1 // SPDX-License-Identifier: GPL-2.0-only
5 * (C) Copyright IBM Corporation 2005.
6 * Author : Ram Pai (linuxram@us.ibm.com)
8 #include <linux/mnt_namespace.h>
9 #include <linux/mount.h>
11 #include <linux/nsproxy.h>
12 #include <uapi/linux/mount.h>
16 /* return the next shared peer mount of @p */
17 static inline struct mount *next_peer(struct mount *p)
19 return list_entry(p->mnt_share.next, struct mount, mnt_share);
22 static inline struct mount *first_slave(struct mount *p)
24 return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
27 static inline struct mount *last_slave(struct mount *p)
29 return list_entry(p->mnt_slave_list.prev, struct mount, mnt_slave);
32 static inline struct mount *next_slave(struct mount *p)
34 return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
37 static struct mount *get_peer_under_root(struct mount *mnt,
38 struct mnt_namespace *ns,
39 const struct path *root)
41 struct mount *m = mnt;
44 /* Check the namespace first for optimization */
45 if (m->mnt_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
55 * Get ID of closest dominating peer group having a representative
56 * under the given root.
58 * Caller must hold namespace_sem
60 int get_dominating_id(struct mount *mnt, const struct path *root)
64 for (m = mnt->mnt_master; m != NULL; m = m->mnt_master) {
65 struct mount *d = get_peer_under_root(m, mnt->mnt_ns, root);
67 return d->mnt_group_id;
73 static int do_make_slave(struct mount *mnt)
75 struct mount *master, *slave_mnt;
77 if (list_empty(&mnt->mnt_share)) {
78 if (IS_MNT_SHARED(mnt)) {
79 mnt_release_group_id(mnt);
80 CLEAR_MNT_SHARED(mnt);
82 master = mnt->mnt_master;
84 struct list_head *p = &mnt->mnt_slave_list;
85 while (!list_empty(p)) {
86 slave_mnt = list_first_entry(p,
87 struct mount, mnt_slave);
88 list_del_init(&slave_mnt->mnt_slave);
89 slave_mnt->mnt_master = NULL;
96 * slave 'mnt' to a peer mount that has the
97 * same root dentry. If none is available then
98 * slave it to anything that is available.
100 for (m = master = next_peer(mnt); m != mnt; m = next_peer(m)) {
101 if (m->mnt.mnt_root == mnt->mnt.mnt_root) {
106 list_del_init(&mnt->mnt_share);
107 mnt->mnt_group_id = 0;
108 CLEAR_MNT_SHARED(mnt);
110 list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
111 slave_mnt->mnt_master = master;
112 list_move(&mnt->mnt_slave, &master->mnt_slave_list);
113 list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
114 INIT_LIST_HEAD(&mnt->mnt_slave_list);
115 mnt->mnt_master = master;
120 * vfsmount lock must be held for write
122 void change_mnt_propagation(struct mount *mnt, int type)
124 if (type == MS_SHARED) {
129 if (type != MS_SLAVE) {
130 list_del_init(&mnt->mnt_slave);
131 mnt->mnt_master = NULL;
132 if (type == MS_UNBINDABLE)
133 mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
135 mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
140 * get the next mount in the propagation tree.
141 * @m: the mount seen last
142 * @origin: the original mount from where the tree walk initiated
144 * Note that peer groups form contiguous segments of slave lists.
145 * We rely on that in get_source() to be able to find out if
146 * vfsmount found while iterating with propagation_next() is
147 * a peer of one we'd found earlier.
149 static struct mount *propagation_next(struct mount *m,
150 struct mount *origin)
152 /* are there any slaves of this mount? */
153 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
154 return first_slave(m);
157 struct mount *master = m->mnt_master;
159 if (master == origin->mnt_master) {
160 struct mount *next = next_peer(m);
161 return (next == origin) ? NULL : next;
162 } else if (m->mnt_slave.next != &master->mnt_slave_list)
163 return next_slave(m);
170 static struct mount *skip_propagation_subtree(struct mount *m,
171 struct mount *origin)
174 * Advance m such that propagation_next will not return
177 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
183 static struct mount *next_group(struct mount *m, struct mount *origin)
188 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
189 return first_slave(m);
191 if (m->mnt_group_id == origin->mnt_group_id) {
194 } else if (m->mnt_slave.next != &next->mnt_slave)
198 /* m is the last peer */
200 struct mount *master = m->mnt_master;
201 if (m->mnt_slave.next != &master->mnt_slave_list)
202 return next_slave(m);
203 m = next_peer(master);
204 if (master->mnt_group_id == origin->mnt_group_id)
206 if (master->mnt_slave.next == &m->mnt_slave)
215 /* all accesses are serialized by namespace_sem */
216 static struct mount *last_dest, *first_source, *last_source, *dest_master;
217 static struct hlist_head *list;
219 static inline bool peers(struct mount *m1, struct mount *m2)
221 return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
224 static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
228 /* skip ones added by this propagate_mnt() */
231 /* skip if mountpoint isn't covered by it */
232 if (!is_subdir(dest_mp->m_dentry, m->mnt.mnt_root))
234 if (peers(m, last_dest)) {
235 type = CL_MAKE_SHARED;
239 for (n = m; ; n = p) {
241 if (p == dest_master || IS_MNT_MARKED(p))
245 struct mount *parent = last_source->mnt_parent;
246 if (peers(last_source, first_source))
248 done = parent->mnt_master == p;
249 if (done && peers(n, parent))
251 last_source = last_source->mnt_master;
255 /* beginning of peer group among the slaves? */
256 if (IS_MNT_SHARED(m))
257 type |= CL_MAKE_SHARED;
260 child = copy_tree(last_source, last_source->mnt.mnt_root, type);
262 return PTR_ERR(child);
263 read_seqlock_excl(&mount_lock);
264 mnt_set_mountpoint(m, dest_mp, child);
265 if (m->mnt_master != dest_master)
266 SET_MNT_MARK(m->mnt_master);
267 read_sequnlock_excl(&mount_lock);
270 hlist_add_head(&child->mnt_hash, list);
271 return count_mounts(m->mnt_ns, child);
275 * mount 'source_mnt' under the destination 'dest_mnt' at
276 * dentry 'dest_dentry'. And propagate that mount to
277 * all the peer and slave mounts of 'dest_mnt'.
278 * Link all the new mounts into a propagation tree headed at
279 * source_mnt. Also link all the new mounts using ->mnt_list
280 * headed at source_mnt's ->mnt_list
282 * @dest_mnt: destination mount.
283 * @dest_dentry: destination dentry.
284 * @source_mnt: source mount.
285 * @tree_list : list of heads of trees to be attached.
287 int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
288 struct mount *source_mnt, struct hlist_head *tree_list)
294 * we don't want to bother passing tons of arguments to
295 * propagate_one(); everything is serialized by namespace_sem,
296 * so globals will do just fine.
298 last_dest = dest_mnt;
299 first_source = source_mnt;
300 last_source = source_mnt;
302 dest_master = dest_mnt->mnt_master;
304 /* all peers of dest_mnt, except dest_mnt itself */
305 for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
306 ret = propagate_one(n, dest_mp);
311 /* all slave groups */
312 for (m = next_group(dest_mnt, dest_mnt); m;
313 m = next_group(m, dest_mnt)) {
314 /* everything in that slave group */
317 ret = propagate_one(n, dest_mp);
324 read_seqlock_excl(&mount_lock);
325 hlist_for_each_entry(n, tree_list, mnt_hash) {
327 if (m->mnt_master != dest_mnt->mnt_master)
328 CLEAR_MNT_MARK(m->mnt_master);
330 read_sequnlock_excl(&mount_lock);
334 static struct mount *find_topper(struct mount *mnt)
336 /* If there is exactly one mount covering mnt completely return it. */
339 if (!list_is_singular(&mnt->mnt_mounts))
342 child = list_first_entry(&mnt->mnt_mounts, struct mount, mnt_child);
343 if (child->mnt_mountpoint != mnt->mnt.mnt_root)
350 * return true if the refcount is greater than count
352 static inline int do_refcount_check(struct mount *mnt, int count)
354 return mnt_get_count(mnt) > count;
358 * check if the mount 'mnt' can be unmounted successfully.
359 * @mnt: the mount to be checked for unmount
360 * NOTE: unmounting 'mnt' would naturally propagate to all
361 * other mounts its parent propagates to.
362 * Check if any of these mounts that **do not have submounts**
363 * have more references than 'refcnt'. If so return busy.
365 * vfsmount lock must be held for write
367 int propagate_mount_busy(struct mount *mnt, int refcnt)
369 struct mount *m, *child, *topper;
370 struct mount *parent = mnt->mnt_parent;
373 return do_refcount_check(mnt, refcnt);
376 * quickly check if the current mount can be unmounted.
377 * If not, we don't have to go checking for all other
380 if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
383 for (m = propagation_next(parent, parent); m;
384 m = propagation_next(m, parent)) {
386 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
390 /* Is there exactly one mount on the child that covers
391 * it completely whose reference should be ignored?
393 topper = find_topper(child);
396 else if (!list_empty(&child->mnt_mounts))
399 if (do_refcount_check(child, count))
406 * Clear MNT_LOCKED when it can be shown to be safe.
408 * mount_lock lock must be held for write
410 void propagate_mount_unlock(struct mount *mnt)
412 struct mount *parent = mnt->mnt_parent;
413 struct mount *m, *child;
415 BUG_ON(parent == mnt);
417 for (m = propagation_next(parent, parent); m;
418 m = propagation_next(m, parent)) {
419 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
421 child->mnt.mnt_flags &= ~MNT_LOCKED;
425 static void umount_one(struct mount *mnt, struct list_head *to_umount)
428 mnt->mnt.mnt_flags |= MNT_UMOUNT;
429 list_del_init(&mnt->mnt_child);
430 list_del_init(&mnt->mnt_umounting);
431 list_move_tail(&mnt->mnt_list, to_umount);
435 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
436 * parent propagates to.
438 static bool __propagate_umount(struct mount *mnt,
439 struct list_head *to_umount,
440 struct list_head *to_restore)
442 bool progress = false;
446 * The state of the parent won't change if this mount is
447 * already unmounted or marked as without children.
449 if (mnt->mnt.mnt_flags & (MNT_UMOUNT | MNT_MARKED))
452 /* Verify topper is the only grandchild that has not been
453 * speculatively unmounted.
455 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
456 if (child->mnt_mountpoint == mnt->mnt.mnt_root)
458 if (!list_empty(&child->mnt_umounting) && IS_MNT_MARKED(child))
460 /* Found a mounted child */
464 /* Mark mounts that can be unmounted if not locked */
468 /* If a mount is without children and not locked umount it. */
469 if (!IS_MNT_LOCKED(mnt)) {
470 umount_one(mnt, to_umount);
473 list_move_tail(&mnt->mnt_umounting, to_restore);
479 static void umount_list(struct list_head *to_umount,
480 struct list_head *to_restore)
482 struct mount *mnt, *child, *tmp;
483 list_for_each_entry(mnt, to_umount, mnt_list) {
484 list_for_each_entry_safe(child, tmp, &mnt->mnt_mounts, mnt_child) {
486 if (child->mnt_mountpoint == mnt->mnt.mnt_root)
487 list_move_tail(&child->mnt_umounting, to_restore);
489 umount_one(child, to_umount);
494 static void restore_mounts(struct list_head *to_restore)
496 /* Restore mounts to a clean working state */
497 while (!list_empty(to_restore)) {
498 struct mount *mnt, *parent;
499 struct mountpoint *mp;
501 mnt = list_first_entry(to_restore, struct mount, mnt_umounting);
503 list_del_init(&mnt->mnt_umounting);
505 /* Should this mount be reparented? */
507 parent = mnt->mnt_parent;
508 while (parent->mnt.mnt_flags & MNT_UMOUNT) {
510 parent = parent->mnt_parent;
512 if (parent != mnt->mnt_parent)
513 mnt_change_mountpoint(parent, mp, mnt);
517 static void cleanup_umount_visitations(struct list_head *visited)
519 while (!list_empty(visited)) {
521 list_first_entry(visited, struct mount, mnt_umounting);
522 list_del_init(&mnt->mnt_umounting);
527 * collect all mounts that receive propagation from the mount in @list,
528 * and return these additional mounts in the same list.
529 * @list: the list of mounts to be unmounted.
531 * vfsmount lock must be held for write
533 int propagate_umount(struct list_head *list)
536 LIST_HEAD(to_restore);
537 LIST_HEAD(to_umount);
540 /* Find candidates for unmounting */
541 list_for_each_entry_reverse(mnt, list, mnt_list) {
542 struct mount *parent = mnt->mnt_parent;
546 * If this mount has already been visited it is known that it's
547 * entire peer group and all of their slaves in the propagation
548 * tree for the mountpoint has already been visited and there is
549 * no need to visit them again.
551 if (!list_empty(&mnt->mnt_umounting))
554 list_add_tail(&mnt->mnt_umounting, &visited);
555 for (m = propagation_next(parent, parent); m;
556 m = propagation_next(m, parent)) {
557 struct mount *child = __lookup_mnt(&m->mnt,
558 mnt->mnt_mountpoint);
562 if (!list_empty(&child->mnt_umounting)) {
564 * If the child has already been visited it is
565 * know that it's entire peer group and all of
566 * their slaves in the propgation tree for the
567 * mountpoint has already been visited and there
568 * is no need to visit this subtree again.
570 m = skip_propagation_subtree(m, parent);
572 } else if (child->mnt.mnt_flags & MNT_UMOUNT) {
574 * We have come accross an partially unmounted
575 * mount in list that has not been visited yet.
576 * Remember it has been visited and continue
577 * about our merry way.
579 list_add_tail(&child->mnt_umounting, &visited);
583 /* Check the child and parents while progress is made */
584 while (__propagate_umount(child,
585 &to_umount, &to_restore)) {
586 /* Is the parent a umount candidate? */
587 child = child->mnt_parent;
588 if (list_empty(&child->mnt_umounting))
594 umount_list(&to_umount, &to_restore);
595 restore_mounts(&to_restore);
596 cleanup_umount_visitations(&visited);
597 list_splice_tail(&to_umount, list);