1/*
2 * linux/fs/pnode.c
3 *
4 * (C) Copyright IBM Corporation 2005.
5 * Released under GPL v2.
6 * Author : Ram Pai (linuxram@us.ibm.com)
7 *
8 */
9#include <linux/mnt_namespace.h>
10#include <linux/mount.h>
11#include <linux/fs.h>
12#include <linux/nsproxy.h>
13#include <uapi/linux/mount.h>
14#include "internal.h"
15#include "pnode.h"
16
17/* return the next shared peer mount of @p */
18static inline struct mount *next_peer(struct mount *p)
19{
20 return list_entry(p->mnt_share.next, struct mount, mnt_share);
21}
22
23static inline struct mount *first_slave(struct mount *p)
24{
25 return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
26}
27
28static inline struct mount *last_slave(struct mount *p)
29{
30 return list_entry(p->mnt_slave_list.prev, struct mount, mnt_slave);
31}
32
33static inline struct mount *next_slave(struct mount *p)
34{
35 return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
36}
37
38static struct mount *get_peer_under_root(struct mount *mnt,
39 struct mnt_namespace *ns,
40 const struct path *root)
41{
42 struct mount *m = mnt;
43
44 do {
45 /* Check the namespace first for optimization */
46 if (m->mnt_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
47 return m;
48
49 m = next_peer(m);
50 } while (m != mnt);
51
52 return NULL;
53}
54
55/*
56 * Get ID of closest dominating peer group having a representative
57 * under the given root.
58 *
59 * Caller must hold namespace_sem
60 */
61int get_dominating_id(struct mount *mnt, const struct path *root)
62{
63 struct mount *m;
64
65 for (m = mnt->mnt_master; m != NULL; m = m->mnt_master) {
66 struct mount *d = get_peer_under_root(m, mnt->mnt_ns, root);
67 if (d)
68 return d->mnt_group_id;
69 }
70
71 return 0;
72}
73
74static int do_make_slave(struct mount *mnt)
75{
76 struct mount *master, *slave_mnt;
77
78 if (list_empty(&mnt->mnt_share)) {
79 if (IS_MNT_SHARED(mnt)) {
80 mnt_release_group_id(mnt);
81 CLEAR_MNT_SHARED(mnt);
82 }
83 master = mnt->mnt_master;
84 if (!master) {
85 struct list_head *p = &mnt->mnt_slave_list;
86 while (!list_empty(p)) {
87 slave_mnt = list_first_entry(p,
88 struct mount, mnt_slave);
89 list_del_init(&slave_mnt->mnt_slave);
90 slave_mnt->mnt_master = NULL;
91 }
92 return 0;
93 }
94 } else {
95 struct mount *m;
96 /*
97 * slave 'mnt' to a peer mount that has the
98 * same root dentry. If none is available then
99 * slave it to anything that is available.
100 */
101 for (m = master = next_peer(mnt); m != mnt; m = next_peer(m)) {
102 if (m->mnt.mnt_root == mnt->mnt.mnt_root) {
103 master = m;
104 break;
105 }
106 }
107 list_del_init(&mnt->mnt_share);
108 mnt->mnt_group_id = 0;
109 CLEAR_MNT_SHARED(mnt);
110 }
111 list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
112 slave_mnt->mnt_master = master;
113 list_move(&mnt->mnt_slave, &master->mnt_slave_list);
114 list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
115 INIT_LIST_HEAD(&mnt->mnt_slave_list);
116 mnt->mnt_master = master;
117 return 0;
118}
119
120/*
121 * vfsmount lock must be held for write
122 */
123void change_mnt_propagation(struct mount *mnt, int type)
124{
125 if (type == MS_SHARED) {
126 set_mnt_shared(mnt);
127 return;
128 }
129 do_make_slave(mnt);
130 if (type != MS_SLAVE) {
131 list_del_init(&mnt->mnt_slave);
132 mnt->mnt_master = NULL;
133 if (type == MS_UNBINDABLE)
134 mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
135 else
136 mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
137 }
138}
139
140/*
141 * get the next mount in the propagation tree.
142 * @m: the mount seen last
143 * @origin: the original mount from where the tree walk initiated
144 *
145 * Note that peer groups form contiguous segments of slave lists.
146 * We rely on that in get_source() to be able to find out if
147 * vfsmount found while iterating with propagation_next() is
148 * a peer of one we'd found earlier.
149 */
150static struct mount *propagation_next(struct mount *m,
151 struct mount *origin)
152{
153 /* are there any slaves of this mount? */
154 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
155 return first_slave(m);
156
157 while (1) {
158 struct mount *master = m->mnt_master;
159
160 if (master == origin->mnt_master) {
161 struct mount *next = next_peer(m);
162 return (next == origin) ? NULL : next;
163 } else if (m->mnt_slave.next != &master->mnt_slave_list)
164 return next_slave(m);
165
166 /* back at master */
167 m = master;
168 }
169}
170
171static struct mount *skip_propagation_subtree(struct mount *m,
172 struct mount *origin)
173{
174 /*
175 * Advance m such that propagation_next will not return
176 * the slaves of m.
177 */
178 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
179 m = last_slave(m);
180
181 return m;
182}
183
184static struct mount *next_group(struct mount *m, struct mount *origin)
185{
186 while (1) {
187 while (1) {
188 struct mount *next;
189 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
190 return first_slave(m);
191 next = next_peer(m);
192 if (m->mnt_group_id == origin->mnt_group_id) {
193 if (next == origin)
194 return NULL;
195 } else if (m->mnt_slave.next != &next->mnt_slave)
196 break;
197 m = next;
198 }
199 /* m is the last peer */
200 while (1) {
201 struct mount *master = m->mnt_master;
202 if (m->mnt_slave.next != &master->mnt_slave_list)
203 return next_slave(m);
204 m = next_peer(master);
205 if (master->mnt_group_id == origin->mnt_group_id)
206 break;
207 if (master->mnt_slave.next == &m->mnt_slave)
208 break;
209 m = master;
210 }
211 if (m == origin)
212 return NULL;
213 }
214}
215
216/* all accesses are serialized by namespace_sem */
217static struct mount *last_dest, *first_source, *last_source, *dest_master;
218static struct mountpoint *mp;
219static struct hlist_head *list;
220
221static inline bool peers(struct mount *m1, struct mount *m2)
222{
223 return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
224}
225
226static int propagate_one(struct mount *m)
227{
228 struct mount *child;
229 int type;
230 /* skip ones added by this propagate_mnt() */
231 if (IS_MNT_NEW(m))
232 return 0;
233 /* skip if mountpoint isn't covered by it */
234 if (!is_subdir(mp->m_dentry, m->mnt.mnt_root))
235 return 0;
236 if (peers(m, last_dest)) {
237 type = CL_MAKE_SHARED;
238 } else {
239 struct mount *n, *p;
240 bool done;
241 for (n = m; ; n = p) {
242 p = n->mnt_master;
243 if (p == dest_master || IS_MNT_MARKED(p))
244 break;
245 }
246 do {
247 struct mount *parent = last_source->mnt_parent;
248 if (last_source == first_source)
249 break;
250 done = parent->mnt_master == p;
251 if (done && peers(n, parent))
252 break;
253 last_source = last_source->mnt_master;
254 } while (!done);
255
256 type = CL_SLAVE;
257 /* beginning of peer group among the slaves? */
258 if (IS_MNT_SHARED(m))
259 type |= CL_MAKE_SHARED;
260 }
261
262 child = copy_tree(last_source, last_source->mnt.mnt_root, type);
263 if (IS_ERR(child))
264 return PTR_ERR(child);
265 child->mnt.mnt_flags &= ~MNT_LOCKED;
266 mnt_set_mountpoint(m, mp, child);
267 last_dest = m;
268 last_source = child;
269 if (m->mnt_master != dest_master) {
270 read_seqlock_excl(&mount_lock);
271 SET_MNT_MARK(m->mnt_master);
272 read_sequnlock_excl(&mount_lock);
273 }
274 hlist_add_head(&child->mnt_hash, list);
275 return count_mounts(m->mnt_ns, 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 mp = dest_mp;
306 list = tree_list;
307 dest_master = dest_mnt->mnt_master;
308
309 /* all peers of dest_mnt, except dest_mnt itself */
310 for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
311 ret = propagate_one(n);
312 if (ret)
313 goto out;
314 }
315
316 /* all slave groups */
317 for (m = next_group(dest_mnt, dest_mnt); m;
318 m = next_group(m, dest_mnt)) {
319 /* everything in that slave group */
320 n = m;
321 do {
322 ret = propagate_one(n);
323 if (ret)
324 goto out;
325 n = next_peer(n);
326 } while (n != m);
327 }
328out:
329 read_seqlock_excl(&mount_lock);
330 hlist_for_each_entry(n, tree_list, mnt_hash) {
331 m = n->mnt_parent;
332 if (m->mnt_master != dest_mnt->mnt_master)
333 CLEAR_MNT_MARK(m->mnt_master);
334 }
335 read_sequnlock_excl(&mount_lock);
336 return ret;
337}
338
339static struct mount *find_topper(struct mount *mnt)
340{
341 /* If there is exactly one mount covering mnt completely return it. */
342 struct mount *child;
343
344 if (!list_is_singular(&mnt->mnt_mounts))
345 return NULL;
346
347 child = list_first_entry(&mnt->mnt_mounts, struct mount, mnt_child);
348 if (child->mnt_mountpoint != mnt->mnt.mnt_root)
349 return NULL;
350
351 return child;
352}
353
354/*
355 * return true if the refcount is greater than count
356 */
357static inline int do_refcount_check(struct mount *mnt, int count)
358{
359 return mnt_get_count(mnt) > count;
360}
361
362/*
363 * check if the mount 'mnt' can be unmounted successfully.
364 * @mnt: the mount to be checked for unmount
365 * NOTE: unmounting 'mnt' would naturally propagate to all
366 * other mounts its parent propagates to.
367 * Check if any of these mounts that **do not have submounts**
368 * have more references than 'refcnt'. If so return busy.
369 *
370 * vfsmount lock must be held for write
371 */
372int propagate_mount_busy(struct mount *mnt, int refcnt)
373{
374 struct mount *m, *child, *topper;
375 struct mount *parent = mnt->mnt_parent;
376
377 if (mnt == parent)
378 return do_refcount_check(mnt, refcnt);
379
380 /*
381 * quickly check if the current mount can be unmounted.
382 * If not, we don't have to go checking for all other
383 * mounts
384 */
385 if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
386 return 1;
387
388 for (m = propagation_next(parent, parent); m;
389 m = propagation_next(m, parent)) {
390 int count = 1;
391 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
392 if (!child)
393 continue;
394
395 /* Is there exactly one mount on the child that covers
396 * it completely whose reference should be ignored?
397 */
398 topper = find_topper(child);
399 if (topper)
400 count += 1;
401 else if (!list_empty(&child->mnt_mounts))
402 continue;
403
404 if (do_refcount_check(child, count))
405 return 1;
406 }
407 return 0;
408}
409
410/*
411 * Clear MNT_LOCKED when it can be shown to be safe.
412 *
413 * mount_lock lock must be held for write
414 */
415void propagate_mount_unlock(struct mount *mnt)
416{
417 struct mount *parent = mnt->mnt_parent;
418 struct mount *m, *child;
419
420 BUG_ON(parent == mnt);
421
422 for (m = propagation_next(parent, parent); m;
423 m = propagation_next(m, parent)) {
424 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
425 if (child)
426 child->mnt.mnt_flags &= ~MNT_LOCKED;
427 }
428}
429
430static void umount_one(struct mount *mnt, struct list_head *to_umount)
431{
432 CLEAR_MNT_MARK(mnt);
433 mnt->mnt.mnt_flags |= MNT_UMOUNT;
434 list_del_init(&mnt->mnt_child);
435 list_del_init(&mnt->mnt_umounting);
436 list_move_tail(&mnt->mnt_list, to_umount);
437}
438
439/*
440 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
441 * parent propagates to.
442 */
443static bool __propagate_umount(struct mount *mnt,
444 struct list_head *to_umount,
445 struct list_head *to_restore)
446{
447 bool progress = false;
448 struct mount *child;
449
450 /*
451 * The state of the parent won't change if this mount is
452 * already unmounted or marked as without children.
453 */
454 if (mnt->mnt.mnt_flags & (MNT_UMOUNT | MNT_MARKED))
455 goto out;
456
457 /* Verify topper is the only grandchild that has not been
458 * speculatively unmounted.
459 */
460 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
461 if (child->mnt_mountpoint == mnt->mnt.mnt_root)
462 continue;
463 if (!list_empty(&child->mnt_umounting) && IS_MNT_MARKED(child))
464 continue;
465 /* Found a mounted child */
466 goto children;
467 }
468
469 /* Mark mounts that can be unmounted if not locked */
470 SET_MNT_MARK(mnt);
471 progress = true;
472
473 /* If a mount is without children and not locked umount it. */
474 if (!IS_MNT_LOCKED(mnt)) {
475 umount_one(mnt, to_umount);
476 } else {
477children:
478 list_move_tail(&mnt->mnt_umounting, to_restore);
479 }
480out:
481 return progress;
482}
483
484static void umount_list(struct list_head *to_umount,
485 struct list_head *to_restore)
486{
487 struct mount *mnt, *child, *tmp;
488 list_for_each_entry(mnt, to_umount, mnt_list) {
489 list_for_each_entry_safe(child, tmp, &mnt->mnt_mounts, mnt_child) {
490 /* topper? */
491 if (child->mnt_mountpoint == mnt->mnt.mnt_root)
492 list_move_tail(&child->mnt_umounting, to_restore);
493 else
494 umount_one(child, to_umount);
495 }
496 }
497}
498
499static void restore_mounts(struct list_head *to_restore)
500{
501 /* Restore mounts to a clean working state */
502 while (!list_empty(to_restore)) {
503 struct mount *mnt, *parent;
504 struct mountpoint *mp;
505
506 mnt = list_first_entry(to_restore, struct mount, mnt_umounting);
507 CLEAR_MNT_MARK(mnt);
508 list_del_init(&mnt->mnt_umounting);
509
510 /* Should this mount be reparented? */
511 mp = mnt->mnt_mp;
512 parent = mnt->mnt_parent;
513 while (parent->mnt.mnt_flags & MNT_UMOUNT) {
514 mp = parent->mnt_mp;
515 parent = parent->mnt_parent;
516 }
517 if (parent != mnt->mnt_parent)
518 mnt_change_mountpoint(parent, mp, mnt);
519 }
520}
521
522static void cleanup_umount_visitations(struct list_head *visited)
523{
524 while (!list_empty(visited)) {
525 struct mount *mnt =
526 list_first_entry(visited, struct mount, mnt_umounting);
527 list_del_init(&mnt->mnt_umounting);
528 }
529}
530
531/*
532 * collect all mounts that receive propagation from the mount in @list,
533 * and return these additional mounts in the same list.
534 * @list: the list of mounts to be unmounted.
535 *
536 * vfsmount lock must be held for write
537 */
538int propagate_umount(struct list_head *list)
539{
540 struct mount *mnt;
541 LIST_HEAD(to_restore);
542 LIST_HEAD(to_umount);
543 LIST_HEAD(visited);
544
545 /* Find candidates for unmounting */
546 list_for_each_entry_reverse(mnt, list, mnt_list) {
547 struct mount *parent = mnt->mnt_parent;
548 struct mount *m;
549
550 /*
551 * If this mount has already been visited it is known that it's
552 * entire peer group and all of their slaves in the propagation
553 * tree for the mountpoint has already been visited and there is
554 * no need to visit them again.
555 */
556 if (!list_empty(&mnt->mnt_umounting))
557 continue;
558
559 list_add_tail(&mnt->mnt_umounting, &visited);
560 for (m = propagation_next(parent, parent); m;
561 m = propagation_next(m, parent)) {
562 struct mount *child = __lookup_mnt(&m->mnt,
563 mnt->mnt_mountpoint);
564 if (!child)
565 continue;
566
567 if (!list_empty(&child->mnt_umounting)) {
568 /*
569 * If the child has already been visited it is
570 * know that it's entire peer group and all of
571 * their slaves in the propgation tree for the
572 * mountpoint has already been visited and there
573 * is no need to visit this subtree again.
574 */
575 m = skip_propagation_subtree(m, parent);
576 continue;
577 } else if (child->mnt.mnt_flags & MNT_UMOUNT) {
578 /*
579 * We have come accross an partially unmounted
580 * mount in list that has not been visited yet.
581 * Remember it has been visited and continue
582 * about our merry way.
583 */
584 list_add_tail(&child->mnt_umounting, &visited);
585 continue;
586 }
587
588 /* Check the child and parents while progress is made */
589 while (__propagate_umount(child,
590 &to_umount, &to_restore)) {
591 /* Is the parent a umount candidate? */
592 child = child->mnt_parent;
593 if (list_empty(&child->mnt_umounting))
594 break;
595 }
596 }
597 }
598
599 umount_list(&to_umount, &to_restore);
600 restore_mounts(&to_restore);
601 cleanup_umount_visitations(&visited);
602 list_splice_tail(&to_umount, list);
603
604 return 0;
605}
606