1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/pnode.c
4 *
5 * (C) Copyright IBM Corporation 2005.
6 * Author : Ram Pai (linuxram@us.ibm.com)
7 */
8#include <linux/mnt_namespace.h>
9#include <linux/mount.h>
10#include <linux/fs.h>
11#include <linux/nsproxy.h>
12#include <uapi/linux/mount.h>
13#include "internal.h"
14#include "pnode.h"
15
16/* return the next shared peer mount of @p */
17static inline struct mount *next_peer(struct mount *p)
18{
19 return list_entry(p->mnt_share.next, struct mount, mnt_share);
20}
21
22static inline struct mount *first_slave(struct mount *p)
23{
24 return hlist_entry(p->mnt_slave_list.first, struct mount, mnt_slave);
25}
26
27static inline struct mount *next_slave(struct mount *p)
28{
29 return hlist_entry(p->mnt_slave.next, struct mount, mnt_slave);
30}
31
32/* locks: namespace_shared && is_mounted(mnt) */
33static struct mount *get_peer_under_root(struct mount *mnt,
34 struct mnt_namespace *ns,
35 const struct path *root)
36{
37 struct mount *m = mnt;
38
39 do {
40 /* Check the namespace first for optimization */
41 if (m->mnt_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
42 return m;
43
44 m = next_peer(p: m);
45 } while (m != mnt);
46
47 return NULL;
48}
49
50/*
51 * Get ID of closest dominating peer group having a representative
52 * under the given root.
53 *
54 * locks: namespace_shared
55 */
56int get_dominating_id(struct mount *mnt, const struct path *root)
57{
58 struct mount *m;
59
60 for (m = mnt->mnt_master; m != NULL; m = m->mnt_master) {
61 struct mount *d = get_peer_under_root(mnt: m, ns: mnt->mnt_ns, root);
62 if (d)
63 return d->mnt_group_id;
64 }
65
66 return 0;
67}
68
69static inline bool will_be_unmounted(struct mount *m)
70{
71 return m->mnt.mnt_flags & MNT_UMOUNT;
72}
73
74static void transfer_propagation(struct mount *mnt, struct mount *to)
75{
76 struct hlist_node *p = NULL, *n;
77 struct mount *m;
78
79 hlist_for_each_entry_safe(m, n, &mnt->mnt_slave_list, mnt_slave) {
80 m->mnt_master = to;
81 if (!to)
82 hlist_del_init(n: &m->mnt_slave);
83 else
84 p = &m->mnt_slave;
85 }
86 if (p)
87 hlist_splice_init(from: &mnt->mnt_slave_list, last: p, to: &to->mnt_slave_list);
88}
89
90/*
91 * EXCL[namespace_sem]
92 */
93void change_mnt_propagation(struct mount *mnt, int type)
94{
95 struct mount *m = mnt->mnt_master;
96
97 if (type == MS_SHARED) {
98 set_mnt_shared(mnt);
99 return;
100 }
101 if (IS_MNT_SHARED(mnt)) {
102 if (list_empty(head: &mnt->mnt_share)) {
103 mnt_release_group_id(mnt);
104 } else {
105 m = next_peer(p: mnt);
106 list_del_init(entry: &mnt->mnt_share);
107 mnt->mnt_group_id = 0;
108 }
109 CLEAR_MNT_SHARED(mnt);
110 transfer_propagation(mnt, to: m);
111 }
112 hlist_del_init(n: &mnt->mnt_slave);
113 if (type == MS_SLAVE) {
114 mnt->mnt_master = m;
115 if (m)
116 hlist_add_head(n: &mnt->mnt_slave, h: &m->mnt_slave_list);
117 } else {
118 mnt->mnt_master = NULL;
119 if (type == MS_UNBINDABLE)
120 mnt->mnt_t_flags |= T_UNBINDABLE;
121 else
122 mnt->mnt_t_flags &= ~T_UNBINDABLE;
123 }
124}
125
126static struct mount *trace_transfers(struct mount *m)
127{
128 while (1) {
129 struct mount *next = next_peer(p: m);
130
131 if (next != m) {
132 list_del_init(entry: &m->mnt_share);
133 m->mnt_group_id = 0;
134 m->mnt_master = next;
135 } else {
136 if (IS_MNT_SHARED(m))
137 mnt_release_group_id(m);
138 next = m->mnt_master;
139 }
140 hlist_del_init(n: &m->mnt_slave);
141 CLEAR_MNT_SHARED(m);
142 SET_MNT_MARK(m);
143
144 if (!next || !will_be_unmounted(m: next))
145 return next;
146 if (IS_MNT_MARKED(next))
147 return next->mnt_master;
148 m = next;
149 }
150}
151
152static void set_destinations(struct mount *m, struct mount *master)
153{
154 struct mount *next;
155
156 while ((next = m->mnt_master) != master) {
157 m->mnt_master = master;
158 m = next;
159 }
160}
161
162void bulk_make_private(struct list_head *set)
163{
164 struct mount *m;
165
166 list_for_each_entry(m, set, mnt_list)
167 if (!IS_MNT_MARKED(m))
168 set_destinations(m, master: trace_transfers(m));
169
170 list_for_each_entry(m, set, mnt_list) {
171 transfer_propagation(mnt: m, to: m->mnt_master);
172 m->mnt_master = NULL;
173 CLEAR_MNT_MARK(m);
174 }
175}
176
177static struct mount *__propagation_next(struct mount *m,
178 struct mount *origin)
179{
180 while (1) {
181 struct mount *master = m->mnt_master;
182
183 if (master == origin->mnt_master) {
184 struct mount *next = next_peer(p: m);
185 return (next == origin) ? NULL : next;
186 } else if (m->mnt_slave.next)
187 return next_slave(p: m);
188
189 /* back at master */
190 m = master;
191 }
192}
193
194/*
195 * get the next mount in the propagation tree.
196 * @m: the mount seen last
197 * @origin: the original mount from where the tree walk initiated
198 *
199 * Note that peer groups form contiguous segments of slave lists.
200 * We rely on that in get_source() to be able to find out if
201 * vfsmount found while iterating with propagation_next() is
202 * a peer of one we'd found earlier.
203 */
204static struct mount *propagation_next(struct mount *m,
205 struct mount *origin)
206{
207 /* are there any slaves of this mount? */
208 if (!IS_MNT_NEW(m) && !hlist_empty(h: &m->mnt_slave_list))
209 return first_slave(p: m);
210
211 return __propagation_next(m, origin);
212}
213
214static struct mount *skip_propagation_subtree(struct mount *m,
215 struct mount *origin)
216{
217 /*
218 * Advance m past everything that gets propagation from it.
219 */
220 struct mount *p = __propagation_next(m, origin);
221
222 while (p && peers(m1: m, m2: p))
223 p = __propagation_next(m: p, origin);
224
225 return p;
226}
227
228static struct mount *next_group(struct mount *m, struct mount *origin)
229{
230 while (1) {
231 while (1) {
232 struct mount *next;
233 if (!IS_MNT_NEW(m) && !hlist_empty(h: &m->mnt_slave_list))
234 return first_slave(p: m);
235 next = next_peer(p: m);
236 if (m->mnt_group_id == origin->mnt_group_id) {
237 if (next == origin)
238 return NULL;
239 } else if (m->mnt_slave.next != &next->mnt_slave)
240 break;
241 m = next;
242 }
243 /* m is the last peer */
244 while (1) {
245 struct mount *master = m->mnt_master;
246 if (m->mnt_slave.next)
247 return next_slave(p: m);
248 m = next_peer(p: master);
249 if (master->mnt_group_id == origin->mnt_group_id)
250 break;
251 if (master->mnt_slave.next == &m->mnt_slave)
252 break;
253 m = master;
254 }
255 if (m == origin)
256 return NULL;
257 }
258}
259
260static bool need_secondary(struct mount *m, struct mountpoint *dest_mp)
261{
262 /* skip ones added by this propagate_mnt() */
263 if (IS_MNT_NEW(m))
264 return false;
265 /* skip if mountpoint isn't visible in m */
266 if (!is_subdir(dest_mp->m_dentry, m->mnt.mnt_root))
267 return false;
268 /* skip if m is in the anon_ns */
269 if (is_anon_ns(ns: m->mnt_ns))
270 return false;
271 return true;
272}
273
274static struct mount *find_master(struct mount *m,
275 struct mount *last_copy,
276 struct mount *original)
277{
278 struct mount *p;
279
280 // ascend until there's a copy for something with the same master
281 for (;;) {
282 p = m->mnt_master;
283 if (!p || IS_MNT_MARKED(p))
284 break;
285 m = p;
286 }
287 while (!peers(m1: last_copy, m2: original)) {
288 struct mount *parent = last_copy->mnt_parent;
289 if (parent->mnt_master == p) {
290 if (!peers(m1: parent, m2: m))
291 last_copy = last_copy->mnt_master;
292 break;
293 }
294 last_copy = last_copy->mnt_master;
295 }
296 return last_copy;
297}
298
299/**
300 * propagate_mnt() - create secondary copies for tree attachment
301 * @dest_mnt: destination mount.
302 * @dest_mp: destination mountpoint.
303 * @source_mnt: source mount.
304 * @tree_list: list of secondaries to be attached.
305 *
306 * Create secondary copies for attaching a tree with root @source_mnt
307 * at mount @dest_mnt with mountpoint @dest_mp. Link all new mounts
308 * into a propagation graph. Set mountpoints for all secondaries,
309 * link their roots into @tree_list via ->mnt_hash.
310 */
311int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
312 struct mount *source_mnt, struct hlist_head *tree_list)
313{
314 struct mount *m, *n, *copy, *this;
315 int err = 0, type;
316
317 if (dest_mnt->mnt_master)
318 SET_MNT_MARK(dest_mnt->mnt_master);
319
320 /* iterate over peer groups, depth first */
321 for (m = dest_mnt; m && !err; m = next_group(m, origin: dest_mnt)) {
322 if (m == dest_mnt) { // have one for dest_mnt itself
323 copy = source_mnt;
324 type = CL_MAKE_SHARED;
325 n = next_peer(p: m);
326 if (n == m)
327 continue;
328 } else {
329 type = CL_SLAVE;
330 /* beginning of peer group among the slaves? */
331 if (IS_MNT_SHARED(m))
332 type |= CL_MAKE_SHARED;
333 n = m;
334 }
335 do {
336 if (!need_secondary(m: n, dest_mp))
337 continue;
338 if (type & CL_SLAVE) // first in this peer group
339 copy = find_master(m: n, last_copy: copy, original: source_mnt);
340 this = copy_tree(copy, copy->mnt.mnt_root, type);
341 if (IS_ERR(ptr: this)) {
342 err = PTR_ERR(ptr: this);
343 break;
344 }
345 scoped_guard(mount_locked_reader)
346 mnt_set_mountpoint(n, dest_mp, this);
347 if (n->mnt_master)
348 SET_MNT_MARK(n->mnt_master);
349 copy = this;
350 hlist_add_head(n: &this->mnt_hash, h: tree_list);
351 err = count_mounts(ns: n->mnt_ns, mnt: this);
352 if (err)
353 break;
354 type = CL_MAKE_SHARED;
355 } while ((n = next_peer(p: n)) != m);
356 }
357
358 hlist_for_each_entry(n, tree_list, mnt_hash) {
359 m = n->mnt_parent;
360 if (m->mnt_master)
361 CLEAR_MNT_MARK(m->mnt_master);
362 }
363 if (dest_mnt->mnt_master)
364 CLEAR_MNT_MARK(dest_mnt->mnt_master);
365 return err;
366}
367
368/*
369 * return true if the refcount is greater than count
370 */
371static inline int do_refcount_check(struct mount *mnt, int count)
372{
373 return mnt_get_count(mnt) > count;
374}
375
376/**
377 * propagation_would_overmount - check whether propagation from @from
378 * would overmount @to
379 * @from: shared mount
380 * @to: mount to check
381 * @mp: future mountpoint of @to on @from
382 *
383 * If @from propagates mounts to @to, @from and @to must either be peers
384 * or one of the masters in the hierarchy of masters of @to must be a
385 * peer of @from.
386 *
387 * If the root of the @to mount is equal to the future mountpoint @mp of
388 * the @to mount on @from then @to will be overmounted by whatever is
389 * propagated to it.
390 *
391 * Context: This function expects namespace_lock() to be held and that
392 * @mp is stable.
393 * Return: If @from overmounts @to, true is returned, false if not.
394 */
395bool propagation_would_overmount(const struct mount *from,
396 const struct mount *to,
397 const struct mountpoint *mp)
398{
399 if (!IS_MNT_SHARED(from))
400 return false;
401
402 if (to->mnt.mnt_root != mp->m_dentry)
403 return false;
404
405 for (const struct mount *m = to; m; m = m->mnt_master) {
406 if (peers(m1: from, m2: m))
407 return true;
408 }
409
410 return false;
411}
412
413/*
414 * check if the mount 'mnt' can be unmounted successfully.
415 * @mnt: the mount to be checked for unmount
416 * NOTE: unmounting 'mnt' would naturally propagate to all
417 * other mounts its parent propagates to.
418 * Check if any of these mounts that **do not have submounts**
419 * have more references than 'refcnt'. If so return busy.
420 *
421 * vfsmount lock must be held for write
422 */
423int propagate_mount_busy(struct mount *mnt, int refcnt)
424{
425 struct mount *parent = mnt->mnt_parent;
426
427 /*
428 * quickly check if the current mount can be unmounted.
429 * If not, we don't have to go checking for all other
430 * mounts
431 */
432 if (!list_empty(head: &mnt->mnt_mounts) || do_refcount_check(mnt, count: refcnt))
433 return 1;
434
435 if (mnt == parent)
436 return 0;
437
438 for (struct mount *m = propagation_next(m: parent, origin: parent); m;
439 m = propagation_next(m, origin: parent)) {
440 struct list_head *head;
441 struct mount *child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
442
443 if (!child)
444 continue;
445
446 head = &child->mnt_mounts;
447 if (!list_empty(head)) {
448 /*
449 * a mount that covers child completely wouldn't prevent
450 * it being pulled out; any other would.
451 */
452 if (!list_is_singular(head) || !child->overmount)
453 continue;
454 }
455 if (do_refcount_check(mnt: child, count: 1))
456 return 1;
457 }
458 return 0;
459}
460
461/*
462 * Clear MNT_LOCKED when it can be shown to be safe.
463 *
464 * mount_lock lock must be held for write
465 */
466void propagate_mount_unlock(struct mount *mnt)
467{
468 struct mount *parent = mnt->mnt_parent;
469 struct mount *m, *child;
470
471 BUG_ON(parent == mnt);
472
473 for (m = propagation_next(m: parent, origin: parent); m;
474 m = propagation_next(m, origin: parent)) {
475 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
476 if (child)
477 child->mnt.mnt_flags &= ~MNT_LOCKED;
478 }
479}
480
481static inline bool is_candidate(struct mount *m)
482{
483 return m->mnt_t_flags & T_UMOUNT_CANDIDATE;
484}
485
486static void umount_one(struct mount *m, struct list_head *to_umount)
487{
488 m->mnt.mnt_flags |= MNT_UMOUNT;
489 list_del_init(entry: &m->mnt_child);
490 move_from_ns(mnt: m);
491 list_add_tail(new: &m->mnt_list, head: to_umount);
492}
493
494static void remove_from_candidate_list(struct mount *m)
495{
496 m->mnt_t_flags &= ~(T_MARKED | T_UMOUNT_CANDIDATE);
497 list_del_init(entry: &m->mnt_list);
498}
499
500static void gather_candidates(struct list_head *set,
501 struct list_head *candidates)
502{
503 struct mount *m, *p, *q;
504
505 list_for_each_entry(m, set, mnt_list) {
506 if (is_candidate(m))
507 continue;
508 m->mnt_t_flags |= T_UMOUNT_CANDIDATE;
509 p = m->mnt_parent;
510 q = propagation_next(m: p, origin: p);
511 while (q) {
512 struct mount *child = __lookup_mnt(&q->mnt,
513 m->mnt_mountpoint);
514 if (child) {
515 /*
516 * We might've already run into this one. That
517 * must've happened on earlier iteration of the
518 * outer loop; in that case we can skip those
519 * parents that get propagation from q - there
520 * will be nothing new on those as well.
521 */
522 if (is_candidate(m: child)) {
523 q = skip_propagation_subtree(m: q, origin: p);
524 continue;
525 }
526 child->mnt_t_flags |= T_UMOUNT_CANDIDATE;
527 if (!will_be_unmounted(m: child))
528 list_add(new: &child->mnt_list, head: candidates);
529 }
530 q = propagation_next(m: q, origin: p);
531 }
532 }
533 list_for_each_entry(m, set, mnt_list)
534 m->mnt_t_flags &= ~T_UMOUNT_CANDIDATE;
535}
536
537/*
538 * We know that some child of @m can't be unmounted. In all places where the
539 * chain of descent of @m has child not overmounting the root of parent,
540 * the parent can't be unmounted either.
541 */
542static void trim_ancestors(struct mount *m)
543{
544 struct mount *p;
545
546 for (p = m->mnt_parent; is_candidate(m: p); m = p, p = p->mnt_parent) {
547 if (IS_MNT_MARKED(m)) // all candidates beneath are overmounts
548 return;
549 SET_MNT_MARK(m);
550 if (m != p->overmount)
551 p->mnt_t_flags &= ~T_UMOUNT_CANDIDATE;
552 }
553}
554
555/*
556 * Find and exclude all umount candidates forbidden by @m
557 * (see Documentation/filesystems/propagate_umount.txt)
558 * If we can immediately tell that @m is OK to unmount (unlocked
559 * and all children are already committed to unmounting) commit
560 * to unmounting it.
561 * Only @m itself might be taken from the candidates list;
562 * anything found by trim_ancestors() is marked non-candidate
563 * and left on the list.
564 */
565static void trim_one(struct mount *m, struct list_head *to_umount)
566{
567 bool remove_this = false, found = false, umount_this = false;
568 struct mount *n;
569
570 if (!is_candidate(m)) { // trim_ancestors() left it on list
571 remove_from_candidate_list(m);
572 return;
573 }
574
575 list_for_each_entry(n, &m->mnt_mounts, mnt_child) {
576 if (!is_candidate(m: n)) {
577 found = true;
578 if (n != m->overmount) {
579 remove_this = true;
580 break;
581 }
582 }
583 }
584 if (found) {
585 trim_ancestors(m);
586 } else if (!IS_MNT_LOCKED(m) && list_empty(head: &m->mnt_mounts)) {
587 remove_this = true;
588 umount_this = true;
589 }
590 if (remove_this) {
591 remove_from_candidate_list(m);
592 if (umount_this)
593 umount_one(m, to_umount);
594 }
595}
596
597static void handle_locked(struct mount *m, struct list_head *to_umount)
598{
599 struct mount *cutoff = m, *p;
600
601 if (!is_candidate(m)) { // trim_ancestors() left it on list
602 remove_from_candidate_list(m);
603 return;
604 }
605 for (p = m; is_candidate(m: p); p = p->mnt_parent) {
606 remove_from_candidate_list(m: p);
607 if (!IS_MNT_LOCKED(p))
608 cutoff = p->mnt_parent;
609 }
610 if (will_be_unmounted(m: p))
611 cutoff = p;
612 while (m != cutoff) {
613 umount_one(m, to_umount);
614 m = m->mnt_parent;
615 }
616}
617
618/*
619 * @m is not to going away, and it overmounts the top of a stack of mounts
620 * that are going away. We know that all of those are fully overmounted
621 * by the one above (@m being the topmost of the chain), so @m can be slid
622 * in place where the bottom of the stack is attached.
623 *
624 * NOTE: here we temporarily violate a constraint - two mounts end up with
625 * the same parent and mountpoint; that will be remedied as soon as we
626 * return from propagate_umount() - its caller (umount_tree()) will detach
627 * the stack from the parent it (and now @m) is attached to. umount_tree()
628 * might choose to keep unmounted pieces stuck to each other, but it always
629 * detaches them from the mounts that remain in the tree.
630 */
631static void reparent(struct mount *m)
632{
633 struct mount *p = m;
634 struct mountpoint *mp;
635
636 do {
637 mp = p->mnt_mp;
638 p = p->mnt_parent;
639 } while (will_be_unmounted(m: p));
640
641 mnt_change_mountpoint(parent: p, mp, mnt: m);
642 mnt_notify_add(m);
643}
644
645/**
646 * propagate_umount - apply propagation rules to the set of mounts for umount()
647 * @set: the list of mounts to be unmounted.
648 *
649 * Collect all mounts that receive propagation from the mount in @set and have
650 * no obstacles to being unmounted. Add these additional mounts to the set.
651 *
652 * See Documentation/filesystems/propagate_umount.txt if you do anything in
653 * this area.
654 *
655 * Locks held:
656 * mount_lock (write_seqlock), namespace_sem (exclusive).
657 */
658void propagate_umount(struct list_head *set)
659{
660 struct mount *m, *p;
661 LIST_HEAD(to_umount); // committed to unmounting
662 LIST_HEAD(candidates); // undecided umount candidates
663
664 // collect all candidates
665 gather_candidates(set, candidates: &candidates);
666
667 // reduce the set until it's non-shifting
668 list_for_each_entry_safe(m, p, &candidates, mnt_list)
669 trim_one(m, to_umount: &to_umount);
670
671 // ... and non-revealing
672 while (!list_empty(head: &candidates)) {
673 m = list_first_entry(&candidates,struct mount, mnt_list);
674 handle_locked(m, to_umount: &to_umount);
675 }
676
677 // now to_umount consists of all acceptable candidates
678 // deal with reparenting of surviving overmounts on those
679 list_for_each_entry(m, &to_umount, mnt_list) {
680 struct mount *over = m->overmount;
681 if (over && !will_be_unmounted(m: over))
682 reparent(m: over);
683 }
684
685 // and fold them into the set
686 list_splice_tail_init(list: &to_umount, head: set);
687}
688

source code of linux/fs/pnode.c