xref: /openbmc/linux/fs/notify/mark.c (revision e7f127b2)
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 
81 struct srcu_struct fsnotify_mark_srcu;
82 struct kmem_cache *fsnotify_mark_connector_cachep;
83 
84 static DEFINE_SPINLOCK(destroy_lock);
85 static LIST_HEAD(destroy_list);
86 static struct fsnotify_mark_connector *connector_destroy_list;
87 
88 static void fsnotify_mark_destroy_workfn(struct work_struct *work);
89 static DECLARE_DELAYED_WORK(reaper_work, fsnotify_mark_destroy_workfn);
90 
91 static void fsnotify_connector_destroy_workfn(struct work_struct *work);
92 static DECLARE_WORK(connector_reaper_work, fsnotify_connector_destroy_workfn);
93 
94 void fsnotify_get_mark(struct fsnotify_mark *mark)
95 {
96 	WARN_ON_ONCE(!refcount_read(&mark->refcnt));
97 	refcount_inc(&mark->refcnt);
98 }
99 
100 static __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 
119 static void __fsnotify_recalc_mask(struct fsnotify_mark_connector *conn)
120 {
121 	u32 new_mask = 0;
122 	struct fsnotify_mark *mark;
123 
124 	assert_spin_locked(&conn->lock);
125 	/* We can get detached connector here when inode is getting unlinked. */
126 	if (!fsnotify_valid_obj_type(conn->type))
127 		return;
128 	hlist_for_each_entry(mark, &conn->list, obj_list) {
129 		if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)
130 			new_mask |= mark->mask;
131 	}
132 	*fsnotify_conn_mask_p(conn) = new_mask;
133 }
134 
135 /*
136  * Calculate mask of events for a list of marks. The caller must make sure
137  * connector and connector->obj cannot disappear under us.  Callers achieve
138  * this by holding a mark->lock or mark->group->mark_mutex for a mark on this
139  * list.
140  */
141 void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn)
142 {
143 	if (!conn)
144 		return;
145 
146 	spin_lock(&conn->lock);
147 	__fsnotify_recalc_mask(conn);
148 	spin_unlock(&conn->lock);
149 	if (conn->type == FSNOTIFY_OBJ_TYPE_INODE)
150 		__fsnotify_update_child_dentry_flags(
151 					fsnotify_conn_inode(conn));
152 }
153 
154 /* Free all connectors queued for freeing once SRCU period ends */
155 static void fsnotify_connector_destroy_workfn(struct work_struct *work)
156 {
157 	struct fsnotify_mark_connector *conn, *free;
158 
159 	spin_lock(&destroy_lock);
160 	conn = connector_destroy_list;
161 	connector_destroy_list = NULL;
162 	spin_unlock(&destroy_lock);
163 
164 	synchronize_srcu(&fsnotify_mark_srcu);
165 	while (conn) {
166 		free = conn;
167 		conn = conn->destroy_next;
168 		kmem_cache_free(fsnotify_mark_connector_cachep, free);
169 	}
170 }
171 
172 static void fsnotify_get_inode_ref(struct inode *inode)
173 {
174 	ihold(inode);
175 	atomic_long_inc(&inode->i_sb->s_fsnotify_connectors);
176 }
177 
178 static void fsnotify_put_inode_ref(struct inode *inode)
179 {
180 	struct super_block *sb = inode->i_sb;
181 
182 	iput(inode);
183 	if (atomic_long_dec_and_test(&sb->s_fsnotify_connectors))
184 		wake_up_var(&sb->s_fsnotify_connectors);
185 }
186 
187 static void fsnotify_get_sb_connectors(struct fsnotify_mark_connector *conn)
188 {
189 	struct super_block *sb = fsnotify_connector_sb(conn);
190 
191 	if (sb)
192 		atomic_long_inc(&sb->s_fsnotify_connectors);
193 }
194 
195 static void fsnotify_put_sb_connectors(struct fsnotify_mark_connector *conn)
196 {
197 	struct super_block *sb = fsnotify_connector_sb(conn);
198 
199 	if (sb && atomic_long_dec_and_test(&sb->s_fsnotify_connectors))
200 		wake_up_var(&sb->s_fsnotify_connectors);
201 }
202 
203 static void *fsnotify_detach_connector_from_object(
204 					struct fsnotify_mark_connector *conn,
205 					unsigned int *type)
206 {
207 	struct inode *inode = NULL;
208 
209 	*type = conn->type;
210 	if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED)
211 		return NULL;
212 
213 	if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) {
214 		inode = fsnotify_conn_inode(conn);
215 		inode->i_fsnotify_mask = 0;
216 	} else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT) {
217 		fsnotify_conn_mount(conn)->mnt_fsnotify_mask = 0;
218 	} else if (conn->type == FSNOTIFY_OBJ_TYPE_SB) {
219 		fsnotify_conn_sb(conn)->s_fsnotify_mask = 0;
220 	}
221 
222 	fsnotify_put_sb_connectors(conn);
223 	rcu_assign_pointer(*(conn->obj), NULL);
224 	conn->obj = NULL;
225 	conn->type = FSNOTIFY_OBJ_TYPE_DETACHED;
226 
227 	return inode;
228 }
229 
230 static void fsnotify_final_mark_destroy(struct fsnotify_mark *mark)
231 {
232 	struct fsnotify_group *group = mark->group;
233 
234 	if (WARN_ON_ONCE(!group))
235 		return;
236 	group->ops->free_mark(mark);
237 	fsnotify_put_group(group);
238 }
239 
240 /* Drop object reference originally held by a connector */
241 static void fsnotify_drop_object(unsigned int type, void *objp)
242 {
243 	if (!objp)
244 		return;
245 	/* Currently only inode references are passed to be dropped */
246 	if (WARN_ON_ONCE(type != FSNOTIFY_OBJ_TYPE_INODE))
247 		return;
248 	fsnotify_put_inode_ref(objp);
249 }
250 
251 void fsnotify_put_mark(struct fsnotify_mark *mark)
252 {
253 	struct fsnotify_mark_connector *conn = READ_ONCE(mark->connector);
254 	void *objp = NULL;
255 	unsigned int type = FSNOTIFY_OBJ_TYPE_DETACHED;
256 	bool free_conn = false;
257 
258 	/* Catch marks that were actually never attached to object */
259 	if (!conn) {
260 		if (refcount_dec_and_test(&mark->refcnt))
261 			fsnotify_final_mark_destroy(mark);
262 		return;
263 	}
264 
265 	/*
266 	 * We have to be careful so that traversals of obj_list under lock can
267 	 * safely grab mark reference.
268 	 */
269 	if (!refcount_dec_and_lock(&mark->refcnt, &conn->lock))
270 		return;
271 
272 	hlist_del_init_rcu(&mark->obj_list);
273 	if (hlist_empty(&conn->list)) {
274 		objp = fsnotify_detach_connector_from_object(conn, &type);
275 		free_conn = true;
276 	} else {
277 		__fsnotify_recalc_mask(conn);
278 	}
279 	WRITE_ONCE(mark->connector, NULL);
280 	spin_unlock(&conn->lock);
281 
282 	fsnotify_drop_object(type, objp);
283 
284 	if (free_conn) {
285 		spin_lock(&destroy_lock);
286 		conn->destroy_next = connector_destroy_list;
287 		connector_destroy_list = conn;
288 		spin_unlock(&destroy_lock);
289 		queue_work(system_unbound_wq, &connector_reaper_work);
290 	}
291 	/*
292 	 * Note that we didn't update flags telling whether inode cares about
293 	 * what's happening with children. We update these flags from
294 	 * __fsnotify_parent() lazily when next event happens on one of our
295 	 * children.
296 	 */
297 	spin_lock(&destroy_lock);
298 	list_add(&mark->g_list, &destroy_list);
299 	spin_unlock(&destroy_lock);
300 	queue_delayed_work(system_unbound_wq, &reaper_work,
301 			   FSNOTIFY_REAPER_DELAY);
302 }
303 EXPORT_SYMBOL_GPL(fsnotify_put_mark);
304 
305 /*
306  * Get mark reference when we found the mark via lockless traversal of object
307  * list. Mark can be already removed from the list by now and on its way to be
308  * destroyed once SRCU period ends.
309  *
310  * Also pin the group so it doesn't disappear under us.
311  */
312 static bool fsnotify_get_mark_safe(struct fsnotify_mark *mark)
313 {
314 	if (!mark)
315 		return true;
316 
317 	if (refcount_inc_not_zero(&mark->refcnt)) {
318 		spin_lock(&mark->lock);
319 		if (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) {
320 			/* mark is attached, group is still alive then */
321 			atomic_inc(&mark->group->user_waits);
322 			spin_unlock(&mark->lock);
323 			return true;
324 		}
325 		spin_unlock(&mark->lock);
326 		fsnotify_put_mark(mark);
327 	}
328 	return false;
329 }
330 
331 /*
332  * Puts marks and wakes up group destruction if necessary.
333  *
334  * Pairs with fsnotify_get_mark_safe()
335  */
336 static void fsnotify_put_mark_wake(struct fsnotify_mark *mark)
337 {
338 	if (mark) {
339 		struct fsnotify_group *group = mark->group;
340 
341 		fsnotify_put_mark(mark);
342 		/*
343 		 * We abuse notification_waitq on group shutdown for waiting for
344 		 * all marks pinned when waiting for userspace.
345 		 */
346 		if (atomic_dec_and_test(&group->user_waits) && group->shutdown)
347 			wake_up(&group->notification_waitq);
348 	}
349 }
350 
351 bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info)
352 	__releases(&fsnotify_mark_srcu)
353 {
354 	int type;
355 
356 	fsnotify_foreach_iter_type(type) {
357 		/* This can fail if mark is being removed */
358 		if (!fsnotify_get_mark_safe(iter_info->marks[type])) {
359 			__release(&fsnotify_mark_srcu);
360 			goto fail;
361 		}
362 	}
363 
364 	/*
365 	 * Now that both marks are pinned by refcount in the inode / vfsmount
366 	 * lists, we can drop SRCU lock, and safely resume the list iteration
367 	 * once userspace returns.
368 	 */
369 	srcu_read_unlock(&fsnotify_mark_srcu, iter_info->srcu_idx);
370 
371 	return true;
372 
373 fail:
374 	for (type--; type >= 0; type--)
375 		fsnotify_put_mark_wake(iter_info->marks[type]);
376 	return false;
377 }
378 
379 void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info)
380 	__acquires(&fsnotify_mark_srcu)
381 {
382 	int type;
383 
384 	iter_info->srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
385 	fsnotify_foreach_iter_type(type)
386 		fsnotify_put_mark_wake(iter_info->marks[type]);
387 }
388 
389 /*
390  * Mark mark as detached, remove it from group list. Mark still stays in object
391  * list until its last reference is dropped. Note that we rely on mark being
392  * removed from group list before corresponding reference to it is dropped. In
393  * particular we rely on mark->connector being valid while we hold
394  * group->mark_mutex if we found the mark through g_list.
395  *
396  * Must be called with group->mark_mutex held. The caller must either hold
397  * reference to the mark or be protected by fsnotify_mark_srcu.
398  */
399 void fsnotify_detach_mark(struct fsnotify_mark *mark)
400 {
401 	struct fsnotify_group *group = mark->group;
402 
403 	WARN_ON_ONCE(!mutex_is_locked(&group->mark_mutex));
404 	WARN_ON_ONCE(!srcu_read_lock_held(&fsnotify_mark_srcu) &&
405 		     refcount_read(&mark->refcnt) < 1 +
406 			!!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED));
407 
408 	spin_lock(&mark->lock);
409 	/* something else already called this function on this mark */
410 	if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
411 		spin_unlock(&mark->lock);
412 		return;
413 	}
414 	mark->flags &= ~FSNOTIFY_MARK_FLAG_ATTACHED;
415 	list_del_init(&mark->g_list);
416 	spin_unlock(&mark->lock);
417 
418 	/* Drop mark reference acquired in fsnotify_add_mark_locked() */
419 	fsnotify_put_mark(mark);
420 }
421 
422 /*
423  * Free fsnotify mark. The mark is actually only marked as being freed.  The
424  * freeing is actually happening only once last reference to the mark is
425  * dropped from a workqueue which first waits for srcu period end.
426  *
427  * Caller must have a reference to the mark or be protected by
428  * fsnotify_mark_srcu.
429  */
430 void fsnotify_free_mark(struct fsnotify_mark *mark)
431 {
432 	struct fsnotify_group *group = mark->group;
433 
434 	spin_lock(&mark->lock);
435 	/* something else already called this function on this mark */
436 	if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
437 		spin_unlock(&mark->lock);
438 		return;
439 	}
440 	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
441 	spin_unlock(&mark->lock);
442 
443 	/*
444 	 * Some groups like to know that marks are being freed.  This is a
445 	 * callback to the group function to let it know that this mark
446 	 * is being freed.
447 	 */
448 	if (group->ops->freeing_mark)
449 		group->ops->freeing_mark(mark, group);
450 }
451 
452 void fsnotify_destroy_mark(struct fsnotify_mark *mark,
453 			   struct fsnotify_group *group)
454 {
455 	mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
456 	fsnotify_detach_mark(mark);
457 	mutex_unlock(&group->mark_mutex);
458 	fsnotify_free_mark(mark);
459 }
460 EXPORT_SYMBOL_GPL(fsnotify_destroy_mark);
461 
462 /*
463  * Sorting function for lists of fsnotify marks.
464  *
465  * Fanotify supports different notification classes (reflected as priority of
466  * notification group). Events shall be passed to notification groups in
467  * decreasing priority order. To achieve this marks in notification lists for
468  * inodes and vfsmounts are sorted so that priorities of corresponding groups
469  * are descending.
470  *
471  * Furthermore correct handling of the ignore mask requires processing inode
472  * and vfsmount marks of each group together. Using the group address as
473  * further sort criterion provides a unique sorting order and thus we can
474  * merge inode and vfsmount lists of marks in linear time and find groups
475  * present in both lists.
476  *
477  * A return value of 1 signifies that b has priority over a.
478  * A return value of 0 signifies that the two marks have to be handled together.
479  * A return value of -1 signifies that a has priority over b.
480  */
481 int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b)
482 {
483 	if (a == b)
484 		return 0;
485 	if (!a)
486 		return 1;
487 	if (!b)
488 		return -1;
489 	if (a->priority < b->priority)
490 		return 1;
491 	if (a->priority > b->priority)
492 		return -1;
493 	if (a < b)
494 		return 1;
495 	return -1;
496 }
497 
498 static int fsnotify_attach_connector_to_object(fsnotify_connp_t *connp,
499 					       unsigned int obj_type,
500 					       __kernel_fsid_t *fsid)
501 {
502 	struct inode *inode = NULL;
503 	struct fsnotify_mark_connector *conn;
504 
505 	conn = kmem_cache_alloc(fsnotify_mark_connector_cachep, GFP_KERNEL);
506 	if (!conn)
507 		return -ENOMEM;
508 	spin_lock_init(&conn->lock);
509 	INIT_HLIST_HEAD(&conn->list);
510 	conn->type = obj_type;
511 	conn->obj = connp;
512 	/* Cache fsid of filesystem containing the object */
513 	if (fsid) {
514 		conn->fsid = *fsid;
515 		conn->flags = FSNOTIFY_CONN_FLAG_HAS_FSID;
516 	} else {
517 		conn->fsid.val[0] = conn->fsid.val[1] = 0;
518 		conn->flags = 0;
519 	}
520 	if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) {
521 		inode = fsnotify_conn_inode(conn);
522 		fsnotify_get_inode_ref(inode);
523 	}
524 	fsnotify_get_sb_connectors(conn);
525 
526 	/*
527 	 * cmpxchg() provides the barrier so that readers of *connp can see
528 	 * only initialized structure
529 	 */
530 	if (cmpxchg(connp, NULL, conn)) {
531 		/* Someone else created list structure for us */
532 		if (inode)
533 			fsnotify_put_inode_ref(inode);
534 		fsnotify_put_sb_connectors(conn);
535 		kmem_cache_free(fsnotify_mark_connector_cachep, conn);
536 	}
537 
538 	return 0;
539 }
540 
541 /*
542  * Get mark connector, make sure it is alive and return with its lock held.
543  * This is for users that get connector pointer from inode or mount. Users that
544  * hold reference to a mark on the list may directly lock connector->lock as
545  * they are sure list cannot go away under them.
546  */
547 static struct fsnotify_mark_connector *fsnotify_grab_connector(
548 						fsnotify_connp_t *connp)
549 {
550 	struct fsnotify_mark_connector *conn;
551 	int idx;
552 
553 	idx = srcu_read_lock(&fsnotify_mark_srcu);
554 	conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
555 	if (!conn)
556 		goto out;
557 	spin_lock(&conn->lock);
558 	if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED) {
559 		spin_unlock(&conn->lock);
560 		srcu_read_unlock(&fsnotify_mark_srcu, idx);
561 		return NULL;
562 	}
563 out:
564 	srcu_read_unlock(&fsnotify_mark_srcu, idx);
565 	return conn;
566 }
567 
568 /*
569  * Add mark into proper place in given list of marks. These marks may be used
570  * for the fsnotify backend to determine which event types should be delivered
571  * to which group and for which inodes. These marks are ordered according to
572  * priority, highest number first, and then by the group's location in memory.
573  */
574 static int fsnotify_add_mark_list(struct fsnotify_mark *mark,
575 				  fsnotify_connp_t *connp,
576 				  unsigned int obj_type,
577 				  int allow_dups, __kernel_fsid_t *fsid)
578 {
579 	struct fsnotify_mark *lmark, *last = NULL;
580 	struct fsnotify_mark_connector *conn;
581 	int cmp;
582 	int err = 0;
583 
584 	if (WARN_ON(!fsnotify_valid_obj_type(obj_type)))
585 		return -EINVAL;
586 
587 	/* Backend is expected to check for zero fsid (e.g. tmpfs) */
588 	if (fsid && WARN_ON_ONCE(!fsid->val[0] && !fsid->val[1]))
589 		return -ENODEV;
590 
591 restart:
592 	spin_lock(&mark->lock);
593 	conn = fsnotify_grab_connector(connp);
594 	if (!conn) {
595 		spin_unlock(&mark->lock);
596 		err = fsnotify_attach_connector_to_object(connp, obj_type,
597 							  fsid);
598 		if (err)
599 			return err;
600 		goto restart;
601 	} else if (fsid && !(conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID)) {
602 		conn->fsid = *fsid;
603 		/* Pairs with smp_rmb() in fanotify_get_fsid() */
604 		smp_wmb();
605 		conn->flags |= FSNOTIFY_CONN_FLAG_HAS_FSID;
606 	} else if (fsid && (conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID) &&
607 		   (fsid->val[0] != conn->fsid.val[0] ||
608 		    fsid->val[1] != conn->fsid.val[1])) {
609 		/*
610 		 * Backend is expected to check for non uniform fsid
611 		 * (e.g. btrfs), but maybe we missed something?
612 		 * Only allow setting conn->fsid once to non zero fsid.
613 		 * inotify and non-fid fanotify groups do not set nor test
614 		 * conn->fsid.
615 		 */
616 		pr_warn_ratelimited("%s: fsid mismatch on object of type %u: "
617 				    "%x.%x != %x.%x\n", __func__, conn->type,
618 				    fsid->val[0], fsid->val[1],
619 				    conn->fsid.val[0], conn->fsid.val[1]);
620 		err = -EXDEV;
621 		goto out_err;
622 	}
623 
624 	/* is mark the first mark? */
625 	if (hlist_empty(&conn->list)) {
626 		hlist_add_head_rcu(&mark->obj_list, &conn->list);
627 		goto added;
628 	}
629 
630 	/* should mark be in the middle of the current list? */
631 	hlist_for_each_entry(lmark, &conn->list, obj_list) {
632 		last = lmark;
633 
634 		if ((lmark->group == mark->group) &&
635 		    (lmark->flags & FSNOTIFY_MARK_FLAG_ATTACHED) &&
636 		    !allow_dups) {
637 			err = -EEXIST;
638 			goto out_err;
639 		}
640 
641 		cmp = fsnotify_compare_groups(lmark->group, mark->group);
642 		if (cmp >= 0) {
643 			hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list);
644 			goto added;
645 		}
646 	}
647 
648 	BUG_ON(last == NULL);
649 	/* mark should be the last entry.  last is the current last entry */
650 	hlist_add_behind_rcu(&mark->obj_list, &last->obj_list);
651 added:
652 	/*
653 	 * Since connector is attached to object using cmpxchg() we are
654 	 * guaranteed that connector initialization is fully visible by anyone
655 	 * seeing mark->connector set.
656 	 */
657 	WRITE_ONCE(mark->connector, conn);
658 out_err:
659 	spin_unlock(&conn->lock);
660 	spin_unlock(&mark->lock);
661 	return err;
662 }
663 
664 /*
665  * Attach an initialized mark to a given group and fs object.
666  * These marks may be used for the fsnotify backend to determine which
667  * event types should be delivered to which group.
668  */
669 int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
670 			     fsnotify_connp_t *connp, unsigned int obj_type,
671 			     int allow_dups, __kernel_fsid_t *fsid)
672 {
673 	struct fsnotify_group *group = mark->group;
674 	int ret = 0;
675 
676 	BUG_ON(!mutex_is_locked(&group->mark_mutex));
677 
678 	/*
679 	 * LOCKING ORDER!!!!
680 	 * group->mark_mutex
681 	 * mark->lock
682 	 * mark->connector->lock
683 	 */
684 	spin_lock(&mark->lock);
685 	mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE | FSNOTIFY_MARK_FLAG_ATTACHED;
686 
687 	list_add(&mark->g_list, &group->marks_list);
688 	fsnotify_get_mark(mark); /* for g_list */
689 	spin_unlock(&mark->lock);
690 
691 	ret = fsnotify_add_mark_list(mark, connp, obj_type, allow_dups, fsid);
692 	if (ret)
693 		goto err;
694 
695 	if (mark->mask)
696 		fsnotify_recalc_mask(mark->connector);
697 
698 	return ret;
699 err:
700 	spin_lock(&mark->lock);
701 	mark->flags &= ~(FSNOTIFY_MARK_FLAG_ALIVE |
702 			 FSNOTIFY_MARK_FLAG_ATTACHED);
703 	list_del_init(&mark->g_list);
704 	spin_unlock(&mark->lock);
705 
706 	fsnotify_put_mark(mark);
707 	return ret;
708 }
709 
710 int fsnotify_add_mark(struct fsnotify_mark *mark, fsnotify_connp_t *connp,
711 		      unsigned int obj_type, int allow_dups,
712 		      __kernel_fsid_t *fsid)
713 {
714 	int ret;
715 	struct fsnotify_group *group = mark->group;
716 
717 	mutex_lock(&group->mark_mutex);
718 	ret = fsnotify_add_mark_locked(mark, connp, obj_type, allow_dups, fsid);
719 	mutex_unlock(&group->mark_mutex);
720 	return ret;
721 }
722 EXPORT_SYMBOL_GPL(fsnotify_add_mark);
723 
724 /*
725  * Given a list of marks, find the mark associated with given group. If found
726  * take a reference to that mark and return it, else return NULL.
727  */
728 struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp,
729 					 struct fsnotify_group *group)
730 {
731 	struct fsnotify_mark_connector *conn;
732 	struct fsnotify_mark *mark;
733 
734 	conn = fsnotify_grab_connector(connp);
735 	if (!conn)
736 		return NULL;
737 
738 	hlist_for_each_entry(mark, &conn->list, obj_list) {
739 		if (mark->group == group &&
740 		    (mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
741 			fsnotify_get_mark(mark);
742 			spin_unlock(&conn->lock);
743 			return mark;
744 		}
745 	}
746 	spin_unlock(&conn->lock);
747 	return NULL;
748 }
749 EXPORT_SYMBOL_GPL(fsnotify_find_mark);
750 
751 /* Clear any marks in a group with given type mask */
752 void fsnotify_clear_marks_by_group(struct fsnotify_group *group,
753 				   unsigned int obj_type)
754 {
755 	struct fsnotify_mark *lmark, *mark;
756 	LIST_HEAD(to_free);
757 	struct list_head *head = &to_free;
758 
759 	/* Skip selection step if we want to clear all marks. */
760 	if (obj_type == FSNOTIFY_OBJ_TYPE_ANY) {
761 		head = &group->marks_list;
762 		goto clear;
763 	}
764 	/*
765 	 * We have to be really careful here. Anytime we drop mark_mutex, e.g.
766 	 * fsnotify_clear_marks_by_inode() can come and free marks. Even in our
767 	 * to_free list so we have to use mark_mutex even when accessing that
768 	 * list. And freeing mark requires us to drop mark_mutex. So we can
769 	 * reliably free only the first mark in the list. That's why we first
770 	 * move marks to free to to_free list in one go and then free marks in
771 	 * to_free list one by one.
772 	 */
773 	mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
774 	list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
775 		if (mark->connector->type == obj_type)
776 			list_move(&mark->g_list, &to_free);
777 	}
778 	mutex_unlock(&group->mark_mutex);
779 
780 clear:
781 	while (1) {
782 		mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
783 		if (list_empty(head)) {
784 			mutex_unlock(&group->mark_mutex);
785 			break;
786 		}
787 		mark = list_first_entry(head, struct fsnotify_mark, g_list);
788 		fsnotify_get_mark(mark);
789 		fsnotify_detach_mark(mark);
790 		mutex_unlock(&group->mark_mutex);
791 		fsnotify_free_mark(mark);
792 		fsnotify_put_mark(mark);
793 	}
794 }
795 
796 /* Destroy all marks attached to an object via connector */
797 void fsnotify_destroy_marks(fsnotify_connp_t *connp)
798 {
799 	struct fsnotify_mark_connector *conn;
800 	struct fsnotify_mark *mark, *old_mark = NULL;
801 	void *objp;
802 	unsigned int type;
803 
804 	conn = fsnotify_grab_connector(connp);
805 	if (!conn)
806 		return;
807 	/*
808 	 * We have to be careful since we can race with e.g.
809 	 * fsnotify_clear_marks_by_group() and once we drop the conn->lock, the
810 	 * list can get modified. However we are holding mark reference and
811 	 * thus our mark cannot be removed from obj_list so we can continue
812 	 * iteration after regaining conn->lock.
813 	 */
814 	hlist_for_each_entry(mark, &conn->list, obj_list) {
815 		fsnotify_get_mark(mark);
816 		spin_unlock(&conn->lock);
817 		if (old_mark)
818 			fsnotify_put_mark(old_mark);
819 		old_mark = mark;
820 		fsnotify_destroy_mark(mark, mark->group);
821 		spin_lock(&conn->lock);
822 	}
823 	/*
824 	 * Detach list from object now so that we don't pin inode until all
825 	 * mark references get dropped. It would lead to strange results such
826 	 * as delaying inode deletion or blocking unmount.
827 	 */
828 	objp = fsnotify_detach_connector_from_object(conn, &type);
829 	spin_unlock(&conn->lock);
830 	if (old_mark)
831 		fsnotify_put_mark(old_mark);
832 	fsnotify_drop_object(type, objp);
833 }
834 
835 /*
836  * Nothing fancy, just initialize lists and locks and counters.
837  */
838 void fsnotify_init_mark(struct fsnotify_mark *mark,
839 			struct fsnotify_group *group)
840 {
841 	memset(mark, 0, sizeof(*mark));
842 	spin_lock_init(&mark->lock);
843 	refcount_set(&mark->refcnt, 1);
844 	fsnotify_get_group(group);
845 	mark->group = group;
846 	WRITE_ONCE(mark->connector, NULL);
847 }
848 EXPORT_SYMBOL_GPL(fsnotify_init_mark);
849 
850 /*
851  * Destroy all marks in destroy_list, waits for SRCU period to finish before
852  * actually freeing marks.
853  */
854 static void fsnotify_mark_destroy_workfn(struct work_struct *work)
855 {
856 	struct fsnotify_mark *mark, *next;
857 	struct list_head private_destroy_list;
858 
859 	spin_lock(&destroy_lock);
860 	/* exchange the list head */
861 	list_replace_init(&destroy_list, &private_destroy_list);
862 	spin_unlock(&destroy_lock);
863 
864 	synchronize_srcu(&fsnotify_mark_srcu);
865 
866 	list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) {
867 		list_del_init(&mark->g_list);
868 		fsnotify_final_mark_destroy(mark);
869 	}
870 }
871 
872 /* Wait for all marks queued for destruction to be actually destroyed */
873 void fsnotify_wait_marks_destroyed(void)
874 {
875 	flush_delayed_work(&reaper_work);
876 }
877 EXPORT_SYMBOL_GPL(fsnotify_wait_marks_destroyed);
878