Line data Source code
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 */
17 4931 : static inline struct mount *next_peer(struct mount *p)
18 : {
19 4931 : return list_entry(p->mnt_share.next, struct mount, mnt_share);
20 : }
21 :
22 185 : static inline struct mount *first_slave(struct mount *p)
23 : {
24 185 : return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
25 : }
26 :
27 0 : static inline struct mount *last_slave(struct mount *p)
28 : {
29 0 : return list_entry(p->mnt_slave_list.prev, struct mount, mnt_slave);
30 : }
31 :
32 1113 : static inline struct mount *next_slave(struct mount *p)
33 : {
34 1113 : return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
35 : }
36 :
37 2893 : static struct mount *get_peer_under_root(struct mount *mnt,
38 : struct mnt_namespace *ns,
39 : const struct path *root)
40 : {
41 2893 : struct mount *m = mnt;
42 :
43 2893 : do {
44 : /* Check the namespace first for optimization */
45 2893 : if (m->mnt_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
46 0 : return m;
47 :
48 2893 : m = next_peer(m);
49 2893 : } 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 : */
60 2893 : int get_dominating_id(struct mount *mnt, const struct path *root)
61 : {
62 2893 : struct mount *m;
63 :
64 5786 : for (m = mnt->mnt_master; m != NULL; m = m->mnt_master) {
65 2893 : struct mount *d = get_peer_under_root(m, mnt->mnt_ns, root);
66 2893 : if (d)
67 0 : return d->mnt_group_id;
68 : }
69 :
70 : return 0;
71 : }
72 :
73 1149 : static int do_make_slave(struct mount *mnt)
74 : {
75 1149 : struct mount *master, *slave_mnt;
76 :
77 1149 : if (list_empty(&mnt->mnt_share)) {
78 420 : if (IS_MNT_SHARED(mnt)) {
79 161 : mnt_release_group_id(mnt);
80 161 : CLEAR_MNT_SHARED(mnt);
81 : }
82 420 : master = mnt->mnt_master;
83 420 : if (!master) {
84 377 : struct list_head *p = &mnt->mnt_slave_list;
85 561 : while (!list_empty(p)) {
86 184 : slave_mnt = list_first_entry(p,
87 : struct mount, mnt_slave);
88 184 : list_del_init(&slave_mnt->mnt_slave);
89 184 : slave_mnt->mnt_master = NULL;
90 : }
91 : return 0;
92 : }
93 : } else {
94 729 : 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 : */
100 729 : for (m = master = next_peer(mnt); m != mnt; m = next_peer(m)) {
101 729 : if (m->mnt.mnt_root == mnt->mnt.mnt_root) {
102 : master = m;
103 : break;
104 : }
105 : }
106 729 : list_del_init(&mnt->mnt_share);
107 729 : mnt->mnt_group_id = 0;
108 729 : CLEAR_MNT_SHARED(mnt);
109 : }
110 772 : list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
111 0 : slave_mnt->mnt_master = master;
112 772 : list_move(&mnt->mnt_slave, &master->mnt_slave_list);
113 772 : list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
114 772 : INIT_LIST_HEAD(&mnt->mnt_slave_list);
115 772 : mnt->mnt_master = master;
116 772 : return 0;
117 : }
118 :
119 : /*
120 : * vfsmount lock must be held for write
121 : */
122 1225 : void change_mnt_propagation(struct mount *mnt, int type)
123 : {
124 1225 : if (type == MS_SHARED) {
125 76 : set_mnt_shared(mnt);
126 76 : return;
127 : }
128 1149 : do_make_slave(mnt);
129 1149 : if (type != MS_SLAVE) {
130 1110 : list_del_init(&mnt->mnt_slave);
131 1110 : mnt->mnt_master = NULL;
132 1110 : if (type == MS_UNBINDABLE)
133 0 : mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
134 : else
135 1110 : 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 : */
149 1421 : static struct mount *propagation_next(struct mount *m,
150 : struct mount *origin)
151 : {
152 : /* are there any slaves of this mount? */
153 1421 : if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
154 138 : return first_slave(m);
155 :
156 1421 : while (1) {
157 1421 : struct mount *master = m->mnt_master;
158 :
159 1421 : if (master == origin->mnt_master) {
160 455 : struct mount *next = next_peer(m);
161 455 : return (next == origin) ? NULL : next;
162 966 : } else if (m->mnt_slave.next != &master->mnt_slave_list)
163 828 : return next_slave(m);
164 :
165 : /* back at master */
166 : m = master;
167 : }
168 : }
169 :
170 0 : static 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 0 : if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
178 0 : m = last_slave(m);
179 :
180 0 : return m;
181 : }
182 :
183 381 : static struct mount *next_group(struct mount *m, struct mount *origin)
184 : {
185 427 : while (1) {
186 427 : while (1) {
187 427 : struct mount *next;
188 427 : if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
189 47 : return first_slave(m);
190 380 : next = next_peer(m);
191 380 : if (m->mnt_group_id == origin->mnt_group_id) {
192 48 : if (next == origin)
193 : return NULL;
194 332 : } else if (m->mnt_slave.next != &next->mnt_slave)
195 : break;
196 : m = next;
197 : }
198 : /* m is the last peer */
199 332 : while (1) {
200 332 : struct mount *master = m->mnt_master;
201 332 : if (m->mnt_slave.next != &master->mnt_slave_list)
202 285 : return next_slave(m);
203 47 : m = next_peer(master);
204 47 : if (master->mnt_group_id == origin->mnt_group_id)
205 : break;
206 0 : if (master->mnt_slave.next == &m->mnt_slave)
207 : break;
208 : m = master;
209 : }
210 47 : if (m == origin)
211 : return NULL;
212 : }
213 : }
214 :
215 : /* all accesses are serialized by namespace_sem */
216 : static struct mount *last_dest, *first_source, *last_source, *dest_master;
217 : static struct mountpoint *mp;
218 : static struct hlist_head *list;
219 :
220 421 : static inline bool peers(struct mount *m1, struct mount *m2)
221 : {
222 233 : return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
223 : }
224 :
225 378 : static int propagate_one(struct mount *m)
226 : {
227 378 : struct mount *child;
228 378 : int type;
229 : /* skip ones added by this propagate_mnt() */
230 378 : if (IS_MNT_NEW(m))
231 : return 0;
232 : /* skip if mountpoint isn't covered by it */
233 378 : if (!is_subdir(mp->m_dentry, m->mnt.mnt_root))
234 : return 0;
235 468 : 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) {
241 235 : p = n->mnt_master;
242 235 : if (p == dest_master || IS_MNT_MARKED(p))
243 : break;
244 : }
245 188 : do {
246 188 : struct mount *parent = last_source->mnt_parent;
247 188 : if (last_source == first_source)
248 : break;
249 187 : done = parent->mnt_master == p;
250 374 : if (done && peers(n, parent))
251 : break;
252 141 : last_source = last_source->mnt_master;
253 141 : } while (!done);
254 :
255 188 : type = CL_SLAVE;
256 : /* beginning of peer group among the slaves? */
257 188 : if (IS_MNT_SHARED(m))
258 94 : type |= CL_MAKE_SHARED;
259 : }
260 :
261 234 : child = copy_tree(last_source, last_source->mnt.mnt_root, type);
262 234 : if (IS_ERR(child))
263 0 : return PTR_ERR(child);
264 234 : read_seqlock_excl(&mount_lock);
265 234 : mnt_set_mountpoint(m, mp, child);
266 234 : if (m->mnt_master != dest_master)
267 188 : SET_MNT_MARK(m->mnt_master);
268 234 : read_sequnlock_excl(&mount_lock);
269 234 : last_dest = m;
270 234 : last_source = child;
271 234 : hlist_add_head(&child->mnt_hash, list);
272 234 : 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 : */
288 49 : int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
289 : struct mount *source_mnt, struct hlist_head *tree_list)
290 : {
291 49 : struct mount *m, *n;
292 49 : 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 49 : last_dest = dest_mnt;
300 49 : first_source = source_mnt;
301 49 : last_source = source_mnt;
302 49 : mp = dest_mp;
303 49 : list = tree_list;
304 49 : dest_master = dest_mnt->mnt_master;
305 :
306 : /* all peers of dest_mnt, except dest_mnt itself */
307 95 : for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
308 46 : ret = propagate_one(n);
309 46 : if (ret)
310 0 : goto out;
311 : }
312 :
313 : /* all slave groups */
314 381 : for (m = next_group(dest_mnt, dest_mnt); m;
315 332 : m = next_group(m, dest_mnt)) {
316 : /* everything in that slave group */
317 : n = m;
318 332 : do {
319 332 : ret = propagate_one(n);
320 332 : if (ret)
321 0 : goto out;
322 332 : n = next_peer(n);
323 332 : } while (n != m);
324 : }
325 49 : out:
326 49 : read_seqlock_excl(&mount_lock);
327 519 : hlist_for_each_entry(n, tree_list, mnt_hash) {
328 234 : m = n->mnt_parent;
329 234 : if (m->mnt_master != dest_mnt->mnt_master)
330 188 : CLEAR_MNT_MARK(m->mnt_master);
331 : }
332 49 : read_sequnlock_excl(&mount_lock);
333 49 : return ret;
334 : }
335 :
336 230 : static struct mount *find_topper(struct mount *mnt)
337 : {
338 : /* If there is exactly one mount covering mnt completely return it. */
339 230 : struct mount *child;
340 :
341 230 : if (!list_is_singular(&mnt->mnt_mounts))
342 : return NULL;
343 :
344 0 : child = list_first_entry(&mnt->mnt_mounts, struct mount, mnt_child);
345 0 : if (child->mnt_mountpoint != mnt->mnt.mnt_root)
346 0 : return NULL;
347 :
348 : return child;
349 : }
350 :
351 : /*
352 : * return true if the refcount is greater than count
353 : */
354 335 : static inline int do_refcount_check(struct mount *mnt, int count)
355 : {
356 105 : 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 : */
369 111 : int propagate_mount_busy(struct mount *mnt, int refcnt)
370 : {
371 111 : struct mount *m, *child, *topper;
372 111 : struct mount *parent = mnt->mnt_parent;
373 :
374 111 : if (mnt == parent)
375 0 : 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 : */
382 216 : if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
383 6 : return 1;
384 :
385 473 : for (m = propagation_next(parent, parent); m;
386 368 : m = propagation_next(m, parent)) {
387 368 : int count = 1;
388 368 : child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
389 368 : if (!child)
390 138 : continue;
391 :
392 : /* Is there exactly one mount on the child that covers
393 : * it completely whose reference should be ignored?
394 : */
395 230 : topper = find_topper(child);
396 230 : if (topper)
397 : count += 1;
398 230 : else if (!list_empty(&child->mnt_mounts))
399 0 : continue;
400 :
401 230 : 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 : */
412 106 : void propagate_mount_unlock(struct mount *mnt)
413 : {
414 106 : struct mount *parent = mnt->mnt_parent;
415 106 : struct mount *m, *child;
416 :
417 106 : BUG_ON(parent == mnt);
418 :
419 474 : for (m = propagation_next(parent, parent); m;
420 368 : m = propagation_next(m, parent)) {
421 368 : child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
422 368 : if (child)
423 230 : child->mnt.mnt_flags &= ~MNT_LOCKED;
424 : }
425 106 : }
426 :
427 230 : static void umount_one(struct mount *mnt, struct list_head *to_umount)
428 : {
429 230 : CLEAR_MNT_MARK(mnt);
430 230 : mnt->mnt.mnt_flags |= MNT_UMOUNT;
431 230 : list_del_init(&mnt->mnt_child);
432 230 : list_del_init(&mnt->mnt_umounting);
433 230 : list_move_tail(&mnt->mnt_list, to_umount);
434 230 : }
435 :
436 : /*
437 : * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
438 : * parent propagates to.
439 : */
440 230 : static bool __propagate_umount(struct mount *mnt,
441 : struct list_head *to_umount,
442 : struct list_head *to_restore)
443 : {
444 230 : bool progress = false;
445 230 : 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 : */
451 230 : if (mnt->mnt.mnt_flags & (MNT_UMOUNT | MNT_MARKED))
452 0 : goto out;
453 :
454 : /* Verify topper is the only grandchild that has not been
455 : * speculatively unmounted.
456 : */
457 230 : list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
458 0 : if (child->mnt_mountpoint == mnt->mnt.mnt_root)
459 0 : continue;
460 0 : if (!list_empty(&child->mnt_umounting) && IS_MNT_MARKED(child))
461 0 : continue;
462 : /* Found a mounted child */
463 0 : goto children;
464 : }
465 :
466 : /* Mark mounts that can be unmounted if not locked */
467 230 : SET_MNT_MARK(mnt);
468 230 : progress = true;
469 :
470 : /* If a mount is without children and not locked umount it. */
471 230 : if (!IS_MNT_LOCKED(mnt)) {
472 230 : umount_one(mnt, to_umount);
473 : } else {
474 0 : children:
475 0 : list_move_tail(&mnt->mnt_umounting, to_restore);
476 : }
477 230 : out:
478 230 : return progress;
479 : }
480 :
481 106 : static void umount_list(struct list_head *to_umount,
482 : struct list_head *to_restore)
483 : {
484 106 : struct mount *mnt, *child, *tmp;
485 336 : list_for_each_entry(mnt, to_umount, mnt_list) {
486 230 : list_for_each_entry_safe(child, tmp, &mnt->mnt_mounts, mnt_child) {
487 : /* topper? */
488 0 : if (child->mnt_mountpoint == mnt->mnt.mnt_root)
489 0 : list_move_tail(&child->mnt_umounting, to_restore);
490 : else
491 0 : umount_one(child, to_umount);
492 : }
493 : }
494 106 : }
495 :
496 106 : static void restore_mounts(struct list_head *to_restore)
497 : {
498 : /* Restore mounts to a clean working state */
499 212 : while (!list_empty(to_restore)) {
500 0 : struct mount *mnt, *parent;
501 0 : struct mountpoint *mp;
502 :
503 0 : mnt = list_first_entry(to_restore, struct mount, mnt_umounting);
504 0 : CLEAR_MNT_MARK(mnt);
505 0 : list_del_init(&mnt->mnt_umounting);
506 :
507 : /* Should this mount be reparented? */
508 0 : mp = mnt->mnt_mp;
509 0 : parent = mnt->mnt_parent;
510 0 : while (parent->mnt.mnt_flags & MNT_UMOUNT) {
511 0 : mp = parent->mnt_mp;
512 0 : parent = parent->mnt_parent;
513 : }
514 0 : if (parent != mnt->mnt_parent)
515 0 : mnt_change_mountpoint(parent, mp, mnt);
516 : }
517 106 : }
518 :
519 106 : static void cleanup_umount_visitations(struct list_head *visited)
520 : {
521 106 : while (!list_empty(visited)) {
522 106 : struct mount *mnt =
523 106 : list_first_entry(visited, struct mount, mnt_umounting);
524 212 : list_del_init(&mnt->mnt_umounting);
525 : }
526 106 : }
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 : */
535 106 : int propagate_umount(struct list_head *list)
536 : {
537 106 : struct mount *mnt;
538 106 : LIST_HEAD(to_restore);
539 106 : LIST_HEAD(to_umount);
540 106 : LIST_HEAD(visited);
541 :
542 : /* Find candidates for unmounting */
543 212 : list_for_each_entry_reverse(mnt, list, mnt_list) {
544 106 : struct mount *parent = mnt->mnt_parent;
545 106 : 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 : */
553 106 : if (!list_empty(&mnt->mnt_umounting))
554 0 : continue;
555 :
556 106 : list_add_tail(&mnt->mnt_umounting, &visited);
557 474 : for (m = propagation_next(parent, parent); m;
558 368 : m = propagation_next(m, parent)) {
559 368 : struct mount *child = __lookup_mnt(&m->mnt,
560 : mnt->mnt_mountpoint);
561 368 : if (!child)
562 138 : continue;
563 :
564 230 : 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 0 : m = skip_propagation_subtree(m, parent);
573 0 : continue;
574 230 : } 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 0 : list_add_tail(&child->mnt_umounting, &visited);
582 0 : continue;
583 : }
584 :
585 : /* Check the child and parents while progress is made */
586 230 : while (__propagate_umount(child,
587 : &to_umount, &to_restore)) {
588 : /* Is the parent a umount candidate? */
589 230 : child = child->mnt_parent;
590 230 : if (list_empty(&child->mnt_umounting))
591 : break;
592 : }
593 : }
594 : }
595 :
596 106 : umount_list(&to_umount, &to_restore);
597 106 : restore_mounts(&to_restore);
598 106 : cleanup_umount_visitations(&visited);
599 106 : list_splice_tail(&to_umount, list);
600 :
601 106 : return 0;
602 : }
|
