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 list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
25}
26
27static inline struct mount *last_slave(struct mount *p)
28{
29 return list_entry(p->mnt_slave_list.prev, struct mount, mnt_slave);
30}
31
32static inline struct mount *next_slave(struct mount *p)
33{
34 return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
35}
36
37static struct mount *get_peer_under_root(struct mount *mnt,
38 struct mnt_namespace *ns,
39 const struct path *root)
40{
41 struct mount *m = mnt;
42
43 do {
44 /* Check the namespace first for optimization */
45 if (m->mnt_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
46 return m;
47
48 m = next_peer(p: m);
49 } while (m != mnt);
50
51 return NULL;
52}
53
54/*
55 * Get ID of closest dominating peer group having a representative
56 * under the given root.
57 *
58 * Caller must hold namespace_sem
59 */
60int get_dominating_id(struct mount *mnt, const struct path *root)
61{
62 struct mount *m;
63
64 for (m = mnt->mnt_master; m != NULL; m = m->mnt_master) {
65 struct mount *d = get_peer_under_root(mnt: m, ns: mnt->mnt_ns, root);
66 if (d)
67 return d->mnt_group_id;
68 }
69
70 return 0;
71}
72
73static int do_make_slave(struct mount *mnt)
74{
75 struct mount *master, *slave_mnt;
76
77 if (list_empty(head: &mnt->mnt_share)) {
78 if (IS_MNT_SHARED(mnt)) {
79 mnt_release_group_id(mnt);
80 CLEAR_MNT_SHARED(mnt);
81 }
82 master = mnt->mnt_master;
83 if (!master) {
84 struct list_head *p = &mnt->mnt_slave_list;
85 while (!list_empty(head: p)) {
86 slave_mnt = list_first_entry(p,
87 struct mount, mnt_slave);
88 list_del_init(entry: &slave_mnt->mnt_slave);
89 slave_mnt->mnt_master = NULL;
90 }
91 return 0;
92 }
93 } else {
94 struct mount *m;
95 /*
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.
99 */
100 for (m = master = next_peer(p: mnt); m != mnt; m = next_peer(p: m)) {
101 if (m->mnt.mnt_root == mnt->mnt.mnt_root) {
102 master = m;
103 break;
104 }
105 }
106 list_del_init(entry: &mnt->mnt_share);
107 mnt->mnt_group_id = 0;
108 CLEAR_MNT_SHARED(mnt);
109 }
110 list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
111 slave_mnt->mnt_master = master;
112 list_move(list: &mnt->mnt_slave, head: &master->mnt_slave_list);
113 list_splice(list: &mnt->mnt_slave_list, head: master->mnt_slave_list.prev);
114 INIT_LIST_HEAD(list: &mnt->mnt_slave_list);
115 mnt->mnt_master = master;
116 return 0;
117}
118
119/*
120 * vfsmount lock must be held for write
121 */
122void change_mnt_propagation(struct mount *mnt, int type)
123{
124 if (type == MS_SHARED) {
125 set_mnt_shared(mnt);
126 return;
127 }
128 do_make_slave(mnt);
129 if (type != MS_SLAVE) {
130 list_del_init(entry: &mnt->mnt_slave);
131 mnt->mnt_master = NULL;
132 if (type == MS_UNBINDABLE)
133 mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
134 else
135 mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
136 }
137}
138
139/*
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
143 *
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.
148 */
149static struct mount *propagation_next(struct mount *m,
150 struct mount *origin)
151{
152 /* are there any slaves of this mount? */
153 if (!IS_MNT_NEW(m) && !list_empty(head: &m->mnt_slave_list))
154 return first_slave(p: m);
155
156 while (1) {
157 struct mount *master = m->mnt_master;
158
159 if (master == origin->mnt_master) {
160 struct mount *next = next_peer(p: m);
161 return (next == origin) ? NULL : next;
162 } else if (m->mnt_slave.next != &master->mnt_slave_list)
163 return next_slave(p: m);
164
165 /* back at master */
166 m = master;
167 }
168}
169
170static struct mount *skip_propagation_subtree(struct mount *m,
171 struct mount *origin)
172{
173 /*
174 * Advance m such that propagation_next will not return
175 * the slaves of m.
176 */
177 if (!IS_MNT_NEW(m) && !list_empty(head: &m->mnt_slave_list))
178 m = last_slave(p: m);
179
180 return m;
181}
182
183static struct mount *next_group(struct mount *m, struct mount *origin)
184{
185 while (1) {
186 while (1) {
187 struct mount *next;
188 if (!IS_MNT_NEW(m) && !list_empty(head: &m->mnt_slave_list))
189 return first_slave(p: m);
190 next = next_peer(p: m);
191 if (m->mnt_group_id == origin->mnt_group_id) {
192 if (next == origin)
193 return NULL;
194 } else if (m->mnt_slave.next != &next->mnt_slave)
195 break;
196 m = next;
197 }
198 /* m is the last peer */
199 while (1) {
200 struct mount *master = m->mnt_master;
201 if (m->mnt_slave.next != &master->mnt_slave_list)
202 return next_slave(p: m);
203 m = next_peer(p: master);
204 if (master->mnt_group_id == origin->mnt_group_id)
205 break;
206 if (master->mnt_slave.next == &m->mnt_slave)
207 break;
208 m = master;
209 }
210 if (m == origin)
211 return NULL;
212 }
213}
214
215/* all accesses are serialized by namespace_sem */
216static struct mount *last_dest, *first_source, *last_source, *dest_master;
217static struct hlist_head *list;
218
219static inline bool peers(const struct mount *m1, const struct mount *m2)
220{
221 return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
222}
223
224static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
225{
226 struct mount *child;
227 int type;
228 /* skip ones added by this propagate_mnt() */
229 if (IS_MNT_NEW(m))
230 return 0;
231 /* skip if mountpoint isn't visible in m */
232 if (!is_subdir(dest_mp->m_dentry, m->mnt.mnt_root))
233 return 0;
234 /* skip if m is in the anon_ns */
235 if (is_anon_ns(ns: m->mnt_ns))
236 return 0;
237
238 if (peers(m1: m, m2: last_dest)) {
239 type = CL_MAKE_SHARED;
240 } else {
241 struct mount *n, *p;
242 bool done;
243 for (n = m; ; n = p) {
244 p = n->mnt_master;
245 if (p == dest_master || IS_MNT_MARKED(p))
246 break;
247 }
248 do {
249 struct mount *parent = last_source->mnt_parent;
250 if (peers(m1: last_source, m2: first_source))
251 break;
252 done = parent->mnt_master == p;
253 if (done && peers(m1: n, m2: parent))
254 break;
255 last_source = last_source->mnt_master;
256 } while (!done);
257
258 type = CL_SLAVE;
259 /* beginning of peer group among the slaves? */
260 if (IS_MNT_SHARED(m))
261 type |= CL_MAKE_SHARED;
262 }
263
264 child = copy_tree(last_source, last_source->mnt.mnt_root, type);
265 if (IS_ERR(ptr: child))
266 return PTR_ERR(ptr: child);
267 read_seqlock_excl(sl: &mount_lock);
268 mnt_set_mountpoint(m, dest_mp, child);
269 if (m->mnt_master != dest_master)
270 SET_MNT_MARK(m->mnt_master);
271 read_sequnlock_excl(sl: &mount_lock);
272 last_dest = m;
273 last_source = child;
274 hlist_add_head(n: &child->mnt_hash, h: list);
275 return count_mounts(ns: m->mnt_ns, mnt: child);
276}
277
278/*
279 * mount 'source_mnt' under the destination 'dest_mnt' at
280 * dentry 'dest_dentry'. And propagate that mount to
281 * all the peer and slave mounts of 'dest_mnt'.
282 * Link all the new mounts into a propagation tree headed at
283 * source_mnt. Also link all the new mounts using ->mnt_list
284 * headed at source_mnt's ->mnt_list
285 *
286 * @dest_mnt: destination mount.
287 * @dest_dentry: destination dentry.
288 * @source_mnt: source mount.
289 * @tree_list : list of heads of trees to be attached.
290 */
291int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
292 struct mount *source_mnt, struct hlist_head *tree_list)
293{
294 struct mount *m, *n;
295 int ret = 0;
296
297 /*
298 * we don't want to bother passing tons of arguments to
299 * propagate_one(); everything is serialized by namespace_sem,
300 * so globals will do just fine.
301 */
302 last_dest = dest_mnt;
303 first_source = source_mnt;
304 last_source = source_mnt;
305 list = tree_list;
306 dest_master = dest_mnt->mnt_master;
307
308 /* all peers of dest_mnt, except dest_mnt itself */
309 for (n = next_peer(p: dest_mnt); n != dest_mnt; n = next_peer(p: n)) {
310 ret = propagate_one(m: n, dest_mp);
311 if (ret)
312 goto out;
313 }
314
315 /* all slave groups */
316 for (m = next_group(m: dest_mnt, origin: dest_mnt); m;
317 m = next_group(m, origin: dest_mnt)) {
318 /* everything in that slave group */
319 n = m;
320 do {
321 ret = propagate_one(m: n, dest_mp);
322 if (ret)
323 goto out;
324 n = next_peer(p: n);
325 } while (n != m);
326 }
327out:
328 read_seqlock_excl(sl: &mount_lock);
329 hlist_for_each_entry(n, tree_list, mnt_hash) {
330 m = n->mnt_parent;
331 if (m->mnt_master != dest_mnt->mnt_master)
332 CLEAR_MNT_MARK(m->mnt_master);
333 }
334 read_sequnlock_excl(sl: &mount_lock);
335 return ret;
336}
337
338static struct mount *find_topper(struct mount *mnt)
339{
340 /* If there is exactly one mount covering mnt completely return it. */
341 struct mount *child;
342
343 if (!list_is_singular(head: &mnt->mnt_mounts))
344 return NULL;
345
346 child = list_first_entry(&mnt->mnt_mounts, struct mount, mnt_child);
347 if (child->mnt_mountpoint != mnt->mnt.mnt_root)
348 return NULL;
349
350 return child;
351}
352
353/*
354 * return true if the refcount is greater than count
355 */
356static inline int do_refcount_check(struct mount *mnt, int count)
357{
358 return mnt_get_count(mnt) > count;
359}
360
361/**
362 * propagation_would_overmount - check whether propagation from @from
363 * would overmount @to
364 * @from: shared mount
365 * @to: mount to check
366 * @mp: future mountpoint of @to on @from
367 *
368 * If @from propagates mounts to @to, @from and @to must either be peers
369 * or one of the masters in the hierarchy of masters of @to must be a
370 * peer of @from.
371 *
372 * If the root of the @to mount is equal to the future mountpoint @mp of
373 * the @to mount on @from then @to will be overmounted by whatever is
374 * propagated to it.
375 *
376 * Context: This function expects namespace_lock() to be held and that
377 * @mp is stable.
378 * Return: If @from overmounts @to, true is returned, false if not.
379 */
380bool propagation_would_overmount(const struct mount *from,
381 const struct mount *to,
382 const struct mountpoint *mp)
383{
384 if (!IS_MNT_SHARED(from))
385 return false;
386
387 if (to->mnt.mnt_root != mp->m_dentry)
388 return false;
389
390 for (const struct mount *m = to; m; m = m->mnt_master) {
391 if (peers(m1: from, m2: m))
392 return true;
393 }
394
395 return false;
396}
397
398/*
399 * check if the mount 'mnt' can be unmounted successfully.
400 * @mnt: the mount to be checked for unmount
401 * NOTE: unmounting 'mnt' would naturally propagate to all
402 * other mounts its parent propagates to.
403 * Check if any of these mounts that **do not have submounts**
404 * have more references than 'refcnt'. If so return busy.
405 *
406 * vfsmount lock must be held for write
407 */
408int propagate_mount_busy(struct mount *mnt, int refcnt)
409{
410 struct mount *m, *child, *topper;
411 struct mount *parent = mnt->mnt_parent;
412
413 if (mnt == parent)
414 return do_refcount_check(mnt, count: refcnt);
415
416 /*
417 * quickly check if the current mount can be unmounted.
418 * If not, we don't have to go checking for all other
419 * mounts
420 */
421 if (!list_empty(head: &mnt->mnt_mounts) || do_refcount_check(mnt, count: refcnt))
422 return 1;
423
424 for (m = propagation_next(m: parent, origin: parent); m;
425 m = propagation_next(m, origin: parent)) {
426 int count = 1;
427 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
428 if (!child)
429 continue;
430
431 /* Is there exactly one mount on the child that covers
432 * it completely whose reference should be ignored?
433 */
434 topper = find_topper(mnt: child);
435 if (topper)
436 count += 1;
437 else if (!list_empty(head: &child->mnt_mounts))
438 continue;
439
440 if (do_refcount_check(mnt: child, count))
441 return 1;
442 }
443 return 0;
444}
445
446/*
447 * Clear MNT_LOCKED when it can be shown to be safe.
448 *
449 * mount_lock lock must be held for write
450 */
451void propagate_mount_unlock(struct mount *mnt)
452{
453 struct mount *parent = mnt->mnt_parent;
454 struct mount *m, *child;
455
456 BUG_ON(parent == mnt);
457
458 for (m = propagation_next(m: parent, origin: parent); m;
459 m = propagation_next(m, origin: parent)) {
460 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
461 if (child)
462 child->mnt.mnt_flags &= ~MNT_LOCKED;
463 }
464}
465
466static void umount_one(struct mount *mnt, struct list_head *to_umount)
467{
468 CLEAR_MNT_MARK(mnt);
469 mnt->mnt.mnt_flags |= MNT_UMOUNT;
470 list_del_init(entry: &mnt->mnt_child);
471 list_del_init(entry: &mnt->mnt_umounting);
472 move_from_ns(mnt, dt_list: to_umount);
473}
474
475/*
476 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
477 * parent propagates to.
478 */
479static bool __propagate_umount(struct mount *mnt,
480 struct list_head *to_umount,
481 struct list_head *to_restore)
482{
483 bool progress = false;
484 struct mount *child;
485
486 /*
487 * The state of the parent won't change if this mount is
488 * already unmounted or marked as without children.
489 */
490 if (mnt->mnt.mnt_flags & (MNT_UMOUNT | MNT_MARKED))
491 goto out;
492
493 /* Verify topper is the only grandchild that has not been
494 * speculatively unmounted.
495 */
496 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
497 if (child->mnt_mountpoint == mnt->mnt.mnt_root)
498 continue;
499 if (!list_empty(head: &child->mnt_umounting) && IS_MNT_MARKED(child))
500 continue;
501 /* Found a mounted child */
502 goto children;
503 }
504
505 /* Mark mounts that can be unmounted if not locked */
506 SET_MNT_MARK(mnt);
507 progress = true;
508
509 /* If a mount is without children and not locked umount it. */
510 if (!IS_MNT_LOCKED(mnt)) {
511 umount_one(mnt, to_umount);
512 } else {
513children:
514 list_move_tail(list: &mnt->mnt_umounting, head: to_restore);
515 }
516out:
517 return progress;
518}
519
520static void umount_list(struct list_head *to_umount,
521 struct list_head *to_restore)
522{
523 struct mount *mnt, *child, *tmp;
524 list_for_each_entry(mnt, to_umount, mnt_list) {
525 list_for_each_entry_safe(child, tmp, &mnt->mnt_mounts, mnt_child) {
526 /* topper? */
527 if (child->mnt_mountpoint == mnt->mnt.mnt_root)
528 list_move_tail(list: &child->mnt_umounting, head: to_restore);
529 else
530 umount_one(mnt: child, to_umount);
531 }
532 }
533}
534
535static void restore_mounts(struct list_head *to_restore)
536{
537 /* Restore mounts to a clean working state */
538 while (!list_empty(head: to_restore)) {
539 struct mount *mnt, *parent;
540 struct mountpoint *mp;
541
542 mnt = list_first_entry(to_restore, struct mount, mnt_umounting);
543 CLEAR_MNT_MARK(mnt);
544 list_del_init(entry: &mnt->mnt_umounting);
545
546 /* Should this mount be reparented? */
547 mp = mnt->mnt_mp;
548 parent = mnt->mnt_parent;
549 while (parent->mnt.mnt_flags & MNT_UMOUNT) {
550 mp = parent->mnt_mp;
551 parent = parent->mnt_parent;
552 }
553 if (parent != mnt->mnt_parent) {
554 mnt_change_mountpoint(parent, mp, mnt);
555 mnt_notify_add(m: mnt);
556 }
557 }
558}
559
560static void cleanup_umount_visitations(struct list_head *visited)
561{
562 while (!list_empty(head: visited)) {
563 struct mount *mnt =
564 list_first_entry(visited, struct mount, mnt_umounting);
565 list_del_init(entry: &mnt->mnt_umounting);
566 }
567}
568
569/*
570 * collect all mounts that receive propagation from the mount in @list,
571 * and return these additional mounts in the same list.
572 * @list: the list of mounts to be unmounted.
573 *
574 * vfsmount lock must be held for write
575 */
576int propagate_umount(struct list_head *list)
577{
578 struct mount *mnt;
579 LIST_HEAD(to_restore);
580 LIST_HEAD(to_umount);
581 LIST_HEAD(visited);
582
583 /* Find candidates for unmounting */
584 list_for_each_entry_reverse(mnt, list, mnt_list) {
585 struct mount *parent = mnt->mnt_parent;
586 struct mount *m;
587
588 /*
589 * If this mount has already been visited it is known that it's
590 * entire peer group and all of their slaves in the propagation
591 * tree for the mountpoint has already been visited and there is
592 * no need to visit them again.
593 */
594 if (!list_empty(head: &mnt->mnt_umounting))
595 continue;
596
597 list_add_tail(new: &mnt->mnt_umounting, head: &visited);
598 for (m = propagation_next(m: parent, origin: parent); m;
599 m = propagation_next(m, origin: parent)) {
600 struct mount *child = __lookup_mnt(&m->mnt,
601 mnt->mnt_mountpoint);
602 if (!child)
603 continue;
604
605 if (!list_empty(head: &child->mnt_umounting)) {
606 /*
607 * If the child has already been visited it is
608 * know that it's entire peer group and all of
609 * their slaves in the propgation tree for the
610 * mountpoint has already been visited and there
611 * is no need to visit this subtree again.
612 */
613 m = skip_propagation_subtree(m, origin: parent);
614 continue;
615 } else if (child->mnt.mnt_flags & MNT_UMOUNT) {
616 /*
617 * We have come across a partially unmounted
618 * mount in a list that has not been visited
619 * yet. Remember it has been visited and
620 * continue about our merry way.
621 */
622 list_add_tail(new: &child->mnt_umounting, head: &visited);
623 continue;
624 }
625
626 /* Check the child and parents while progress is made */
627 while (__propagate_umount(mnt: child,
628 to_umount: &to_umount, to_restore: &to_restore)) {
629 /* Is the parent a umount candidate? */
630 child = child->mnt_parent;
631 if (list_empty(head: &child->mnt_umounting))
632 break;
633 }
634 }
635 }
636
637 umount_list(to_umount: &to_umount, to_restore: &to_restore);
638 restore_mounts(to_restore: &to_restore);
639 cleanup_umount_visitations(visited: &visited);
640 list_splice_tail(list: &to_umount, head: list);
641
642 return 0;
643}
644

source code of linux/fs/pnode.c

Morty Proxy This is a proxified and sanitized view of the page, visit original site.