Loading...
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(&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
90void 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
108void 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 */
120void 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 */
154static 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? */
169static 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(
187 READ_ONCE(inode->i_fsnotify_mask));
188 marks_mask |= fsnotify_parent_needed_mask(
189 READ_ONCE(inode->i_sb->s_fsnotify_mask));
190 marks_mask |= fsnotify_parent_needed_mask(mnt_mask);
191
192 /* Did they subscribe for this event with parent/name info? */
193 return mask & marks_mask;
194}
195
196/* Are there any inode/mount/sb objects that are interested in this event? */
197static inline bool fsnotify_object_watched(struct inode *inode, __u32 mnt_mask,
198 __u32 mask)
199{
200 __u32 marks_mask = READ_ONCE(inode->i_fsnotify_mask) | mnt_mask |
201 READ_ONCE(inode->i_sb->s_fsnotify_mask);
202
203 return mask & marks_mask & ALL_FSNOTIFY_EVENTS;
204}
205
206/*
207 * Notify this dentry's parent about a child's events with child name info
208 * if parent is watching or if inode/sb/mount are interested in events with
209 * parent and name info.
210 *
211 * Notify only the child without name info if parent is not watching and
212 * inode/sb/mount are not interested in events with parent and name info.
213 */
214int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
215 int data_type)
216{
217 const struct path *path = fsnotify_data_path(data, data_type);
218 __u32 mnt_mask = path ?
219 READ_ONCE(real_mount(path->mnt)->mnt_fsnotify_mask) : 0;
220 struct inode *inode = d_inode(dentry);
221 struct dentry *parent;
222 bool parent_watched = dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED;
223 bool parent_needed, parent_interested;
224 __u32 p_mask;
225 struct inode *p_inode = NULL;
226 struct name_snapshot name;
227 struct qstr *file_name = NULL;
228 int ret = 0;
229
230 /* Optimize the likely case of nobody watching this path */
231 if (likely(!parent_watched &&
232 !fsnotify_object_watched(inode, mnt_mask, mask)))
233 return 0;
234
235 parent = NULL;
236 parent_needed = fsnotify_event_needs_parent(inode, mnt_mask, mask);
237 if (!parent_watched && !parent_needed)
238 goto notify;
239
240 /* Does parent inode care about events on children? */
241 parent = dget_parent(dentry);
242 p_inode = parent->d_inode;
243 p_mask = fsnotify_inode_watches_children(p_inode);
244 if (unlikely(parent_watched && !p_mask))
245 fsnotify_clear_child_dentry_flag(p_inode, dentry);
246
247 /*
248 * Include parent/name in notification either if some notification
249 * groups require parent info or the parent is interested in this event.
250 */
251 parent_interested = mask & p_mask & ALL_FSNOTIFY_EVENTS;
252 if (parent_needed || parent_interested) {
253 /* When notifying parent, child should be passed as data */
254 WARN_ON_ONCE(inode != fsnotify_data_inode(data, data_type));
255
256 /* Notify both parent and child with child name info */
257 take_dentry_name_snapshot(&name, dentry);
258 file_name = &name.name;
259 if (parent_interested)
260 mask |= FS_EVENT_ON_CHILD;
261 }
262
263notify:
264 ret = fsnotify(mask, data, data_type, p_inode, file_name, inode, 0);
265
266 if (file_name)
267 release_dentry_name_snapshot(&name);
268 dput(parent);
269
270 return ret;
271}
272EXPORT_SYMBOL_GPL(__fsnotify_parent);
273
274static int fsnotify_handle_inode_event(struct fsnotify_group *group,
275 struct fsnotify_mark *inode_mark,
276 u32 mask, const void *data, int data_type,
277 struct inode *dir, const struct qstr *name,
278 u32 cookie)
279{
280 const struct path *path = fsnotify_data_path(data, data_type);
281 struct inode *inode = fsnotify_data_inode(data, data_type);
282 const struct fsnotify_ops *ops = group->ops;
283
284 if (WARN_ON_ONCE(!ops->handle_inode_event))
285 return 0;
286
287 if (WARN_ON_ONCE(!inode && !dir))
288 return 0;
289
290 if ((inode_mark->flags & FSNOTIFY_MARK_FLAG_EXCL_UNLINK) &&
291 path && d_unlinked(path->dentry))
292 return 0;
293
294 /* Check interest of this mark in case event was sent with two marks */
295 if (!(mask & inode_mark->mask & ALL_FSNOTIFY_EVENTS))
296 return 0;
297
298 return ops->handle_inode_event(inode_mark, mask, inode, dir, name, cookie);
299}
300
301static int fsnotify_handle_event(struct fsnotify_group *group, __u32 mask,
302 const void *data, int data_type,
303 struct inode *dir, const struct qstr *name,
304 u32 cookie, struct fsnotify_iter_info *iter_info)
305{
306 struct fsnotify_mark *inode_mark = fsnotify_iter_inode_mark(iter_info);
307 struct fsnotify_mark *parent_mark = fsnotify_iter_parent_mark(iter_info);
308 int ret;
309
310 if (WARN_ON_ONCE(fsnotify_iter_sb_mark(iter_info)) ||
311 WARN_ON_ONCE(fsnotify_iter_vfsmount_mark(iter_info)))
312 return 0;
313
314 /*
315 * For FS_RENAME, 'dir' is old dir and 'data' is new dentry.
316 * The only ->handle_inode_event() backend that supports FS_RENAME is
317 * dnotify, where it means file was renamed within same parent.
318 */
319 if (mask & FS_RENAME) {
320 struct dentry *moved = fsnotify_data_dentry(data, data_type);
321
322 if (dir != moved->d_parent->d_inode)
323 return 0;
324 }
325
326 if (parent_mark) {
327 ret = fsnotify_handle_inode_event(group, parent_mark, mask,
328 data, data_type, dir, name, 0);
329 if (ret)
330 return ret;
331 }
332
333 if (!inode_mark)
334 return 0;
335
336 /*
337 * Some events can be sent on both parent dir and child marks (e.g.
338 * FS_ATTRIB). If both parent dir and child are watching, report the
339 * event once to parent dir with name (if interested) and once to child
340 * without name (if interested).
341 *
342 * In any case regardless whether the parent is watching or not, the
343 * child watcher is expecting an event without the FS_EVENT_ON_CHILD
344 * flag. The file name is expected if and only if this is a directory
345 * event.
346 */
347 mask &= ~FS_EVENT_ON_CHILD;
348 if (!(mask & ALL_FSNOTIFY_DIRENT_EVENTS)) {
349 dir = NULL;
350 name = NULL;
351 }
352
353 return fsnotify_handle_inode_event(group, inode_mark, mask, data, data_type,
354 dir, name, cookie);
355}
356
357static int send_to_group(__u32 mask, const void *data, int data_type,
358 struct inode *dir, const struct qstr *file_name,
359 u32 cookie, struct fsnotify_iter_info *iter_info)
360{
361 struct fsnotify_group *group = NULL;
362 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
363 __u32 marks_mask = 0;
364 __u32 marks_ignore_mask = 0;
365 bool is_dir = mask & FS_ISDIR;
366 struct fsnotify_mark *mark;
367 int type;
368
369 if (!iter_info->report_mask)
370 return 0;
371
372 /* clear ignored on inode modification */
373 if (mask & FS_MODIFY) {
374 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
375 if (!(mark->flags &
376 FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
377 mark->ignore_mask = 0;
378 }
379 }
380
381 /* Are any of the group marks interested in this event? */
382 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
383 group = mark->group;
384 marks_mask |= mark->mask;
385 marks_ignore_mask |=
386 fsnotify_effective_ignore_mask(mark, is_dir, type);
387 }
388
389 pr_debug("%s: group=%p mask=%x marks_mask=%x marks_ignore_mask=%x data=%p data_type=%d dir=%p cookie=%d\n",
390 __func__, group, mask, marks_mask, marks_ignore_mask,
391 data, data_type, dir, cookie);
392
393 if (!(test_mask & marks_mask & ~marks_ignore_mask))
394 return 0;
395
396 if (group->ops->handle_event) {
397 return group->ops->handle_event(group, mask, data, data_type, dir,
398 file_name, cookie, iter_info);
399 }
400
401 return fsnotify_handle_event(group, mask, data, data_type, dir,
402 file_name, cookie, iter_info);
403}
404
405static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
406{
407 struct fsnotify_mark_connector *conn;
408 struct hlist_node *node = NULL;
409
410 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
411 if (conn)
412 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
413
414 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
415}
416
417static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
418{
419 struct hlist_node *node = NULL;
420
421 if (mark)
422 node = srcu_dereference(mark->obj_list.next,
423 &fsnotify_mark_srcu);
424
425 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
426}
427
428/*
429 * iter_info is a multi head priority queue of marks.
430 * Pick a subset of marks from queue heads, all with the same group
431 * and set the report_mask to a subset of the selected marks.
432 * Returns false if there are no more groups to iterate.
433 */
434static bool fsnotify_iter_select_report_types(
435 struct fsnotify_iter_info *iter_info)
436{
437 struct fsnotify_group *max_prio_group = NULL;
438 struct fsnotify_mark *mark;
439 int type;
440
441 /* Choose max prio group among groups of all queue heads */
442 fsnotify_foreach_iter_type(type) {
443 mark = iter_info->marks[type];
444 if (mark &&
445 fsnotify_compare_groups(max_prio_group, mark->group) > 0)
446 max_prio_group = mark->group;
447 }
448
449 if (!max_prio_group)
450 return false;
451
452 /* Set the report mask for marks from same group as max prio group */
453 iter_info->current_group = max_prio_group;
454 iter_info->report_mask = 0;
455 fsnotify_foreach_iter_type(type) {
456 mark = iter_info->marks[type];
457 if (mark && mark->group == iter_info->current_group) {
458 /*
459 * FSNOTIFY_ITER_TYPE_PARENT indicates that this inode
460 * is watching children and interested in this event,
461 * which is an event possible on child.
462 * But is *this mark* watching children?
463 */
464 if (type == FSNOTIFY_ITER_TYPE_PARENT &&
465 !(mark->mask & FS_EVENT_ON_CHILD) &&
466 !(fsnotify_ignore_mask(mark) & FS_EVENT_ON_CHILD))
467 continue;
468
469 fsnotify_iter_set_report_type(iter_info, type);
470 }
471 }
472
473 return true;
474}
475
476/*
477 * Pop from iter_info multi head queue, the marks that belong to the group of
478 * current iteration step.
479 */
480static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
481{
482 struct fsnotify_mark *mark;
483 int type;
484
485 /*
486 * We cannot use fsnotify_foreach_iter_mark_type() here because we
487 * may need to advance a mark of type X that belongs to current_group
488 * but was not selected for reporting.
489 */
490 fsnotify_foreach_iter_type(type) {
491 mark = iter_info->marks[type];
492 if (mark && mark->group == iter_info->current_group)
493 iter_info->marks[type] =
494 fsnotify_next_mark(iter_info->marks[type]);
495 }
496}
497
498/*
499 * fsnotify - This is the main call to fsnotify.
500 *
501 * The VFS calls into hook specific functions in linux/fsnotify.h.
502 * Those functions then in turn call here. Here will call out to all of the
503 * registered fsnotify_group. Those groups can then use the notification event
504 * in whatever means they feel necessary.
505 *
506 * @mask: event type and flags
507 * @data: object that event happened on
508 * @data_type: type of object for fanotify_data_XXX() accessors
509 * @dir: optional directory associated with event -
510 * if @file_name is not NULL, this is the directory that
511 * @file_name is relative to
512 * @file_name: optional file name associated with event
513 * @inode: optional inode associated with event -
514 * If @dir and @inode are both non-NULL, event may be
515 * reported to both.
516 * @cookie: inotify rename cookie
517 */
518int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
519 const struct qstr *file_name, struct inode *inode, u32 cookie)
520{
521 const struct path *path = fsnotify_data_path(data, data_type);
522 struct super_block *sb = fsnotify_data_sb(data, data_type);
523 struct fsnotify_sb_info *sbinfo = fsnotify_sb_info(sb);
524 struct fsnotify_iter_info iter_info = {};
525 struct mount *mnt = NULL;
526 struct inode *inode2 = NULL;
527 struct dentry *moved;
528 int inode2_type;
529 int ret = 0;
530 __u32 test_mask, marks_mask;
531
532 if (path)
533 mnt = real_mount(path->mnt);
534
535 if (!inode) {
536 /* Dirent event - report on TYPE_INODE to dir */
537 inode = dir;
538 /* For FS_RENAME, inode is old_dir and inode2 is new_dir */
539 if (mask & FS_RENAME) {
540 moved = fsnotify_data_dentry(data, data_type);
541 inode2 = moved->d_parent->d_inode;
542 inode2_type = FSNOTIFY_ITER_TYPE_INODE2;
543 }
544 } else if (mask & FS_EVENT_ON_CHILD) {
545 /*
546 * Event on child - report on TYPE_PARENT to dir if it is
547 * watching children and on TYPE_INODE to child.
548 */
549 inode2 = dir;
550 inode2_type = FSNOTIFY_ITER_TYPE_PARENT;
551 }
552
553 /*
554 * Optimization: srcu_read_lock() has a memory barrier which can
555 * be expensive. It protects walking the *_fsnotify_marks lists.
556 * However, if we do not walk the lists, we do not have to do
557 * SRCU because we have no references to any objects and do not
558 * need SRCU to keep them "alive".
559 */
560 if ((!sbinfo || !sbinfo->sb_marks) &&
561 (!mnt || !mnt->mnt_fsnotify_marks) &&
562 (!inode || !inode->i_fsnotify_marks) &&
563 (!inode2 || !inode2->i_fsnotify_marks))
564 return 0;
565
566 marks_mask = READ_ONCE(sb->s_fsnotify_mask);
567 if (mnt)
568 marks_mask |= READ_ONCE(mnt->mnt_fsnotify_mask);
569 if (inode)
570 marks_mask |= READ_ONCE(inode->i_fsnotify_mask);
571 if (inode2)
572 marks_mask |= READ_ONCE(inode2->i_fsnotify_mask);
573
574
575 /*
576 * If this is a modify event we may need to clear some ignore masks.
577 * In that case, the object with ignore masks will have the FS_MODIFY
578 * event in its mask.
579 * Otherwise, return if none of the marks care about this type of event.
580 */
581 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
582 if (!(test_mask & marks_mask))
583 return 0;
584
585 iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
586
587 if (sbinfo) {
588 iter_info.marks[FSNOTIFY_ITER_TYPE_SB] =
589 fsnotify_first_mark(&sbinfo->sb_marks);
590 }
591 if (mnt) {
592 iter_info.marks[FSNOTIFY_ITER_TYPE_VFSMOUNT] =
593 fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
594 }
595 if (inode) {
596 iter_info.marks[FSNOTIFY_ITER_TYPE_INODE] =
597 fsnotify_first_mark(&inode->i_fsnotify_marks);
598 }
599 if (inode2) {
600 iter_info.marks[inode2_type] =
601 fsnotify_first_mark(&inode2->i_fsnotify_marks);
602 }
603
604 /*
605 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
606 * ignore masks are properly reflected for mount/sb mark notifications.
607 * That's why this traversal is so complicated...
608 */
609 while (fsnotify_iter_select_report_types(&iter_info)) {
610 ret = send_to_group(mask, data, data_type, dir, file_name,
611 cookie, &iter_info);
612
613 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
614 goto out;
615
616 fsnotify_iter_next(&iter_info);
617 }
618 ret = 0;
619out:
620 srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
621
622 return ret;
623}
624EXPORT_SYMBOL_GPL(fsnotify);
625
626static __init int fsnotify_init(void)
627{
628 int ret;
629
630 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 23);
631
632 ret = init_srcu_struct(&fsnotify_mark_srcu);
633 if (ret)
634 panic("initializing fsnotify_mark_srcu");
635
636 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
637 SLAB_PANIC);
638
639 return 0;
640}
641core_initcall(fsnotify_init);
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(&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 if (iput_inode)
74 iput(iput_inode);
75
76 /* for each watch, send FS_UNMOUNT and then remove it */
77 fsnotify_inode(inode, FS_UNMOUNT);
78
79 fsnotify_inode_delete(inode);
80
81 iput_inode = inode;
82
83 cond_resched();
84 spin_lock(&sb->s_inode_list_lock);
85 }
86 spin_unlock(&sb->s_inode_list_lock);
87
88 if (iput_inode)
89 iput(iput_inode);
90 /* Wait for outstanding inode references from connectors */
91 wait_var_event(&sb->s_fsnotify_inode_refs,
92 !atomic_long_read(&sb->s_fsnotify_inode_refs));
93}
94
95void fsnotify_sb_delete(struct super_block *sb)
96{
97 fsnotify_unmount_inodes(sb);
98 fsnotify_clear_marks_by_sb(sb);
99}
100
101/*
102 * Given an inode, first check if we care what happens to our children. Inotify
103 * and dnotify both tell their parents about events. If we care about any event
104 * on a child we run all of our children and set a dentry flag saying that the
105 * parent cares. Thus when an event happens on a child it can quickly tell if
106 * if there is a need to find a parent and send the event to the parent.
107 */
108void __fsnotify_update_child_dentry_flags(struct inode *inode)
109{
110 struct dentry *alias;
111 int watched;
112
113 if (!S_ISDIR(inode->i_mode))
114 return;
115
116 /* determine if the children should tell inode about their events */
117 watched = fsnotify_inode_watches_children(inode);
118
119 spin_lock(&inode->i_lock);
120 /* run all of the dentries associated with this inode. Since this is a
121 * directory, there damn well better only be one item on this list */
122 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
123 struct dentry *child;
124
125 /* run all of the children of the original inode and fix their
126 * d_flags to indicate parental interest (their parent is the
127 * original inode) */
128 spin_lock(&alias->d_lock);
129 list_for_each_entry(child, &alias->d_subdirs, d_child) {
130 if (!child->d_inode)
131 continue;
132
133 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
134 if (watched)
135 child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
136 else
137 child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
138 spin_unlock(&child->d_lock);
139 }
140 spin_unlock(&alias->d_lock);
141 }
142 spin_unlock(&inode->i_lock);
143}
144
145/* Are inode/sb/mount interested in parent and name info with this event? */
146static bool fsnotify_event_needs_parent(struct inode *inode, struct mount *mnt,
147 __u32 mask)
148{
149 __u32 marks_mask = 0;
150
151 /* We only send parent/name to inode/sb/mount for events on non-dir */
152 if (mask & FS_ISDIR)
153 return false;
154
155 /* Did either inode/sb/mount subscribe for events with parent/name? */
156 marks_mask |= fsnotify_parent_needed_mask(inode->i_fsnotify_mask);
157 marks_mask |= fsnotify_parent_needed_mask(inode->i_sb->s_fsnotify_mask);
158 if (mnt)
159 marks_mask |= fsnotify_parent_needed_mask(mnt->mnt_fsnotify_mask);
160
161 /* Did they subscribe for this event with parent/name info? */
162 return mask & marks_mask;
163}
164
165/*
166 * Notify this dentry's parent about a child's events with child name info
167 * if parent is watching or if inode/sb/mount are interested in events with
168 * parent and name info.
169 *
170 * Notify only the child without name info if parent is not watching and
171 * inode/sb/mount are not interested in events with parent and name info.
172 */
173int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
174 int data_type)
175{
176 const struct path *path = fsnotify_data_path(data, data_type);
177 struct mount *mnt = path ? real_mount(path->mnt) : NULL;
178 struct inode *inode = d_inode(dentry);
179 struct dentry *parent;
180 bool parent_watched = dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED;
181 __u32 p_mask;
182 struct inode *p_inode = NULL;
183 struct name_snapshot name;
184 struct qstr *file_name = NULL;
185 int ret = 0;
186
187 /*
188 * Do inode/sb/mount care about parent and name info on non-dir?
189 * Do they care about any event at all?
190 */
191 if (!inode->i_fsnotify_marks && !inode->i_sb->s_fsnotify_marks &&
192 (!mnt || !mnt->mnt_fsnotify_marks) && !parent_watched)
193 return 0;
194
195 parent = NULL;
196 if (!parent_watched && !fsnotify_event_needs_parent(inode, mnt, mask))
197 goto notify;
198
199 /* Does parent inode care about events on children? */
200 parent = dget_parent(dentry);
201 p_inode = parent->d_inode;
202 p_mask = fsnotify_inode_watches_children(p_inode);
203 if (unlikely(parent_watched && !p_mask))
204 __fsnotify_update_child_dentry_flags(p_inode);
205
206 /*
207 * Include parent/name in notification either if some notification
208 * groups require parent info (!parent_watched case) or the parent is
209 * interested in this event.
210 */
211 if (!parent_watched || (mask & p_mask & ALL_FSNOTIFY_EVENTS)) {
212 /* When notifying parent, child should be passed as data */
213 WARN_ON_ONCE(inode != fsnotify_data_inode(data, data_type));
214
215 /* Notify both parent and child with child name info */
216 take_dentry_name_snapshot(&name, dentry);
217 file_name = &name.name;
218 if (parent_watched)
219 mask |= FS_EVENT_ON_CHILD;
220 }
221
222notify:
223 ret = fsnotify(mask, data, data_type, p_inode, file_name, inode, 0);
224
225 if (file_name)
226 release_dentry_name_snapshot(&name);
227 dput(parent);
228
229 return ret;
230}
231EXPORT_SYMBOL_GPL(__fsnotify_parent);
232
233static int fsnotify_handle_event(struct fsnotify_group *group, __u32 mask,
234 const void *data, int data_type,
235 struct inode *dir, const struct qstr *name,
236 u32 cookie, struct fsnotify_iter_info *iter_info)
237{
238 struct fsnotify_mark *inode_mark = fsnotify_iter_inode_mark(iter_info);
239 struct fsnotify_mark *child_mark = fsnotify_iter_child_mark(iter_info);
240 struct inode *inode = fsnotify_data_inode(data, data_type);
241 const struct fsnotify_ops *ops = group->ops;
242 int ret;
243
244 if (WARN_ON_ONCE(!ops->handle_inode_event))
245 return 0;
246
247 if (WARN_ON_ONCE(fsnotify_iter_sb_mark(iter_info)) ||
248 WARN_ON_ONCE(fsnotify_iter_vfsmount_mark(iter_info)))
249 return 0;
250
251 /*
252 * An event can be sent on child mark iterator instead of inode mark
253 * iterator because of other groups that have interest of this inode
254 * and have marks on both parent and child. We can simplify this case.
255 */
256 if (!inode_mark) {
257 inode_mark = child_mark;
258 child_mark = NULL;
259 dir = NULL;
260 name = NULL;
261 }
262
263 ret = ops->handle_inode_event(inode_mark, mask, inode, dir, name);
264 if (ret || !child_mark)
265 return ret;
266
267 /*
268 * Some events can be sent on both parent dir and child marks
269 * (e.g. FS_ATTRIB). If both parent dir and child are watching,
270 * report the event once to parent dir with name and once to child
271 * without name.
272 */
273 return ops->handle_inode_event(child_mark, mask, inode, NULL, NULL);
274}
275
276static int send_to_group(__u32 mask, const void *data, int data_type,
277 struct inode *dir, const struct qstr *file_name,
278 u32 cookie, struct fsnotify_iter_info *iter_info)
279{
280 struct fsnotify_group *group = NULL;
281 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
282 __u32 marks_mask = 0;
283 __u32 marks_ignored_mask = 0;
284 struct fsnotify_mark *mark;
285 int type;
286
287 if (WARN_ON(!iter_info->report_mask))
288 return 0;
289
290 /* clear ignored on inode modification */
291 if (mask & FS_MODIFY) {
292 fsnotify_foreach_obj_type(type) {
293 if (!fsnotify_iter_should_report_type(iter_info, type))
294 continue;
295 mark = iter_info->marks[type];
296 if (mark &&
297 !(mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
298 mark->ignored_mask = 0;
299 }
300 }
301
302 fsnotify_foreach_obj_type(type) {
303 if (!fsnotify_iter_should_report_type(iter_info, type))
304 continue;
305 mark = iter_info->marks[type];
306 /* does the object mark tell us to do something? */
307 if (mark) {
308 group = mark->group;
309 marks_mask |= mark->mask;
310 marks_ignored_mask |= mark->ignored_mask;
311 }
312 }
313
314 pr_debug("%s: group=%p mask=%x marks_mask=%x marks_ignored_mask=%x data=%p data_type=%d dir=%p cookie=%d\n",
315 __func__, group, mask, marks_mask, marks_ignored_mask,
316 data, data_type, dir, cookie);
317
318 if (!(test_mask & marks_mask & ~marks_ignored_mask))
319 return 0;
320
321 if (group->ops->handle_event) {
322 return group->ops->handle_event(group, mask, data, data_type, dir,
323 file_name, cookie, iter_info);
324 }
325
326 return fsnotify_handle_event(group, mask, data, data_type, dir,
327 file_name, cookie, iter_info);
328}
329
330static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
331{
332 struct fsnotify_mark_connector *conn;
333 struct hlist_node *node = NULL;
334
335 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
336 if (conn)
337 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
338
339 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
340}
341
342static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
343{
344 struct hlist_node *node = NULL;
345
346 if (mark)
347 node = srcu_dereference(mark->obj_list.next,
348 &fsnotify_mark_srcu);
349
350 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
351}
352
353/*
354 * iter_info is a multi head priority queue of marks.
355 * Pick a subset of marks from queue heads, all with the
356 * same group and set the report_mask for selected subset.
357 * Returns the report_mask of the selected subset.
358 */
359static unsigned int fsnotify_iter_select_report_types(
360 struct fsnotify_iter_info *iter_info)
361{
362 struct fsnotify_group *max_prio_group = NULL;
363 struct fsnotify_mark *mark;
364 int type;
365
366 /* Choose max prio group among groups of all queue heads */
367 fsnotify_foreach_obj_type(type) {
368 mark = iter_info->marks[type];
369 if (mark &&
370 fsnotify_compare_groups(max_prio_group, mark->group) > 0)
371 max_prio_group = mark->group;
372 }
373
374 if (!max_prio_group)
375 return 0;
376
377 /* Set the report mask for marks from same group as max prio group */
378 iter_info->report_mask = 0;
379 fsnotify_foreach_obj_type(type) {
380 mark = iter_info->marks[type];
381 if (mark &&
382 fsnotify_compare_groups(max_prio_group, mark->group) == 0)
383 fsnotify_iter_set_report_type(iter_info, type);
384 }
385
386 return iter_info->report_mask;
387}
388
389/*
390 * Pop from iter_info multi head queue, the marks that were iterated in the
391 * current iteration step.
392 */
393static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
394{
395 int type;
396
397 fsnotify_foreach_obj_type(type) {
398 if (fsnotify_iter_should_report_type(iter_info, type))
399 iter_info->marks[type] =
400 fsnotify_next_mark(iter_info->marks[type]);
401 }
402}
403
404/*
405 * fsnotify - This is the main call to fsnotify.
406 *
407 * The VFS calls into hook specific functions in linux/fsnotify.h.
408 * Those functions then in turn call here. Here will call out to all of the
409 * registered fsnotify_group. Those groups can then use the notification event
410 * in whatever means they feel necessary.
411 *
412 * @mask: event type and flags
413 * @data: object that event happened on
414 * @data_type: type of object for fanotify_data_XXX() accessors
415 * @dir: optional directory associated with event -
416 * if @file_name is not NULL, this is the directory that
417 * @file_name is relative to
418 * @file_name: optional file name associated with event
419 * @inode: optional inode associated with event -
420 * either @dir or @inode must be non-NULL.
421 * if both are non-NULL event may be reported to both.
422 * @cookie: inotify rename cookie
423 */
424int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir,
425 const struct qstr *file_name, struct inode *inode, u32 cookie)
426{
427 const struct path *path = fsnotify_data_path(data, data_type);
428 struct fsnotify_iter_info iter_info = {};
429 struct super_block *sb;
430 struct mount *mnt = NULL;
431 struct inode *child = NULL;
432 int ret = 0;
433 __u32 test_mask, marks_mask;
434
435 if (path)
436 mnt = real_mount(path->mnt);
437
438 if (!inode) {
439 /* Dirent event - report on TYPE_INODE to dir */
440 inode = dir;
441 } else if (mask & FS_EVENT_ON_CHILD) {
442 /*
443 * Event on child - report on TYPE_INODE to dir if it is
444 * watching children and on TYPE_CHILD to child.
445 */
446 child = inode;
447 inode = dir;
448 }
449 sb = inode->i_sb;
450
451 /*
452 * Optimization: srcu_read_lock() has a memory barrier which can
453 * be expensive. It protects walking the *_fsnotify_marks lists.
454 * However, if we do not walk the lists, we do not have to do
455 * SRCU because we have no references to any objects and do not
456 * need SRCU to keep them "alive".
457 */
458 if (!sb->s_fsnotify_marks &&
459 (!mnt || !mnt->mnt_fsnotify_marks) &&
460 (!inode || !inode->i_fsnotify_marks) &&
461 (!child || !child->i_fsnotify_marks))
462 return 0;
463
464 marks_mask = sb->s_fsnotify_mask;
465 if (mnt)
466 marks_mask |= mnt->mnt_fsnotify_mask;
467 if (inode)
468 marks_mask |= inode->i_fsnotify_mask;
469 if (child)
470 marks_mask |= child->i_fsnotify_mask;
471
472
473 /*
474 * if this is a modify event we may need to clear the ignored masks
475 * otherwise return if none of the marks care about this type of event.
476 */
477 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
478 if (!(mask & FS_MODIFY) && !(test_mask & marks_mask))
479 return 0;
480
481 iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
482
483 iter_info.marks[FSNOTIFY_OBJ_TYPE_SB] =
484 fsnotify_first_mark(&sb->s_fsnotify_marks);
485 if (mnt) {
486 iter_info.marks[FSNOTIFY_OBJ_TYPE_VFSMOUNT] =
487 fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
488 }
489 if (inode) {
490 iter_info.marks[FSNOTIFY_OBJ_TYPE_INODE] =
491 fsnotify_first_mark(&inode->i_fsnotify_marks);
492 }
493 if (child) {
494 iter_info.marks[FSNOTIFY_OBJ_TYPE_CHILD] =
495 fsnotify_first_mark(&child->i_fsnotify_marks);
496 }
497
498 /*
499 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
500 * ignore masks are properly reflected for mount/sb mark notifications.
501 * That's why this traversal is so complicated...
502 */
503 while (fsnotify_iter_select_report_types(&iter_info)) {
504 ret = send_to_group(mask, data, data_type, dir, file_name,
505 cookie, &iter_info);
506
507 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
508 goto out;
509
510 fsnotify_iter_next(&iter_info);
511 }
512 ret = 0;
513out:
514 srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
515
516 return ret;
517}
518EXPORT_SYMBOL_GPL(fsnotify);
519
520static __init int fsnotify_init(void)
521{
522 int ret;
523
524 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 25);
525
526 ret = init_srcu_struct(&fsnotify_mark_srcu);
527 if (ret)
528 panic("initializing fsnotify_mark_srcu");
529
530 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
531 SLAB_PANIC);
532
533 return 0;
534}
535core_initcall(fsnotify_init);