projects/linux·blob v6.17 · refs

650 lines · 16.3 KiB

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