Loading...
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(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(m, 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(&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(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;
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(mnt); m != mnt; m = next_peer(m)) {
101 if (m->mnt.mnt_root == mnt->mnt.mnt_root) {
102 master = m;
103 break;
104 }
105 }
106 list_del_init(&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(&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;
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(&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(&m->mnt_slave_list))
154 return first_slave(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(m);
161 return (next == origin) ? NULL : next;
162 } else if (m->mnt_slave.next != &master->mnt_slave_list)
163 return next_slave(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(&m->mnt_slave_list))
178 m = last_slave(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(&m->mnt_slave_list))
189 return first_slave(m);
190 next = next_peer(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(m);
203 m = next_peer(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 mountpoint *mp;
218static struct hlist_head *list;
219
220static inline bool peers(struct mount *m1, struct mount *m2)
221{
222 return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
223}
224
225static int propagate_one(struct mount *m)
226{
227 struct mount *child;
228 int type;
229 /* skip ones added by this propagate_mnt() */
230 if (IS_MNT_NEW(m))
231 return 0;
232 /* skip if mountpoint isn't covered by it */
233 if (!is_subdir(mp->m_dentry, m->mnt.mnt_root))
234 return 0;
235 if (peers(m, last_dest)) {
236 type = CL_MAKE_SHARED;
237 } else {
238 struct mount *n, *p;
239 bool done;
240 for (n = m; ; n = p) {
241 p = n->mnt_master;
242 if (p == dest_master || IS_MNT_MARKED(p))
243 break;
244 }
245 do {
246 struct mount *parent = last_source->mnt_parent;
247 if (last_source == first_source)
248 break;
249 done = parent->mnt_master == p;
250 if (done && peers(n, parent))
251 break;
252 last_source = last_source->mnt_master;
253 } while (!done);
254
255 type = CL_SLAVE;
256 /* beginning of peer group among the slaves? */
257 if (IS_MNT_SHARED(m))
258 type |= CL_MAKE_SHARED;
259 }
260
261 child = copy_tree(last_source, last_source->mnt.mnt_root, type);
262 if (IS_ERR(child))
263 return PTR_ERR(child);
264 read_seqlock_excl(&mount_lock);
265 mnt_set_mountpoint(m, mp, child);
266 if (m->mnt_master != dest_master)
267 SET_MNT_MARK(m->mnt_master);
268 read_sequnlock_excl(&mount_lock);
269 last_dest = m;
270 last_source = child;
271 hlist_add_head(&child->mnt_hash, list);
272 return count_mounts(m->mnt_ns, child);
273}
274
275/*
276 * mount 'source_mnt' under the destination 'dest_mnt' at
277 * dentry 'dest_dentry'. And propagate that mount to
278 * all the peer and slave mounts of 'dest_mnt'.
279 * Link all the new mounts into a propagation tree headed at
280 * source_mnt. Also link all the new mounts using ->mnt_list
281 * headed at source_mnt's ->mnt_list
282 *
283 * @dest_mnt: destination mount.
284 * @dest_dentry: destination dentry.
285 * @source_mnt: source mount.
286 * @tree_list : list of heads of trees to be attached.
287 */
288int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
289 struct mount *source_mnt, struct hlist_head *tree_list)
290{
291 struct mount *m, *n;
292 int ret = 0;
293
294 /*
295 * we don't want to bother passing tons of arguments to
296 * propagate_one(); everything is serialized by namespace_sem,
297 * so globals will do just fine.
298 */
299 last_dest = dest_mnt;
300 first_source = source_mnt;
301 last_source = source_mnt;
302 mp = dest_mp;
303 list = tree_list;
304 dest_master = dest_mnt->mnt_master;
305
306 /* all peers of dest_mnt, except dest_mnt itself */
307 for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
308 ret = propagate_one(n);
309 if (ret)
310 goto out;
311 }
312
313 /* all slave groups */
314 for (m = next_group(dest_mnt, dest_mnt); m;
315 m = next_group(m, dest_mnt)) {
316 /* everything in that slave group */
317 n = m;
318 do {
319 ret = propagate_one(n);
320 if (ret)
321 goto out;
322 n = next_peer(n);
323 } while (n != m);
324 }
325out:
326 read_seqlock_excl(&mount_lock);
327 hlist_for_each_entry(n, tree_list, mnt_hash) {
328 m = n->mnt_parent;
329 if (m->mnt_master != dest_mnt->mnt_master)
330 CLEAR_MNT_MARK(m->mnt_master);
331 }
332 read_sequnlock_excl(&mount_lock);
333 return ret;
334}
335
336static struct mount *find_topper(struct mount *mnt)
337{
338 /* If there is exactly one mount covering mnt completely return it. */
339 struct mount *child;
340
341 if (!list_is_singular(&mnt->mnt_mounts))
342 return NULL;
343
344 child = list_first_entry(&mnt->mnt_mounts, struct mount, mnt_child);
345 if (child->mnt_mountpoint != mnt->mnt.mnt_root)
346 return NULL;
347
348 return child;
349}
350
351/*
352 * return true if the refcount is greater than count
353 */
354static inline int do_refcount_check(struct mount *mnt, int count)
355{
356 return mnt_get_count(mnt) > count;
357}
358
359/*
360 * check if the mount 'mnt' can be unmounted successfully.
361 * @mnt: the mount to be checked for unmount
362 * NOTE: unmounting 'mnt' would naturally propagate to all
363 * other mounts its parent propagates to.
364 * Check if any of these mounts that **do not have submounts**
365 * have more references than 'refcnt'. If so return busy.
366 *
367 * vfsmount lock must be held for write
368 */
369int propagate_mount_busy(struct mount *mnt, int refcnt)
370{
371 struct mount *m, *child, *topper;
372 struct mount *parent = mnt->mnt_parent;
373
374 if (mnt == parent)
375 return do_refcount_check(mnt, refcnt);
376
377 /*
378 * quickly check if the current mount can be unmounted.
379 * If not, we don't have to go checking for all other
380 * mounts
381 */
382 if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
383 return 1;
384
385 for (m = propagation_next(parent, parent); m;
386 m = propagation_next(m, parent)) {
387 int count = 1;
388 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
389 if (!child)
390 continue;
391
392 /* Is there exactly one mount on the child that covers
393 * it completely whose reference should be ignored?
394 */
395 topper = find_topper(child);
396 if (topper)
397 count += 1;
398 else if (!list_empty(&child->mnt_mounts))
399 continue;
400
401 if (do_refcount_check(child, count))
402 return 1;
403 }
404 return 0;
405}
406
407/*
408 * Clear MNT_LOCKED when it can be shown to be safe.
409 *
410 * mount_lock lock must be held for write
411 */
412void propagate_mount_unlock(struct mount *mnt)
413{
414 struct mount *parent = mnt->mnt_parent;
415 struct mount *m, *child;
416
417 BUG_ON(parent == mnt);
418
419 for (m = propagation_next(parent, parent); m;
420 m = propagation_next(m, parent)) {
421 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
422 if (child)
423 child->mnt.mnt_flags &= ~MNT_LOCKED;
424 }
425}
426
427static void umount_one(struct mount *mnt, struct list_head *to_umount)
428{
429 CLEAR_MNT_MARK(mnt);
430 mnt->mnt.mnt_flags |= MNT_UMOUNT;
431 list_del_init(&mnt->mnt_child);
432 list_del_init(&mnt->mnt_umounting);
433 list_move_tail(&mnt->mnt_list, to_umount);
434}
435
436/*
437 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
438 * parent propagates to.
439 */
440static bool __propagate_umount(struct mount *mnt,
441 struct list_head *to_umount,
442 struct list_head *to_restore)
443{
444 bool progress = false;
445 struct mount *child;
446
447 /*
448 * The state of the parent won't change if this mount is
449 * already unmounted or marked as without children.
450 */
451 if (mnt->mnt.mnt_flags & (MNT_UMOUNT | MNT_MARKED))
452 goto out;
453
454 /* Verify topper is the only grandchild that has not been
455 * speculatively unmounted.
456 */
457 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
458 if (child->mnt_mountpoint == mnt->mnt.mnt_root)
459 continue;
460 if (!list_empty(&child->mnt_umounting) && IS_MNT_MARKED(child))
461 continue;
462 /* Found a mounted child */
463 goto children;
464 }
465
466 /* Mark mounts that can be unmounted if not locked */
467 SET_MNT_MARK(mnt);
468 progress = true;
469
470 /* If a mount is without children and not locked umount it. */
471 if (!IS_MNT_LOCKED(mnt)) {
472 umount_one(mnt, to_umount);
473 } else {
474children:
475 list_move_tail(&mnt->mnt_umounting, to_restore);
476 }
477out:
478 return progress;
479}
480
481static void umount_list(struct list_head *to_umount,
482 struct list_head *to_restore)
483{
484 struct mount *mnt, *child, *tmp;
485 list_for_each_entry(mnt, to_umount, mnt_list) {
486 list_for_each_entry_safe(child, tmp, &mnt->mnt_mounts, mnt_child) {
487 /* topper? */
488 if (child->mnt_mountpoint == mnt->mnt.mnt_root)
489 list_move_tail(&child->mnt_umounting, to_restore);
490 else
491 umount_one(child, to_umount);
492 }
493 }
494}
495
496static void restore_mounts(struct list_head *to_restore)
497{
498 /* Restore mounts to a clean working state */
499 while (!list_empty(to_restore)) {
500 struct mount *mnt, *parent;
501 struct mountpoint *mp;
502
503 mnt = list_first_entry(to_restore, struct mount, mnt_umounting);
504 CLEAR_MNT_MARK(mnt);
505 list_del_init(&mnt->mnt_umounting);
506
507 /* Should this mount be reparented? */
508 mp = mnt->mnt_mp;
509 parent = mnt->mnt_parent;
510 while (parent->mnt.mnt_flags & MNT_UMOUNT) {
511 mp = parent->mnt_mp;
512 parent = parent->mnt_parent;
513 }
514 if (parent != mnt->mnt_parent)
515 mnt_change_mountpoint(parent, mp, mnt);
516 }
517}
518
519static void cleanup_umount_visitations(struct list_head *visited)
520{
521 while (!list_empty(visited)) {
522 struct mount *mnt =
523 list_first_entry(visited, struct mount, mnt_umounting);
524 list_del_init(&mnt->mnt_umounting);
525 }
526}
527
528/*
529 * collect all mounts that receive propagation from the mount in @list,
530 * and return these additional mounts in the same list.
531 * @list: the list of mounts to be unmounted.
532 *
533 * vfsmount lock must be held for write
534 */
535int propagate_umount(struct list_head *list)
536{
537 struct mount *mnt;
538 LIST_HEAD(to_restore);
539 LIST_HEAD(to_umount);
540 LIST_HEAD(visited);
541
542 /* Find candidates for unmounting */
543 list_for_each_entry_reverse(mnt, list, mnt_list) {
544 struct mount *parent = mnt->mnt_parent;
545 struct mount *m;
546
547 /*
548 * If this mount has already been visited it is known that it's
549 * entire peer group and all of their slaves in the propagation
550 * tree for the mountpoint has already been visited and there is
551 * no need to visit them again.
552 */
553 if (!list_empty(&mnt->mnt_umounting))
554 continue;
555
556 list_add_tail(&mnt->mnt_umounting, &visited);
557 for (m = propagation_next(parent, parent); m;
558 m = propagation_next(m, parent)) {
559 struct mount *child = __lookup_mnt(&m->mnt,
560 mnt->mnt_mountpoint);
561 if (!child)
562 continue;
563
564 if (!list_empty(&child->mnt_umounting)) {
565 /*
566 * If the child has already been visited it is
567 * know that it's entire peer group and all of
568 * their slaves in the propgation tree for the
569 * mountpoint has already been visited and there
570 * is no need to visit this subtree again.
571 */
572 m = skip_propagation_subtree(m, parent);
573 continue;
574 } else if (child->mnt.mnt_flags & MNT_UMOUNT) {
575 /*
576 * We have come accross an partially unmounted
577 * mount in list that has not been visited yet.
578 * Remember it has been visited and continue
579 * about our merry way.
580 */
581 list_add_tail(&child->mnt_umounting, &visited);
582 continue;
583 }
584
585 /* Check the child and parents while progress is made */
586 while (__propagate_umount(child,
587 &to_umount, &to_restore)) {
588 /* Is the parent a umount candidate? */
589 child = child->mnt_parent;
590 if (list_empty(&child->mnt_umounting))
591 break;
592 }
593 }
594 }
595
596 umount_list(&to_umount, &to_restore);
597 restore_mounts(&to_restore);
598 cleanup_umount_visitations(&visited);
599 list_splice_tail(&to_umount, list);
600
601 return 0;
602}