct super_block *sb = fsnotify_data_sb(data, data_type);
517 struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
518 struct fsnotify_iter_info iter_info = {};
519 struct mount *mnt = NULL;
520 struct inode *inode2 = NULL;
521 struct dentry *moved;
522 int inode2_type;
523 int ret = 0;
524 __u32 test_mask, marks_mask;
525
526 if (path)
527 mnt = real_mount(path->mnt);
528
529 if (!inode) {
530 /* Dirent event - report on TYPE_INODE to dir */
531 inode = dir;
532 /* For FS_RENAME, inode is old_dir and inode2 is new_dir */
533 if (mask & FS_RENAME) {
534 moved = fsnotify_data_dentry(data, data_type);
535 inode2 = moved->d_parent->d_inode;
536 inode2_type = FSNOTIFY_ITER_TYPE_INODE2;
537 }
538 } else if (mask & FS_EVENT_ON_CHILD) {
539 /*
540 * Event on child - report on TYPE_PARENT to dir if it is
541 * watching children and on TYPE_INODE to child.
542 */
543 inode2 = dir;
544 inode2_type = FSNOTIFY_ITER_TYPE_PARENT;
545 }
546
547 /*
548 * Optimization: srcu_read_lock() has a memory barrier which can
549 * be expensive. It protects walking the *_fsnotify_marks lists.
550 * However, if we do not walk the lists, we do not have to do
551 * SRCU because we have no references to any objects and do not
552 * need SRCU to keep them "alive".
553 */
554 if ((!sbinfo || !sbinfo->sb_marks) &&
555 (!mnt || !mnt->mnt_fsnotify_marks) &&
556 (!inode || !inode->i_fsnotify_marks) &&
557 (!inode2 || !inode2->i_fsnotify_marks))
558 return 0;
559
560 marks_mask = sb->s_fsnotify_mask;
561 if (mnt)
562 marks_mask |= mnt->mnt_fsnotify_mask;
563 if (inode)
564 marks_mask |= inode->i_fsnotify_mask;
565 if (inode2)
566 marks_mask |= inode2->i_fsnotify_mask;
567
568
569 /*
570 * If this is a modify event we may need to clear some ignore masks.
571 * In that case, the object with ignore masks will have the FS_MODIFY
572 * event in its mask.
573 * Otherwise, return if none of the marks care about this type of event.
574 */
575 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
576 if (!(test_mask & marks_mask))
577 return 0;
578
579 iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
580
581 if (sbinfo) {
582 iter_info.marks[FSNOTIFY_ITER_TYPE_SB] =
583 fsnotify_first_mark(&sbinfo->sb_marks);
584 }
585 if (mnt) {
586 iter_info.marks[FSNOTIFY_ITER_TYPE_VFSMOUNT] =
587 fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
588 }
589 if (inode) {
590 iter_info.marks[FSNOTIFY_ITER_TYPE_INODE] =
591 fsnotify_first_mark(&inode->i_fsnotify_marks);
592 }
593 if (inode2) {
594 iter_info.marks[inode2_type] =
595 fsnotify_first_mark(&inode2->i_fsnotify_marks);
596 }
597
598 /*
599 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
600 * ignore masks are properly reflected for mount/sb mark notifications.
601 * That's why this traversal is so complicated...
602 */
603 while (fsnotify_iter_select_report_types(&iter_info)) {
604 ret = send_to_group(mask, data, data_type, dir, file_name,
605 cookie, &iter_info);
606
607 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
608 goto out;
609
610 fsnotify_iter_next(&iter_info);
611 }
612 ret = 0;
613 out:
614 srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
615
616 return ret;
617 }
618 EXPORT_SYMBOL_GPL(fsnotify);
619
620 static __init int fsnotify_init(void)
621 {
622 int ret;
623
624 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 23);
625
626 ret = init_srcu_struct(&fsnotify_mark_srcu);
627 if (ret)
628 panic("initializing fsnotify_mark_srcu");
629
630 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
631 SLAB_PANIC);
632
633 return 0;
634 }
635 core_initcall(fsnotify_init);
636
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.