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/*
7 * fsnotify inode mark locking/lifetime/and refcnting
8 *
9 * REFCNT:
10 * The group->recnt and mark->refcnt tell how many "things" in the kernel
11 * currently are referencing the objects. Both kind of objects typically will
12 * live inside the kernel with a refcnt of 2, one for its creation and one for
13 * the reference a group and a mark hold to each other.
14 * If you are holding the appropriate locks, you can take a reference and the
15 * object itself is guaranteed to survive until the reference is dropped.
16 *
17 * LOCKING:
18 * There are 3 locks involved with fsnotify inode marks and they MUST be taken
19 * in order as follows:
20 *
21 * group->mark_mutex
22 * mark->lock
23 * mark->connector->lock
24 *
25 * group->mark_mutex protects the marks_list anchored inside a given group and
26 * each mark is hooked via the g_list. It also protects the groups private
27 * data (i.e group limits).
28
29 * mark->lock protects the marks attributes like its masks and flags.
30 * Furthermore it protects the access to a reference of the group that the mark
31 * is assigned to as well as the access to a reference of the inode/vfsmount
32 * that is being watched by the mark.
33 *
34 * mark->connector->lock protects the list of marks anchored inside an
35 * inode / vfsmount and each mark is hooked via the i_list.
36 *
37 * A list of notification marks relating to inode / mnt is contained in
38 * fsnotify_mark_connector. That structure is alive as long as there are any
39 * marks in the list and is also protected by fsnotify_mark_srcu. A mark gets
40 * detached from fsnotify_mark_connector when last reference to the mark is
41 * dropped. Thus having mark reference is enough to protect mark->connector
42 * pointer and to make sure fsnotify_mark_connector cannot disappear. Also
43 * because we remove mark from g_list before dropping mark reference associated
44 * with that, any mark found through g_list is guaranteed to have
45 * mark->connector set until we drop group->mark_mutex.
46 *
47 * LIFETIME:
48 * Inode marks survive between when they are added to an inode and when their
49 * refcnt==0. Marks are also protected by fsnotify_mark_srcu.
50 *
51 * The inode mark can be cleared for a number of different reasons including:
52 * - The inode is unlinked for the last time. (fsnotify_inode_remove)
53 * - The inode is being evicted from cache. (fsnotify_inode_delete)
54 * - The fs the inode is on is unmounted. (fsnotify_inode_delete/fsnotify_unmount_inodes)
55 * - Something explicitly requests that it be removed. (fsnotify_destroy_mark)
56 * - The fsnotify_group associated with the mark is going away and all such marks
57 * need to be cleaned up. (fsnotify_clear_marks_by_group)
58 *
59 * This has the very interesting property of being able to run concurrently with
60 * any (or all) other directions.
61 */
62
63#include <linux/fs.h>
64#include <linux/init.h>
65#include <linux/kernel.h>
66#include <linux/kthread.h>
67#include <linux/module.h>
68#include <linux/mutex.h>
69#include <linux/slab.h>
70#include <linux/spinlock.h>
71#include <linux/srcu.h>
72#include <linux/ratelimit.h>
73
74#include <linux/atomic.h>
75
76#include <linux/fsnotify_backend.h>
77#include "fsnotify.h"
78
79#define FSNOTIFY_REAPER_DELAY (1) /* 1 jiffy */
80
81struct srcu_struct fsnotify_mark_srcu;
82struct kmem_cache *fsnotify_mark_connector_cachep;
83
84static DEFINE_SPINLOCK(destroy_lock);
85static LIST_HEAD(destroy_list);
86static struct fsnotify_mark_connector *connector_destroy_list;
87
88static void fsnotify_mark_destroy_workfn(struct work_struct *work);
89static DECLARE_DELAYED_WORK(reaper_work, fsnotify_mark_destroy_workfn);
90
91static void fsnotify_connector_destroy_workfn(struct work_struct *work);
92static DECLARE_WORK(connector_reaper_work, fsnotify_connector_destroy_workfn);
93
94void fsnotify_get_mark(struct fsnotify_mark *mark)
95{
96 WARN_ON_ONCE(!refcount_read(&mark->refcnt));
97 refcount_inc(r: &mark->refcnt);
98}
99
100static __u32 *fsnotify_conn_mask_p(struct fsnotify_mark_connector *conn)
101{
102 if (conn->type == FSNOTIFY_OBJ_TYPE_INODE)
103 return &fsnotify_conn_inode(conn)->i_fsnotify_mask;
104 else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT)
105 return &fsnotify_conn_mount(conn)->mnt_fsnotify_mask;
106 else if (conn->type == FSNOTIFY_OBJ_TYPE_SB)
107 return &fsnotify_conn_sb(conn)->s_fsnotify_mask;
108 return NULL;
109}
110
111__u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn)
112{
113 if (WARN_ON(!fsnotify_valid_obj_type(conn->type)))
114 return 0;
115
116 return *fsnotify_conn_mask_p(conn);
117}
118
119static void fsnotify_get_inode_ref(struct inode *inode)
120{
121 ihold(inode);
122 atomic_long_inc(v: &inode->i_sb->s_fsnotify_connectors);
123}
124
125/*
126 * Grab or drop inode reference for the connector if needed.
127 *
128 * When it's time to drop the reference, we only clear the HAS_IREF flag and
129 * return the inode object. fsnotify_drop_object() will be resonsible for doing
130 * iput() outside of spinlocks. This happens when last mark that wanted iref is
131 * detached.
132 */
133static struct inode *fsnotify_update_iref(struct fsnotify_mark_connector *conn,
134 bool want_iref)
135{
136 bool has_iref = conn->flags & FSNOTIFY_CONN_FLAG_HAS_IREF;
137 struct inode *inode = NULL;
138
139 if (conn->type != FSNOTIFY_OBJ_TYPE_INODE ||
140 want_iref == has_iref)
141 return NULL;
142
143 if (want_iref) {
144 /* Pin inode if any mark wants inode refcount held */
145 fsnotify_get_inode_ref(inode: fsnotify_conn_inode(conn));
146 conn->flags |= FSNOTIFY_CONN_FLAG_HAS_IREF;
147 } else {
148 /* Unpin inode after detach of last mark that wanted iref */
149 inode = fsnotify_conn_inode(conn);
150 conn->flags &= ~FSNOTIFY_CONN_FLAG_HAS_IREF;
151 }
152
153 return inode;
154}
155
156static void *__fsnotify_recalc_mask(struct fsnotify_mark_connector *conn)
157{
158 u32 new_mask = 0;
159 bool want_iref = false;
160 struct fsnotify_mark *mark;
161
162 assert_spin_locked(&conn->lock);
163 /* We can get detached connector here when inode is getting unlinked. */
164 if (!fsnotify_valid_obj_type(obj_type: conn->type))
165 return NULL;
166 hlist_for_each_entry(mark, &conn->list, obj_list) {
167 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED))
168 continue;
169 new_mask |= fsnotify_calc_mask(mark);
170 if (conn->type == FSNOTIFY_OBJ_TYPE_INODE &&
171 !(mark->flags & FSNOTIFY_MARK_FLAG_NO_IREF))
172 want_iref = true;
173 }
174 *fsnotify_conn_mask_p(conn) = new_mask;
175
176 return fsnotify_update_iref(conn, want_iref);
177}
178
179/*
180 * Calculate mask of events for a list of marks. The caller must make sure
181 * connector and connector->obj cannot disappear under us. Callers achieve
182 * this by holding a mark->lock or mark->group->mark_mutex for a mark on this
183 * list.
184 */
185void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn)
186{
187 if (!conn)
188 return;
189
190 spin_lock(lock: &conn->lock);
191 __fsnotify_recalc_mask(conn);
192 spin_unlock(lock: &conn->lock);
193 if (conn->type == FSNOTIFY_OBJ_TYPE_INODE)
194 __fsnotify_update_child_dentry_flags(
195 inode: fsnotify_conn_inode(conn));
196}
197
198/* Free all connectors queued for freeing once SRCU period ends */
199static void fsnotify_connector_destroy_workfn(struct work_struct *work)
200{
201 struct fsnotify_mark_connector *conn, *free;
202
203 spin_lock(lock: &destroy_lock);
204 conn = connector_destroy_list;
205 connector_destroy_list = NULL;
206 spin_unlock(lock: &destroy_lock);
207
208 synchronize_srcu(ssp: &fsnotify_mark_srcu);
209 while (conn) {
210 free = conn;
211 conn = conn->destroy_next;
212 kmem_cache_free(s: fsnotify_mark_connector_cachep, objp: free);
213 }
214}
215
216static void fsnotify_put_inode_ref(struct inode *inode)
217{
218 struct super_block *sb = inode->i_sb;
219
220 iput(inode);
221 if (atomic_long_dec_and_test(v: &sb->s_fsnotify_connectors))
222 wake_up_var(var: &sb->s_fsnotify_connectors);
223}
224
225static void fsnotify_get_sb_connectors(struct fsnotify_mark_connector *conn)
226{
227 struct super_block *sb = fsnotify_connector_sb(conn);
228
229 if (sb)
230 atomic_long_inc(v: &sb->s_fsnotify_connectors);
231}
232
233static void fsnotify_put_sb_connectors(struct fsnotify_mark_connector *conn)
234{
235 struct super_block *sb = fsnotify_connector_sb(conn);
236
237 if (sb && atomic_long_dec_and_test(v: &sb->s_fsnotify_connectors))
238 wake_up_var(var: &sb->s_fsnotify_connectors);
239}
240
241static void *fsnotify_detach_connector_from_object(
242 struct fsnotify_mark_connector *conn,
243 unsigned int *type)
244{
245 struct inode *inode = NULL;
246
247 *type = conn->type;
248 if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED)
249 return NULL;
250
251 if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) {
252 inode = fsnotify_conn_inode(conn);
253 inode->i_fsnotify_mask = 0;
254
255 /* Unpin inode when detaching from connector */
256 if (!(conn->flags & FSNOTIFY_CONN_FLAG_HAS_IREF))
257 inode = NULL;
258 } else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT) {
259 fsnotify_conn_mount(conn)->mnt_fsnotify_mask = 0;
260 } else if (conn->type == FSNOTIFY_OBJ_TYPE_SB) {
261 fsnotify_conn_sb(conn)->s_fsnotify_mask = 0;
262 }
263
264 fsnotify_put_sb_connectors(conn);
265 rcu_assign_pointer(*(conn->obj), NULL);
266 conn->obj = NULL;
267 conn->type = FSNOTIFY_OBJ_TYPE_DETACHED;
268
269 return inode;
270}
271
272static void fsnotify_final_mark_destroy(struct fsnotify_mark *mark)
273{
274 struct fsnotify_group *group = mark->group;
275
276 if (WARN_ON_ONCE(!group))
277 return;
278 group->ops->free_mark(mark);
279 fsnotify_put_group(group);
280}
281
282/* Drop object reference originally held by a connector */
283static void fsnotify_drop_object(unsigned int type, void *objp)
284{
285 if (!objp)
286 return;
287 /* Currently only inode references are passed to be dropped */
288 if (WARN_ON_ONCE(type != FSNOTIFY_OBJ_TYPE_INODE))
289 return;
290 fsnotify_put_inode_ref(inode: objp);
291}
292
293void fsnotify_put_mark(struct fsnotify_mark *mark)
294{
295 struct fsnotify_mark_connector *conn = READ_ONCE(mark->connector);
296 void *objp = NULL;
297 unsigned int type = FSNOTIFY_OBJ_TYPE_DETACHED;
298 bool free_conn = false;
299
300 /* Catch marks that were actually never attached to object */
301 if (!conn) {
302 if (refcount_dec_and_test(r: &mark->refcnt))
303 fsnotify_final_mark_destroy(mark);
304 return;
305 }
306
307 /*
308 * We have to be careful so that traversals of obj_list under lock can
309 * safely grab mark reference.
310 */
311 if (!refcount_dec_and_lock(r: &mark->refcnt, lock: &conn->lock))
312 return;
313
314 hlist_del_init_rcu(n: &mark->obj_list);
315 if (hlist_empty(h: &conn->list)) {
316 objp = fsnotify_detach_connector_from_object(conn, type: &type);
317 free_conn = true;
318 } else {
319 objp = __fsnotify_recalc_mask(conn);
320 type = conn->type;
321 }
322 WRITE_ONCE(mark->connector, NULL);
323 spin_unlock(lock: &conn->lock);
324
325 fsnotify_drop_object(type, objp);
326
327 if (free_conn) {
328 spin_lock(lock: &destroy_lock);
329 conn->destroy_next = connector_destroy_list;
330 connector_destroy_list = conn;
331 spin_unlock(lock: &destroy_lock);
332 queue_work(wq: system_unbound_wq, work: &connector_reaper_work);
333 }
334 /*
335 * Note that we didn't update flags telling whether inode cares about
336 * what's happening with children. We update these flags from
337 * __fsnotify_parent() lazily when next event happens on one of our
338 * children.
339 */
340 spin_lock(lock: &destroy_lock);
341 list_add(new: &mark->g_list, head: &destroy_list);
342 spin_unlock(lock: &destroy_lock);
343 queue_delayed_work(wq: system_unbound_wq, dwork: &reaper_work,
344 FSNOTIFY_REAPER_DELAY);
345}
346EXPORT_SYMBOL_GPL(fsnotify_put_mark);
347
348/*
349 * Get mark reference when we found the mark via lockless traversal of object
350 * list. Mark can be already removed from the list by now and on its way to be
351 * destroyed once SRCU period ends.
352 *
353 * Also pin the group so it doesn't disappear under us.
354 */
355static bool fsnotify_get_mark_safe(struct fsnotify_mark *mark)
356{
357 if (!mark)
358 return true;
359
360 if (refcount_inc_not_zero(r: &mark->refcnt)) {
361 spin_lock(lock: &mark->lock);
362 if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) {
363 /* mark is attached, group is still alive then */
364 atomic_inc(v: &mark->group->user_waits);
365 spin_unlock(lock: &mark->lock);
366 return true;
367 }
368 spin_unlock(lock: &mark->lock);
369 fsnotify_put_mark(mark);
370 }
371 return false;
372}
373
374/*
375 * Puts marks and wakes up group destruction if necessary.
376 *
377 * Pairs with fsnotify_get_mark_safe()
378 */
379static void fsnotify_put_mark_wake(struct fsnotify_mark *mark)
380{
381 if (mark) {
382 struct fsnotify_group *group = mark->group;
383
384 fsnotify_put_mark(mark);
385 /*
386 * We abuse notification_waitq on group shutdown for waiting for
387 * all marks pinned when waiting for userspace.
388 */
389 if (atomic_dec_and_test(v: &group->user_waits) && group->shutdown)
390 wake_up(&group->notification_waitq);
391 }
392}
393
394bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info)
395 __releases(&fsnotify_mark_srcu)
396{
397 int type;
398
399 fsnotify_foreach_iter_type(type) {
400 /* This can fail if mark is being removed */
401 if (!fsnotify_get_mark_safe(mark: iter_info->marks[type])) {
402 __release(&fsnotify_mark_srcu);
403 goto fail;
404 }
405 }
406
407 /*
408 * Now that both marks are pinned by refcount in the inode / vfsmount
409 * lists, we can drop SRCU lock, and safely resume the list iteration
410 * once userspace returns.
411 */
412 srcu_read_unlock(ssp: &fsnotify_mark_srcu, idx: iter_info->srcu_idx);
413
414 return true;
415
416fail:
417 for (type--; type >= 0; type--)
418 fsnotify_put_mark_wake(mark: iter_info->marks[type]);
419 return false;
420}
421
422void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info)
423 __acquires(&fsnotify_mark_srcu)
424{
425 int type;
426
427 iter_info->srcu_idx = srcu_read_lock(ssp: &fsnotify_mark_srcu);
428 fsnotify_foreach_iter_type(type)
429 fsnotify_put_mark_wake(mark: iter_info->marks[type]);
430}
431
432/*
433 * Mark mark as detached, remove it from group list. Mark still stays in object
434 * list until its last reference is dropped. Note that we rely on mark being
435 * removed from group list before corresponding reference to it is dropped. In
436 * particular we rely on mark->connector being valid while we hold
437 * group->mark_mutex if we found the mark through g_list.
438 *
439 * Must be called with group->mark_mutex held. The caller must either hold
440 * reference to the mark or be protected by fsnotify_mark_srcu.
441 */
442void fsnotify_detach_mark(struct fsnotify_mark *mark)
443{
444 fsnotify_group_assert_locked(group: mark->group);
445 WARN_ON_ONCE(!srcu_read_lock_held(&fsnotify_mark_srcu) &&
446 refcount_read(&mark->refcnt) < 1 +
447 !!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED));
448
449 spin_lock(lock: &mark->lock);
450 /* something else already called this function on this mark */
451 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
452 spin_unlock(lock: &mark->lock);
453 return;
454 }
455 mark->flags &= ~FSNOTIFY_MARK_FLAG_ATTACHED;
456 list_del_init(entry: &mark->g_list);
457 spin_unlock(lock: &mark->lock);
458
459 /* Drop mark reference acquired in fsnotify_add_mark_locked() */
460 fsnotify_put_mark(mark);
461}
462
463/*
464 * Free fsnotify mark. The mark is actually only marked as being freed. The
465 * freeing is actually happening only once last reference to the mark is
466 * dropped from a workqueue which first waits for srcu period end.
467 *
468 * Caller must have a reference to the mark or be protected by
469 * fsnotify_mark_srcu.
470 */
471void fsnotify_free_mark(struct fsnotify_mark *mark)
472{
473 struct fsnotify_group *group = mark->group;
474
475 spin_lock(lock: &mark->lock);
476 /* something else already called this function on this mark */
477 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
478 spin_unlock(lock: &mark->lock);
479 return;
480 }
481 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
482 spin_unlock(lock: &mark->lock);
483
484 /*
485 * Some groups like to know that marks are being freed. This is a
486 * callback to the group function to let it know that this mark
487 * is being freed.
488 */
489 if (group->ops->freeing_mark)
490 group->ops->freeing_mark(mark, group);
491}
492
493void fsnotify_destroy_mark(struct fsnotify_mark *mark,
494 struct fsnotify_group *group)
495{
496 fsnotify_group_lock(group);
497 fsnotify_detach_mark(mark);
498 fsnotify_group_unlock(group);
499 fsnotify_free_mark(mark);
500}
501EXPORT_SYMBOL_GPL(fsnotify_destroy_mark);
502
503/*
504 * Sorting function for lists of fsnotify marks.
505 *
506 * Fanotify supports different notification classes (reflected as priority of
507 * notification group). Events shall be passed to notification groups in
508 * decreasing priority order. To achieve this marks in notification lists for
509 * inodes and vfsmounts are sorted so that priorities of corresponding groups
510 * are descending.
511 *
512 * Furthermore correct handling of the ignore mask requires processing inode
513 * and vfsmount marks of each group together. Using the group address as
514 * further sort criterion provides a unique sorting order and thus we can
515 * merge inode and vfsmount lists of marks in linear time and find groups
516 * present in both lists.
517 *
518 * A return value of 1 signifies that b has priority over a.
519 * A return value of 0 signifies that the two marks have to be handled together.
520 * A return value of -1 signifies that a has priority over b.
521 */
522int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b)
523{
524 if (a == b)
525 return 0;
526 if (!a)
527 return 1;
528 if (!b)
529 return -1;
530 if (a->priority < b->priority)
531 return 1;
532 if (a->priority > b->priority)
533 return -1;
534 if (a < b)
535 return 1;
536 return -1;
537}
538
539static int fsnotify_attach_connector_to_object(fsnotify_connp_t *connp,
540 unsigned int obj_type,
541 __kernel_fsid_t *fsid)
542{
543 struct fsnotify_mark_connector *conn;
544
545 conn = kmem_cache_alloc(cachep: fsnotify_mark_connector_cachep, GFP_KERNEL);
546 if (!conn)
547 return -ENOMEM;
548 spin_lock_init(&conn->lock);
549 INIT_HLIST_HEAD(&conn->list);
550 conn->flags = 0;
551 conn->type = obj_type;
552 conn->obj = connp;
553 /* Cache fsid of filesystem containing the object */
554 if (fsid) {
555 conn->fsid = *fsid;
556 conn->flags = FSNOTIFY_CONN_FLAG_HAS_FSID;
557 } else {
558 conn->fsid.val[0] = conn->fsid.val[1] = 0;
559 conn->flags = 0;
560 }
561 fsnotify_get_sb_connectors(conn);
562
563 /*
564 * cmpxchg() provides the barrier so that readers of *connp can see
565 * only initialized structure
566 */
567 if (cmpxchg(connp, NULL, conn)) {
568 /* Someone else created list structure for us */
569 fsnotify_put_sb_connectors(conn);
570 kmem_cache_free(s: fsnotify_mark_connector_cachep, objp: conn);
571 }
572
573 return 0;
574}
575
576/*
577 * Get mark connector, make sure it is alive and return with its lock held.
578 * This is for users that get connector pointer from inode or mount. Users that
579 * hold reference to a mark on the list may directly lock connector->lock as
580 * they are sure list cannot go away under them.
581 */
582static struct fsnotify_mark_connector *fsnotify_grab_connector(
583 fsnotify_connp_t *connp)
584{
585 struct fsnotify_mark_connector *conn;
586 int idx;
587
588 idx = srcu_read_lock(ssp: &fsnotify_mark_srcu);
589 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
590 if (!conn)
591 goto out;
592 spin_lock(lock: &conn->lock);
593 if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED) {
594 spin_unlock(lock: &conn->lock);
595 srcu_read_unlock(ssp: &fsnotify_mark_srcu, idx);
596 return NULL;
597 }
598out:
599 srcu_read_unlock(ssp: &fsnotify_mark_srcu, idx);
600 return conn;
601}
602
603/*
604 * Add mark into proper place in given list of marks. These marks may be used
605 * for the fsnotify backend to determine which event types should be delivered
606 * to which group and for which inodes. These marks are ordered according to
607 * priority, highest number first, and then by the group's location in memory.
608 */
609static int fsnotify_add_mark_list(struct fsnotify_mark *mark,
610 fsnotify_connp_t *connp,
611 unsigned int obj_type,
612 int add_flags, __kernel_fsid_t *fsid)
613{
614 struct fsnotify_mark *lmark, *last = NULL;
615 struct fsnotify_mark_connector *conn;
616 int cmp;
617 int err = 0;
618
619 if (WARN_ON(!fsnotify_valid_obj_type(obj_type)))
620 return -EINVAL;
621
622 /* Backend is expected to check for zero fsid (e.g. tmpfs) */
623 if (fsid && WARN_ON_ONCE(!fsid->val[0] && !fsid->val[1]))
624 return -ENODEV;
625
626restart:
627 spin_lock(lock: &mark->lock);
628 conn = fsnotify_grab_connector(connp);
629 if (!conn) {
630 spin_unlock(lock: &mark->lock);
631 err = fsnotify_attach_connector_to_object(connp, obj_type,
632 fsid);
633 if (err)
634 return err;
635 goto restart;
636 } else if (fsid && !(conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID)) {
637 conn->fsid = *fsid;
638 /* Pairs with smp_rmb() in fanotify_get_fsid() */
639 smp_wmb();
640 conn->flags |= FSNOTIFY_CONN_FLAG_HAS_FSID;
641 } else if (fsid && (conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID) &&
642 (fsid->val[0] != conn->fsid.val[0] ||
643 fsid->val[1] != conn->fsid.val[1])) {
644 /*
645 * Backend is expected to check for non uniform fsid
646 * (e.g. btrfs), but maybe we missed something?
647 * Only allow setting conn->fsid once to non zero fsid.
648 * inotify and non-fid fanotify groups do not set nor test
649 * conn->fsid.
650 */
651 pr_warn_ratelimited("%s: fsid mismatch on object of type %u: "
652 "%x.%x != %x.%x\n", __func__, conn->type,
653 fsid->val[0], fsid->val[1],
654 conn->fsid.val[0], conn->fsid.val[1]);
655 err = -EXDEV;
656 goto out_err;
657 }
658
659 /* is mark the first mark? */
660 if (hlist_empty(h: &conn->list)) {
661 hlist_add_head_rcu(n: &mark->obj_list, h: &conn->list);
662 goto added;
663 }
664
665 /* should mark be in the middle of the current list? */
666 hlist_for_each_entry(lmark, &conn->list, obj_list) {
667 last = lmark;
668
669 if ((lmark->group == mark->group) &&
670 (lmark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) &&
671 !(mark->group->flags & FSNOTIFY_GROUP_DUPS)) {
672 err = -EEXIST;
673 goto out_err;
674 }
675
676 cmp = fsnotify_compare_groups(a: lmark->group, b: mark->group);
677 if (cmp >= 0) {
678 hlist_add_before_rcu(n: &mark->obj_list, next: &lmark->obj_list);
679 goto added;
680 }
681 }
682
683 BUG_ON(last == NULL);
684 /* mark should be the last entry. last is the current last entry */
685 hlist_add_behind_rcu(n: &mark->obj_list, prev: &last->obj_list);
686added:
687 /*
688 * Since connector is attached to object using cmpxchg() we are
689 * guaranteed that connector initialization is fully visible by anyone
690 * seeing mark->connector set.
691 */
692 WRITE_ONCE(mark->connector, conn);
693out_err:
694 spin_unlock(lock: &conn->lock);
695 spin_unlock(lock: &mark->lock);
696 return err;
697}
698
699/*
700 * Attach an initialized mark to a given group and fs object.
701 * These marks may be used for the fsnotify backend to determine which
702 * event types should be delivered to which group.
703 */
704int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
705 fsnotify_connp_t *connp, unsigned int obj_type,
706 int add_flags, __kernel_fsid_t *fsid)
707{
708 struct fsnotify_group *group = mark->group;
709 int ret = 0;
710
711 fsnotify_group_assert_locked(group);
712
713 /*
714 * LOCKING ORDER!!!!
715 * group->mark_mutex
716 * mark->lock
717 * mark->connector->lock
718 */
719 spin_lock(lock: &mark->lock);
720 mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE | FSNOTIFY_MARK_FLAG_ATTACHED;
721
722 list_add(new: &mark->g_list, head: &group->marks_list);
723 fsnotify_get_mark(mark); /* for g_list */
724 spin_unlock(lock: &mark->lock);
725
726 ret = fsnotify_add_mark_list(mark, connp, obj_type, add_flags, fsid);
727 if (ret)
728 goto err;
729
730 fsnotify_recalc_mask(conn: mark->connector);
731
732 return ret;
733err:
734 spin_lock(lock: &mark->lock);
735 mark->flags &= ~(FSNOTIFY_MARK_FLAG_ALIVE |
736 FSNOTIFY_MARK_FLAG_ATTACHED);
737 list_del_init(entry: &mark->g_list);
738 spin_unlock(lock: &mark->lock);
739
740 fsnotify_put_mark(mark);
741 return ret;
742}
743
744int fsnotify_add_mark(struct fsnotify_mark *mark, fsnotify_connp_t *connp,
745 unsigned int obj_type, int add_flags,
746 __kernel_fsid_t *fsid)
747{
748 int ret;
749 struct fsnotify_group *group = mark->group;
750
751 fsnotify_group_lock(group);
752 ret = fsnotify_add_mark_locked(mark, connp, obj_type, add_flags, fsid);
753 fsnotify_group_unlock(group);
754 return ret;
755}
756EXPORT_SYMBOL_GPL(fsnotify_add_mark);
757
758/*
759 * Given a list of marks, find the mark associated with given group. If found
760 * take a reference to that mark and return it, else return NULL.
761 */
762struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp,
763 struct fsnotify_group *group)
764{
765 struct fsnotify_mark_connector *conn;
766 struct fsnotify_mark *mark;
767
768 conn = fsnotify_grab_connector(connp);
769 if (!conn)
770 return NULL;
771
772 hlist_for_each_entry(mark, &conn->list, obj_list) {
773 if (mark->group == group &&
774 (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
775 fsnotify_get_mark(mark);
776 spin_unlock(lock: &conn->lock);
777 return mark;
778 }
779 }
780 spin_unlock(lock: &conn->lock);
781 return NULL;
782}
783EXPORT_SYMBOL_GPL(fsnotify_find_mark);
784
785/* Clear any marks in a group with given type mask */
786void fsnotify_clear_marks_by_group(struct fsnotify_group *group,
787 unsigned int obj_type)
788{
789 struct fsnotify_mark *lmark, *mark;
790 LIST_HEAD(to_free);
791 struct list_head *head = &to_free;
792
793 /* Skip selection step if we want to clear all marks. */
794 if (obj_type == FSNOTIFY_OBJ_TYPE_ANY) {
795 head = &group->marks_list;
796 goto clear;
797 }
798 /*
799 * We have to be really careful here. Anytime we drop mark_mutex, e.g.
800 * fsnotify_clear_marks_by_inode() can come and free marks. Even in our
801 * to_free list so we have to use mark_mutex even when accessing that
802 * list. And freeing mark requires us to drop mark_mutex. So we can
803 * reliably free only the first mark in the list. That's why we first
804 * move marks to free to to_free list in one go and then free marks in
805 * to_free list one by one.
806 */
807 fsnotify_group_lock(group);
808 list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
809 if (mark->connector->type == obj_type)
810 list_move(list: &mark->g_list, head: &to_free);
811 }
812 fsnotify_group_unlock(group);
813
814clear:
815 while (1) {
816 fsnotify_group_lock(group);
817 if (list_empty(head)) {
818 fsnotify_group_unlock(group);
819 break;
820 }
821 mark = list_first_entry(head, struct fsnotify_mark, g_list);
822 fsnotify_get_mark(mark);
823 fsnotify_detach_mark(mark);
824 fsnotify_group_unlock(group);
825 fsnotify_free_mark(mark);
826 fsnotify_put_mark(mark);
827 }
828}
829
830/* Destroy all marks attached to an object via connector */
831void fsnotify_destroy_marks(fsnotify_connp_t *connp)
832{
833 struct fsnotify_mark_connector *conn;
834 struct fsnotify_mark *mark, *old_mark = NULL;
835 void *objp;
836 unsigned int type;
837
838 conn = fsnotify_grab_connector(connp);
839 if (!conn)
840 return;
841 /*
842 * We have to be careful since we can race with e.g.
843 * fsnotify_clear_marks_by_group() and once we drop the conn->lock, the
844 * list can get modified. However we are holding mark reference and
845 * thus our mark cannot be removed from obj_list so we can continue
846 * iteration after regaining conn->lock.
847 */
848 hlist_for_each_entry(mark, &conn->list, obj_list) {
849 fsnotify_get_mark(mark);
850 spin_unlock(lock: &conn->lock);
851 if (old_mark)
852 fsnotify_put_mark(old_mark);
853 old_mark = mark;
854 fsnotify_destroy_mark(mark, mark->group);
855 spin_lock(lock: &conn->lock);
856 }
857 /*
858 * Detach list from object now so that we don't pin inode until all
859 * mark references get dropped. It would lead to strange results such
860 * as delaying inode deletion or blocking unmount.
861 */
862 objp = fsnotify_detach_connector_from_object(conn, type: &type);
863 spin_unlock(lock: &conn->lock);
864 if (old_mark)
865 fsnotify_put_mark(old_mark);
866 fsnotify_drop_object(type, objp);
867}
868
869/*
870 * Nothing fancy, just initialize lists and locks and counters.
871 */
872void fsnotify_init_mark(struct fsnotify_mark *mark,
873 struct fsnotify_group *group)
874{
875 memset(mark, 0, sizeof(*mark));
876 spin_lock_init(&mark->lock);
877 refcount_set(r: &mark->refcnt, n: 1);
878 fsnotify_get_group(group);
879 mark->group = group;
880 WRITE_ONCE(mark->connector, NULL);
881}
882EXPORT_SYMBOL_GPL(fsnotify_init_mark);
883
884/*
885 * Destroy all marks in destroy_list, waits for SRCU period to finish before
886 * actually freeing marks.
887 */
888static void fsnotify_mark_destroy_workfn(struct work_struct *work)
889{
890 struct fsnotify_mark *mark, *next;
891 struct list_head private_destroy_list;
892
893 spin_lock(lock: &destroy_lock);
894 /* exchange the list head */
895 list_replace_init(old: &destroy_list, new: &private_destroy_list);
896 spin_unlock(lock: &destroy_lock);
897
898 synchronize_srcu(ssp: &fsnotify_mark_srcu);
899
900 list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) {
901 list_del_init(entry: &mark->g_list);
902 fsnotify_final_mark_destroy(mark);
903 }
904}
905
906/* Wait for all marks queued for destruction to be actually destroyed */
907void fsnotify_wait_marks_destroyed(void)
908{
909 flush_delayed_work(dwork: &reaper_work);
910}
911EXPORT_SYMBOL_GPL(fsnotify_wait_marks_destroyed);
912

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